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 "compiler/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 whether or not a variable is part of a uniform or
436 * shader storage block.
438 inline bool is_in_buffer_block() const
440 return (this->data
.mode
== ir_var_uniform
||
441 this->data
.mode
== ir_var_shader_storage
) &&
442 this->interface_type
!= NULL
;
446 * Determine whether or not a variable is part of a shader storage block.
448 inline bool is_in_shader_storage_block() const
450 return this->data
.mode
== ir_var_shader_storage
&&
451 this->interface_type
!= NULL
;
455 * Determine whether or not a variable is the declaration of an interface
458 * For the first declaration below, there will be an \c ir_variable named
459 * "instance" whose type and whose instance_type will be the same
460 * \cglsl_type. For the second declaration, there will be an \c ir_variable
461 * named "f" whose type is float and whose instance_type is B2.
463 * "instance" is an interface instance variable, but "f" is not.
473 inline bool is_interface_instance() const
475 return this->type
->without_array() == this->interface_type
;
479 * Set this->interface_type on a newly created variable.
481 void init_interface_type(const struct glsl_type
*type
)
483 assert(this->interface_type
== NULL
);
484 this->interface_type
= type
;
485 if (this->is_interface_instance()) {
486 this->u
.max_ifc_array_access
=
487 ralloc_array(this, int, type
->length
);
488 for (unsigned i
= 0; i
< type
->length
; i
++) {
489 this->u
.max_ifc_array_access
[i
] = -1;
495 * Change this->interface_type on a variable that previously had a
496 * different, but compatible, interface_type. This is used during linking
497 * to set the size of arrays in interface blocks.
499 void change_interface_type(const struct glsl_type
*type
)
501 if (this->u
.max_ifc_array_access
!= NULL
) {
502 /* max_ifc_array_access has already been allocated, so make sure the
503 * new interface has the same number of fields as the old one.
505 assert(this->interface_type
->length
== type
->length
);
507 this->interface_type
= type
;
511 * Change this->interface_type on a variable that previously had a
512 * different, and incompatible, interface_type. This is used during
513 * compilation to handle redeclaration of the built-in gl_PerVertex
516 void reinit_interface_type(const struct glsl_type
*type
)
518 if (this->u
.max_ifc_array_access
!= NULL
) {
520 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
521 * it defines have been accessed yet; so it's safe to throw away the
522 * old max_ifc_array_access pointer, since all of its values are
525 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
526 assert(this->u
.max_ifc_array_access
[i
] == -1);
528 ralloc_free(this->u
.max_ifc_array_access
);
529 this->u
.max_ifc_array_access
= NULL
;
531 this->interface_type
= NULL
;
532 init_interface_type(type
);
535 const glsl_type
*get_interface_type() const
537 return this->interface_type
;
540 enum glsl_interface_packing
get_interface_type_packing() const
542 return this->interface_type
->get_interface_packing();
545 * Get the max_ifc_array_access pointer
547 * A "set" function is not needed because the array is dynmically allocated
550 inline int *get_max_ifc_array_access()
552 assert(this->data
._num_state_slots
== 0);
553 return this->u
.max_ifc_array_access
;
556 inline unsigned get_num_state_slots() const
558 assert(!this->is_interface_instance()
559 || this->data
._num_state_slots
== 0);
560 return this->data
._num_state_slots
;
563 inline void set_num_state_slots(unsigned n
)
565 assert(!this->is_interface_instance()
567 this->data
._num_state_slots
= n
;
570 inline ir_state_slot
*get_state_slots()
572 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
575 inline const ir_state_slot
*get_state_slots() const
577 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
580 inline ir_state_slot
*allocate_state_slots(unsigned n
)
582 assert(!this->is_interface_instance());
584 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
585 this->data
._num_state_slots
= 0;
587 if (this->u
.state_slots
!= NULL
)
588 this->data
._num_state_slots
= n
;
590 return this->u
.state_slots
;
593 inline bool is_interpolation_flat() const
595 return this->data
.interpolation
== INTERP_MODE_FLAT
||
596 this->type
->contains_integer() ||
597 this->type
->contains_double();
600 inline bool is_name_ralloced() const
602 return this->name
!= ir_variable::tmp_name
;
606 * Enable emitting extension warnings for this variable
608 void enable_extension_warning(const char *extension
);
611 * Get the extension warning string for this variable
613 * If warnings are not enabled, \c NULL is returned.
615 const char *get_extension_warning() const;
618 * Declared type of the variable
620 const struct glsl_type
*type
;
623 * Declared name of the variable
627 struct ir_variable_data
{
630 * Is the variable read-only?
632 * This is set for variables declared as \c const, shader inputs,
635 unsigned read_only
:1;
639 unsigned invariant
:1;
643 * Has this variable been used for reading or writing?
645 * Several GLSL semantic checks require knowledge of whether or not a
646 * variable has been used. For example, it is an error to redeclare a
647 * variable as invariant after it has been used.
649 * This is only maintained in the ast_to_hir.cpp path, not in
650 * Mesa's fixed function or ARB program paths.
655 * Has this variable been statically assigned?
657 * This answers whether the variable was assigned in any path of
658 * the shader during ast_to_hir. This doesn't answer whether it is
659 * still written after dead code removal, nor is it maintained in
660 * non-ast_to_hir.cpp (GLSL parsing) paths.
665 * When separate shader programs are enabled, only input/outputs between
666 * the stages of a multi-stage separate program can be safely removed
667 * from the shader interface. Other input/outputs must remains active.
669 unsigned always_active_io
:1;
672 * Enum indicating how the variable was declared. See
673 * ir_var_declaration_type.
675 * This is used to detect certain kinds of illegal variable redeclarations.
677 unsigned how_declared
:2;
680 * Storage class of the variable.
682 * \sa ir_variable_mode
687 * Interpolation mode for shader inputs / outputs
689 * \sa glsl_interp_mode
691 unsigned interpolation
:2;
694 * \name ARB_fragment_coord_conventions
697 unsigned origin_upper_left
:1;
698 unsigned pixel_center_integer
:1;
702 * Was the location explicitly set in the shader?
704 * If the location is explicitly set in the shader, it \b cannot be changed
705 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
708 unsigned explicit_location
:1;
709 unsigned explicit_index
:1;
712 * Was an initial binding explicitly set in the shader?
714 * If so, constant_value contains an integer ir_constant representing the
715 * initial binding point.
717 unsigned explicit_binding
:1;
720 * Was an initial component explicitly set in the shader?
722 unsigned explicit_component
:1;
725 * Does this variable have an initializer?
727 * This is used by the linker to cross-validiate initializers of global
730 unsigned has_initializer
:1;
733 * Is this variable a generic output or input that has not yet been matched
734 * up to a variable in another stage of the pipeline?
736 * This is used by the linker as scratch storage while assigning locations
737 * to generic inputs and outputs.
739 unsigned is_unmatched_generic_inout
:1;
742 * Is this varying used only by transform feedback?
744 * This is used by the linker to decide if its safe to pack the varying.
746 unsigned is_xfb_only
:1;
749 * Was a transfor feedback buffer set in the shader?
751 unsigned explicit_xfb_buffer
:1;
754 * Was a transfor feedback offset set in the shader?
756 unsigned explicit_xfb_offset
:1;
759 * Was a transfor feedback stride set in the shader?
761 unsigned explicit_xfb_stride
:1;
764 * If non-zero, then this variable may be packed along with other variables
765 * into a single varying slot, so this offset should be applied when
766 * accessing components. For example, an offset of 1 means that the x
767 * component of this variable is actually stored in component y of the
768 * location specified by \c location.
770 unsigned location_frac
:2;
773 * Layout of the matrix. Uses glsl_matrix_layout values.
775 unsigned matrix_layout
:2;
778 * Non-zero if this variable was created by lowering a named interface
781 unsigned from_named_ifc_block
:1;
784 * Non-zero if the variable must be a shader input. This is useful for
785 * constraints on function parameters.
787 unsigned must_be_shader_input
:1;
790 * Output index for dual source blending.
793 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
799 * Precision qualifier.
801 * In desktop GLSL we do not care about precision qualifiers at all, in
802 * fact, the spec says that precision qualifiers are ignored.
804 * To make things easy, we make it so that this field is always
805 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
806 * have the same precision value and the checks we add in the compiler
807 * for this field will never break a desktop shader compile.
809 unsigned precision
:2;
812 * \brief Layout qualifier for gl_FragDepth.
814 * This is not equal to \c ir_depth_layout_none if and only if this
815 * variable is \c gl_FragDepth and a layout qualifier is specified.
817 ir_depth_layout depth_layout
:3;
820 * ARB_shader_image_load_store qualifiers.
822 unsigned image_read_only
:1; /**< "readonly" qualifier. */
823 unsigned image_write_only
:1; /**< "writeonly" qualifier. */
824 unsigned image_coherent
:1;
825 unsigned image_volatile
:1;
826 unsigned image_restrict
:1;
829 * ARB_shader_storage_buffer_object
831 unsigned from_ssbo_unsized_array
:1; /**< unsized array buffer variable. */
833 unsigned implicit_sized_array
:1;
835 * Emit a warning if this variable is accessed.
838 uint8_t warn_extension_index
;
841 /** Image internal format if specified explicitly, otherwise GL_NONE. */
842 uint16_t image_format
;
846 * Number of state slots used
849 * This could be stored in as few as 7-bits, if necessary. If it is made
850 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
853 uint16_t _num_state_slots
;
857 * Initial binding point for a sampler, atomic, or UBO.
859 * For array types, this represents the binding point for the first element.
864 * Storage location of the base of this variable
866 * The precise meaning of this field depends on the nature of the variable.
868 * - Vertex shader input: one of the values from \c gl_vert_attrib.
869 * - Vertex shader output: one of the values from \c gl_varying_slot.
870 * - Geometry shader input: one of the values from \c gl_varying_slot.
871 * - Geometry shader output: one of the values from \c gl_varying_slot.
872 * - Fragment shader input: one of the values from \c gl_varying_slot.
873 * - Fragment shader output: one of the values from \c gl_frag_result.
874 * - Uniforms: Per-stage uniform slot number for default uniform block.
875 * - Uniforms: Index within the uniform block definition for UBO members.
876 * - Non-UBO Uniforms: explicit location until linking then reused to
877 * store uniform slot number.
878 * - Other: This field is not currently used.
880 * If the variable is a uniform, shader input, or shader output, and the
881 * slot has not been assigned, the value will be -1.
886 * for glsl->tgsi/mesa IR we need to store the index into the
887 * parameters for uniforms, initially the code overloaded location
888 * but this causes problems with indirect samplers and AoA.
889 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
894 * Vertex stream output identifier.
899 * Atomic, transform feedback or block member offset.
904 * Highest element accessed with a constant expression array index
906 * Not used for non-array variables. -1 is never accessed.
908 int max_array_access
;
911 * Transform feedback buffer.
916 * Transform feedback stride.
921 * Allow (only) ir_variable direct access private members.
923 friend class ir_variable
;
927 * Value assigned in the initializer of a variable declared "const"
929 ir_constant
*constant_value
;
932 * Constant expression assigned in the initializer of the variable
935 * This field and \c ::constant_value are distinct. Even if the two fields
936 * refer to constants with the same value, they must point to separate
939 ir_constant
*constant_initializer
;
942 static const char *const warn_extension_table
[];
946 * For variables which satisfy the is_interface_instance() predicate,
947 * this points to an array of integers such that if the ith member of
948 * the interface block is an array, max_ifc_array_access[i] is the
949 * maximum array element of that member that has been accessed. If the
950 * ith member of the interface block is not an array,
951 * max_ifc_array_access[i] is unused.
953 * For variables whose type is not an interface block, this pointer is
956 int *max_ifc_array_access
;
959 * Built-in state that backs this uniform
961 * Once set at variable creation, \c state_slots must remain invariant.
963 * If the variable is not a uniform, \c _num_state_slots will be zero
964 * and \c state_slots will be \c NULL.
966 ir_state_slot
*state_slots
;
970 * For variables that are in an interface block or are an instance of an
971 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
973 * \sa ir_variable::location
975 const glsl_type
*interface_type
;
978 * Name used for anonymous compiler temporaries
980 static const char tmp_name
[];
984 * Should the construct keep names for ir_var_temporary variables?
986 * When this global is false, names passed to the constructor for
987 * \c ir_var_temporary variables will be dropped. Instead, the variable will
988 * be named "compiler_temp". This name will be in static storage.
991 * \b NEVER change the mode of an \c ir_var_temporary.
994 * This variable is \b not thread-safe. It is global, \b not
995 * per-context. It begins life false. A context can, at some point, make
996 * it true. From that point on, it will be true forever. This should be
997 * okay since it will only be set true while debugging.
999 static bool temporaries_allocate_names
;
1003 * A function that returns whether a built-in function is available in the
1004 * current shading language (based on version, ES or desktop, and extensions).
1006 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
1010 * The representation of a function instance; may be the full definition or
1011 * simply a prototype.
1013 class ir_function_signature
: public ir_instruction
{
1014 /* An ir_function_signature will be part of the list of signatures in
1018 ir_function_signature(const glsl_type
*return_type
,
1019 builtin_available_predicate builtin_avail
= NULL
);
1021 virtual ir_function_signature
*clone(void *mem_ctx
,
1022 struct hash_table
*ht
) const;
1023 ir_function_signature
*clone_prototype(void *mem_ctx
,
1024 struct hash_table
*ht
) const;
1026 virtual void accept(ir_visitor
*v
)
1031 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1034 * Attempt to evaluate this function as a constant expression,
1035 * given a list of the actual parameters and the variable context.
1036 * Returns NULL for non-built-ins.
1038 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
1041 * Get the name of the function for which this is a signature
1043 const char *function_name() const;
1046 * Get a handle to the function for which this is a signature
1048 * There is no setter function, this function returns a \c const pointer,
1049 * and \c ir_function_signature::_function is private for a reason. The
1050 * only way to make a connection between a function and function signature
1051 * is via \c ir_function::add_signature. This helps ensure that certain
1052 * invariants (i.e., a function signature is in the list of signatures for
1053 * its \c _function) are met.
1055 * \sa ir_function::add_signature
1057 inline const class ir_function
*function() const
1059 return this->_function
;
1063 * Check whether the qualifiers match between this signature's parameters
1064 * and the supplied parameter list. If not, returns the name of the first
1065 * parameter with mismatched qualifiers (for use in error messages).
1067 const char *qualifiers_match(exec_list
*params
);
1070 * Replace the current parameter list with the given one. This is useful
1071 * if the current information came from a prototype, and either has invalid
1072 * or missing parameter names.
1074 void replace_parameters(exec_list
*new_params
);
1077 * Function return type.
1079 * \note This discards the optional precision qualifier.
1081 const struct glsl_type
*return_type
;
1084 * List of ir_variable of function parameters.
1086 * This represents the storage. The paramaters passed in a particular
1087 * call will be in ir_call::actual_paramaters.
1089 struct exec_list parameters
;
1091 /** Whether or not this function has a body (which may be empty). */
1092 unsigned is_defined
:1;
1094 /** Whether or not this function signature is a built-in. */
1095 bool is_builtin() const;
1098 * Whether or not this function is an intrinsic to be implemented
1103 /** Whether or not a built-in is available for this shader. */
1104 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1106 /** Body of instructions in the function. */
1107 struct exec_list body
;
1111 * A function pointer to a predicate that answers whether a built-in
1112 * function is available in the current shader. NULL if not a built-in.
1114 builtin_available_predicate builtin_avail
;
1116 /** Function of which this signature is one overload. */
1117 class ir_function
*_function
;
1119 /** Function signature of which this one is a prototype clone */
1120 const ir_function_signature
*origin
;
1122 friend class ir_function
;
1125 * Helper function to run a list of instructions for constant
1126 * expression evaluation.
1128 * The hash table represents the values of the visible variables.
1129 * There are no scoping issues because the table is indexed on
1130 * ir_variable pointers, not variable names.
1132 * Returns false if the expression is not constant, true otherwise,
1133 * and the value in *result if result is non-NULL.
1135 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1136 struct hash_table
*variable_context
,
1137 ir_constant
**result
);
1142 * Header for tracking multiple overloaded functions with the same name.
1143 * Contains a list of ir_function_signatures representing each of the
1146 class ir_function
: public ir_instruction
{
1148 ir_function(const char *name
);
1150 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1152 virtual void accept(ir_visitor
*v
)
1157 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1159 void add_signature(ir_function_signature
*sig
)
1161 sig
->_function
= this;
1162 this->signatures
.push_tail(sig
);
1166 * Find a signature that matches a set of actual parameters, taking implicit
1167 * conversions into account. Also flags whether the match was exact.
1169 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1170 const exec_list
*actual_param
,
1171 bool allow_builtins
,
1172 bool *match_is_exact
);
1175 * Find a signature that matches a set of actual parameters, taking implicit
1176 * conversions into account.
1178 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1179 const exec_list
*actual_param
,
1180 bool allow_builtins
);
1183 * Find a signature that exactly matches a set of actual parameters without
1184 * any implicit type conversions.
1186 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1187 const exec_list
*actual_ps
);
1190 * Name of the function.
1194 /** Whether or not this function has a signature that isn't a built-in. */
1195 bool has_user_signature();
1198 * List of ir_function_signature for each overloaded function with this name.
1200 struct exec_list signatures
;
1203 * is this function a subroutine type declaration
1204 * e.g. subroutine void type1(float arg1);
1209 * is this function associated to a subroutine type
1210 * e.g. subroutine (type1, type2) function_name { function_body };
1211 * would have num_subroutine_types 2,
1212 * and pointers to the type1 and type2 types.
1214 int num_subroutine_types
;
1215 const struct glsl_type
**subroutine_types
;
1217 int subroutine_index
;
1220 inline const char *ir_function_signature::function_name() const
1222 return this->_function
->name
;
1228 * IR instruction representing high-level if-statements
1230 class ir_if
: public ir_instruction
{
1232 ir_if(ir_rvalue
*condition
)
1233 : ir_instruction(ir_type_if
), condition(condition
)
1237 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1239 virtual void accept(ir_visitor
*v
)
1244 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1246 ir_rvalue
*condition
;
1247 /** List of ir_instruction for the body of the then branch */
1248 exec_list then_instructions
;
1249 /** List of ir_instruction for the body of the else branch */
1250 exec_list else_instructions
;
1255 * IR instruction representing a high-level loop structure.
1257 class ir_loop
: public ir_instruction
{
1261 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1263 virtual void accept(ir_visitor
*v
)
1268 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1270 /** List of ir_instruction that make up the body of the loop. */
1271 exec_list body_instructions
;
1275 class ir_assignment
: public ir_instruction
{
1277 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1280 * Construct an assignment with an explicit write mask
1283 * Since a write mask is supplied, the LHS must already be a bare
1284 * \c ir_dereference. The cannot be any swizzles in the LHS.
1286 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1287 unsigned write_mask
);
1289 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1291 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1293 virtual void accept(ir_visitor
*v
)
1298 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1301 * Get a whole variable written by an assignment
1303 * If the LHS of the assignment writes a whole variable, the variable is
1304 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1307 * - Assigning to a scalar
1308 * - Assigning to all components of a vector
1309 * - Whole array (or matrix) assignment
1310 * - Whole structure assignment
1312 ir_variable
*whole_variable_written();
1315 * Set the LHS of an assignment
1317 void set_lhs(ir_rvalue
*lhs
);
1320 * Left-hand side of the assignment.
1322 * This should be treated as read only. If you need to set the LHS of an
1323 * assignment, use \c ir_assignment::set_lhs.
1325 ir_dereference
*lhs
;
1328 * Value being assigned
1333 * Optional condition for the assignment.
1335 ir_rvalue
*condition
;
1339 * Component mask written
1341 * For non-vector types in the LHS, this field will be zero. For vector
1342 * types, a bit will be set for each component that is written. Note that
1343 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1345 * A partially-set write mask means that each enabled channel gets
1346 * the value from a consecutive channel of the rhs. For example,
1347 * to write just .xyw of gl_FrontColor with color:
1349 * (assign (constant bool (1)) (xyw)
1350 * (var_ref gl_FragColor)
1351 * (swiz xyw (var_ref color)))
1353 unsigned write_mask
:4;
1356 /* Update ir_expression::get_num_operands() and operator_strs when
1357 * updating this list.
1359 enum ir_expression_operation
{
1368 ir_unop_exp
, /**< Log base e on gentype */
1369 ir_unop_log
, /**< Natural log on gentype */
1372 ir_unop_f2i
, /**< Float-to-integer conversion. */
1373 ir_unop_f2u
, /**< Float-to-unsigned conversion. */
1374 ir_unop_i2f
, /**< Integer-to-float conversion. */
1375 ir_unop_f2b
, /**< Float-to-boolean conversion */
1376 ir_unop_b2f
, /**< Boolean-to-float conversion */
1377 ir_unop_i2b
, /**< int-to-boolean conversion */
1378 ir_unop_b2i
, /**< Boolean-to-int conversion */
1379 ir_unop_u2f
, /**< Unsigned-to-float conversion. */
1380 ir_unop_i2u
, /**< Integer-to-unsigned conversion. */
1381 ir_unop_u2i
, /**< Unsigned-to-integer conversion. */
1382 ir_unop_d2f
, /**< Double-to-float conversion. */
1383 ir_unop_f2d
, /**< Float-to-double conversion. */
1384 ir_unop_d2i
, /**< Double-to-integer conversion. */
1385 ir_unop_i2d
, /**< Integer-to-double conversion. */
1386 ir_unop_d2u
, /**< Double-to-unsigned conversion. */
1387 ir_unop_u2d
, /**< Unsigned-to-double conversion. */
1388 ir_unop_d2b
, /**< Double-to-boolean conversion. */
1389 ir_unop_bitcast_i2f
, /**< Bit-identical int-to-float "conversion" */
1390 ir_unop_bitcast_f2i
, /**< Bit-identical float-to-int "conversion" */
1391 ir_unop_bitcast_u2f
, /**< Bit-identical uint-to-float "conversion" */
1392 ir_unop_bitcast_f2u
, /**< Bit-identical float-to-uint "conversion" */
1395 * \name Unary floating-point rounding operations.
1406 * \name Trigonometric operations.
1414 * \name Partial derivatives.
1418 ir_unop_dFdx_coarse
,
1421 ir_unop_dFdy_coarse
,
1426 * \name Floating point pack and unpack operations.
1429 ir_unop_pack_snorm_2x16
,
1430 ir_unop_pack_snorm_4x8
,
1431 ir_unop_pack_unorm_2x16
,
1432 ir_unop_pack_unorm_4x8
,
1433 ir_unop_pack_half_2x16
,
1434 ir_unop_unpack_snorm_2x16
,
1435 ir_unop_unpack_snorm_4x8
,
1436 ir_unop_unpack_unorm_2x16
,
1437 ir_unop_unpack_unorm_4x8
,
1438 ir_unop_unpack_half_2x16
,
1442 * \name Bit operations, part of ARB_gpu_shader5.
1445 ir_unop_bitfield_reverse
,
1454 * \name Double packing, part of ARB_gpu_shader_fp64.
1457 ir_unop_pack_double_2x32
,
1458 ir_unop_unpack_double_2x32
,
1466 ir_unop_subroutine_to_int
,
1468 * Interpolate fs input at centroid
1470 * operand0 is the fs input.
1472 ir_unop_interpolate_at_centroid
,
1475 * Ask the driver for the total size of a buffer block.
1477 * operand0 is the ir_constant buffer block index in the linked shader.
1479 ir_unop_get_buffer_size
,
1482 * Calculate length of an unsized array inside a buffer block.
1483 * This opcode is going to be replaced in a lowering pass inside
1486 * operand0 is the unsized array's ir_value for the calculation
1489 ir_unop_ssbo_unsized_array_length
,
1492 * Vote among threads on the value of the boolean argument.
1499 * A sentinel marking the last of the unary operations.
1501 ir_last_unop
= ir_unop_vote_eq
,
1505 ir_binop_mul
, /**< Floating-point or low 32-bit integer multiply. */
1506 ir_binop_imul_high
, /**< Calculates the high 32-bits of a 64-bit multiply. */
1510 * Returns the carry resulting from the addition of the two arguments.
1517 * Returns the borrow resulting from the subtraction of the second argument
1518 * from the first argument.
1525 * Takes one of two combinations of arguments:
1528 * - mod(vecN, float)
1530 * Does not take integer types.
1535 * \name Binary comparison operators which return a boolean vector.
1536 * The type of both operands must be equal.
1546 * Returns single boolean for whether all components of operands[0]
1547 * equal the components of operands[1].
1551 * Returns single boolean for whether any component of operands[0]
1552 * is not equal to the corresponding component of operands[1].
1554 ir_binop_any_nequal
,
1558 * \name Bit-wise binary operations.
1579 * Load a value the size of a given GLSL type from a uniform block.
1581 * operand0 is the ir_constant uniform block index in the linked shader.
1582 * operand1 is a byte offset within the uniform block.
1587 * \name Multiplies a number by two to a power, part of ARB_gpu_shader5.
1594 * Extract a scalar from a vector
1596 * operand0 is the vector
1597 * operand1 is the index of the field to read from operand0
1599 ir_binop_vector_extract
,
1602 * Interpolate fs input at offset
1604 * operand0 is the fs input
1605 * operand1 is the offset from the pixel center
1607 ir_binop_interpolate_at_offset
,
1610 * Interpolate fs input at sample position
1612 * operand0 is the fs input
1613 * operand1 is the sample ID
1615 ir_binop_interpolate_at_sample
,
1618 * A sentinel marking the last of the binary operations.
1620 ir_last_binop
= ir_binop_interpolate_at_sample
,
1623 * \name Fused floating-point multiply-add, part of ARB_gpu_shader5.
1632 * \name Conditional Select
1634 * A vector conditional select instruction (like ?:, but operating per-
1635 * component on vectors).
1637 * \see lower_instructions_visitor::ldexp_to_arith
1643 ir_triop_bitfield_extract
,
1646 * Generate a value with one field of a vector changed
1648 * operand0 is the vector
1649 * operand1 is the value to write into the vector result
1650 * operand2 is the index in operand0 to be modified
1652 ir_triop_vector_insert
,
1655 * A sentinel marking the last of the ternary operations.
1657 ir_last_triop
= ir_triop_vector_insert
,
1659 ir_quadop_bitfield_insert
,
1664 * A sentinel marking the last of the ternary operations.
1666 ir_last_quadop
= ir_quadop_vector
,
1669 * A sentinel marking the last of all operations.
1671 ir_last_opcode
= ir_quadop_vector
1674 class ir_expression
: public ir_rvalue
{
1676 ir_expression(int op
, const struct glsl_type
*type
,
1677 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1678 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1681 * Constructor for unary operation expressions
1683 ir_expression(int op
, ir_rvalue
*);
1686 * Constructor for binary operation expressions
1688 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1691 * Constructor for ternary operation expressions
1693 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1695 virtual bool equals(const ir_instruction
*ir
,
1696 enum ir_node_type ignore
= ir_type_unset
) const;
1698 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1701 * Attempt to constant-fold the expression
1703 * The "variable_context" hash table links ir_variable * to ir_constant *
1704 * that represent the variables' values. \c NULL represents an empty
1707 * If the expression cannot be constant folded, this method will return
1710 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1713 * Determine the number of operands used by an expression
1715 static unsigned int get_num_operands(ir_expression_operation
);
1718 * Determine the number of operands used by an expression
1720 unsigned int get_num_operands() const
1722 return (this->operation
== ir_quadop_vector
)
1723 ? this->type
->vector_elements
: get_num_operands(operation
);
1727 * Return whether the expression operates on vectors horizontally.
1729 bool is_horizontal() const
1731 return operation
== ir_binop_all_equal
||
1732 operation
== ir_binop_any_nequal
||
1733 operation
== ir_binop_dot
||
1734 operation
== ir_binop_vector_extract
||
1735 operation
== ir_triop_vector_insert
||
1736 operation
== ir_binop_ubo_load
||
1737 operation
== ir_quadop_vector
;
1741 * Return a string representing this expression's operator.
1743 const char *operator_string();
1746 * Return a string representing this expression's operator.
1748 static const char *operator_string(ir_expression_operation
);
1752 * Do a reverse-lookup to translate the given string into an operator.
1754 static ir_expression_operation
get_operator(const char *);
1756 virtual void accept(ir_visitor
*v
)
1761 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1763 virtual ir_variable
*variable_referenced() const;
1765 ir_expression_operation operation
;
1766 ir_rvalue
*operands
[4];
1771 * HIR instruction representing a high-level function call, containing a list
1772 * of parameters and returning a value in the supplied temporary.
1774 class ir_call
: public ir_instruction
{
1776 ir_call(ir_function_signature
*callee
,
1777 ir_dereference_variable
*return_deref
,
1778 exec_list
*actual_parameters
)
1779 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(NULL
), array_idx(NULL
)
1781 assert(callee
->return_type
!= NULL
);
1782 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1783 this->use_builtin
= callee
->is_builtin();
1786 ir_call(ir_function_signature
*callee
,
1787 ir_dereference_variable
*return_deref
,
1788 exec_list
*actual_parameters
,
1789 ir_variable
*var
, ir_rvalue
*array_idx
)
1790 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(var
), array_idx(array_idx
)
1792 assert(callee
->return_type
!= NULL
);
1793 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1794 this->use_builtin
= callee
->is_builtin();
1797 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1799 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1801 virtual void accept(ir_visitor
*v
)
1806 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1809 * Get the name of the function being called.
1811 const char *callee_name() const
1813 return callee
->function_name();
1817 * Generates an inline version of the function before @ir,
1818 * storing the return value in return_deref.
1820 void generate_inline(ir_instruction
*ir
);
1823 * Storage for the function's return value.
1824 * This must be NULL if the return type is void.
1826 ir_dereference_variable
*return_deref
;
1829 * The specific function signature being called.
1831 ir_function_signature
*callee
;
1833 /* List of ir_rvalue of paramaters passed in this call. */
1834 exec_list actual_parameters
;
1836 /** Should this call only bind to a built-in function? */
1840 * ARB_shader_subroutine support -
1841 * the subroutine uniform variable and array index
1842 * rvalue to be used in the lowering pass later.
1844 ir_variable
*sub_var
;
1845 ir_rvalue
*array_idx
;
1850 * \name Jump-like IR instructions.
1852 * These include \c break, \c continue, \c return, and \c discard.
1855 class ir_jump
: public ir_instruction
{
1857 ir_jump(enum ir_node_type t
)
1863 class ir_return
: public ir_jump
{
1866 : ir_jump(ir_type_return
), value(NULL
)
1870 ir_return(ir_rvalue
*value
)
1871 : ir_jump(ir_type_return
), value(value
)
1875 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1877 ir_rvalue
*get_value() const
1882 virtual void accept(ir_visitor
*v
)
1887 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1894 * Jump instructions used inside loops
1896 * These include \c break and \c continue. The \c break within a loop is
1897 * different from the \c break within a switch-statement.
1899 * \sa ir_switch_jump
1901 class ir_loop_jump
: public ir_jump
{
1908 ir_loop_jump(jump_mode mode
)
1909 : ir_jump(ir_type_loop_jump
)
1914 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1916 virtual void accept(ir_visitor
*v
)
1921 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1923 bool is_break() const
1925 return mode
== jump_break
;
1928 bool is_continue() const
1930 return mode
== jump_continue
;
1933 /** Mode selector for the jump instruction. */
1934 enum jump_mode mode
;
1938 * IR instruction representing discard statements.
1940 class ir_discard
: public ir_jump
{
1943 : ir_jump(ir_type_discard
)
1945 this->condition
= NULL
;
1948 ir_discard(ir_rvalue
*cond
)
1949 : ir_jump(ir_type_discard
)
1951 this->condition
= cond
;
1954 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1956 virtual void accept(ir_visitor
*v
)
1961 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1963 ir_rvalue
*condition
;
1969 * Texture sampling opcodes used in ir_texture
1971 enum ir_texture_opcode
{
1972 ir_tex
, /**< Regular texture look-up */
1973 ir_txb
, /**< Texture look-up with LOD bias */
1974 ir_txl
, /**< Texture look-up with explicit LOD */
1975 ir_txd
, /**< Texture look-up with partial derivatvies */
1976 ir_txf
, /**< Texel fetch with explicit LOD */
1977 ir_txf_ms
, /**< Multisample texture fetch */
1978 ir_txs
, /**< Texture size */
1979 ir_lod
, /**< Texture lod query */
1980 ir_tg4
, /**< Texture gather */
1981 ir_query_levels
, /**< Texture levels query */
1982 ir_texture_samples
, /**< Texture samples query */
1983 ir_samples_identical
, /**< Query whether all samples are definitely identical. */
1988 * IR instruction to sample a texture
1990 * The specific form of the IR instruction depends on the \c mode value
1991 * selected from \c ir_texture_opcodes. In the printed IR, these will
1994 * Texel offset (0 or an expression)
1995 * | Projection divisor
1996 * | | Shadow comparitor
1999 * (tex <type> <sampler> <coordinate> 0 1 ( ))
2000 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
2001 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
2002 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
2003 * (txf <type> <sampler> <coordinate> 0 <lod>)
2005 * <type> <sampler> <coordinate> <sample_index>)
2006 * (txs <type> <sampler> <lod>)
2007 * (lod <type> <sampler> <coordinate>)
2008 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
2009 * (query_levels <type> <sampler>)
2010 * (samples_identical <sampler> <coordinate>)
2012 class ir_texture
: public ir_rvalue
{
2014 ir_texture(enum ir_texture_opcode op
)
2015 : ir_rvalue(ir_type_texture
),
2016 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
2017 shadow_comparitor(NULL
), offset(NULL
)
2019 memset(&lod_info
, 0, sizeof(lod_info
));
2022 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
2024 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2026 virtual void accept(ir_visitor
*v
)
2031 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2033 virtual bool equals(const ir_instruction
*ir
,
2034 enum ir_node_type ignore
= ir_type_unset
) const;
2037 * Return a string representing the ir_texture_opcode.
2039 const char *opcode_string();
2041 /** Set the sampler and type. */
2042 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
2045 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
2047 static ir_texture_opcode
get_opcode(const char *);
2049 enum ir_texture_opcode op
;
2051 /** Sampler to use for the texture access. */
2052 ir_dereference
*sampler
;
2054 /** Texture coordinate to sample */
2055 ir_rvalue
*coordinate
;
2058 * Value used for projective divide.
2060 * If there is no projective divide (the common case), this will be
2061 * \c NULL. Optimization passes should check for this to point to a constant
2062 * of 1.0 and replace that with \c NULL.
2064 ir_rvalue
*projector
;
2067 * Coordinate used for comparison on shadow look-ups.
2069 * If there is no shadow comparison, this will be \c NULL. For the
2070 * \c ir_txf opcode, this *must* be \c NULL.
2072 ir_rvalue
*shadow_comparitor
;
2074 /** Texel offset. */
2078 ir_rvalue
*lod
; /**< Floating point LOD */
2079 ir_rvalue
*bias
; /**< Floating point LOD bias */
2080 ir_rvalue
*sample_index
; /**< MSAA sample index */
2081 ir_rvalue
*component
; /**< Gather component selector */
2083 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
2084 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
2090 struct ir_swizzle_mask
{
2097 * Number of components in the swizzle.
2099 unsigned num_components
:3;
2102 * Does the swizzle contain duplicate components?
2104 * L-value swizzles cannot contain duplicate components.
2106 unsigned has_duplicates
:1;
2110 class ir_swizzle
: public ir_rvalue
{
2112 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
2115 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
2117 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
2119 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
2121 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2124 * Construct an ir_swizzle from the textual representation. Can fail.
2126 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
2128 virtual void accept(ir_visitor
*v
)
2133 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2135 virtual bool equals(const ir_instruction
*ir
,
2136 enum ir_node_type ignore
= ir_type_unset
) const;
2138 bool is_lvalue() const
2140 return val
->is_lvalue() && !mask
.has_duplicates
;
2144 * Get the variable that is ultimately referenced by an r-value
2146 virtual ir_variable
*variable_referenced() const;
2149 ir_swizzle_mask mask
;
2153 * Initialize the mask component of a swizzle
2155 * This is used by the \c ir_swizzle constructors.
2157 void init_mask(const unsigned *components
, unsigned count
);
2161 class ir_dereference
: public ir_rvalue
{
2163 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2165 bool is_lvalue() const;
2168 * Get the variable that is ultimately referenced by an r-value
2170 virtual ir_variable
*variable_referenced() const = 0;
2173 ir_dereference(enum ir_node_type t
)
2180 class ir_dereference_variable
: public ir_dereference
{
2182 ir_dereference_variable(ir_variable
*var
);
2184 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2185 struct hash_table
*) const;
2187 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2189 virtual bool equals(const ir_instruction
*ir
,
2190 enum ir_node_type ignore
= ir_type_unset
) const;
2193 * Get the variable that is ultimately referenced by an r-value
2195 virtual ir_variable
*variable_referenced() const
2200 virtual ir_variable
*whole_variable_referenced()
2202 /* ir_dereference_variable objects always dereference the entire
2203 * variable. However, if this dereference is dereferenced by anything
2204 * else, the complete deferefernce chain is not a whole-variable
2205 * dereference. This method should only be called on the top most
2206 * ir_rvalue in a dereference chain.
2211 virtual void accept(ir_visitor
*v
)
2216 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2219 * Object being dereferenced.
2225 class ir_dereference_array
: public ir_dereference
{
2227 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2229 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2231 virtual ir_dereference_array
*clone(void *mem_ctx
,
2232 struct hash_table
*) const;
2234 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2236 virtual bool equals(const ir_instruction
*ir
,
2237 enum ir_node_type ignore
= ir_type_unset
) const;
2240 * Get the variable that is ultimately referenced by an r-value
2242 virtual ir_variable
*variable_referenced() const
2244 return this->array
->variable_referenced();
2247 virtual void accept(ir_visitor
*v
)
2252 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2255 ir_rvalue
*array_index
;
2258 void set_array(ir_rvalue
*value
);
2262 class ir_dereference_record
: public ir_dereference
{
2264 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2266 ir_dereference_record(ir_variable
*var
, const char *field
);
2268 virtual ir_dereference_record
*clone(void *mem_ctx
,
2269 struct hash_table
*) const;
2271 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2274 * Get the variable that is ultimately referenced by an r-value
2276 virtual ir_variable
*variable_referenced() const
2278 return this->record
->variable_referenced();
2281 virtual void accept(ir_visitor
*v
)
2286 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2294 * Data stored in an ir_constant
2296 union ir_constant_data
{
2305 class ir_constant
: public ir_rvalue
{
2307 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2308 ir_constant(bool b
, unsigned vector_elements
=1);
2309 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2310 ir_constant(int i
, unsigned vector_elements
=1);
2311 ir_constant(float f
, unsigned vector_elements
=1);
2312 ir_constant(double d
, unsigned vector_elements
=1);
2315 * Construct an ir_constant from a list of ir_constant values
2317 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2320 * Construct an ir_constant from a scalar component of another ir_constant
2322 * The new \c ir_constant inherits the type of the component from the
2326 * In the case of a matrix constant, the new constant is a scalar, \b not
2329 ir_constant(const ir_constant
*c
, unsigned i
);
2332 * Return a new ir_constant of the specified type containing all zeros.
2334 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2336 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2338 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2340 virtual void accept(ir_visitor
*v
)
2345 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2347 virtual bool equals(const ir_instruction
*ir
,
2348 enum ir_node_type ignore
= ir_type_unset
) const;
2351 * Get a particular component of a constant as a specific type
2353 * This is useful, for example, to get a value from an integer constant
2354 * as a float or bool. This appears frequently when constructors are
2355 * called with all constant parameters.
2358 bool get_bool_component(unsigned i
) const;
2359 float get_float_component(unsigned i
) const;
2360 double get_double_component(unsigned i
) const;
2361 int get_int_component(unsigned i
) const;
2362 unsigned get_uint_component(unsigned i
) const;
2365 ir_constant
*get_array_element(unsigned i
) const;
2367 ir_constant
*get_record_field(const char *name
);
2370 * Copy the values on another constant at a given offset.
2372 * The offset is ignored for array or struct copies, it's only for
2373 * scalars or vectors into vectors or matrices.
2375 * With identical types on both sides and zero offset it's clone()
2376 * without creating a new object.
2379 void copy_offset(ir_constant
*src
, int offset
);
2382 * Copy the values on another constant at a given offset and
2383 * following an assign-like mask.
2385 * The mask is ignored for scalars.
2387 * Note that this function only handles what assign can handle,
2388 * i.e. at most a vector as source and a column of a matrix as
2392 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2395 * Determine whether a constant has the same value as another constant
2397 * \sa ir_constant::is_zero, ir_constant::is_one,
2398 * ir_constant::is_negative_one
2400 bool has_value(const ir_constant
*) const;
2403 * Return true if this ir_constant represents the given value.
2405 * For vectors, this checks that each component is the given value.
2407 virtual bool is_value(float f
, int i
) const;
2408 virtual bool is_zero() const;
2409 virtual bool is_one() const;
2410 virtual bool is_negative_one() const;
2413 * Return true for constants that could be stored as 16-bit unsigned values.
2415 * Note that this will return true even for signed integer ir_constants, as
2416 * long as the value is non-negative and fits in 16-bits.
2418 virtual bool is_uint16_constant() const;
2421 * Value of the constant.
2423 * The field used to back the values supplied by the constant is determined
2424 * by the type associated with the \c ir_instruction. Constants may be
2425 * scalars, vectors, or matrices.
2427 union ir_constant_data value
;
2429 /* Array elements */
2430 ir_constant
**array_elements
;
2432 /* Structure fields */
2433 exec_list components
;
2437 * Parameterless constructor only used by the clone method
2443 * IR instruction to emit a vertex in a geometry shader.
2445 class ir_emit_vertex
: public ir_instruction
{
2447 ir_emit_vertex(ir_rvalue
*stream
)
2448 : ir_instruction(ir_type_emit_vertex
),
2454 virtual void accept(ir_visitor
*v
)
2459 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2461 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2464 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2466 int stream_id() const
2468 return stream
->as_constant()->value
.i
[0];
2475 * IR instruction to complete the current primitive and start a new one in a
2478 class ir_end_primitive
: public ir_instruction
{
2480 ir_end_primitive(ir_rvalue
*stream
)
2481 : ir_instruction(ir_type_end_primitive
),
2487 virtual void accept(ir_visitor
*v
)
2492 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2494 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2497 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2499 int stream_id() const
2501 return stream
->as_constant()->value
.i
[0];
2508 * IR instruction for tessellation control and compute shader barrier.
2510 class ir_barrier
: public ir_instruction
{
2513 : ir_instruction(ir_type_barrier
)
2517 virtual void accept(ir_visitor
*v
)
2522 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2524 return new(mem_ctx
) ir_barrier();
2527 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2533 * Apply a visitor to each IR node in a list
2536 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2539 * Validate invariants on each IR node in a list
2541 void validate_ir_tree(exec_list
*instructions
);
2543 struct _mesa_glsl_parse_state
;
2544 struct gl_shader_program
;
2547 * Detect whether an unlinked shader contains static recursion
2549 * If the list of instructions is determined to contain static recursion,
2550 * \c _mesa_glsl_error will be called to emit error messages for each function
2551 * that is in the recursion cycle.
2554 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2555 exec_list
*instructions
);
2558 * Detect whether a linked shader contains static recursion
2560 * If the list of instructions is determined to contain static recursion,
2561 * \c link_error_printf will be called to emit error messages for each function
2562 * that is in the recursion cycle. In addition,
2563 * \c gl_shader_program::LinkStatus will be set to false.
2566 detect_recursion_linked(struct gl_shader_program
*prog
,
2567 exec_list
*instructions
);
2570 * Make a clone of each IR instruction in a list
2572 * \param in List of IR instructions that are to be cloned
2573 * \param out List to hold the cloned instructions
2576 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2579 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2580 struct _mesa_glsl_parse_state
*state
);
2583 _mesa_glsl_initialize_derived_variables(struct gl_context
*ctx
,
2587 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state
*state
);
2590 _mesa_glsl_initialize_builtin_functions();
2592 extern ir_function_signature
*
2593 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2594 const char *name
, exec_list
*actual_parameters
);
2596 extern ir_function
*
2597 _mesa_glsl_find_builtin_function_by_name(const char *name
);
2600 _mesa_glsl_get_builtin_function_shader(void);
2602 extern ir_function_signature
*
2603 _mesa_get_main_function_signature(glsl_symbol_table
*symbols
);
2606 _mesa_glsl_release_functions(void);
2609 _mesa_glsl_release_builtin_functions(void);
2612 reparent_ir(exec_list
*list
, void *mem_ctx
);
2614 struct glsl_symbol_table
;
2617 import_prototypes(const exec_list
*source
, exec_list
*dest
,
2618 struct glsl_symbol_table
*symbols
, void *mem_ctx
);
2621 ir_has_call(ir_instruction
*ir
);
2624 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2625 gl_shader_stage shader_stage
);
2628 prototype_string(const glsl_type
*return_type
, const char *name
,
2629 exec_list
*parameters
);
2632 mode_string(const ir_variable
*var
);
2635 * Built-in / reserved GL variables names start with "gl_"
2638 is_gl_identifier(const char *s
)
2640 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2644 #endif /* __cplusplus */
2646 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2647 struct _mesa_glsl_parse_state
*state
);
2650 fprint_ir(FILE *f
, const void *instruction
);
2652 extern const struct gl_builtin_uniform_desc
*
2653 _mesa_glsl_get_builtin_uniform_desc(const char *name
);
2660 vertices_per_prim(GLenum prim
);