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.
31 #include "util/ralloc.h"
32 #include "compiler/glsl_types.h"
34 #include "ir_visitor.h"
35 #include "ir_hierarchical_visitor.h"
40 * \defgroup IR Intermediate representation nodes
48 * Each concrete class derived from \c ir_instruction has a value in this
49 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
50 * by the constructor. While using type tags is not very C++, it is extremely
51 * convenient. For example, during debugging you can simply inspect
52 * \c ir_instruction::ir_type to find out the actual type of the object.
54 * In addition, it is possible to use a switch-statement based on \c
55 * \c ir_instruction::ir_type to select different behavior for different object
56 * types. For functions that have only slight differences for several object
57 * types, this allows writing very straightforward, readable code.
60 ir_type_dereference_array
,
61 ir_type_dereference_record
,
62 ir_type_dereference_variable
,
71 ir_type_function_signature
,
78 ir_type_end_primitive
,
80 ir_type_max
, /**< maximum ir_type enum number, for validation */
81 ir_type_unset
= ir_type_max
86 * Base class of all IR instructions
88 class ir_instruction
: public exec_node
{
90 enum ir_node_type ir_type
;
93 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
94 * there's a virtual destructor present. Because we almost
95 * universally use ralloc for our memory management of
96 * ir_instructions, the destructor doesn't need to do any work.
98 virtual ~ir_instruction()
102 /** ir_print_visitor helper for debugging. */
103 void print(void) const;
104 void fprint(FILE *f
) const;
106 virtual void accept(ir_visitor
*) = 0;
107 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*) = 0;
108 virtual ir_instruction
*clone(void *mem_ctx
,
109 struct hash_table
*ht
) const = 0;
111 bool is_rvalue() const
113 return ir_type
== ir_type_dereference_array
||
114 ir_type
== ir_type_dereference_record
||
115 ir_type
== ir_type_dereference_variable
||
116 ir_type
== ir_type_constant
||
117 ir_type
== ir_type_expression
||
118 ir_type
== ir_type_swizzle
||
119 ir_type
== ir_type_texture
;
122 bool is_dereference() const
124 return ir_type
== ir_type_dereference_array
||
125 ir_type
== ir_type_dereference_record
||
126 ir_type
== ir_type_dereference_variable
;
131 return ir_type
== ir_type_loop_jump
||
132 ir_type
== ir_type_return
||
133 ir_type
== ir_type_discard
;
137 * \name IR instruction downcast functions
139 * These functions either cast the object to a derived class or return
140 * \c NULL if the object's type does not match the specified derived class.
141 * Additional downcast functions will be added as needed.
144 #define AS_BASE(TYPE) \
145 class ir_##TYPE *as_##TYPE() \
147 assume(this != NULL); \
148 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
150 const class ir_##TYPE *as_##TYPE() const \
152 assume(this != NULL); \
153 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
161 #define AS_CHILD(TYPE) \
162 class ir_##TYPE * as_##TYPE() \
164 assume(this != NULL); \
165 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
167 const class ir_##TYPE * as_##TYPE() const \
169 assume(this != NULL); \
170 return ir_type == ir_type_##TYPE ? (const ir_##TYPE *) this : NULL; \
174 AS_CHILD(dereference_array
)
175 AS_CHILD(dereference_variable
)
176 AS_CHILD(dereference_record
)
191 * IR equality method: Return true if the referenced instruction would
192 * return the same value as this one.
194 * This intended to be used for CSE and algebraic optimizations, on rvalues
195 * in particular. No support for other instruction types (assignments,
196 * jumps, calls, etc.) is planned.
198 virtual bool equals(const ir_instruction
*ir
,
199 enum ir_node_type ignore
= ir_type_unset
) const;
202 ir_instruction(enum ir_node_type t
)
210 assert(!"Should not get here.");
216 * The base class for all "values"/expression trees.
218 class ir_rvalue
: public ir_instruction
{
220 const struct glsl_type
*type
;
222 virtual ir_rvalue
*clone(void *mem_ctx
, struct hash_table
*) const;
224 virtual void accept(ir_visitor
*v
)
229 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
231 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
232 struct hash_table
*variable_context
= NULL
);
234 ir_rvalue
*as_rvalue_to_saturate();
236 virtual bool is_lvalue(const struct _mesa_glsl_parse_state
* = NULL
) const
242 * Get the variable that is ultimately referenced by an r-value
244 virtual ir_variable
*variable_referenced() const
251 * If an r-value is a reference to a whole variable, get that variable
254 * Pointer to a variable that is completely dereferenced by the r-value. If
255 * the r-value is not a dereference or the dereference does not access the
256 * entire variable (i.e., it's just one array element, struct field), \c NULL
259 virtual ir_variable
*whole_variable_referenced()
265 * Determine if an r-value has the value zero
267 * The base implementation of this function always returns \c false. The
268 * \c ir_constant class over-rides this function to return \c true \b only
269 * for vector and scalar types that have all elements set to the value
270 * zero (or \c false for booleans).
272 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
274 virtual bool is_zero() const;
277 * Determine if an r-value has the value one
279 * The base implementation of this function always returns \c false. The
280 * \c ir_constant class over-rides this function to return \c true \b only
281 * for vector and scalar types that have all elements set to the value
282 * one (or \c true for booleans).
284 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
286 virtual bool is_one() const;
289 * Determine if an r-value has the value negative one
291 * The base implementation of this function always returns \c false. The
292 * \c ir_constant class over-rides this function to return \c true \b only
293 * for vector and scalar types that have all elements set to the value
294 * negative one. For boolean types, the result is always \c false.
296 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
298 virtual bool is_negative_one() const;
301 * Determine if an r-value is an unsigned integer constant which can be
304 * \sa ir_constant::is_uint16_constant.
306 virtual bool is_uint16_constant() const { return false; }
309 * Return a generic value of error_type.
311 * Allocation will be performed with 'mem_ctx' as ralloc owner.
313 static ir_rvalue
*error_value(void *mem_ctx
);
316 ir_rvalue(enum ir_node_type t
);
321 * Variable storage classes
323 enum ir_variable_mode
{
324 ir_var_auto
= 0, /**< Function local variables and globals. */
325 ir_var_uniform
, /**< Variable declared as a uniform. */
326 ir_var_shader_storage
, /**< Variable declared as an ssbo. */
327 ir_var_shader_shared
, /**< Variable declared as shared. */
332 ir_var_function_inout
,
333 ir_var_const_in
, /**< "in" param that must be a constant expression */
334 ir_var_system_value
, /**< Ex: front-face, instance-id, etc. */
335 ir_var_temporary
, /**< Temporary variable generated during compilation. */
336 ir_var_mode_count
/**< Number of variable modes */
340 * Enum keeping track of how a variable was declared. For error checking of
341 * the gl_PerVertex redeclaration rules.
343 enum ir_var_declaration_type
{
345 * Normal declaration (for most variables, this means an explicit
346 * declaration. Exception: temporaries are always implicitly declared, but
347 * they still use ir_var_declared_normally).
349 * Note: an ir_variable that represents a named interface block uses
350 * ir_var_declared_normally.
352 ir_var_declared_normally
= 0,
355 * Variable was explicitly declared (or re-declared) in an unnamed
358 ir_var_declared_in_block
,
361 * Variable is an implicitly declared built-in that has not been explicitly
362 * re-declared by the shader.
364 ir_var_declared_implicitly
,
367 * Variable is implicitly generated by the compiler and should not be
368 * visible via the API.
374 * \brief Layout qualifiers for gl_FragDepth.
376 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
377 * with a layout qualifier.
379 enum ir_depth_layout
{
380 ir_depth_layout_none
, /**< No depth layout is specified. */
382 ir_depth_layout_greater
,
383 ir_depth_layout_less
,
384 ir_depth_layout_unchanged
388 * \brief Convert depth layout qualifier to string.
391 depth_layout_string(ir_depth_layout layout
);
394 * Description of built-in state associated with a uniform
396 * \sa ir_variable::state_slots
398 struct ir_state_slot
{
399 gl_state_index16 tokens
[STATE_LENGTH
];
405 * Get the string value for an interpolation qualifier
407 * \return The string that would be used in a shader to specify \c
408 * mode will be returned.
410 * This function is used to generate error messages of the form "shader
411 * uses %s interpolation qualifier", so in the case where there is no
412 * interpolation qualifier, it returns "no".
414 * This function should only be used on a shader input or output variable.
416 const char *interpolation_string(unsigned interpolation
);
419 class ir_variable
: public ir_instruction
{
421 ir_variable(const struct glsl_type
*, const char *, ir_variable_mode
);
423 virtual ir_variable
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
425 virtual void accept(ir_visitor
*v
)
430 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
434 * Determine whether or not a variable is part of a uniform or
435 * shader storage block.
437 inline bool is_in_buffer_block() const
439 return (this->data
.mode
== ir_var_uniform
||
440 this->data
.mode
== ir_var_shader_storage
) &&
441 this->interface_type
!= NULL
;
445 * Determine whether or not a variable is part of a shader storage block.
447 inline bool is_in_shader_storage_block() const
449 return this->data
.mode
== ir_var_shader_storage
&&
450 this->interface_type
!= NULL
;
454 * Determine whether or not a variable is the declaration of an interface
457 * For the first declaration below, there will be an \c ir_variable named
458 * "instance" whose type and whose instance_type will be the same
459 * \c glsl_type. For the second declaration, there will be an \c ir_variable
460 * named "f" whose type is float and whose instance_type is B2.
462 * "instance" is an interface instance variable, but "f" is not.
472 inline bool is_interface_instance() const
474 return this->type
->without_array() == this->interface_type
;
478 * Return whether this variable contains a bindless sampler/image.
480 inline bool contains_bindless() const
482 if (!this->type
->contains_sampler() && !this->type
->contains_image())
485 return this->data
.bindless
|| this->data
.mode
!= ir_var_uniform
;
489 * Set this->interface_type on a newly created variable.
491 void init_interface_type(const struct glsl_type
*type
)
493 assert(this->interface_type
== NULL
);
494 this->interface_type
= type
;
495 if (this->is_interface_instance()) {
496 this->u
.max_ifc_array_access
=
497 ralloc_array(this, int, type
->length
);
498 for (unsigned i
= 0; i
< type
->length
; i
++) {
499 this->u
.max_ifc_array_access
[i
] = -1;
505 * Change this->interface_type on a variable that previously had a
506 * different, but compatible, interface_type. This is used during linking
507 * to set the size of arrays in interface blocks.
509 void change_interface_type(const struct glsl_type
*type
)
511 if (this->u
.max_ifc_array_access
!= NULL
) {
512 /* max_ifc_array_access has already been allocated, so make sure the
513 * new interface has the same number of fields as the old one.
515 assert(this->interface_type
->length
== type
->length
);
517 this->interface_type
= type
;
521 * Change this->interface_type on a variable that previously had a
522 * different, and incompatible, interface_type. This is used during
523 * compilation to handle redeclaration of the built-in gl_PerVertex
526 void reinit_interface_type(const struct glsl_type
*type
)
528 if (this->u
.max_ifc_array_access
!= NULL
) {
530 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
531 * it defines have been accessed yet; so it's safe to throw away the
532 * old max_ifc_array_access pointer, since all of its values are
535 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
536 assert(this->u
.max_ifc_array_access
[i
] == -1);
538 ralloc_free(this->u
.max_ifc_array_access
);
539 this->u
.max_ifc_array_access
= NULL
;
541 this->interface_type
= NULL
;
542 init_interface_type(type
);
545 const glsl_type
*get_interface_type() const
547 return this->interface_type
;
550 enum glsl_interface_packing
get_interface_type_packing() const
552 return this->interface_type
->get_interface_packing();
555 * Get the max_ifc_array_access pointer
557 * A "set" function is not needed because the array is dynmically allocated
560 inline int *get_max_ifc_array_access()
562 assert(this->data
._num_state_slots
== 0);
563 return this->u
.max_ifc_array_access
;
566 inline unsigned get_num_state_slots() const
568 assert(!this->is_interface_instance()
569 || this->data
._num_state_slots
== 0);
570 return this->data
._num_state_slots
;
573 inline void set_num_state_slots(unsigned n
)
575 assert(!this->is_interface_instance()
577 this->data
._num_state_slots
= n
;
580 inline ir_state_slot
*get_state_slots()
582 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
585 inline const ir_state_slot
*get_state_slots() const
587 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
590 inline ir_state_slot
*allocate_state_slots(unsigned n
)
592 assert(!this->is_interface_instance());
594 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
595 this->data
._num_state_slots
= 0;
597 if (this->u
.state_slots
!= NULL
)
598 this->data
._num_state_slots
= n
;
600 return this->u
.state_slots
;
603 inline bool is_interpolation_flat() const
605 return this->data
.interpolation
== INTERP_MODE_FLAT
||
606 this->type
->contains_integer() ||
607 this->type
->contains_double();
610 inline bool is_name_ralloced() const
612 return this->name
!= ir_variable::tmp_name
&&
613 this->name
!= this->name_storage
;
617 * Enable emitting extension warnings for this variable
619 void enable_extension_warning(const char *extension
);
622 * Get the extension warning string for this variable
624 * If warnings are not enabled, \c NULL is returned.
626 const char *get_extension_warning() const;
629 * Declared type of the variable
631 const struct glsl_type
*type
;
634 * Declared name of the variable
640 * If the name length fits into name_storage, it's used, otherwise
641 * the name is ralloc'd. shader-db mining showed that 70% of variables
642 * fit here. This is a win over ralloc where only ralloc_header has
643 * 20 bytes on 64-bit (28 bytes with DEBUG), and we can also skip malloc.
645 char name_storage
[16];
648 struct ir_variable_data
{
651 * Is the variable read-only?
653 * This is set for variables declared as \c const, shader inputs,
656 unsigned read_only
:1;
661 * Was an 'invariant' qualifier explicitly set in the shader?
663 * This is used to cross validate qualifiers.
665 unsigned explicit_invariant
:1;
667 * Is the variable invariant?
669 * It can happen either by having the 'invariant' qualifier
670 * explicitly set in the shader or by being used in calculations
671 * of other invariant variables.
673 unsigned invariant
:1;
677 * Has this variable been used for reading or writing?
679 * Several GLSL semantic checks require knowledge of whether or not a
680 * variable has been used. For example, it is an error to redeclare a
681 * variable as invariant after it has been used.
683 * This is maintained in the ast_to_hir.cpp path and during linking,
684 * but not in Mesa's fixed function or ARB program paths.
689 * Has this variable been statically assigned?
691 * This answers whether the variable was assigned in any path of
692 * the shader during ast_to_hir. This doesn't answer whether it is
693 * still written after dead code removal, nor is it maintained in
694 * non-ast_to_hir.cpp (GLSL parsing) paths.
699 * When separate shader programs are enabled, only input/outputs between
700 * the stages of a multi-stage separate program can be safely removed
701 * from the shader interface. Other input/outputs must remains active.
703 unsigned always_active_io
:1;
706 * Enum indicating how the variable was declared. See
707 * ir_var_declaration_type.
709 * This is used to detect certain kinds of illegal variable redeclarations.
711 unsigned how_declared
:2;
714 * Storage class of the variable.
716 * \sa ir_variable_mode
721 * Interpolation mode for shader inputs / outputs
723 * \sa glsl_interp_mode
725 unsigned interpolation
:2;
728 * Was the location explicitly set in the shader?
730 * If the location is explicitly set in the shader, it \b cannot be changed
731 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
734 unsigned explicit_location
:1;
735 unsigned explicit_index
:1;
738 * Was an initial binding explicitly set in the shader?
740 * If so, constant_value contains an integer ir_constant representing the
741 * initial binding point.
743 unsigned explicit_binding
:1;
746 * Was an initial component explicitly set in the shader?
748 unsigned explicit_component
:1;
751 * Does this variable have an initializer?
753 * This is used by the linker to cross-validiate initializers of global
756 unsigned has_initializer
:1;
759 * Is this variable a generic output or input that has not yet been matched
760 * up to a variable in another stage of the pipeline?
762 * This is used by the linker as scratch storage while assigning locations
763 * to generic inputs and outputs.
765 unsigned is_unmatched_generic_inout
:1;
768 * Is this varying used only by transform feedback?
770 * This is used by the linker to decide if its safe to pack the varying.
772 unsigned is_xfb_only
:1;
775 * Was a transform feedback buffer set in the shader?
777 unsigned explicit_xfb_buffer
:1;
780 * Was a transform feedback offset set in the shader?
782 unsigned explicit_xfb_offset
:1;
785 * Was a transform feedback stride set in the shader?
787 unsigned explicit_xfb_stride
:1;
790 * If non-zero, then this variable may be packed along with other variables
791 * into a single varying slot, so this offset should be applied when
792 * accessing components. For example, an offset of 1 means that the x
793 * component of this variable is actually stored in component y of the
794 * location specified by \c location.
796 unsigned location_frac
:2;
799 * Layout of the matrix. Uses glsl_matrix_layout values.
801 unsigned matrix_layout
:2;
804 * Non-zero if this variable was created by lowering a named interface
807 unsigned from_named_ifc_block
:1;
810 * Non-zero if the variable must be a shader input. This is useful for
811 * constraints on function parameters.
813 unsigned must_be_shader_input
:1;
816 * Output index for dual source blending.
819 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
825 * Precision qualifier.
827 * In desktop GLSL we do not care about precision qualifiers at all, in
828 * fact, the spec says that precision qualifiers are ignored.
830 * To make things easy, we make it so that this field is always
831 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
832 * have the same precision value and the checks we add in the compiler
833 * for this field will never break a desktop shader compile.
835 unsigned precision
:2;
838 * \brief Layout qualifier for gl_FragDepth.
840 * This is not equal to \c ir_depth_layout_none if and only if this
841 * variable is \c gl_FragDepth and a layout qualifier is specified.
843 ir_depth_layout depth_layout
:3;
848 unsigned memory_read_only
:1; /**< "readonly" qualifier. */
849 unsigned memory_write_only
:1; /**< "writeonly" qualifier. */
850 unsigned memory_coherent
:1;
851 unsigned memory_volatile
:1;
852 unsigned memory_restrict
:1;
855 * ARB_shader_storage_buffer_object
857 unsigned from_ssbo_unsized_array
:1; /**< unsized array buffer variable. */
859 unsigned implicit_sized_array
:1;
862 * Whether this is a fragment shader output implicitly initialized with
863 * the previous contents of the specified render target at the
864 * framebuffer location corresponding to this shader invocation.
866 unsigned fb_fetch_output
:1;
869 * Non-zero if this variable is considered bindless as defined by
870 * ARB_bindless_texture.
875 * Non-zero if this variable is considered bound as defined by
876 * ARB_bindless_texture.
881 * Emit a warning if this variable is accessed.
884 uint8_t warn_extension_index
;
887 /** Image internal format if specified explicitly, otherwise GL_NONE. */
888 uint16_t image_format
;
892 * Number of state slots used
895 * This could be stored in as few as 7-bits, if necessary. If it is made
896 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
899 uint16_t _num_state_slots
;
903 * Initial binding point for a sampler, atomic, or UBO.
905 * For array types, this represents the binding point for the first element.
910 * Storage location of the base of this variable
912 * The precise meaning of this field depends on the nature of the variable.
914 * - Vertex shader input: one of the values from \c gl_vert_attrib.
915 * - Vertex shader output: one of the values from \c gl_varying_slot.
916 * - Geometry shader input: one of the values from \c gl_varying_slot.
917 * - Geometry shader output: one of the values from \c gl_varying_slot.
918 * - Fragment shader input: one of the values from \c gl_varying_slot.
919 * - Fragment shader output: one of the values from \c gl_frag_result.
920 * - Uniforms: Per-stage uniform slot number for default uniform block.
921 * - Uniforms: Index within the uniform block definition for UBO members.
922 * - Non-UBO Uniforms: explicit location until linking then reused to
923 * store uniform slot number.
924 * - Other: This field is not currently used.
926 * If the variable is a uniform, shader input, or shader output, and the
927 * slot has not been assigned, the value will be -1.
932 * for glsl->tgsi/mesa IR we need to store the index into the
933 * parameters for uniforms, initially the code overloaded location
934 * but this causes problems with indirect samplers and AoA.
935 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
940 * Vertex stream output identifier.
942 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
943 * stream of the i-th component.
948 * Atomic, transform feedback or block member offset.
953 * Highest element accessed with a constant expression array index
955 * Not used for non-array variables. -1 is never accessed.
957 int max_array_access
;
960 * Transform feedback buffer.
965 * Transform feedback stride.
970 * Allow (only) ir_variable direct access private members.
972 friend class ir_variable
;
976 * Value assigned in the initializer of a variable declared "const"
978 ir_constant
*constant_value
;
981 * Constant expression assigned in the initializer of the variable
984 * This field and \c ::constant_value are distinct. Even if the two fields
985 * refer to constants with the same value, they must point to separate
988 ir_constant
*constant_initializer
;
991 static const char *const warn_extension_table
[];
995 * For variables which satisfy the is_interface_instance() predicate,
996 * this points to an array of integers such that if the ith member of
997 * the interface block is an array, max_ifc_array_access[i] is the
998 * maximum array element of that member that has been accessed. If the
999 * ith member of the interface block is not an array,
1000 * max_ifc_array_access[i] is unused.
1002 * For variables whose type is not an interface block, this pointer is
1005 int *max_ifc_array_access
;
1008 * Built-in state that backs this uniform
1010 * Once set at variable creation, \c state_slots must remain invariant.
1012 * If the variable is not a uniform, \c _num_state_slots will be zero
1013 * and \c state_slots will be \c NULL.
1015 ir_state_slot
*state_slots
;
1019 * For variables that are in an interface block or are an instance of an
1020 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
1022 * \sa ir_variable::location
1024 const glsl_type
*interface_type
;
1027 * Name used for anonymous compiler temporaries
1029 static const char tmp_name
[];
1033 * Should the construct keep names for ir_var_temporary variables?
1035 * When this global is false, names passed to the constructor for
1036 * \c ir_var_temporary variables will be dropped. Instead, the variable will
1037 * be named "compiler_temp". This name will be in static storage.
1040 * \b NEVER change the mode of an \c ir_var_temporary.
1043 * This variable is \b not thread-safe. It is global, \b not
1044 * per-context. It begins life false. A context can, at some point, make
1045 * it true. From that point on, it will be true forever. This should be
1046 * okay since it will only be set true while debugging.
1048 static bool temporaries_allocate_names
;
1052 * A function that returns whether a built-in function is available in the
1053 * current shading language (based on version, ES or desktop, and extensions).
1055 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
1057 #define MAKE_INTRINSIC_FOR_TYPE(op, t) \
1058 ir_intrinsic_generic_ ## op - ir_intrinsic_generic_load + ir_intrinsic_ ## t ## _ ## load
1060 #define MAP_INTRINSIC_TO_TYPE(i, t) \
1061 ir_intrinsic_id(int(i) - int(ir_intrinsic_generic_load) + int(ir_intrinsic_ ## t ## _ ## load))
1063 enum ir_intrinsic_id
{
1064 ir_intrinsic_invalid
= 0,
1067 * \name Generic intrinsics
1069 * Each of these intrinsics has a specific version for shared variables and
1073 ir_intrinsic_generic_load
,
1074 ir_intrinsic_generic_store
,
1075 ir_intrinsic_generic_atomic_add
,
1076 ir_intrinsic_generic_atomic_and
,
1077 ir_intrinsic_generic_atomic_or
,
1078 ir_intrinsic_generic_atomic_xor
,
1079 ir_intrinsic_generic_atomic_min
,
1080 ir_intrinsic_generic_atomic_max
,
1081 ir_intrinsic_generic_atomic_exchange
,
1082 ir_intrinsic_generic_atomic_comp_swap
,
1085 ir_intrinsic_atomic_counter_read
,
1086 ir_intrinsic_atomic_counter_increment
,
1087 ir_intrinsic_atomic_counter_predecrement
,
1088 ir_intrinsic_atomic_counter_add
,
1089 ir_intrinsic_atomic_counter_and
,
1090 ir_intrinsic_atomic_counter_or
,
1091 ir_intrinsic_atomic_counter_xor
,
1092 ir_intrinsic_atomic_counter_min
,
1093 ir_intrinsic_atomic_counter_max
,
1094 ir_intrinsic_atomic_counter_exchange
,
1095 ir_intrinsic_atomic_counter_comp_swap
,
1097 ir_intrinsic_image_load
,
1098 ir_intrinsic_image_store
,
1099 ir_intrinsic_image_atomic_add
,
1100 ir_intrinsic_image_atomic_and
,
1101 ir_intrinsic_image_atomic_or
,
1102 ir_intrinsic_image_atomic_xor
,
1103 ir_intrinsic_image_atomic_min
,
1104 ir_intrinsic_image_atomic_max
,
1105 ir_intrinsic_image_atomic_exchange
,
1106 ir_intrinsic_image_atomic_comp_swap
,
1107 ir_intrinsic_image_size
,
1108 ir_intrinsic_image_samples
,
1109 ir_intrinsic_image_atomic_inc_wrap
,
1110 ir_intrinsic_image_atomic_dec_wrap
,
1112 ir_intrinsic_ssbo_load
,
1113 ir_intrinsic_ssbo_store
= MAKE_INTRINSIC_FOR_TYPE(store
, ssbo
),
1114 ir_intrinsic_ssbo_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, ssbo
),
1115 ir_intrinsic_ssbo_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, ssbo
),
1116 ir_intrinsic_ssbo_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, ssbo
),
1117 ir_intrinsic_ssbo_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, ssbo
),
1118 ir_intrinsic_ssbo_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, ssbo
),
1119 ir_intrinsic_ssbo_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, ssbo
),
1120 ir_intrinsic_ssbo_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, ssbo
),
1121 ir_intrinsic_ssbo_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, ssbo
),
1123 ir_intrinsic_memory_barrier
,
1124 ir_intrinsic_shader_clock
,
1125 ir_intrinsic_group_memory_barrier
,
1126 ir_intrinsic_memory_barrier_atomic_counter
,
1127 ir_intrinsic_memory_barrier_buffer
,
1128 ir_intrinsic_memory_barrier_image
,
1129 ir_intrinsic_memory_barrier_shared
,
1130 ir_intrinsic_begin_invocation_interlock
,
1131 ir_intrinsic_end_invocation_interlock
,
1133 ir_intrinsic_vote_all
,
1134 ir_intrinsic_vote_any
,
1135 ir_intrinsic_vote_eq
,
1136 ir_intrinsic_ballot
,
1137 ir_intrinsic_read_invocation
,
1138 ir_intrinsic_read_first_invocation
,
1140 ir_intrinsic_shared_load
,
1141 ir_intrinsic_shared_store
= MAKE_INTRINSIC_FOR_TYPE(store
, shared
),
1142 ir_intrinsic_shared_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, shared
),
1143 ir_intrinsic_shared_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, shared
),
1144 ir_intrinsic_shared_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, shared
),
1145 ir_intrinsic_shared_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, shared
),
1146 ir_intrinsic_shared_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, shared
),
1147 ir_intrinsic_shared_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, shared
),
1148 ir_intrinsic_shared_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, shared
),
1149 ir_intrinsic_shared_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, shared
),
1154 * The representation of a function instance; may be the full definition or
1155 * simply a prototype.
1157 class ir_function_signature
: public ir_instruction
{
1158 /* An ir_function_signature will be part of the list of signatures in
1162 ir_function_signature(const glsl_type
*return_type
,
1163 builtin_available_predicate builtin_avail
= NULL
);
1165 virtual ir_function_signature
*clone(void *mem_ctx
,
1166 struct hash_table
*ht
) const;
1167 ir_function_signature
*clone_prototype(void *mem_ctx
,
1168 struct hash_table
*ht
) const;
1170 virtual void accept(ir_visitor
*v
)
1175 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1178 * Attempt to evaluate this function as a constant expression,
1179 * given a list of the actual parameters and the variable context.
1180 * Returns NULL for non-built-ins.
1182 ir_constant
*constant_expression_value(void *mem_ctx
,
1183 exec_list
*actual_parameters
,
1184 struct hash_table
*variable_context
);
1187 * Get the name of the function for which this is a signature
1189 const char *function_name() const;
1192 * Get a handle to the function for which this is a signature
1194 * There is no setter function, this function returns a \c const pointer,
1195 * and \c ir_function_signature::_function is private for a reason. The
1196 * only way to make a connection between a function and function signature
1197 * is via \c ir_function::add_signature. This helps ensure that certain
1198 * invariants (i.e., a function signature is in the list of signatures for
1199 * its \c _function) are met.
1201 * \sa ir_function::add_signature
1203 inline const class ir_function
*function() const
1205 return this->_function
;
1209 * Check whether the qualifiers match between this signature's parameters
1210 * and the supplied parameter list. If not, returns the name of the first
1211 * parameter with mismatched qualifiers (for use in error messages).
1213 const char *qualifiers_match(exec_list
*params
);
1216 * Replace the current parameter list with the given one. This is useful
1217 * if the current information came from a prototype, and either has invalid
1218 * or missing parameter names.
1220 void replace_parameters(exec_list
*new_params
);
1223 * Function return type.
1225 * \note This discards the optional precision qualifier.
1227 const struct glsl_type
*return_type
;
1230 * List of ir_variable of function parameters.
1232 * This represents the storage. The paramaters passed in a particular
1233 * call will be in ir_call::actual_paramaters.
1235 struct exec_list parameters
;
1237 /** Whether or not this function has a body (which may be empty). */
1238 unsigned is_defined
:1;
1240 /** Whether or not this function signature is a built-in. */
1241 bool is_builtin() const;
1244 * Whether or not this function is an intrinsic to be implemented
1247 inline bool is_intrinsic() const
1249 return intrinsic_id
!= ir_intrinsic_invalid
;
1252 /** Indentifier for this intrinsic. */
1253 enum ir_intrinsic_id intrinsic_id
;
1255 /** Whether or not a built-in is available for this shader. */
1256 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1258 /** Body of instructions in the function. */
1259 struct exec_list body
;
1263 * A function pointer to a predicate that answers whether a built-in
1264 * function is available in the current shader. NULL if not a built-in.
1266 builtin_available_predicate builtin_avail
;
1268 /** Function of which this signature is one overload. */
1269 class ir_function
*_function
;
1271 /** Function signature of which this one is a prototype clone */
1272 const ir_function_signature
*origin
;
1274 friend class ir_function
;
1277 * Helper function to run a list of instructions for constant
1278 * expression evaluation.
1280 * The hash table represents the values of the visible variables.
1281 * There are no scoping issues because the table is indexed on
1282 * ir_variable pointers, not variable names.
1284 * Returns false if the expression is not constant, true otherwise,
1285 * and the value in *result if result is non-NULL.
1287 bool constant_expression_evaluate_expression_list(void *mem_ctx
,
1288 const struct exec_list
&body
,
1289 struct hash_table
*variable_context
,
1290 ir_constant
**result
);
1295 * Header for tracking multiple overloaded functions with the same name.
1296 * Contains a list of ir_function_signatures representing each of the
1299 class ir_function
: public ir_instruction
{
1301 ir_function(const char *name
);
1303 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1305 virtual void accept(ir_visitor
*v
)
1310 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1312 void add_signature(ir_function_signature
*sig
)
1314 sig
->_function
= this;
1315 this->signatures
.push_tail(sig
);
1319 * Find a signature that matches a set of actual parameters, taking implicit
1320 * conversions into account. Also flags whether the match was exact.
1322 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1323 const exec_list
*actual_param
,
1324 bool allow_builtins
,
1325 bool *match_is_exact
);
1328 * Find a signature that matches a set of actual parameters, taking implicit
1329 * conversions into account.
1331 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1332 const exec_list
*actual_param
,
1333 bool allow_builtins
);
1336 * Find a signature that exactly matches a set of actual parameters without
1337 * any implicit type conversions.
1339 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1340 const exec_list
*actual_ps
);
1343 * Name of the function.
1347 /** Whether or not this function has a signature that isn't a built-in. */
1348 bool has_user_signature();
1351 * List of ir_function_signature for each overloaded function with this name.
1353 struct exec_list signatures
;
1356 * is this function a subroutine type declaration
1357 * e.g. subroutine void type1(float arg1);
1362 * is this function associated to a subroutine type
1363 * e.g. subroutine (type1, type2) function_name { function_body };
1364 * would have num_subroutine_types 2,
1365 * and pointers to the type1 and type2 types.
1367 int num_subroutine_types
;
1368 const struct glsl_type
**subroutine_types
;
1370 int subroutine_index
;
1373 inline const char *ir_function_signature::function_name() const
1375 return this->_function
->name
;
1381 * IR instruction representing high-level if-statements
1383 class ir_if
: public ir_instruction
{
1385 ir_if(ir_rvalue
*condition
)
1386 : ir_instruction(ir_type_if
), condition(condition
)
1390 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1392 virtual void accept(ir_visitor
*v
)
1397 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1399 ir_rvalue
*condition
;
1400 /** List of ir_instruction for the body of the then branch */
1401 exec_list then_instructions
;
1402 /** List of ir_instruction for the body of the else branch */
1403 exec_list else_instructions
;
1408 * IR instruction representing a high-level loop structure.
1410 class ir_loop
: public ir_instruction
{
1414 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1416 virtual void accept(ir_visitor
*v
)
1421 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1423 /** List of ir_instruction that make up the body of the loop. */
1424 exec_list body_instructions
;
1428 class ir_assignment
: public ir_instruction
{
1430 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1433 * Construct an assignment with an explicit write mask
1436 * Since a write mask is supplied, the LHS must already be a bare
1437 * \c ir_dereference. The cannot be any swizzles in the LHS.
1439 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1440 unsigned write_mask
);
1442 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1444 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1445 struct hash_table
*variable_context
= NULL
);
1447 virtual void accept(ir_visitor
*v
)
1452 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1455 * Get a whole variable written by an assignment
1457 * If the LHS of the assignment writes a whole variable, the variable is
1458 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1461 * - Assigning to a scalar
1462 * - Assigning to all components of a vector
1463 * - Whole array (or matrix) assignment
1464 * - Whole structure assignment
1466 ir_variable
*whole_variable_written();
1469 * Set the LHS of an assignment
1471 void set_lhs(ir_rvalue
*lhs
);
1474 * Left-hand side of the assignment.
1476 * This should be treated as read only. If you need to set the LHS of an
1477 * assignment, use \c ir_assignment::set_lhs.
1479 ir_dereference
*lhs
;
1482 * Value being assigned
1487 * Optional condition for the assignment.
1489 ir_rvalue
*condition
;
1493 * Component mask written
1495 * For non-vector types in the LHS, this field will be zero. For vector
1496 * types, a bit will be set for each component that is written. Note that
1497 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1499 * A partially-set write mask means that each enabled channel gets
1500 * the value from a consecutive channel of the rhs. For example,
1501 * to write just .xyw of gl_FrontColor with color:
1503 * (assign (constant bool (1)) (xyw)
1504 * (var_ref gl_FragColor)
1505 * (swiz xyw (var_ref color)))
1507 unsigned write_mask
:4;
1510 #include "ir_expression_operation.h"
1512 extern const char *const ir_expression_operation_strings
[ir_last_opcode
+ 1];
1513 extern const char *const ir_expression_operation_enum_strings
[ir_last_opcode
+ 1];
1515 class ir_expression
: public ir_rvalue
{
1517 ir_expression(int op
, const struct glsl_type
*type
,
1518 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1519 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1522 * Constructor for unary operation expressions
1524 ir_expression(int op
, ir_rvalue
*);
1527 * Constructor for binary operation expressions
1529 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1532 * Constructor for ternary operation expressions
1534 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1536 virtual bool equals(const ir_instruction
*ir
,
1537 enum ir_node_type ignore
= ir_type_unset
) const;
1539 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1542 * Attempt to constant-fold the expression
1544 * The "variable_context" hash table links ir_variable * to ir_constant *
1545 * that represent the variables' values. \c NULL represents an empty
1548 * If the expression cannot be constant folded, this method will return
1551 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1552 struct hash_table
*variable_context
= NULL
);
1555 * This is only here for ir_reader to used for testing purposes please use
1556 * the precomputed num_operands field if you need the number of operands.
1558 static unsigned get_num_operands(ir_expression_operation
);
1561 * Return whether the expression operates on vectors horizontally.
1563 bool is_horizontal() const
1565 return operation
== ir_binop_all_equal
||
1566 operation
== ir_binop_any_nequal
||
1567 operation
== ir_binop_dot
||
1568 operation
== ir_binop_vector_extract
||
1569 operation
== ir_triop_vector_insert
||
1570 operation
== ir_binop_ubo_load
||
1571 operation
== ir_quadop_vector
;
1575 * Do a reverse-lookup to translate the given string into an operator.
1577 static ir_expression_operation
get_operator(const char *);
1579 virtual void accept(ir_visitor
*v
)
1584 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1586 virtual ir_variable
*variable_referenced() const;
1589 * Determine the number of operands used by an expression
1591 void init_num_operands()
1593 if (operation
== ir_quadop_vector
) {
1594 num_operands
= this->type
->vector_elements
;
1596 num_operands
= get_num_operands(operation
);
1600 ir_expression_operation operation
;
1601 ir_rvalue
*operands
[4];
1602 uint8_t num_operands
;
1607 * HIR instruction representing a high-level function call, containing a list
1608 * of parameters and returning a value in the supplied temporary.
1610 class ir_call
: public ir_instruction
{
1612 ir_call(ir_function_signature
*callee
,
1613 ir_dereference_variable
*return_deref
,
1614 exec_list
*actual_parameters
)
1615 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(NULL
), array_idx(NULL
)
1617 assert(callee
->return_type
!= NULL
);
1618 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1621 ir_call(ir_function_signature
*callee
,
1622 ir_dereference_variable
*return_deref
,
1623 exec_list
*actual_parameters
,
1624 ir_variable
*var
, ir_rvalue
*array_idx
)
1625 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(var
), array_idx(array_idx
)
1627 assert(callee
->return_type
!= NULL
);
1628 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1631 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1633 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1634 struct hash_table
*variable_context
= NULL
);
1636 virtual void accept(ir_visitor
*v
)
1641 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1644 * Get the name of the function being called.
1646 const char *callee_name() const
1648 return callee
->function_name();
1652 * Generates an inline version of the function before @ir,
1653 * storing the return value in return_deref.
1655 void generate_inline(ir_instruction
*ir
);
1658 * Storage for the function's return value.
1659 * This must be NULL if the return type is void.
1661 ir_dereference_variable
*return_deref
;
1664 * The specific function signature being called.
1666 ir_function_signature
*callee
;
1668 /* List of ir_rvalue of paramaters passed in this call. */
1669 exec_list actual_parameters
;
1672 * ARB_shader_subroutine support -
1673 * the subroutine uniform variable and array index
1674 * rvalue to be used in the lowering pass later.
1676 ir_variable
*sub_var
;
1677 ir_rvalue
*array_idx
;
1682 * \name Jump-like IR instructions.
1684 * These include \c break, \c continue, \c return, and \c discard.
1687 class ir_jump
: public ir_instruction
{
1689 ir_jump(enum ir_node_type t
)
1695 class ir_return
: public ir_jump
{
1698 : ir_jump(ir_type_return
), value(NULL
)
1702 ir_return(ir_rvalue
*value
)
1703 : ir_jump(ir_type_return
), value(value
)
1707 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1709 ir_rvalue
*get_value() const
1714 virtual void accept(ir_visitor
*v
)
1719 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1726 * Jump instructions used inside loops
1728 * These include \c break and \c continue. The \c break within a loop is
1729 * different from the \c break within a switch-statement.
1731 * \sa ir_switch_jump
1733 class ir_loop_jump
: public ir_jump
{
1740 ir_loop_jump(jump_mode mode
)
1741 : ir_jump(ir_type_loop_jump
)
1746 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1748 virtual void accept(ir_visitor
*v
)
1753 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1755 bool is_break() const
1757 return mode
== jump_break
;
1760 bool is_continue() const
1762 return mode
== jump_continue
;
1765 /** Mode selector for the jump instruction. */
1766 enum jump_mode mode
;
1770 * IR instruction representing discard statements.
1772 class ir_discard
: public ir_jump
{
1775 : ir_jump(ir_type_discard
)
1777 this->condition
= NULL
;
1780 ir_discard(ir_rvalue
*cond
)
1781 : ir_jump(ir_type_discard
)
1783 this->condition
= cond
;
1786 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1788 virtual void accept(ir_visitor
*v
)
1793 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1795 ir_rvalue
*condition
;
1801 * Texture sampling opcodes used in ir_texture
1803 enum ir_texture_opcode
{
1804 ir_tex
, /**< Regular texture look-up */
1805 ir_txb
, /**< Texture look-up with LOD bias */
1806 ir_txl
, /**< Texture look-up with explicit LOD */
1807 ir_txd
, /**< Texture look-up with partial derivatvies */
1808 ir_txf
, /**< Texel fetch with explicit LOD */
1809 ir_txf_ms
, /**< Multisample texture fetch */
1810 ir_txs
, /**< Texture size */
1811 ir_lod
, /**< Texture lod query */
1812 ir_tg4
, /**< Texture gather */
1813 ir_query_levels
, /**< Texture levels query */
1814 ir_texture_samples
, /**< Texture samples query */
1815 ir_samples_identical
, /**< Query whether all samples are definitely identical. */
1820 * IR instruction to sample a texture
1822 * The specific form of the IR instruction depends on the \c mode value
1823 * selected from \c ir_texture_opcodes. In the printed IR, these will
1826 * Texel offset (0 or an expression)
1827 * | Projection divisor
1828 * | | Shadow comparator
1831 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1832 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1833 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1834 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1835 * (txf <type> <sampler> <coordinate> 0 <lod>)
1837 * <type> <sampler> <coordinate> <sample_index>)
1838 * (txs <type> <sampler> <lod>)
1839 * (lod <type> <sampler> <coordinate>)
1840 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1841 * (query_levels <type> <sampler>)
1842 * (samples_identical <sampler> <coordinate>)
1844 class ir_texture
: public ir_rvalue
{
1846 ir_texture(enum ir_texture_opcode op
)
1847 : ir_rvalue(ir_type_texture
),
1848 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1849 shadow_comparator(NULL
), offset(NULL
)
1851 memset(&lod_info
, 0, sizeof(lod_info
));
1854 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1856 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1857 struct hash_table
*variable_context
= NULL
);
1859 virtual void accept(ir_visitor
*v
)
1864 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1866 virtual bool equals(const ir_instruction
*ir
,
1867 enum ir_node_type ignore
= ir_type_unset
) const;
1870 * Return a string representing the ir_texture_opcode.
1872 const char *opcode_string();
1874 /** Set the sampler and type. */
1875 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1878 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1880 static ir_texture_opcode
get_opcode(const char *);
1882 enum ir_texture_opcode op
;
1884 /** Sampler to use for the texture access. */
1885 ir_dereference
*sampler
;
1887 /** Texture coordinate to sample */
1888 ir_rvalue
*coordinate
;
1891 * Value used for projective divide.
1893 * If there is no projective divide (the common case), this will be
1894 * \c NULL. Optimization passes should check for this to point to a constant
1895 * of 1.0 and replace that with \c NULL.
1897 ir_rvalue
*projector
;
1900 * Coordinate used for comparison on shadow look-ups.
1902 * If there is no shadow comparison, this will be \c NULL. For the
1903 * \c ir_txf opcode, this *must* be \c NULL.
1905 ir_rvalue
*shadow_comparator
;
1907 /** Texel offset. */
1911 ir_rvalue
*lod
; /**< Floating point LOD */
1912 ir_rvalue
*bias
; /**< Floating point LOD bias */
1913 ir_rvalue
*sample_index
; /**< MSAA sample index */
1914 ir_rvalue
*component
; /**< Gather component selector */
1916 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1917 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1923 struct ir_swizzle_mask
{
1930 * Number of components in the swizzle.
1932 unsigned num_components
:3;
1935 * Does the swizzle contain duplicate components?
1937 * L-value swizzles cannot contain duplicate components.
1939 unsigned has_duplicates
:1;
1943 class ir_swizzle
: public ir_rvalue
{
1945 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
1948 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
1950 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
1952 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
1954 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1955 struct hash_table
*variable_context
= NULL
);
1958 * Construct an ir_swizzle from the textual representation. Can fail.
1960 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
1962 virtual void accept(ir_visitor
*v
)
1967 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1969 virtual bool equals(const ir_instruction
*ir
,
1970 enum ir_node_type ignore
= ir_type_unset
) const;
1972 bool is_lvalue(const struct _mesa_glsl_parse_state
*state
) const
1974 return val
->is_lvalue(state
) && !mask
.has_duplicates
;
1978 * Get the variable that is ultimately referenced by an r-value
1980 virtual ir_variable
*variable_referenced() const;
1983 ir_swizzle_mask mask
;
1987 * Initialize the mask component of a swizzle
1989 * This is used by the \c ir_swizzle constructors.
1991 void init_mask(const unsigned *components
, unsigned count
);
1995 class ir_dereference
: public ir_rvalue
{
1997 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
1999 bool is_lvalue(const struct _mesa_glsl_parse_state
*state
) const;
2002 * Get the variable that is ultimately referenced by an r-value
2004 virtual ir_variable
*variable_referenced() const = 0;
2007 ir_dereference(enum ir_node_type t
)
2014 class ir_dereference_variable
: public ir_dereference
{
2016 ir_dereference_variable(ir_variable
*var
);
2018 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2019 struct hash_table
*) const;
2021 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2022 struct hash_table
*variable_context
= NULL
);
2024 virtual bool equals(const ir_instruction
*ir
,
2025 enum ir_node_type ignore
= ir_type_unset
) const;
2028 * Get the variable that is ultimately referenced by an r-value
2030 virtual ir_variable
*variable_referenced() const
2035 virtual ir_variable
*whole_variable_referenced()
2037 /* ir_dereference_variable objects always dereference the entire
2038 * variable. However, if this dereference is dereferenced by anything
2039 * else, the complete deferefernce chain is not a whole-variable
2040 * dereference. This method should only be called on the top most
2041 * ir_rvalue in a dereference chain.
2046 virtual void accept(ir_visitor
*v
)
2051 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2054 * Object being dereferenced.
2060 class ir_dereference_array
: public ir_dereference
{
2062 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2064 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2066 virtual ir_dereference_array
*clone(void *mem_ctx
,
2067 struct hash_table
*) const;
2069 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2070 struct hash_table
*variable_context
= NULL
);
2072 virtual bool equals(const ir_instruction
*ir
,
2073 enum ir_node_type ignore
= ir_type_unset
) const;
2076 * Get the variable that is ultimately referenced by an r-value
2078 virtual ir_variable
*variable_referenced() const
2080 return this->array
->variable_referenced();
2083 virtual void accept(ir_visitor
*v
)
2088 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2091 ir_rvalue
*array_index
;
2094 void set_array(ir_rvalue
*value
);
2098 class ir_dereference_record
: public ir_dereference
{
2100 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2102 ir_dereference_record(ir_variable
*var
, const char *field
);
2104 virtual ir_dereference_record
*clone(void *mem_ctx
,
2105 struct hash_table
*) const;
2107 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2108 struct hash_table
*variable_context
= NULL
);
2111 * Get the variable that is ultimately referenced by an r-value
2113 virtual ir_variable
*variable_referenced() const
2115 return this->record
->variable_referenced();
2118 virtual void accept(ir_visitor
*v
)
2123 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2131 * Data stored in an ir_constant
2133 union ir_constant_data
{
2144 class ir_constant
: public ir_rvalue
{
2146 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2147 ir_constant(bool b
, unsigned vector_elements
=1);
2148 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2149 ir_constant(int i
, unsigned vector_elements
=1);
2150 ir_constant(float f
, unsigned vector_elements
=1);
2151 ir_constant(double d
, unsigned vector_elements
=1);
2152 ir_constant(uint64_t u64
, unsigned vector_elements
=1);
2153 ir_constant(int64_t i64
, unsigned vector_elements
=1);
2156 * Construct an ir_constant from a list of ir_constant values
2158 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2161 * Construct an ir_constant from a scalar component of another ir_constant
2163 * The new \c ir_constant inherits the type of the component from the
2167 * In the case of a matrix constant, the new constant is a scalar, \b not
2170 ir_constant(const ir_constant
*c
, unsigned i
);
2173 * Return a new ir_constant of the specified type containing all zeros.
2175 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2177 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2179 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2180 struct hash_table
*variable_context
= NULL
);
2182 virtual void accept(ir_visitor
*v
)
2187 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2189 virtual bool equals(const ir_instruction
*ir
,
2190 enum ir_node_type ignore
= ir_type_unset
) const;
2193 * Get a particular component of a constant as a specific type
2195 * This is useful, for example, to get a value from an integer constant
2196 * as a float or bool. This appears frequently when constructors are
2197 * called with all constant parameters.
2200 bool get_bool_component(unsigned i
) const;
2201 float get_float_component(unsigned i
) const;
2202 double get_double_component(unsigned i
) const;
2203 int get_int_component(unsigned i
) const;
2204 unsigned get_uint_component(unsigned i
) const;
2205 int64_t get_int64_component(unsigned i
) const;
2206 uint64_t get_uint64_component(unsigned i
) const;
2209 ir_constant
*get_array_element(unsigned i
) const;
2211 ir_constant
*get_record_field(int idx
);
2214 * Copy the values on another constant at a given offset.
2216 * The offset is ignored for array or struct copies, it's only for
2217 * scalars or vectors into vectors or matrices.
2219 * With identical types on both sides and zero offset it's clone()
2220 * without creating a new object.
2223 void copy_offset(ir_constant
*src
, int offset
);
2226 * Copy the values on another constant at a given offset and
2227 * following an assign-like mask.
2229 * The mask is ignored for scalars.
2231 * Note that this function only handles what assign can handle,
2232 * i.e. at most a vector as source and a column of a matrix as
2236 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2239 * Determine whether a constant has the same value as another constant
2241 * \sa ir_constant::is_zero, ir_constant::is_one,
2242 * ir_constant::is_negative_one
2244 bool has_value(const ir_constant
*) const;
2247 * Return true if this ir_constant represents the given value.
2249 * For vectors, this checks that each component is the given value.
2251 virtual bool is_value(float f
, int i
) const;
2252 virtual bool is_zero() const;
2253 virtual bool is_one() const;
2254 virtual bool is_negative_one() const;
2257 * Return true for constants that could be stored as 16-bit unsigned values.
2259 * Note that this will return true even for signed integer ir_constants, as
2260 * long as the value is non-negative and fits in 16-bits.
2262 virtual bool is_uint16_constant() const;
2265 * Value of the constant.
2267 * The field used to back the values supplied by the constant is determined
2268 * by the type associated with the \c ir_instruction. Constants may be
2269 * scalars, vectors, or matrices.
2271 union ir_constant_data value
;
2273 /* Array elements and structure fields */
2274 ir_constant
**const_elements
;
2278 * Parameterless constructor only used by the clone method
2284 * IR instruction to emit a vertex in a geometry shader.
2286 class ir_emit_vertex
: public ir_instruction
{
2288 ir_emit_vertex(ir_rvalue
*stream
)
2289 : ir_instruction(ir_type_emit_vertex
),
2295 virtual void accept(ir_visitor
*v
)
2300 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2302 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2305 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2307 int stream_id() const
2309 return stream
->as_constant()->value
.i
[0];
2316 * IR instruction to complete the current primitive and start a new one in a
2319 class ir_end_primitive
: public ir_instruction
{
2321 ir_end_primitive(ir_rvalue
*stream
)
2322 : ir_instruction(ir_type_end_primitive
),
2328 virtual void accept(ir_visitor
*v
)
2333 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2335 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2338 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2340 int stream_id() const
2342 return stream
->as_constant()->value
.i
[0];
2349 * IR instruction for tessellation control and compute shader barrier.
2351 class ir_barrier
: public ir_instruction
{
2354 : ir_instruction(ir_type_barrier
)
2358 virtual void accept(ir_visitor
*v
)
2363 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2365 return new(mem_ctx
) ir_barrier();
2368 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2374 * Apply a visitor to each IR node in a list
2377 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2380 * Validate invariants on each IR node in a list
2382 void validate_ir_tree(exec_list
*instructions
);
2384 struct _mesa_glsl_parse_state
;
2385 struct gl_shader_program
;
2388 * Detect whether an unlinked shader contains static recursion
2390 * If the list of instructions is determined to contain static recursion,
2391 * \c _mesa_glsl_error will be called to emit error messages for each function
2392 * that is in the recursion cycle.
2395 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2396 exec_list
*instructions
);
2399 * Detect whether a linked shader contains static recursion
2401 * If the list of instructions is determined to contain static recursion,
2402 * \c link_error_printf will be called to emit error messages for each function
2403 * that is in the recursion cycle. In addition,
2404 * \c gl_shader_program::LinkStatus will be set to false.
2407 detect_recursion_linked(struct gl_shader_program
*prog
,
2408 exec_list
*instructions
);
2411 * Make a clone of each IR instruction in a list
2413 * \param in List of IR instructions that are to be cloned
2414 * \param out List to hold the cloned instructions
2417 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2420 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2421 struct _mesa_glsl_parse_state
*state
);
2424 reparent_ir(exec_list
*list
, void *mem_ctx
);
2427 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2428 gl_shader_stage shader_stage
);
2431 prototype_string(const glsl_type
*return_type
, const char *name
,
2432 exec_list
*parameters
);
2435 mode_string(const ir_variable
*var
);
2438 * Built-in / reserved GL variables names start with "gl_"
2441 is_gl_identifier(const char *s
)
2443 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2447 #endif /* __cplusplus */
2449 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2450 struct _mesa_glsl_parse_state
*state
);
2453 fprint_ir(FILE *f
, const void *instruction
);
2455 extern const struct gl_builtin_uniform_desc
*
2456 _mesa_glsl_get_builtin_uniform_desc(const char *name
);
2463 vertices_per_prim(GLenum prim
);