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;
836 * Whether this is a fragment shader output implicitly initialized with
837 * the previous contents of the specified render target at the
838 * framebuffer location corresponding to this shader invocation.
840 unsigned fb_fetch_output
:1;
843 * Emit a warning if this variable is accessed.
846 uint8_t warn_extension_index
;
849 /** Image internal format if specified explicitly, otherwise GL_NONE. */
850 uint16_t image_format
;
854 * Number of state slots used
857 * This could be stored in as few as 7-bits, if necessary. If it is made
858 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
861 uint16_t _num_state_slots
;
865 * Initial binding point for a sampler, atomic, or UBO.
867 * For array types, this represents the binding point for the first element.
872 * Storage location of the base of this variable
874 * The precise meaning of this field depends on the nature of the variable.
876 * - Vertex shader input: one of the values from \c gl_vert_attrib.
877 * - Vertex shader output: one of the values from \c gl_varying_slot.
878 * - Geometry shader input: one of the values from \c gl_varying_slot.
879 * - Geometry shader output: one of the values from \c gl_varying_slot.
880 * - Fragment shader input: one of the values from \c gl_varying_slot.
881 * - Fragment shader output: one of the values from \c gl_frag_result.
882 * - Uniforms: Per-stage uniform slot number for default uniform block.
883 * - Uniforms: Index within the uniform block definition for UBO members.
884 * - Non-UBO Uniforms: explicit location until linking then reused to
885 * store uniform slot number.
886 * - Other: This field is not currently used.
888 * If the variable is a uniform, shader input, or shader output, and the
889 * slot has not been assigned, the value will be -1.
894 * for glsl->tgsi/mesa IR we need to store the index into the
895 * parameters for uniforms, initially the code overloaded location
896 * but this causes problems with indirect samplers and AoA.
897 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
902 * Vertex stream output identifier.
907 * Atomic, transform feedback or block member offset.
912 * Highest element accessed with a constant expression array index
914 * Not used for non-array variables. -1 is never accessed.
916 int max_array_access
;
919 * Transform feedback buffer.
924 * Transform feedback stride.
929 * Allow (only) ir_variable direct access private members.
931 friend class ir_variable
;
935 * Value assigned in the initializer of a variable declared "const"
937 ir_constant
*constant_value
;
940 * Constant expression assigned in the initializer of the variable
943 * This field and \c ::constant_value are distinct. Even if the two fields
944 * refer to constants with the same value, they must point to separate
947 ir_constant
*constant_initializer
;
950 static const char *const warn_extension_table
[];
954 * For variables which satisfy the is_interface_instance() predicate,
955 * this points to an array of integers such that if the ith member of
956 * the interface block is an array, max_ifc_array_access[i] is the
957 * maximum array element of that member that has been accessed. If the
958 * ith member of the interface block is not an array,
959 * max_ifc_array_access[i] is unused.
961 * For variables whose type is not an interface block, this pointer is
964 int *max_ifc_array_access
;
967 * Built-in state that backs this uniform
969 * Once set at variable creation, \c state_slots must remain invariant.
971 * If the variable is not a uniform, \c _num_state_slots will be zero
972 * and \c state_slots will be \c NULL.
974 ir_state_slot
*state_slots
;
978 * For variables that are in an interface block or are an instance of an
979 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
981 * \sa ir_variable::location
983 const glsl_type
*interface_type
;
986 * Name used for anonymous compiler temporaries
988 static const char tmp_name
[];
992 * Should the construct keep names for ir_var_temporary variables?
994 * When this global is false, names passed to the constructor for
995 * \c ir_var_temporary variables will be dropped. Instead, the variable will
996 * be named "compiler_temp". This name will be in static storage.
999 * \b NEVER change the mode of an \c ir_var_temporary.
1002 * This variable is \b not thread-safe. It is global, \b not
1003 * per-context. It begins life false. A context can, at some point, make
1004 * it true. From that point on, it will be true forever. This should be
1005 * okay since it will only be set true while debugging.
1007 static bool temporaries_allocate_names
;
1011 * A function that returns whether a built-in function is available in the
1012 * current shading language (based on version, ES or desktop, and extensions).
1014 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
1016 #define MAKE_INTRINSIC_FOR_TYPE(op, t) \
1017 ir_intrinsic_generic_ ## op - ir_intrinsic_generic_load + ir_intrinsic_ ## t ## _ ## load
1019 #define MAP_INTRINSIC_TO_TYPE(i, t) \
1020 ir_intrinsic_id(int(i) - int(ir_intrinsic_generic_load) + int(ir_intrinsic_ ## t ## _ ## load))
1022 enum ir_intrinsic_id
{
1023 ir_intrinsic_invalid
= 0,
1026 * \name Generic intrinsics
1028 * Each of these intrinsics has a specific version for shared variables and
1032 ir_intrinsic_generic_load
,
1033 ir_intrinsic_generic_store
,
1034 ir_intrinsic_generic_atomic_add
,
1035 ir_intrinsic_generic_atomic_and
,
1036 ir_intrinsic_generic_atomic_or
,
1037 ir_intrinsic_generic_atomic_xor
,
1038 ir_intrinsic_generic_atomic_min
,
1039 ir_intrinsic_generic_atomic_max
,
1040 ir_intrinsic_generic_atomic_exchange
,
1041 ir_intrinsic_generic_atomic_comp_swap
,
1044 ir_intrinsic_atomic_counter_read
,
1045 ir_intrinsic_atomic_counter_increment
,
1046 ir_intrinsic_atomic_counter_predecrement
,
1047 ir_intrinsic_atomic_counter_add
,
1048 ir_intrinsic_atomic_counter_sub
,
1049 ir_intrinsic_atomic_counter_and
,
1050 ir_intrinsic_atomic_counter_or
,
1051 ir_intrinsic_atomic_counter_xor
,
1052 ir_intrinsic_atomic_counter_min
,
1053 ir_intrinsic_atomic_counter_max
,
1054 ir_intrinsic_atomic_counter_exchange
,
1055 ir_intrinsic_atomic_counter_comp_swap
,
1057 ir_intrinsic_image_load
,
1058 ir_intrinsic_image_store
,
1059 ir_intrinsic_image_atomic_add
,
1060 ir_intrinsic_image_atomic_and
,
1061 ir_intrinsic_image_atomic_or
,
1062 ir_intrinsic_image_atomic_xor
,
1063 ir_intrinsic_image_atomic_min
,
1064 ir_intrinsic_image_atomic_max
,
1065 ir_intrinsic_image_atomic_exchange
,
1066 ir_intrinsic_image_atomic_comp_swap
,
1067 ir_intrinsic_image_size
,
1068 ir_intrinsic_image_samples
,
1070 ir_intrinsic_ssbo_load
,
1071 ir_intrinsic_ssbo_store
= MAKE_INTRINSIC_FOR_TYPE(store
, ssbo
),
1072 ir_intrinsic_ssbo_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, ssbo
),
1073 ir_intrinsic_ssbo_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, ssbo
),
1074 ir_intrinsic_ssbo_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, ssbo
),
1075 ir_intrinsic_ssbo_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, ssbo
),
1076 ir_intrinsic_ssbo_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, ssbo
),
1077 ir_intrinsic_ssbo_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, ssbo
),
1078 ir_intrinsic_ssbo_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, ssbo
),
1079 ir_intrinsic_ssbo_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, ssbo
),
1081 ir_intrinsic_memory_barrier
,
1082 ir_intrinsic_shader_clock
,
1083 ir_intrinsic_group_memory_barrier
,
1084 ir_intrinsic_memory_barrier_atomic_counter
,
1085 ir_intrinsic_memory_barrier_buffer
,
1086 ir_intrinsic_memory_barrier_image
,
1087 ir_intrinsic_memory_barrier_shared
,
1089 ir_intrinsic_shared_load
,
1090 ir_intrinsic_shared_store
= MAKE_INTRINSIC_FOR_TYPE(store
, shared
),
1091 ir_intrinsic_shared_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, shared
),
1092 ir_intrinsic_shared_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, shared
),
1093 ir_intrinsic_shared_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, shared
),
1094 ir_intrinsic_shared_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, shared
),
1095 ir_intrinsic_shared_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, shared
),
1096 ir_intrinsic_shared_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, shared
),
1097 ir_intrinsic_shared_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, shared
),
1098 ir_intrinsic_shared_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, shared
),
1103 * The representation of a function instance; may be the full definition or
1104 * simply a prototype.
1106 class ir_function_signature
: public ir_instruction
{
1107 /* An ir_function_signature will be part of the list of signatures in
1111 ir_function_signature(const glsl_type
*return_type
,
1112 builtin_available_predicate builtin_avail
= NULL
);
1114 virtual ir_function_signature
*clone(void *mem_ctx
,
1115 struct hash_table
*ht
) const;
1116 ir_function_signature
*clone_prototype(void *mem_ctx
,
1117 struct hash_table
*ht
) const;
1119 virtual void accept(ir_visitor
*v
)
1124 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1127 * Attempt to evaluate this function as a constant expression,
1128 * given a list of the actual parameters and the variable context.
1129 * Returns NULL for non-built-ins.
1131 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
1134 * Get the name of the function for which this is a signature
1136 const char *function_name() const;
1139 * Get a handle to the function for which this is a signature
1141 * There is no setter function, this function returns a \c const pointer,
1142 * and \c ir_function_signature::_function is private for a reason. The
1143 * only way to make a connection between a function and function signature
1144 * is via \c ir_function::add_signature. This helps ensure that certain
1145 * invariants (i.e., a function signature is in the list of signatures for
1146 * its \c _function) are met.
1148 * \sa ir_function::add_signature
1150 inline const class ir_function
*function() const
1152 return this->_function
;
1156 * Check whether the qualifiers match between this signature's parameters
1157 * and the supplied parameter list. If not, returns the name of the first
1158 * parameter with mismatched qualifiers (for use in error messages).
1160 const char *qualifiers_match(exec_list
*params
);
1163 * Replace the current parameter list with the given one. This is useful
1164 * if the current information came from a prototype, and either has invalid
1165 * or missing parameter names.
1167 void replace_parameters(exec_list
*new_params
);
1170 * Function return type.
1172 * \note This discards the optional precision qualifier.
1174 const struct glsl_type
*return_type
;
1177 * List of ir_variable of function parameters.
1179 * This represents the storage. The paramaters passed in a particular
1180 * call will be in ir_call::actual_paramaters.
1182 struct exec_list parameters
;
1184 /** Whether or not this function has a body (which may be empty). */
1185 unsigned is_defined
:1;
1187 /** Whether or not this function signature is a built-in. */
1188 bool is_builtin() const;
1191 * Whether or not this function is an intrinsic to be implemented
1196 /** Indentifier for this intrinsic. */
1197 enum ir_intrinsic_id intrinsic_id
;
1199 /** Whether or not a built-in is available for this shader. */
1200 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1202 /** Body of instructions in the function. */
1203 struct exec_list body
;
1207 * A function pointer to a predicate that answers whether a built-in
1208 * function is available in the current shader. NULL if not a built-in.
1210 builtin_available_predicate builtin_avail
;
1212 /** Function of which this signature is one overload. */
1213 class ir_function
*_function
;
1215 /** Function signature of which this one is a prototype clone */
1216 const ir_function_signature
*origin
;
1218 friend class ir_function
;
1221 * Helper function to run a list of instructions for constant
1222 * expression evaluation.
1224 * The hash table represents the values of the visible variables.
1225 * There are no scoping issues because the table is indexed on
1226 * ir_variable pointers, not variable names.
1228 * Returns false if the expression is not constant, true otherwise,
1229 * and the value in *result if result is non-NULL.
1231 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1232 struct hash_table
*variable_context
,
1233 ir_constant
**result
);
1238 * Header for tracking multiple overloaded functions with the same name.
1239 * Contains a list of ir_function_signatures representing each of the
1242 class ir_function
: public ir_instruction
{
1244 ir_function(const char *name
);
1246 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1248 virtual void accept(ir_visitor
*v
)
1253 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1255 void add_signature(ir_function_signature
*sig
)
1257 sig
->_function
= this;
1258 this->signatures
.push_tail(sig
);
1262 * Find a signature that matches a set of actual parameters, taking implicit
1263 * conversions into account. Also flags whether the match was exact.
1265 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1266 const exec_list
*actual_param
,
1267 bool allow_builtins
,
1268 bool *match_is_exact
);
1271 * Find a signature that matches a set of actual parameters, taking implicit
1272 * conversions into account.
1274 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1275 const exec_list
*actual_param
,
1276 bool allow_builtins
);
1279 * Find a signature that exactly matches a set of actual parameters without
1280 * any implicit type conversions.
1282 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1283 const exec_list
*actual_ps
);
1286 * Name of the function.
1290 /** Whether or not this function has a signature that isn't a built-in. */
1291 bool has_user_signature();
1294 * List of ir_function_signature for each overloaded function with this name.
1296 struct exec_list signatures
;
1299 * is this function a subroutine type declaration
1300 * e.g. subroutine void type1(float arg1);
1305 * is this function associated to a subroutine type
1306 * e.g. subroutine (type1, type2) function_name { function_body };
1307 * would have num_subroutine_types 2,
1308 * and pointers to the type1 and type2 types.
1310 int num_subroutine_types
;
1311 const struct glsl_type
**subroutine_types
;
1313 int subroutine_index
;
1316 inline const char *ir_function_signature::function_name() const
1318 return this->_function
->name
;
1324 * IR instruction representing high-level if-statements
1326 class ir_if
: public ir_instruction
{
1328 ir_if(ir_rvalue
*condition
)
1329 : ir_instruction(ir_type_if
), condition(condition
)
1333 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1335 virtual void accept(ir_visitor
*v
)
1340 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1342 ir_rvalue
*condition
;
1343 /** List of ir_instruction for the body of the then branch */
1344 exec_list then_instructions
;
1345 /** List of ir_instruction for the body of the else branch */
1346 exec_list else_instructions
;
1351 * IR instruction representing a high-level loop structure.
1353 class ir_loop
: public ir_instruction
{
1357 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1359 virtual void accept(ir_visitor
*v
)
1364 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1366 /** List of ir_instruction that make up the body of the loop. */
1367 exec_list body_instructions
;
1371 class ir_assignment
: public ir_instruction
{
1373 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1376 * Construct an assignment with an explicit write mask
1379 * Since a write mask is supplied, the LHS must already be a bare
1380 * \c ir_dereference. The cannot be any swizzles in the LHS.
1382 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1383 unsigned write_mask
);
1385 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1387 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1389 virtual void accept(ir_visitor
*v
)
1394 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1397 * Get a whole variable written by an assignment
1399 * If the LHS of the assignment writes a whole variable, the variable is
1400 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1403 * - Assigning to a scalar
1404 * - Assigning to all components of a vector
1405 * - Whole array (or matrix) assignment
1406 * - Whole structure assignment
1408 ir_variable
*whole_variable_written();
1411 * Set the LHS of an assignment
1413 void set_lhs(ir_rvalue
*lhs
);
1416 * Left-hand side of the assignment.
1418 * This should be treated as read only. If you need to set the LHS of an
1419 * assignment, use \c ir_assignment::set_lhs.
1421 ir_dereference
*lhs
;
1424 * Value being assigned
1429 * Optional condition for the assignment.
1431 ir_rvalue
*condition
;
1435 * Component mask written
1437 * For non-vector types in the LHS, this field will be zero. For vector
1438 * types, a bit will be set for each component that is written. Note that
1439 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1441 * A partially-set write mask means that each enabled channel gets
1442 * the value from a consecutive channel of the rhs. For example,
1443 * to write just .xyw of gl_FrontColor with color:
1445 * (assign (constant bool (1)) (xyw)
1446 * (var_ref gl_FragColor)
1447 * (swiz xyw (var_ref color)))
1449 unsigned write_mask
:4;
1452 #include "ir_expression_operation.h"
1454 extern const char *const ir_expression_operation_strings
[ir_last_opcode
+ 1];
1456 class ir_expression
: public ir_rvalue
{
1458 ir_expression(int op
, const struct glsl_type
*type
,
1459 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1460 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1463 * Constructor for unary operation expressions
1465 ir_expression(int op
, ir_rvalue
*);
1468 * Constructor for binary operation expressions
1470 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1473 * Constructor for ternary operation expressions
1475 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1477 virtual bool equals(const ir_instruction
*ir
,
1478 enum ir_node_type ignore
= ir_type_unset
) const;
1480 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1483 * Attempt to constant-fold the expression
1485 * The "variable_context" hash table links ir_variable * to ir_constant *
1486 * that represent the variables' values. \c NULL represents an empty
1489 * If the expression cannot be constant folded, this method will return
1492 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1495 * Determine the number of operands used by an expression
1497 static unsigned int get_num_operands(ir_expression_operation
);
1500 * Determine the number of operands used by an expression
1502 unsigned int get_num_operands() const
1504 return (this->operation
== ir_quadop_vector
)
1505 ? this->type
->vector_elements
: get_num_operands(operation
);
1509 * Return whether the expression operates on vectors horizontally.
1511 bool is_horizontal() const
1513 return operation
== ir_binop_all_equal
||
1514 operation
== ir_binop_any_nequal
||
1515 operation
== ir_binop_dot
||
1516 operation
== ir_binop_vector_extract
||
1517 operation
== ir_triop_vector_insert
||
1518 operation
== ir_binop_ubo_load
||
1519 operation
== ir_quadop_vector
;
1523 * Do a reverse-lookup to translate the given string into an operator.
1525 static ir_expression_operation
get_operator(const char *);
1527 virtual void accept(ir_visitor
*v
)
1532 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1534 virtual ir_variable
*variable_referenced() const;
1536 ir_expression_operation operation
;
1537 ir_rvalue
*operands
[4];
1542 * HIR instruction representing a high-level function call, containing a list
1543 * of parameters and returning a value in the supplied temporary.
1545 class ir_call
: public ir_instruction
{
1547 ir_call(ir_function_signature
*callee
,
1548 ir_dereference_variable
*return_deref
,
1549 exec_list
*actual_parameters
)
1550 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(NULL
), array_idx(NULL
)
1552 assert(callee
->return_type
!= NULL
);
1553 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1554 this->use_builtin
= callee
->is_builtin();
1557 ir_call(ir_function_signature
*callee
,
1558 ir_dereference_variable
*return_deref
,
1559 exec_list
*actual_parameters
,
1560 ir_variable
*var
, ir_rvalue
*array_idx
)
1561 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(var
), array_idx(array_idx
)
1563 assert(callee
->return_type
!= NULL
);
1564 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1565 this->use_builtin
= callee
->is_builtin();
1568 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1570 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1572 virtual void accept(ir_visitor
*v
)
1577 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1580 * Get the name of the function being called.
1582 const char *callee_name() const
1584 return callee
->function_name();
1588 * Generates an inline version of the function before @ir,
1589 * storing the return value in return_deref.
1591 void generate_inline(ir_instruction
*ir
);
1594 * Storage for the function's return value.
1595 * This must be NULL if the return type is void.
1597 ir_dereference_variable
*return_deref
;
1600 * The specific function signature being called.
1602 ir_function_signature
*callee
;
1604 /* List of ir_rvalue of paramaters passed in this call. */
1605 exec_list actual_parameters
;
1607 /** Should this call only bind to a built-in function? */
1611 * ARB_shader_subroutine support -
1612 * the subroutine uniform variable and array index
1613 * rvalue to be used in the lowering pass later.
1615 ir_variable
*sub_var
;
1616 ir_rvalue
*array_idx
;
1621 * \name Jump-like IR instructions.
1623 * These include \c break, \c continue, \c return, and \c discard.
1626 class ir_jump
: public ir_instruction
{
1628 ir_jump(enum ir_node_type t
)
1634 class ir_return
: public ir_jump
{
1637 : ir_jump(ir_type_return
), value(NULL
)
1641 ir_return(ir_rvalue
*value
)
1642 : ir_jump(ir_type_return
), value(value
)
1646 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1648 ir_rvalue
*get_value() const
1653 virtual void accept(ir_visitor
*v
)
1658 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1665 * Jump instructions used inside loops
1667 * These include \c break and \c continue. The \c break within a loop is
1668 * different from the \c break within a switch-statement.
1670 * \sa ir_switch_jump
1672 class ir_loop_jump
: public ir_jump
{
1679 ir_loop_jump(jump_mode mode
)
1680 : ir_jump(ir_type_loop_jump
)
1685 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1687 virtual void accept(ir_visitor
*v
)
1692 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1694 bool is_break() const
1696 return mode
== jump_break
;
1699 bool is_continue() const
1701 return mode
== jump_continue
;
1704 /** Mode selector for the jump instruction. */
1705 enum jump_mode mode
;
1709 * IR instruction representing discard statements.
1711 class ir_discard
: public ir_jump
{
1714 : ir_jump(ir_type_discard
)
1716 this->condition
= NULL
;
1719 ir_discard(ir_rvalue
*cond
)
1720 : ir_jump(ir_type_discard
)
1722 this->condition
= cond
;
1725 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1727 virtual void accept(ir_visitor
*v
)
1732 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1734 ir_rvalue
*condition
;
1740 * Texture sampling opcodes used in ir_texture
1742 enum ir_texture_opcode
{
1743 ir_tex
, /**< Regular texture look-up */
1744 ir_txb
, /**< Texture look-up with LOD bias */
1745 ir_txl
, /**< Texture look-up with explicit LOD */
1746 ir_txd
, /**< Texture look-up with partial derivatvies */
1747 ir_txf
, /**< Texel fetch with explicit LOD */
1748 ir_txf_ms
, /**< Multisample texture fetch */
1749 ir_txs
, /**< Texture size */
1750 ir_lod
, /**< Texture lod query */
1751 ir_tg4
, /**< Texture gather */
1752 ir_query_levels
, /**< Texture levels query */
1753 ir_texture_samples
, /**< Texture samples query */
1754 ir_samples_identical
, /**< Query whether all samples are definitely identical. */
1759 * IR instruction to sample a texture
1761 * The specific form of the IR instruction depends on the \c mode value
1762 * selected from \c ir_texture_opcodes. In the printed IR, these will
1765 * Texel offset (0 or an expression)
1766 * | Projection divisor
1767 * | | Shadow comparitor
1770 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1771 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1772 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1773 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1774 * (txf <type> <sampler> <coordinate> 0 <lod>)
1776 * <type> <sampler> <coordinate> <sample_index>)
1777 * (txs <type> <sampler> <lod>)
1778 * (lod <type> <sampler> <coordinate>)
1779 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1780 * (query_levels <type> <sampler>)
1781 * (samples_identical <sampler> <coordinate>)
1783 class ir_texture
: public ir_rvalue
{
1785 ir_texture(enum ir_texture_opcode op
)
1786 : ir_rvalue(ir_type_texture
),
1787 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1788 shadow_comparitor(NULL
), offset(NULL
)
1790 memset(&lod_info
, 0, sizeof(lod_info
));
1793 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1795 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1797 virtual void accept(ir_visitor
*v
)
1802 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1804 virtual bool equals(const ir_instruction
*ir
,
1805 enum ir_node_type ignore
= ir_type_unset
) const;
1808 * Return a string representing the ir_texture_opcode.
1810 const char *opcode_string();
1812 /** Set the sampler and type. */
1813 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1816 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1818 static ir_texture_opcode
get_opcode(const char *);
1820 enum ir_texture_opcode op
;
1822 /** Sampler to use for the texture access. */
1823 ir_dereference
*sampler
;
1825 /** Texture coordinate to sample */
1826 ir_rvalue
*coordinate
;
1829 * Value used for projective divide.
1831 * If there is no projective divide (the common case), this will be
1832 * \c NULL. Optimization passes should check for this to point to a constant
1833 * of 1.0 and replace that with \c NULL.
1835 ir_rvalue
*projector
;
1838 * Coordinate used for comparison on shadow look-ups.
1840 * If there is no shadow comparison, this will be \c NULL. For the
1841 * \c ir_txf opcode, this *must* be \c NULL.
1843 ir_rvalue
*shadow_comparitor
;
1845 /** Texel offset. */
1849 ir_rvalue
*lod
; /**< Floating point LOD */
1850 ir_rvalue
*bias
; /**< Floating point LOD bias */
1851 ir_rvalue
*sample_index
; /**< MSAA sample index */
1852 ir_rvalue
*component
; /**< Gather component selector */
1854 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1855 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1861 struct ir_swizzle_mask
{
1868 * Number of components in the swizzle.
1870 unsigned num_components
:3;
1873 * Does the swizzle contain duplicate components?
1875 * L-value swizzles cannot contain duplicate components.
1877 unsigned has_duplicates
:1;
1881 class ir_swizzle
: public ir_rvalue
{
1883 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
1886 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
1888 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
1890 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
1892 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1895 * Construct an ir_swizzle from the textual representation. Can fail.
1897 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
1899 virtual void accept(ir_visitor
*v
)
1904 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1906 virtual bool equals(const ir_instruction
*ir
,
1907 enum ir_node_type ignore
= ir_type_unset
) const;
1909 bool is_lvalue() const
1911 return val
->is_lvalue() && !mask
.has_duplicates
;
1915 * Get the variable that is ultimately referenced by an r-value
1917 virtual ir_variable
*variable_referenced() const;
1920 ir_swizzle_mask mask
;
1924 * Initialize the mask component of a swizzle
1926 * This is used by the \c ir_swizzle constructors.
1928 void init_mask(const unsigned *components
, unsigned count
);
1932 class ir_dereference
: public ir_rvalue
{
1934 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
1936 bool is_lvalue() const;
1939 * Get the variable that is ultimately referenced by an r-value
1941 virtual ir_variable
*variable_referenced() const = 0;
1944 ir_dereference(enum ir_node_type t
)
1951 class ir_dereference_variable
: public ir_dereference
{
1953 ir_dereference_variable(ir_variable
*var
);
1955 virtual ir_dereference_variable
*clone(void *mem_ctx
,
1956 struct hash_table
*) const;
1958 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1960 virtual bool equals(const ir_instruction
*ir
,
1961 enum ir_node_type ignore
= ir_type_unset
) const;
1964 * Get the variable that is ultimately referenced by an r-value
1966 virtual ir_variable
*variable_referenced() const
1971 virtual ir_variable
*whole_variable_referenced()
1973 /* ir_dereference_variable objects always dereference the entire
1974 * variable. However, if this dereference is dereferenced by anything
1975 * else, the complete deferefernce chain is not a whole-variable
1976 * dereference. This method should only be called on the top most
1977 * ir_rvalue in a dereference chain.
1982 virtual void accept(ir_visitor
*v
)
1987 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1990 * Object being dereferenced.
1996 class ir_dereference_array
: public ir_dereference
{
1998 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2000 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2002 virtual ir_dereference_array
*clone(void *mem_ctx
,
2003 struct hash_table
*) const;
2005 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2007 virtual bool equals(const ir_instruction
*ir
,
2008 enum ir_node_type ignore
= ir_type_unset
) const;
2011 * Get the variable that is ultimately referenced by an r-value
2013 virtual ir_variable
*variable_referenced() const
2015 return this->array
->variable_referenced();
2018 virtual void accept(ir_visitor
*v
)
2023 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2026 ir_rvalue
*array_index
;
2029 void set_array(ir_rvalue
*value
);
2033 class ir_dereference_record
: public ir_dereference
{
2035 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2037 ir_dereference_record(ir_variable
*var
, const char *field
);
2039 virtual ir_dereference_record
*clone(void *mem_ctx
,
2040 struct hash_table
*) const;
2042 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2045 * Get the variable that is ultimately referenced by an r-value
2047 virtual ir_variable
*variable_referenced() const
2049 return this->record
->variable_referenced();
2052 virtual void accept(ir_visitor
*v
)
2057 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2065 * Data stored in an ir_constant
2067 union ir_constant_data
{
2076 class ir_constant
: public ir_rvalue
{
2078 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2079 ir_constant(bool b
, unsigned vector_elements
=1);
2080 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2081 ir_constant(int i
, unsigned vector_elements
=1);
2082 ir_constant(float f
, unsigned vector_elements
=1);
2083 ir_constant(double d
, unsigned vector_elements
=1);
2086 * Construct an ir_constant from a list of ir_constant values
2088 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2091 * Construct an ir_constant from a scalar component of another ir_constant
2093 * The new \c ir_constant inherits the type of the component from the
2097 * In the case of a matrix constant, the new constant is a scalar, \b not
2100 ir_constant(const ir_constant
*c
, unsigned i
);
2103 * Return a new ir_constant of the specified type containing all zeros.
2105 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2107 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2109 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2111 virtual void accept(ir_visitor
*v
)
2116 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2118 virtual bool equals(const ir_instruction
*ir
,
2119 enum ir_node_type ignore
= ir_type_unset
) const;
2122 * Get a particular component of a constant as a specific type
2124 * This is useful, for example, to get a value from an integer constant
2125 * as a float or bool. This appears frequently when constructors are
2126 * called with all constant parameters.
2129 bool get_bool_component(unsigned i
) const;
2130 float get_float_component(unsigned i
) const;
2131 double get_double_component(unsigned i
) const;
2132 int get_int_component(unsigned i
) const;
2133 unsigned get_uint_component(unsigned i
) const;
2136 ir_constant
*get_array_element(unsigned i
) const;
2138 ir_constant
*get_record_field(const char *name
);
2141 * Copy the values on another constant at a given offset.
2143 * The offset is ignored for array or struct copies, it's only for
2144 * scalars or vectors into vectors or matrices.
2146 * With identical types on both sides and zero offset it's clone()
2147 * without creating a new object.
2150 void copy_offset(ir_constant
*src
, int offset
);
2153 * Copy the values on another constant at a given offset and
2154 * following an assign-like mask.
2156 * The mask is ignored for scalars.
2158 * Note that this function only handles what assign can handle,
2159 * i.e. at most a vector as source and a column of a matrix as
2163 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2166 * Determine whether a constant has the same value as another constant
2168 * \sa ir_constant::is_zero, ir_constant::is_one,
2169 * ir_constant::is_negative_one
2171 bool has_value(const ir_constant
*) const;
2174 * Return true if this ir_constant represents the given value.
2176 * For vectors, this checks that each component is the given value.
2178 virtual bool is_value(float f
, int i
) const;
2179 virtual bool is_zero() const;
2180 virtual bool is_one() const;
2181 virtual bool is_negative_one() const;
2184 * Return true for constants that could be stored as 16-bit unsigned values.
2186 * Note that this will return true even for signed integer ir_constants, as
2187 * long as the value is non-negative and fits in 16-bits.
2189 virtual bool is_uint16_constant() const;
2192 * Value of the constant.
2194 * The field used to back the values supplied by the constant is determined
2195 * by the type associated with the \c ir_instruction. Constants may be
2196 * scalars, vectors, or matrices.
2198 union ir_constant_data value
;
2200 /* Array elements */
2201 ir_constant
**array_elements
;
2203 /* Structure fields */
2204 exec_list components
;
2208 * Parameterless constructor only used by the clone method
2214 * IR instruction to emit a vertex in a geometry shader.
2216 class ir_emit_vertex
: public ir_instruction
{
2218 ir_emit_vertex(ir_rvalue
*stream
)
2219 : ir_instruction(ir_type_emit_vertex
),
2225 virtual void accept(ir_visitor
*v
)
2230 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2232 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2235 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2237 int stream_id() const
2239 return stream
->as_constant()->value
.i
[0];
2246 * IR instruction to complete the current primitive and start a new one in a
2249 class ir_end_primitive
: public ir_instruction
{
2251 ir_end_primitive(ir_rvalue
*stream
)
2252 : ir_instruction(ir_type_end_primitive
),
2258 virtual void accept(ir_visitor
*v
)
2263 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2265 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2268 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2270 int stream_id() const
2272 return stream
->as_constant()->value
.i
[0];
2279 * IR instruction for tessellation control and compute shader barrier.
2281 class ir_barrier
: public ir_instruction
{
2284 : ir_instruction(ir_type_barrier
)
2288 virtual void accept(ir_visitor
*v
)
2293 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2295 return new(mem_ctx
) ir_barrier();
2298 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2304 * Apply a visitor to each IR node in a list
2307 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2310 * Validate invariants on each IR node in a list
2312 void validate_ir_tree(exec_list
*instructions
);
2314 struct _mesa_glsl_parse_state
;
2315 struct gl_shader_program
;
2318 * Detect whether an unlinked shader contains static recursion
2320 * If the list of instructions is determined to contain static recursion,
2321 * \c _mesa_glsl_error will be called to emit error messages for each function
2322 * that is in the recursion cycle.
2325 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2326 exec_list
*instructions
);
2329 * Detect whether a linked shader contains static recursion
2331 * If the list of instructions is determined to contain static recursion,
2332 * \c link_error_printf will be called to emit error messages for each function
2333 * that is in the recursion cycle. In addition,
2334 * \c gl_shader_program::LinkStatus will be set to false.
2337 detect_recursion_linked(struct gl_shader_program
*prog
,
2338 exec_list
*instructions
);
2341 * Make a clone of each IR instruction in a list
2343 * \param in List of IR instructions that are to be cloned
2344 * \param out List to hold the cloned instructions
2347 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2350 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2351 struct _mesa_glsl_parse_state
*state
);
2354 _mesa_glsl_initialize_derived_variables(struct gl_context
*ctx
,
2358 _mesa_glsl_initialize_functions(_mesa_glsl_parse_state
*state
);
2361 _mesa_glsl_initialize_builtin_functions();
2363 extern ir_function_signature
*
2364 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2365 const char *name
, exec_list
*actual_parameters
);
2367 extern ir_function
*
2368 _mesa_glsl_find_builtin_function_by_name(const char *name
);
2371 _mesa_glsl_get_builtin_function_shader(void);
2373 extern ir_function_signature
*
2374 _mesa_get_main_function_signature(glsl_symbol_table
*symbols
);
2377 _mesa_glsl_release_functions(void);
2380 _mesa_glsl_release_builtin_functions(void);
2383 reparent_ir(exec_list
*list
, void *mem_ctx
);
2385 struct glsl_symbol_table
;
2388 import_prototypes(const exec_list
*source
, exec_list
*dest
,
2389 struct glsl_symbol_table
*symbols
, void *mem_ctx
);
2392 ir_has_call(ir_instruction
*ir
);
2395 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2396 gl_shader_stage shader_stage
);
2399 prototype_string(const glsl_type
*return_type
, const char *name
,
2400 exec_list
*parameters
);
2403 mode_string(const ir_variable
*var
);
2406 * Built-in / reserved GL variables names start with "gl_"
2409 is_gl_identifier(const char *s
)
2411 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2415 #endif /* __cplusplus */
2417 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2418 struct _mesa_glsl_parse_state
*state
);
2421 fprint_ir(FILE *f
, const void *instruction
);
2423 extern const struct gl_builtin_uniform_desc
*
2424 _mesa_glsl_get_builtin_uniform_desc(const char *name
);
2431 vertices_per_prim(GLenum prim
);