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"
36 #include "main/mtypes.h"
41 * \defgroup IR Intermediate representation nodes
49 * Each concrete class derived from \c ir_instruction has a value in this
50 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
51 * by the constructor. While using type tags is not very C++, it is extremely
52 * convenient. For example, during debugging you can simply inspect
53 * \c ir_instruction::ir_type to find out the actual type of the object.
55 * In addition, it is possible to use a switch-statement based on \c
56 * \c ir_instruction::ir_type to select different behavior for different object
57 * types. For functions that have only slight differences for several object
58 * types, this allows writing very straightforward, readable code.
61 ir_type_dereference_array
,
62 ir_type_dereference_record
,
63 ir_type_dereference_variable
,
72 ir_type_function_signature
,
79 ir_type_end_primitive
,
81 ir_type_max
, /**< maximum ir_type enum number, for validation */
82 ir_type_unset
= ir_type_max
87 * Base class of all IR instructions
89 class ir_instruction
: public exec_node
{
91 enum ir_node_type ir_type
;
94 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
95 * there's a virtual destructor present. Because we almost
96 * universally use ralloc for our memory management of
97 * ir_instructions, the destructor doesn't need to do any work.
99 virtual ~ir_instruction()
103 /** ir_print_visitor helper for debugging. */
104 void print(void) const;
105 void fprint(FILE *f
) const;
107 virtual void accept(ir_visitor
*) = 0;
108 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*) = 0;
109 virtual ir_instruction
*clone(void *mem_ctx
,
110 struct hash_table
*ht
) const = 0;
112 bool is_rvalue() const
114 return ir_type
== ir_type_dereference_array
||
115 ir_type
== ir_type_dereference_record
||
116 ir_type
== ir_type_dereference_variable
||
117 ir_type
== ir_type_constant
||
118 ir_type
== ir_type_expression
||
119 ir_type
== ir_type_swizzle
||
120 ir_type
== ir_type_texture
;
123 bool is_dereference() const
125 return ir_type
== ir_type_dereference_array
||
126 ir_type
== ir_type_dereference_record
||
127 ir_type
== ir_type_dereference_variable
;
132 return ir_type
== ir_type_loop_jump
||
133 ir_type
== ir_type_return
||
134 ir_type
== ir_type_discard
;
138 * \name IR instruction downcast functions
140 * These functions either cast the object to a derived class or return
141 * \c NULL if the object's type does not match the specified derived class.
142 * Additional downcast functions will be added as needed.
145 #define AS_BASE(TYPE) \
146 class ir_##TYPE *as_##TYPE() \
148 assume(this != NULL); \
149 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
151 const class ir_##TYPE *as_##TYPE() const \
153 assume(this != NULL); \
154 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
162 #define AS_CHILD(TYPE) \
163 class ir_##TYPE * as_##TYPE() \
165 assume(this != NULL); \
166 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
168 const class ir_##TYPE * as_##TYPE() const \
170 assume(this != NULL); \
171 return ir_type == ir_type_##TYPE ? (const ir_##TYPE *) this : NULL; \
175 AS_CHILD(dereference_array
)
176 AS_CHILD(dereference_variable
)
177 AS_CHILD(dereference_record
)
192 * IR equality method: Return true if the referenced instruction would
193 * return the same value as this one.
195 * This intended to be used for CSE and algebraic optimizations, on rvalues
196 * in particular. No support for other instruction types (assignments,
197 * jumps, calls, etc.) is planned.
199 virtual bool equals(const ir_instruction
*ir
,
200 enum ir_node_type ignore
= ir_type_unset
) const;
203 ir_instruction(enum ir_node_type t
)
211 assert(!"Should not get here.");
217 * The base class for all "values"/expression trees.
219 class ir_rvalue
: public ir_instruction
{
221 const struct glsl_type
*type
;
223 virtual ir_rvalue
*clone(void *mem_ctx
, struct hash_table
*) const;
225 virtual void accept(ir_visitor
*v
)
230 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
232 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
234 ir_rvalue
*as_rvalue_to_saturate();
236 virtual bool is_lvalue() 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
{
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 * \cglsl_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 * Set this->interface_type on a newly created variable.
480 void init_interface_type(const struct glsl_type
*type
)
482 assert(this->interface_type
== NULL
);
483 this->interface_type
= type
;
484 if (this->is_interface_instance()) {
485 this->u
.max_ifc_array_access
=
486 ralloc_array(this, int, type
->length
);
487 for (unsigned i
= 0; i
< type
->length
; i
++) {
488 this->u
.max_ifc_array_access
[i
] = -1;
494 * Change this->interface_type on a variable that previously had a
495 * different, but compatible, interface_type. This is used during linking
496 * to set the size of arrays in interface blocks.
498 void change_interface_type(const struct glsl_type
*type
)
500 if (this->u
.max_ifc_array_access
!= NULL
) {
501 /* max_ifc_array_access has already been allocated, so make sure the
502 * new interface has the same number of fields as the old one.
504 assert(this->interface_type
->length
== type
->length
);
506 this->interface_type
= type
;
510 * Change this->interface_type on a variable that previously had a
511 * different, and incompatible, interface_type. This is used during
512 * compilation to handle redeclaration of the built-in gl_PerVertex
515 void reinit_interface_type(const struct glsl_type
*type
)
517 if (this->u
.max_ifc_array_access
!= NULL
) {
519 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
520 * it defines have been accessed yet; so it's safe to throw away the
521 * old max_ifc_array_access pointer, since all of its values are
524 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
525 assert(this->u
.max_ifc_array_access
[i
] == -1);
527 ralloc_free(this->u
.max_ifc_array_access
);
528 this->u
.max_ifc_array_access
= NULL
;
530 this->interface_type
= NULL
;
531 init_interface_type(type
);
534 const glsl_type
*get_interface_type() const
536 return this->interface_type
;
539 enum glsl_interface_packing
get_interface_type_packing() const
541 return this->interface_type
->get_interface_packing();
544 * Get the max_ifc_array_access pointer
546 * A "set" function is not needed because the array is dynmically allocated
549 inline int *get_max_ifc_array_access()
551 assert(this->data
._num_state_slots
== 0);
552 return this->u
.max_ifc_array_access
;
555 inline unsigned get_num_state_slots() const
557 assert(!this->is_interface_instance()
558 || this->data
._num_state_slots
== 0);
559 return this->data
._num_state_slots
;
562 inline void set_num_state_slots(unsigned n
)
564 assert(!this->is_interface_instance()
566 this->data
._num_state_slots
= n
;
569 inline ir_state_slot
*get_state_slots()
571 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
574 inline const ir_state_slot
*get_state_slots() const
576 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
579 inline ir_state_slot
*allocate_state_slots(unsigned n
)
581 assert(!this->is_interface_instance());
583 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
584 this->data
._num_state_slots
= 0;
586 if (this->u
.state_slots
!= NULL
)
587 this->data
._num_state_slots
= n
;
589 return this->u
.state_slots
;
592 inline bool is_interpolation_flat() const
594 return this->data
.interpolation
== INTERP_MODE_FLAT
||
595 this->type
->contains_integer() ||
596 this->type
->contains_double();
599 inline bool is_name_ralloced() const
601 return this->name
!= ir_variable::tmp_name
&&
602 this->name
!= this->name_storage
;
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
629 * If the name length fits into name_storage, it's used, otherwise
630 * the name is ralloc'd. shader-db mining showed that 70% of variables
631 * fit here. This is a win over ralloc where only ralloc_header has
632 * 20 bytes on 64-bit (28 bytes with DEBUG), and we can also skip malloc.
634 char name_storage
[16];
637 struct ir_variable_data
{
640 * Is the variable read-only?
642 * This is set for variables declared as \c const, shader inputs,
645 unsigned read_only
:1;
649 unsigned invariant
:1;
653 * Has this variable been used for reading or writing?
655 * Several GLSL semantic checks require knowledge of whether or not a
656 * variable has been used. For example, it is an error to redeclare a
657 * variable as invariant after it has been used.
659 * This is only maintained in the ast_to_hir.cpp path, not in
660 * Mesa's fixed function or ARB program paths.
665 * Has this variable been statically assigned?
667 * This answers whether the variable was assigned in any path of
668 * the shader during ast_to_hir. This doesn't answer whether it is
669 * still written after dead code removal, nor is it maintained in
670 * non-ast_to_hir.cpp (GLSL parsing) paths.
675 * When separate shader programs are enabled, only input/outputs between
676 * the stages of a multi-stage separate program can be safely removed
677 * from the shader interface. Other input/outputs must remains active.
679 unsigned always_active_io
:1;
682 * Enum indicating how the variable was declared. See
683 * ir_var_declaration_type.
685 * This is used to detect certain kinds of illegal variable redeclarations.
687 unsigned how_declared
:2;
690 * Storage class of the variable.
692 * \sa ir_variable_mode
697 * Interpolation mode for shader inputs / outputs
699 * \sa glsl_interp_mode
701 unsigned interpolation
:2;
704 * \name ARB_fragment_coord_conventions
707 unsigned origin_upper_left
:1;
708 unsigned pixel_center_integer
:1;
712 * Was the location explicitly set in the shader?
714 * If the location is explicitly set in the shader, it \b cannot be changed
715 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
718 unsigned explicit_location
:1;
719 unsigned explicit_index
:1;
722 * Was an initial binding explicitly set in the shader?
724 * If so, constant_value contains an integer ir_constant representing the
725 * initial binding point.
727 unsigned explicit_binding
:1;
730 * Was an initial component explicitly set in the shader?
732 unsigned explicit_component
:1;
735 * Does this variable have an initializer?
737 * This is used by the linker to cross-validiate initializers of global
740 unsigned has_initializer
:1;
743 * Is this variable a generic output or input that has not yet been matched
744 * up to a variable in another stage of the pipeline?
746 * This is used by the linker as scratch storage while assigning locations
747 * to generic inputs and outputs.
749 unsigned is_unmatched_generic_inout
:1;
752 * Is this varying used only by transform feedback?
754 * This is used by the linker to decide if its safe to pack the varying.
756 unsigned is_xfb_only
:1;
759 * Was a transfor feedback buffer set in the shader?
761 unsigned explicit_xfb_buffer
:1;
764 * Was a transfor feedback offset set in the shader?
766 unsigned explicit_xfb_offset
:1;
769 * Was a transfor feedback stride set in the shader?
771 unsigned explicit_xfb_stride
:1;
774 * If non-zero, then this variable may be packed along with other variables
775 * into a single varying slot, so this offset should be applied when
776 * accessing components. For example, an offset of 1 means that the x
777 * component of this variable is actually stored in component y of the
778 * location specified by \c location.
780 unsigned location_frac
:2;
783 * Layout of the matrix. Uses glsl_matrix_layout values.
785 unsigned matrix_layout
:2;
788 * Non-zero if this variable was created by lowering a named interface
791 unsigned from_named_ifc_block
:1;
794 * Non-zero if the variable must be a shader input. This is useful for
795 * constraints on function parameters.
797 unsigned must_be_shader_input
:1;
800 * Output index for dual source blending.
803 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
809 * Precision qualifier.
811 * In desktop GLSL we do not care about precision qualifiers at all, in
812 * fact, the spec says that precision qualifiers are ignored.
814 * To make things easy, we make it so that this field is always
815 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
816 * have the same precision value and the checks we add in the compiler
817 * for this field will never break a desktop shader compile.
819 unsigned precision
:2;
822 * \brief Layout qualifier for gl_FragDepth.
824 * This is not equal to \c ir_depth_layout_none if and only if this
825 * variable is \c gl_FragDepth and a layout qualifier is specified.
827 ir_depth_layout depth_layout
:3;
832 unsigned memory_read_only
:1; /**< "readonly" qualifier. */
833 unsigned memory_write_only
:1; /**< "writeonly" qualifier. */
834 unsigned memory_coherent
:1;
835 unsigned memory_volatile
:1;
836 unsigned memory_restrict
:1;
839 * ARB_shader_storage_buffer_object
841 unsigned from_ssbo_unsized_array
:1; /**< unsized array buffer variable. */
843 unsigned implicit_sized_array
:1;
846 * Whether this is a fragment shader output implicitly initialized with
847 * the previous contents of the specified render target at the
848 * framebuffer location corresponding to this shader invocation.
850 unsigned fb_fetch_output
:1;
853 * Non-zero if this variable is considered bindless as defined by
854 * ARB_bindless_texture.
859 * Non-zero if this variable is considered bound as defined by
860 * ARB_bindless_texture.
865 * Emit a warning if this variable is accessed.
868 uint8_t warn_extension_index
;
871 /** Image internal format if specified explicitly, otherwise GL_NONE. */
872 uint16_t image_format
;
876 * Number of state slots used
879 * This could be stored in as few as 7-bits, if necessary. If it is made
880 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
883 uint16_t _num_state_slots
;
887 * Initial binding point for a sampler, atomic, or UBO.
889 * For array types, this represents the binding point for the first element.
894 * Storage location of the base of this variable
896 * The precise meaning of this field depends on the nature of the variable.
898 * - Vertex shader input: one of the values from \c gl_vert_attrib.
899 * - Vertex shader output: one of the values from \c gl_varying_slot.
900 * - Geometry shader input: one of the values from \c gl_varying_slot.
901 * - Geometry shader output: one of the values from \c gl_varying_slot.
902 * - Fragment shader input: one of the values from \c gl_varying_slot.
903 * - Fragment shader output: one of the values from \c gl_frag_result.
904 * - Uniforms: Per-stage uniform slot number for default uniform block.
905 * - Uniforms: Index within the uniform block definition for UBO members.
906 * - Non-UBO Uniforms: explicit location until linking then reused to
907 * store uniform slot number.
908 * - Other: This field is not currently used.
910 * If the variable is a uniform, shader input, or shader output, and the
911 * slot has not been assigned, the value will be -1.
916 * for glsl->tgsi/mesa IR we need to store the index into the
917 * parameters for uniforms, initially the code overloaded location
918 * but this causes problems with indirect samplers and AoA.
919 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
924 * Vertex stream output identifier.
926 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
927 * stream of the i-th component.
932 * Atomic, transform feedback or block member offset.
937 * Highest element accessed with a constant expression array index
939 * Not used for non-array variables. -1 is never accessed.
941 int max_array_access
;
944 * Transform feedback buffer.
949 * Transform feedback stride.
954 * Allow (only) ir_variable direct access private members.
956 friend class ir_variable
;
960 * Value assigned in the initializer of a variable declared "const"
962 ir_constant
*constant_value
;
965 * Constant expression assigned in the initializer of the variable
968 * This field and \c ::constant_value are distinct. Even if the two fields
969 * refer to constants with the same value, they must point to separate
972 ir_constant
*constant_initializer
;
975 static const char *const warn_extension_table
[];
979 * For variables which satisfy the is_interface_instance() predicate,
980 * this points to an array of integers such that if the ith member of
981 * the interface block is an array, max_ifc_array_access[i] is the
982 * maximum array element of that member that has been accessed. If the
983 * ith member of the interface block is not an array,
984 * max_ifc_array_access[i] is unused.
986 * For variables whose type is not an interface block, this pointer is
989 int *max_ifc_array_access
;
992 * Built-in state that backs this uniform
994 * Once set at variable creation, \c state_slots must remain invariant.
996 * If the variable is not a uniform, \c _num_state_slots will be zero
997 * and \c state_slots will be \c NULL.
999 ir_state_slot
*state_slots
;
1003 * For variables that are in an interface block or are an instance of an
1004 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
1006 * \sa ir_variable::location
1008 const glsl_type
*interface_type
;
1011 * Name used for anonymous compiler temporaries
1013 static const char tmp_name
[];
1017 * Should the construct keep names for ir_var_temporary variables?
1019 * When this global is false, names passed to the constructor for
1020 * \c ir_var_temporary variables will be dropped. Instead, the variable will
1021 * be named "compiler_temp". This name will be in static storage.
1024 * \b NEVER change the mode of an \c ir_var_temporary.
1027 * This variable is \b not thread-safe. It is global, \b not
1028 * per-context. It begins life false. A context can, at some point, make
1029 * it true. From that point on, it will be true forever. This should be
1030 * okay since it will only be set true while debugging.
1032 static bool temporaries_allocate_names
;
1036 * A function that returns whether a built-in function is available in the
1037 * current shading language (based on version, ES or desktop, and extensions).
1039 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
1041 #define MAKE_INTRINSIC_FOR_TYPE(op, t) \
1042 ir_intrinsic_generic_ ## op - ir_intrinsic_generic_load + ir_intrinsic_ ## t ## _ ## load
1044 #define MAP_INTRINSIC_TO_TYPE(i, t) \
1045 ir_intrinsic_id(int(i) - int(ir_intrinsic_generic_load) + int(ir_intrinsic_ ## t ## _ ## load))
1047 enum ir_intrinsic_id
{
1048 ir_intrinsic_invalid
= 0,
1051 * \name Generic intrinsics
1053 * Each of these intrinsics has a specific version for shared variables and
1057 ir_intrinsic_generic_load
,
1058 ir_intrinsic_generic_store
,
1059 ir_intrinsic_generic_atomic_add
,
1060 ir_intrinsic_generic_atomic_and
,
1061 ir_intrinsic_generic_atomic_or
,
1062 ir_intrinsic_generic_atomic_xor
,
1063 ir_intrinsic_generic_atomic_min
,
1064 ir_intrinsic_generic_atomic_max
,
1065 ir_intrinsic_generic_atomic_exchange
,
1066 ir_intrinsic_generic_atomic_comp_swap
,
1069 ir_intrinsic_atomic_counter_read
,
1070 ir_intrinsic_atomic_counter_increment
,
1071 ir_intrinsic_atomic_counter_predecrement
,
1072 ir_intrinsic_atomic_counter_add
,
1073 ir_intrinsic_atomic_counter_and
,
1074 ir_intrinsic_atomic_counter_or
,
1075 ir_intrinsic_atomic_counter_xor
,
1076 ir_intrinsic_atomic_counter_min
,
1077 ir_intrinsic_atomic_counter_max
,
1078 ir_intrinsic_atomic_counter_exchange
,
1079 ir_intrinsic_atomic_counter_comp_swap
,
1081 ir_intrinsic_image_load
,
1082 ir_intrinsic_image_store
,
1083 ir_intrinsic_image_atomic_add
,
1084 ir_intrinsic_image_atomic_and
,
1085 ir_intrinsic_image_atomic_or
,
1086 ir_intrinsic_image_atomic_xor
,
1087 ir_intrinsic_image_atomic_min
,
1088 ir_intrinsic_image_atomic_max
,
1089 ir_intrinsic_image_atomic_exchange
,
1090 ir_intrinsic_image_atomic_comp_swap
,
1091 ir_intrinsic_image_size
,
1092 ir_intrinsic_image_samples
,
1094 ir_intrinsic_ssbo_load
,
1095 ir_intrinsic_ssbo_store
= MAKE_INTRINSIC_FOR_TYPE(store
, ssbo
),
1096 ir_intrinsic_ssbo_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, ssbo
),
1097 ir_intrinsic_ssbo_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, ssbo
),
1098 ir_intrinsic_ssbo_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, ssbo
),
1099 ir_intrinsic_ssbo_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, ssbo
),
1100 ir_intrinsic_ssbo_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, ssbo
),
1101 ir_intrinsic_ssbo_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, ssbo
),
1102 ir_intrinsic_ssbo_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, ssbo
),
1103 ir_intrinsic_ssbo_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, ssbo
),
1105 ir_intrinsic_memory_barrier
,
1106 ir_intrinsic_shader_clock
,
1107 ir_intrinsic_group_memory_barrier
,
1108 ir_intrinsic_memory_barrier_atomic_counter
,
1109 ir_intrinsic_memory_barrier_buffer
,
1110 ir_intrinsic_memory_barrier_image
,
1111 ir_intrinsic_memory_barrier_shared
,
1113 ir_intrinsic_vote_all
,
1114 ir_intrinsic_vote_any
,
1115 ir_intrinsic_vote_eq
,
1116 ir_intrinsic_ballot
,
1117 ir_intrinsic_read_invocation
,
1118 ir_intrinsic_read_first_invocation
,
1120 ir_intrinsic_shared_load
,
1121 ir_intrinsic_shared_store
= MAKE_INTRINSIC_FOR_TYPE(store
, shared
),
1122 ir_intrinsic_shared_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, shared
),
1123 ir_intrinsic_shared_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, shared
),
1124 ir_intrinsic_shared_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, shared
),
1125 ir_intrinsic_shared_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, shared
),
1126 ir_intrinsic_shared_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, shared
),
1127 ir_intrinsic_shared_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, shared
),
1128 ir_intrinsic_shared_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, shared
),
1129 ir_intrinsic_shared_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, shared
),
1134 * The representation of a function instance; may be the full definition or
1135 * simply a prototype.
1137 class ir_function_signature
: public ir_instruction
{
1138 /* An ir_function_signature will be part of the list of signatures in
1142 ir_function_signature(const glsl_type
*return_type
,
1143 builtin_available_predicate builtin_avail
= NULL
);
1145 virtual ir_function_signature
*clone(void *mem_ctx
,
1146 struct hash_table
*ht
) const;
1147 ir_function_signature
*clone_prototype(void *mem_ctx
,
1148 struct hash_table
*ht
) const;
1150 virtual void accept(ir_visitor
*v
)
1155 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1158 * Attempt to evaluate this function as a constant expression,
1159 * given a list of the actual parameters and the variable context.
1160 * Returns NULL for non-built-ins.
1162 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
1165 * Get the name of the function for which this is a signature
1167 const char *function_name() const;
1170 * Get a handle to the function for which this is a signature
1172 * There is no setter function, this function returns a \c const pointer,
1173 * and \c ir_function_signature::_function is private for a reason. The
1174 * only way to make a connection between a function and function signature
1175 * is via \c ir_function::add_signature. This helps ensure that certain
1176 * invariants (i.e., a function signature is in the list of signatures for
1177 * its \c _function) are met.
1179 * \sa ir_function::add_signature
1181 inline const class ir_function
*function() const
1183 return this->_function
;
1187 * Check whether the qualifiers match between this signature's parameters
1188 * and the supplied parameter list. If not, returns the name of the first
1189 * parameter with mismatched qualifiers (for use in error messages).
1191 const char *qualifiers_match(exec_list
*params
);
1194 * Replace the current parameter list with the given one. This is useful
1195 * if the current information came from a prototype, and either has invalid
1196 * or missing parameter names.
1198 void replace_parameters(exec_list
*new_params
);
1201 * Function return type.
1203 * \note This discards the optional precision qualifier.
1205 const struct glsl_type
*return_type
;
1208 * List of ir_variable of function parameters.
1210 * This represents the storage. The paramaters passed in a particular
1211 * call will be in ir_call::actual_paramaters.
1213 struct exec_list parameters
;
1215 /** Whether or not this function has a body (which may be empty). */
1216 unsigned is_defined
:1;
1218 /** Whether or not this function signature is a built-in. */
1219 bool is_builtin() const;
1222 * Whether or not this function is an intrinsic to be implemented
1225 inline bool is_intrinsic() const
1227 return intrinsic_id
!= ir_intrinsic_invalid
;
1230 /** Indentifier for this intrinsic. */
1231 enum ir_intrinsic_id intrinsic_id
;
1233 /** Whether or not a built-in is available for this shader. */
1234 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1236 /** Body of instructions in the function. */
1237 struct exec_list body
;
1241 * A function pointer to a predicate that answers whether a built-in
1242 * function is available in the current shader. NULL if not a built-in.
1244 builtin_available_predicate builtin_avail
;
1246 /** Function of which this signature is one overload. */
1247 class ir_function
*_function
;
1249 /** Function signature of which this one is a prototype clone */
1250 const ir_function_signature
*origin
;
1252 friend class ir_function
;
1255 * Helper function to run a list of instructions for constant
1256 * expression evaluation.
1258 * The hash table represents the values of the visible variables.
1259 * There are no scoping issues because the table is indexed on
1260 * ir_variable pointers, not variable names.
1262 * Returns false if the expression is not constant, true otherwise,
1263 * and the value in *result if result is non-NULL.
1265 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1266 struct hash_table
*variable_context
,
1267 ir_constant
**result
);
1272 * Header for tracking multiple overloaded functions with the same name.
1273 * Contains a list of ir_function_signatures representing each of the
1276 class ir_function
: public ir_instruction
{
1278 ir_function(const char *name
);
1280 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1282 virtual void accept(ir_visitor
*v
)
1287 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1289 void add_signature(ir_function_signature
*sig
)
1291 sig
->_function
= this;
1292 this->signatures
.push_tail(sig
);
1296 * Find a signature that matches a set of actual parameters, taking implicit
1297 * conversions into account. Also flags whether the match was exact.
1299 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1300 const exec_list
*actual_param
,
1301 bool allow_builtins
,
1302 bool *match_is_exact
);
1305 * Find a signature that matches a set of actual parameters, taking implicit
1306 * conversions into account.
1308 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1309 const exec_list
*actual_param
,
1310 bool allow_builtins
);
1313 * Find a signature that exactly matches a set of actual parameters without
1314 * any implicit type conversions.
1316 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1317 const exec_list
*actual_ps
);
1320 * Name of the function.
1324 /** Whether or not this function has a signature that isn't a built-in. */
1325 bool has_user_signature();
1328 * List of ir_function_signature for each overloaded function with this name.
1330 struct exec_list signatures
;
1333 * is this function a subroutine type declaration
1334 * e.g. subroutine void type1(float arg1);
1339 * is this function associated to a subroutine type
1340 * e.g. subroutine (type1, type2) function_name { function_body };
1341 * would have num_subroutine_types 2,
1342 * and pointers to the type1 and type2 types.
1344 int num_subroutine_types
;
1345 const struct glsl_type
**subroutine_types
;
1347 int subroutine_index
;
1350 inline const char *ir_function_signature::function_name() const
1352 return this->_function
->name
;
1358 * IR instruction representing high-level if-statements
1360 class ir_if
: public ir_instruction
{
1362 ir_if(ir_rvalue
*condition
)
1363 : ir_instruction(ir_type_if
), condition(condition
)
1367 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1369 virtual void accept(ir_visitor
*v
)
1374 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1376 ir_rvalue
*condition
;
1377 /** List of ir_instruction for the body of the then branch */
1378 exec_list then_instructions
;
1379 /** List of ir_instruction for the body of the else branch */
1380 exec_list else_instructions
;
1385 * IR instruction representing a high-level loop structure.
1387 class ir_loop
: public ir_instruction
{
1391 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1393 virtual void accept(ir_visitor
*v
)
1398 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1400 /** List of ir_instruction that make up the body of the loop. */
1401 exec_list body_instructions
;
1405 class ir_assignment
: public ir_instruction
{
1407 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1410 * Construct an assignment with an explicit write mask
1413 * Since a write mask is supplied, the LHS must already be a bare
1414 * \c ir_dereference. The cannot be any swizzles in the LHS.
1416 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1417 unsigned write_mask
);
1419 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1421 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1423 virtual void accept(ir_visitor
*v
)
1428 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1431 * Get a whole variable written by an assignment
1433 * If the LHS of the assignment writes a whole variable, the variable is
1434 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1437 * - Assigning to a scalar
1438 * - Assigning to all components of a vector
1439 * - Whole array (or matrix) assignment
1440 * - Whole structure assignment
1442 ir_variable
*whole_variable_written();
1445 * Set the LHS of an assignment
1447 void set_lhs(ir_rvalue
*lhs
);
1450 * Left-hand side of the assignment.
1452 * This should be treated as read only. If you need to set the LHS of an
1453 * assignment, use \c ir_assignment::set_lhs.
1455 ir_dereference
*lhs
;
1458 * Value being assigned
1463 * Optional condition for the assignment.
1465 ir_rvalue
*condition
;
1469 * Component mask written
1471 * For non-vector types in the LHS, this field will be zero. For vector
1472 * types, a bit will be set for each component that is written. Note that
1473 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1475 * A partially-set write mask means that each enabled channel gets
1476 * the value from a consecutive channel of the rhs. For example,
1477 * to write just .xyw of gl_FrontColor with color:
1479 * (assign (constant bool (1)) (xyw)
1480 * (var_ref gl_FragColor)
1481 * (swiz xyw (var_ref color)))
1483 unsigned write_mask
:4;
1486 #include "ir_expression_operation.h"
1488 extern const char *const ir_expression_operation_strings
[ir_last_opcode
+ 1];
1489 extern const char *const ir_expression_operation_enum_strings
[ir_last_opcode
+ 1];
1491 class ir_expression
: public ir_rvalue
{
1493 ir_expression(int op
, const struct glsl_type
*type
,
1494 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1495 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1498 * Constructor for unary operation expressions
1500 ir_expression(int op
, ir_rvalue
*);
1503 * Constructor for binary operation expressions
1505 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1508 * Constructor for ternary operation expressions
1510 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1512 virtual bool equals(const ir_instruction
*ir
,
1513 enum ir_node_type ignore
= ir_type_unset
) const;
1515 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1518 * Attempt to constant-fold the expression
1520 * The "variable_context" hash table links ir_variable * to ir_constant *
1521 * that represent the variables' values. \c NULL represents an empty
1524 * If the expression cannot be constant folded, this method will return
1527 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1530 * Determine the number of operands used by an expression
1532 static unsigned int get_num_operands(ir_expression_operation
);
1535 * Determine the number of operands used by an expression
1537 unsigned int get_num_operands() const
1539 return (this->operation
== ir_quadop_vector
)
1540 ? this->type
->vector_elements
: get_num_operands(operation
);
1544 * Return whether the expression operates on vectors horizontally.
1546 bool is_horizontal() const
1548 return operation
== ir_binop_all_equal
||
1549 operation
== ir_binop_any_nequal
||
1550 operation
== ir_binop_dot
||
1551 operation
== ir_binop_vector_extract
||
1552 operation
== ir_triop_vector_insert
||
1553 operation
== ir_binop_ubo_load
||
1554 operation
== ir_quadop_vector
;
1558 * Do a reverse-lookup to translate the given string into an operator.
1560 static ir_expression_operation
get_operator(const char *);
1562 virtual void accept(ir_visitor
*v
)
1567 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1569 virtual ir_variable
*variable_referenced() const;
1571 ir_expression_operation operation
;
1572 ir_rvalue
*operands
[4];
1577 * HIR instruction representing a high-level function call, containing a list
1578 * of parameters and returning a value in the supplied temporary.
1580 class ir_call
: public ir_instruction
{
1582 ir_call(ir_function_signature
*callee
,
1583 ir_dereference_variable
*return_deref
,
1584 exec_list
*actual_parameters
)
1585 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(NULL
), array_idx(NULL
)
1587 assert(callee
->return_type
!= NULL
);
1588 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1589 this->use_builtin
= callee
->is_builtin();
1592 ir_call(ir_function_signature
*callee
,
1593 ir_dereference_variable
*return_deref
,
1594 exec_list
*actual_parameters
,
1595 ir_variable
*var
, ir_rvalue
*array_idx
)
1596 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(var
), array_idx(array_idx
)
1598 assert(callee
->return_type
!= NULL
);
1599 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1600 this->use_builtin
= callee
->is_builtin();
1603 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1605 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1607 virtual void accept(ir_visitor
*v
)
1612 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1615 * Get the name of the function being called.
1617 const char *callee_name() const
1619 return callee
->function_name();
1623 * Generates an inline version of the function before @ir,
1624 * storing the return value in return_deref.
1626 void generate_inline(ir_instruction
*ir
);
1629 * Storage for the function's return value.
1630 * This must be NULL if the return type is void.
1632 ir_dereference_variable
*return_deref
;
1635 * The specific function signature being called.
1637 ir_function_signature
*callee
;
1639 /* List of ir_rvalue of paramaters passed in this call. */
1640 exec_list actual_parameters
;
1642 /** Should this call only bind to a built-in function? */
1646 * ARB_shader_subroutine support -
1647 * the subroutine uniform variable and array index
1648 * rvalue to be used in the lowering pass later.
1650 ir_variable
*sub_var
;
1651 ir_rvalue
*array_idx
;
1656 * \name Jump-like IR instructions.
1658 * These include \c break, \c continue, \c return, and \c discard.
1661 class ir_jump
: public ir_instruction
{
1663 ir_jump(enum ir_node_type t
)
1669 class ir_return
: public ir_jump
{
1672 : ir_jump(ir_type_return
), value(NULL
)
1676 ir_return(ir_rvalue
*value
)
1677 : ir_jump(ir_type_return
), value(value
)
1681 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1683 ir_rvalue
*get_value() const
1688 virtual void accept(ir_visitor
*v
)
1693 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1700 * Jump instructions used inside loops
1702 * These include \c break and \c continue. The \c break within a loop is
1703 * different from the \c break within a switch-statement.
1705 * \sa ir_switch_jump
1707 class ir_loop_jump
: public ir_jump
{
1714 ir_loop_jump(jump_mode mode
)
1715 : ir_jump(ir_type_loop_jump
)
1720 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1722 virtual void accept(ir_visitor
*v
)
1727 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1729 bool is_break() const
1731 return mode
== jump_break
;
1734 bool is_continue() const
1736 return mode
== jump_continue
;
1739 /** Mode selector for the jump instruction. */
1740 enum jump_mode mode
;
1744 * IR instruction representing discard statements.
1746 class ir_discard
: public ir_jump
{
1749 : ir_jump(ir_type_discard
)
1751 this->condition
= NULL
;
1754 ir_discard(ir_rvalue
*cond
)
1755 : ir_jump(ir_type_discard
)
1757 this->condition
= cond
;
1760 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1762 virtual void accept(ir_visitor
*v
)
1767 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1769 ir_rvalue
*condition
;
1775 * Texture sampling opcodes used in ir_texture
1777 enum ir_texture_opcode
{
1778 ir_tex
, /**< Regular texture look-up */
1779 ir_txb
, /**< Texture look-up with LOD bias */
1780 ir_txl
, /**< Texture look-up with explicit LOD */
1781 ir_txd
, /**< Texture look-up with partial derivatvies */
1782 ir_txf
, /**< Texel fetch with explicit LOD */
1783 ir_txf_ms
, /**< Multisample texture fetch */
1784 ir_txs
, /**< Texture size */
1785 ir_lod
, /**< Texture lod query */
1786 ir_tg4
, /**< Texture gather */
1787 ir_query_levels
, /**< Texture levels query */
1788 ir_texture_samples
, /**< Texture samples query */
1789 ir_samples_identical
, /**< Query whether all samples are definitely identical. */
1794 * IR instruction to sample a texture
1796 * The specific form of the IR instruction depends on the \c mode value
1797 * selected from \c ir_texture_opcodes. In the printed IR, these will
1800 * Texel offset (0 or an expression)
1801 * | Projection divisor
1802 * | | Shadow comparator
1805 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1806 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1807 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1808 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1809 * (txf <type> <sampler> <coordinate> 0 <lod>)
1811 * <type> <sampler> <coordinate> <sample_index>)
1812 * (txs <type> <sampler> <lod>)
1813 * (lod <type> <sampler> <coordinate>)
1814 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1815 * (query_levels <type> <sampler>)
1816 * (samples_identical <sampler> <coordinate>)
1818 class ir_texture
: public ir_rvalue
{
1820 ir_texture(enum ir_texture_opcode op
)
1821 : ir_rvalue(ir_type_texture
),
1822 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1823 shadow_comparator(NULL
), offset(NULL
)
1825 memset(&lod_info
, 0, sizeof(lod_info
));
1828 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1830 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1832 virtual void accept(ir_visitor
*v
)
1837 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1839 virtual bool equals(const ir_instruction
*ir
,
1840 enum ir_node_type ignore
= ir_type_unset
) const;
1843 * Return a string representing the ir_texture_opcode.
1845 const char *opcode_string();
1847 /** Set the sampler and type. */
1848 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1851 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1853 static ir_texture_opcode
get_opcode(const char *);
1855 enum ir_texture_opcode op
;
1857 /** Sampler to use for the texture access. */
1858 ir_dereference
*sampler
;
1860 /** Texture coordinate to sample */
1861 ir_rvalue
*coordinate
;
1864 * Value used for projective divide.
1866 * If there is no projective divide (the common case), this will be
1867 * \c NULL. Optimization passes should check for this to point to a constant
1868 * of 1.0 and replace that with \c NULL.
1870 ir_rvalue
*projector
;
1873 * Coordinate used for comparison on shadow look-ups.
1875 * If there is no shadow comparison, this will be \c NULL. For the
1876 * \c ir_txf opcode, this *must* be \c NULL.
1878 ir_rvalue
*shadow_comparator
;
1880 /** Texel offset. */
1884 ir_rvalue
*lod
; /**< Floating point LOD */
1885 ir_rvalue
*bias
; /**< Floating point LOD bias */
1886 ir_rvalue
*sample_index
; /**< MSAA sample index */
1887 ir_rvalue
*component
; /**< Gather component selector */
1889 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1890 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1896 struct ir_swizzle_mask
{
1903 * Number of components in the swizzle.
1905 unsigned num_components
:3;
1908 * Does the swizzle contain duplicate components?
1910 * L-value swizzles cannot contain duplicate components.
1912 unsigned has_duplicates
:1;
1916 class ir_swizzle
: public ir_rvalue
{
1918 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
1921 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
1923 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
1925 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
1927 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1930 * Construct an ir_swizzle from the textual representation. Can fail.
1932 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
1934 virtual void accept(ir_visitor
*v
)
1939 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1941 virtual bool equals(const ir_instruction
*ir
,
1942 enum ir_node_type ignore
= ir_type_unset
) const;
1944 bool is_lvalue() const
1946 return val
->is_lvalue() && !mask
.has_duplicates
;
1950 * Get the variable that is ultimately referenced by an r-value
1952 virtual ir_variable
*variable_referenced() const;
1955 ir_swizzle_mask mask
;
1959 * Initialize the mask component of a swizzle
1961 * This is used by the \c ir_swizzle constructors.
1963 void init_mask(const unsigned *components
, unsigned count
);
1967 class ir_dereference
: public ir_rvalue
{
1969 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
1971 bool is_lvalue() const;
1974 * Get the variable that is ultimately referenced by an r-value
1976 virtual ir_variable
*variable_referenced() const = 0;
1979 ir_dereference(enum ir_node_type t
)
1986 class ir_dereference_variable
: public ir_dereference
{
1988 ir_dereference_variable(ir_variable
*var
);
1990 virtual ir_dereference_variable
*clone(void *mem_ctx
,
1991 struct hash_table
*) const;
1993 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1995 virtual bool equals(const ir_instruction
*ir
,
1996 enum ir_node_type ignore
= ir_type_unset
) const;
1999 * Get the variable that is ultimately referenced by an r-value
2001 virtual ir_variable
*variable_referenced() const
2006 virtual ir_variable
*whole_variable_referenced()
2008 /* ir_dereference_variable objects always dereference the entire
2009 * variable. However, if this dereference is dereferenced by anything
2010 * else, the complete deferefernce chain is not a whole-variable
2011 * dereference. This method should only be called on the top most
2012 * ir_rvalue in a dereference chain.
2017 virtual void accept(ir_visitor
*v
)
2022 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2025 * Object being dereferenced.
2031 class ir_dereference_array
: public ir_dereference
{
2033 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2035 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2037 virtual ir_dereference_array
*clone(void *mem_ctx
,
2038 struct hash_table
*) const;
2040 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2042 virtual bool equals(const ir_instruction
*ir
,
2043 enum ir_node_type ignore
= ir_type_unset
) const;
2046 * Get the variable that is ultimately referenced by an r-value
2048 virtual ir_variable
*variable_referenced() const
2050 return this->array
->variable_referenced();
2053 virtual void accept(ir_visitor
*v
)
2058 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2061 ir_rvalue
*array_index
;
2064 void set_array(ir_rvalue
*value
);
2068 class ir_dereference_record
: public ir_dereference
{
2070 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2072 ir_dereference_record(ir_variable
*var
, const char *field
);
2074 virtual ir_dereference_record
*clone(void *mem_ctx
,
2075 struct hash_table
*) const;
2077 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2080 * Get the variable that is ultimately referenced by an r-value
2082 virtual ir_variable
*variable_referenced() const
2084 return this->record
->variable_referenced();
2087 virtual void accept(ir_visitor
*v
)
2092 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2100 * Data stored in an ir_constant
2102 union ir_constant_data
{
2113 class ir_constant
: public ir_rvalue
{
2115 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2116 ir_constant(bool b
, unsigned vector_elements
=1);
2117 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2118 ir_constant(int i
, unsigned vector_elements
=1);
2119 ir_constant(float f
, unsigned vector_elements
=1);
2120 ir_constant(double d
, unsigned vector_elements
=1);
2121 ir_constant(uint64_t u64
, unsigned vector_elements
=1);
2122 ir_constant(int64_t i64
, unsigned vector_elements
=1);
2125 * Construct an ir_constant from a list of ir_constant values
2127 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2130 * Construct an ir_constant from a scalar component of another ir_constant
2132 * The new \c ir_constant inherits the type of the component from the
2136 * In the case of a matrix constant, the new constant is a scalar, \b not
2139 ir_constant(const ir_constant
*c
, unsigned i
);
2142 * Return a new ir_constant of the specified type containing all zeros.
2144 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2146 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2148 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2150 virtual void accept(ir_visitor
*v
)
2155 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2157 virtual bool equals(const ir_instruction
*ir
,
2158 enum ir_node_type ignore
= ir_type_unset
) const;
2161 * Get a particular component of a constant as a specific type
2163 * This is useful, for example, to get a value from an integer constant
2164 * as a float or bool. This appears frequently when constructors are
2165 * called with all constant parameters.
2168 bool get_bool_component(unsigned i
) const;
2169 float get_float_component(unsigned i
) const;
2170 double get_double_component(unsigned i
) const;
2171 int get_int_component(unsigned i
) const;
2172 unsigned get_uint_component(unsigned i
) const;
2173 int64_t get_int64_component(unsigned i
) const;
2174 uint64_t get_uint64_component(unsigned i
) const;
2177 ir_constant
*get_array_element(unsigned i
) const;
2179 ir_constant
*get_record_field(const char *name
);
2182 * Copy the values on another constant at a given offset.
2184 * The offset is ignored for array or struct copies, it's only for
2185 * scalars or vectors into vectors or matrices.
2187 * With identical types on both sides and zero offset it's clone()
2188 * without creating a new object.
2191 void copy_offset(ir_constant
*src
, int offset
);
2194 * Copy the values on another constant at a given offset and
2195 * following an assign-like mask.
2197 * The mask is ignored for scalars.
2199 * Note that this function only handles what assign can handle,
2200 * i.e. at most a vector as source and a column of a matrix as
2204 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2207 * Determine whether a constant has the same value as another constant
2209 * \sa ir_constant::is_zero, ir_constant::is_one,
2210 * ir_constant::is_negative_one
2212 bool has_value(const ir_constant
*) const;
2215 * Return true if this ir_constant represents the given value.
2217 * For vectors, this checks that each component is the given value.
2219 virtual bool is_value(float f
, int i
) const;
2220 virtual bool is_zero() const;
2221 virtual bool is_one() const;
2222 virtual bool is_negative_one() const;
2225 * Return true for constants that could be stored as 16-bit unsigned values.
2227 * Note that this will return true even for signed integer ir_constants, as
2228 * long as the value is non-negative and fits in 16-bits.
2230 virtual bool is_uint16_constant() const;
2233 * Value of the constant.
2235 * The field used to back the values supplied by the constant is determined
2236 * by the type associated with the \c ir_instruction. Constants may be
2237 * scalars, vectors, or matrices.
2239 union ir_constant_data value
;
2241 /* Array elements */
2242 ir_constant
**array_elements
;
2244 /* Structure fields */
2245 exec_list components
;
2249 * Parameterless constructor only used by the clone method
2255 * IR instruction to emit a vertex in a geometry shader.
2257 class ir_emit_vertex
: public ir_instruction
{
2259 ir_emit_vertex(ir_rvalue
*stream
)
2260 : ir_instruction(ir_type_emit_vertex
),
2266 virtual void accept(ir_visitor
*v
)
2271 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2273 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2276 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2278 int stream_id() const
2280 return stream
->as_constant()->value
.i
[0];
2287 * IR instruction to complete the current primitive and start a new one in a
2290 class ir_end_primitive
: public ir_instruction
{
2292 ir_end_primitive(ir_rvalue
*stream
)
2293 : ir_instruction(ir_type_end_primitive
),
2299 virtual void accept(ir_visitor
*v
)
2304 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2306 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2309 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2311 int stream_id() const
2313 return stream
->as_constant()->value
.i
[0];
2320 * IR instruction for tessellation control and compute shader barrier.
2322 class ir_barrier
: public ir_instruction
{
2325 : ir_instruction(ir_type_barrier
)
2329 virtual void accept(ir_visitor
*v
)
2334 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2336 return new(mem_ctx
) ir_barrier();
2339 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2345 * Apply a visitor to each IR node in a list
2348 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2351 * Validate invariants on each IR node in a list
2353 void validate_ir_tree(exec_list
*instructions
);
2355 struct _mesa_glsl_parse_state
;
2356 struct gl_shader_program
;
2359 * Detect whether an unlinked shader contains static recursion
2361 * If the list of instructions is determined to contain static recursion,
2362 * \c _mesa_glsl_error will be called to emit error messages for each function
2363 * that is in the recursion cycle.
2366 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2367 exec_list
*instructions
);
2370 * Detect whether a linked shader contains static recursion
2372 * If the list of instructions is determined to contain static recursion,
2373 * \c link_error_printf will be called to emit error messages for each function
2374 * that is in the recursion cycle. In addition,
2375 * \c gl_shader_program::LinkStatus will be set to false.
2378 detect_recursion_linked(struct gl_shader_program
*prog
,
2379 exec_list
*instructions
);
2382 * Make a clone of each IR instruction in a list
2384 * \param in List of IR instructions that are to be cloned
2385 * \param out List to hold the cloned instructions
2388 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2391 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2392 struct _mesa_glsl_parse_state
*state
);
2395 _mesa_glsl_initialize_derived_variables(struct gl_context
*ctx
,
2399 reparent_ir(exec_list
*list
, void *mem_ctx
);
2402 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2403 gl_shader_stage shader_stage
);
2406 prototype_string(const glsl_type
*return_type
, const char *name
,
2407 exec_list
*parameters
);
2410 mode_string(const ir_variable
*var
);
2413 * Built-in / reserved GL variables names start with "gl_"
2416 is_gl_identifier(const char *s
)
2418 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2422 #endif /* __cplusplus */
2424 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2425 struct _mesa_glsl_parse_state
*state
);
2428 fprint_ir(FILE *f
, const void *instruction
);
2430 extern const struct gl_builtin_uniform_desc
*
2431 _mesa_glsl_get_builtin_uniform_desc(const char *name
);
2438 vertices_per_prim(GLenum prim
);