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 "util/format/u_format.h"
33 #include "compiler/glsl_types.h"
35 #include "ir_visitor.h"
36 #include "ir_hierarchical_visitor.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
,
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(void *mem_ctx
,
234 struct hash_table
*variable_context
= NULL
);
236 ir_rvalue
*as_rvalue_to_saturate();
238 virtual bool is_lvalue(const struct _mesa_glsl_parse_state
* = NULL
) const
244 * Get the variable that is ultimately referenced by an r-value
246 virtual ir_variable
*variable_referenced() const
253 * If an r-value is a reference to a whole variable, get that variable
256 * Pointer to a variable that is completely dereferenced by the r-value. If
257 * the r-value is not a dereference or the dereference does not access the
258 * entire variable (i.e., it's just one array element, struct field), \c NULL
261 virtual ir_variable
*whole_variable_referenced()
267 * Determine if an r-value has the value zero
269 * The base implementation of this function always returns \c false. The
270 * \c ir_constant class over-rides this function to return \c true \b only
271 * for vector and scalar types that have all elements set to the value
272 * zero (or \c false for booleans).
274 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
276 virtual bool is_zero() const;
279 * Determine if an r-value has the value one
281 * The base implementation of this function always returns \c false. The
282 * \c ir_constant class over-rides this function to return \c true \b only
283 * for vector and scalar types that have all elements set to the value
284 * one (or \c true for booleans).
286 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
288 virtual bool is_one() const;
291 * Determine if an r-value has the value negative one
293 * The base implementation of this function always returns \c false. The
294 * \c ir_constant class over-rides this function to return \c true \b only
295 * for vector and scalar types that have all elements set to the value
296 * negative one. For boolean types, the result is always \c false.
298 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
300 virtual bool is_negative_one() const;
303 * Determine if an r-value is an unsigned integer constant which can be
306 * \sa ir_constant::is_uint16_constant.
308 virtual bool is_uint16_constant() const { return false; }
311 * Return a generic value of error_type.
313 * Allocation will be performed with 'mem_ctx' as ralloc owner.
315 static ir_rvalue
*error_value(void *mem_ctx
);
318 ir_rvalue(enum ir_node_type t
);
323 * Variable storage classes
325 enum ir_variable_mode
{
326 ir_var_auto
= 0, /**< Function local variables and globals. */
327 ir_var_uniform
, /**< Variable declared as a uniform. */
328 ir_var_shader_storage
, /**< Variable declared as an ssbo. */
329 ir_var_shader_shared
, /**< Variable declared as shared. */
334 ir_var_function_inout
,
335 ir_var_const_in
, /**< "in" param that must be a constant expression */
336 ir_var_system_value
, /**< Ex: front-face, instance-id, etc. */
337 ir_var_temporary
, /**< Temporary variable generated during compilation. */
338 ir_var_mode_count
/**< Number of variable modes */
342 * Enum keeping track of how a variable was declared. For error checking of
343 * the gl_PerVertex redeclaration rules.
345 enum ir_var_declaration_type
{
347 * Normal declaration (for most variables, this means an explicit
348 * declaration. Exception: temporaries are always implicitly declared, but
349 * they still use ir_var_declared_normally).
351 * Note: an ir_variable that represents a named interface block uses
352 * ir_var_declared_normally.
354 ir_var_declared_normally
= 0,
357 * Variable was explicitly declared (or re-declared) in an unnamed
360 ir_var_declared_in_block
,
363 * Variable is an implicitly declared built-in that has not been explicitly
364 * re-declared by the shader.
366 ir_var_declared_implicitly
,
369 * Variable is implicitly generated by the compiler and should not be
370 * visible via the API.
376 * \brief Layout qualifiers for gl_FragDepth.
378 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
379 * with a layout qualifier.
381 enum ir_depth_layout
{
382 ir_depth_layout_none
, /**< No depth layout is specified. */
384 ir_depth_layout_greater
,
385 ir_depth_layout_less
,
386 ir_depth_layout_unchanged
390 * \brief Convert depth layout qualifier to string.
393 depth_layout_string(ir_depth_layout layout
);
396 * Description of built-in state associated with a uniform
398 * \sa ir_variable::state_slots
400 struct ir_state_slot
{
401 gl_state_index16 tokens
[STATE_LENGTH
];
407 * Get the string value for an interpolation qualifier
409 * \return The string that would be used in a shader to specify \c
410 * mode will be returned.
412 * This function is used to generate error messages of the form "shader
413 * uses %s interpolation qualifier", so in the case where there is no
414 * interpolation qualifier, it returns "no".
416 * This function should only be used on a shader input or output variable.
418 const char *interpolation_string(unsigned interpolation
);
421 class ir_variable
: public ir_instruction
{
423 ir_variable(const struct glsl_type
*, const char *, ir_variable_mode
);
425 virtual ir_variable
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
427 virtual void accept(ir_visitor
*v
)
432 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
436 * Determine whether or not a variable is part of a uniform or
437 * shader storage block.
439 inline bool is_in_buffer_block() const
441 return (this->data
.mode
== ir_var_uniform
||
442 this->data
.mode
== ir_var_shader_storage
) &&
443 this->interface_type
!= NULL
;
447 * Determine whether or not a variable is part of a shader storage block.
449 inline bool is_in_shader_storage_block() const
451 return this->data
.mode
== ir_var_shader_storage
&&
452 this->interface_type
!= NULL
;
456 * Determine whether or not a variable is the declaration of an interface
459 * For the first declaration below, there will be an \c ir_variable named
460 * "instance" whose type and whose instance_type will be the same
461 * \c glsl_type. For the second declaration, there will be an \c ir_variable
462 * named "f" whose type is float and whose instance_type is B2.
464 * "instance" is an interface instance variable, but "f" is not.
474 inline bool is_interface_instance() const
476 return this->type
->without_array() == this->interface_type
;
480 * Return whether this variable contains a bindless sampler/image.
482 inline bool contains_bindless() const
484 if (!this->type
->contains_sampler() && !this->type
->contains_image())
487 return this->data
.bindless
|| this->data
.mode
!= ir_var_uniform
;
491 * Set this->interface_type on a newly created variable.
493 void init_interface_type(const struct glsl_type
*type
)
495 assert(this->interface_type
== NULL
);
496 this->interface_type
= type
;
497 if (this->is_interface_instance()) {
498 this->u
.max_ifc_array_access
=
499 ralloc_array(this, int, type
->length
);
500 for (unsigned i
= 0; i
< type
->length
; i
++) {
501 this->u
.max_ifc_array_access
[i
] = -1;
507 * Change this->interface_type on a variable that previously had a
508 * different, but compatible, interface_type. This is used during linking
509 * to set the size of arrays in interface blocks.
511 void change_interface_type(const struct glsl_type
*type
)
513 if (this->u
.max_ifc_array_access
!= NULL
) {
514 /* max_ifc_array_access has already been allocated, so make sure the
515 * new interface has the same number of fields as the old one.
517 assert(this->interface_type
->length
== type
->length
);
519 this->interface_type
= type
;
523 * Change this->interface_type on a variable that previously had a
524 * different, and incompatible, interface_type. This is used during
525 * compilation to handle redeclaration of the built-in gl_PerVertex
528 void reinit_interface_type(const struct glsl_type
*type
)
530 if (this->u
.max_ifc_array_access
!= NULL
) {
532 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
533 * it defines have been accessed yet; so it's safe to throw away the
534 * old max_ifc_array_access pointer, since all of its values are
537 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
538 assert(this->u
.max_ifc_array_access
[i
] == -1);
540 ralloc_free(this->u
.max_ifc_array_access
);
541 this->u
.max_ifc_array_access
= NULL
;
543 this->interface_type
= NULL
;
544 init_interface_type(type
);
547 const glsl_type
*get_interface_type() const
549 return this->interface_type
;
552 enum glsl_interface_packing
get_interface_type_packing() const
554 return this->interface_type
->get_interface_packing();
557 * Get the max_ifc_array_access pointer
559 * A "set" function is not needed because the array is dynmically allocated
562 inline int *get_max_ifc_array_access()
564 assert(this->data
._num_state_slots
== 0);
565 return this->u
.max_ifc_array_access
;
568 inline unsigned get_num_state_slots() const
570 assert(!this->is_interface_instance()
571 || this->data
._num_state_slots
== 0);
572 return this->data
._num_state_slots
;
575 inline void set_num_state_slots(unsigned n
)
577 assert(!this->is_interface_instance()
579 this->data
._num_state_slots
= n
;
582 inline ir_state_slot
*get_state_slots()
584 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
587 inline const ir_state_slot
*get_state_slots() const
589 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
592 inline ir_state_slot
*allocate_state_slots(unsigned n
)
594 assert(!this->is_interface_instance());
596 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
597 this->data
._num_state_slots
= 0;
599 if (this->u
.state_slots
!= NULL
)
600 this->data
._num_state_slots
= n
;
602 return this->u
.state_slots
;
605 inline bool is_interpolation_flat() const
607 return this->data
.interpolation
== INTERP_MODE_FLAT
||
608 this->type
->contains_integer() ||
609 this->type
->contains_double();
612 inline bool is_name_ralloced() const
614 return this->name
!= ir_variable::tmp_name
&&
615 this->name
!= this->name_storage
;
619 * Enable emitting extension warnings for this variable
621 void enable_extension_warning(const char *extension
);
624 * Get the extension warning string for this variable
626 * If warnings are not enabled, \c NULL is returned.
628 const char *get_extension_warning() const;
631 * Declared type of the variable
633 const struct glsl_type
*type
;
636 * Declared name of the variable
642 * If the name length fits into name_storage, it's used, otherwise
643 * the name is ralloc'd. shader-db mining showed that 70% of variables
644 * fit here. This is a win over ralloc where only ralloc_header has
645 * 20 bytes on 64-bit (28 bytes with DEBUG), and we can also skip malloc.
647 char name_storage
[16];
650 struct ir_variable_data
{
653 * Is the variable read-only?
655 * This is set for variables declared as \c const, shader inputs,
658 unsigned read_only
:1;
663 * Was an 'invariant' qualifier explicitly set in the shader?
665 * This is used to cross validate qualifiers.
667 unsigned explicit_invariant
:1;
669 * Is the variable invariant?
671 * It can happen either by having the 'invariant' qualifier
672 * explicitly set in the shader or by being used in calculations
673 * of other invariant variables.
675 unsigned invariant
:1;
679 * Has this variable been used for reading or writing?
681 * Several GLSL semantic checks require knowledge of whether or not a
682 * variable has been used. For example, it is an error to redeclare a
683 * variable as invariant after it has been used.
685 * This is maintained in the ast_to_hir.cpp path and during linking,
686 * but not in Mesa's fixed function or ARB program paths.
691 * Has this variable been statically assigned?
693 * This answers whether the variable was assigned in any path of
694 * the shader during ast_to_hir. This doesn't answer whether it is
695 * still written after dead code removal, nor is it maintained in
696 * non-ast_to_hir.cpp (GLSL parsing) paths.
701 * When separate shader programs are enabled, only input/outputs between
702 * the stages of a multi-stage separate program can be safely removed
703 * from the shader interface. Other input/outputs must remains active.
705 unsigned always_active_io
:1;
708 * Enum indicating how the variable was declared. See
709 * ir_var_declaration_type.
711 * This is used to detect certain kinds of illegal variable redeclarations.
713 unsigned how_declared
:2;
716 * Storage class of the variable.
718 * \sa ir_variable_mode
723 * Interpolation mode for shader inputs / outputs
725 * \sa glsl_interp_mode
727 unsigned interpolation
:2;
730 * Was the location explicitly set in the shader?
732 * If the location is explicitly set in the shader, it \b cannot be changed
733 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
736 unsigned explicit_location
:1;
737 unsigned explicit_index
:1;
740 * Was an initial binding explicitly set in the shader?
742 * If so, constant_value contains an integer ir_constant representing the
743 * initial binding point.
745 unsigned explicit_binding
:1;
748 * Was an initial component explicitly set in the shader?
750 unsigned explicit_component
:1;
753 * Does this variable have an initializer?
755 * This is used by the linker to cross-validiate initializers of global
758 unsigned has_initializer
:1;
761 * Is this variable a generic output or input that has not yet been matched
762 * up to a variable in another stage of the pipeline?
764 * This is used by the linker as scratch storage while assigning locations
765 * to generic inputs and outputs.
767 unsigned is_unmatched_generic_inout
:1;
770 * Is this varying used by transform feedback?
772 * This is used by the linker to decide if it's safe to pack the varying.
777 * Is this varying used only by transform feedback?
779 * This is used by the linker to decide if its safe to pack the varying.
781 unsigned is_xfb_only
:1;
784 * Was a transform feedback buffer set in the shader?
786 unsigned explicit_xfb_buffer
:1;
789 * Was a transform feedback offset set in the shader?
791 unsigned explicit_xfb_offset
:1;
794 * Was a transform feedback stride set in the shader?
796 unsigned explicit_xfb_stride
:1;
799 * If non-zero, then this variable may be packed along with other variables
800 * into a single varying slot, so this offset should be applied when
801 * accessing components. For example, an offset of 1 means that the x
802 * component of this variable is actually stored in component y of the
803 * location specified by \c location.
805 unsigned location_frac
:2;
808 * Layout of the matrix. Uses glsl_matrix_layout values.
810 unsigned matrix_layout
:2;
813 * Non-zero if this variable was created by lowering a named interface
816 unsigned from_named_ifc_block
:1;
819 * Non-zero if the variable must be a shader input. This is useful for
820 * constraints on function parameters.
822 unsigned must_be_shader_input
:1;
825 * Output index for dual source blending.
828 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
834 * Precision qualifier.
836 * In desktop GLSL we do not care about precision qualifiers at all, in
837 * fact, the spec says that precision qualifiers are ignored.
839 * To make things easy, we make it so that this field is always
840 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
841 * have the same precision value and the checks we add in the compiler
842 * for this field will never break a desktop shader compile.
844 unsigned precision
:2;
847 * \brief Layout qualifier for gl_FragDepth.
849 * This is not equal to \c ir_depth_layout_none if and only if this
850 * variable is \c gl_FragDepth and a layout qualifier is specified.
852 ir_depth_layout depth_layout
:3;
857 unsigned memory_read_only
:1; /**< "readonly" qualifier. */
858 unsigned memory_write_only
:1; /**< "writeonly" qualifier. */
859 unsigned memory_coherent
:1;
860 unsigned memory_volatile
:1;
861 unsigned memory_restrict
:1;
864 * ARB_shader_storage_buffer_object
866 unsigned from_ssbo_unsized_array
:1; /**< unsized array buffer variable. */
868 unsigned implicit_sized_array
:1;
871 * Whether this is a fragment shader output implicitly initialized with
872 * the previous contents of the specified render target at the
873 * framebuffer location corresponding to this shader invocation.
875 unsigned fb_fetch_output
:1;
878 * Non-zero if this variable is considered bindless as defined by
879 * ARB_bindless_texture.
884 * Non-zero if this variable is considered bound as defined by
885 * ARB_bindless_texture.
890 * Emit a warning if this variable is accessed.
893 uint8_t warn_extension_index
;
897 * Image internal format if specified explicitly, otherwise
900 enum pipe_format image_format
;
904 * Number of state slots used
907 * This could be stored in as few as 7-bits, if necessary. If it is made
908 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
911 uint16_t _num_state_slots
;
915 * Initial binding point for a sampler, atomic, or UBO.
917 * For array types, this represents the binding point for the first element.
922 * Storage location of the base of this variable
924 * The precise meaning of this field depends on the nature of the variable.
926 * - Vertex shader input: one of the values from \c gl_vert_attrib.
927 * - Vertex shader output: one of the values from \c gl_varying_slot.
928 * - Geometry shader input: one of the values from \c gl_varying_slot.
929 * - Geometry shader output: one of the values from \c gl_varying_slot.
930 * - Fragment shader input: one of the values from \c gl_varying_slot.
931 * - Fragment shader output: one of the values from \c gl_frag_result.
932 * - Uniforms: Per-stage uniform slot number for default uniform block.
933 * - Uniforms: Index within the uniform block definition for UBO members.
934 * - Non-UBO Uniforms: explicit location until linking then reused to
935 * store uniform slot number.
936 * - Other: This field is not currently used.
938 * If the variable is a uniform, shader input, or shader output, and the
939 * slot has not been assigned, the value will be -1.
944 * for glsl->tgsi/mesa IR we need to store the index into the
945 * parameters for uniforms, initially the code overloaded location
946 * but this causes problems with indirect samplers and AoA.
947 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
952 * Vertex stream output identifier.
954 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
955 * stream of the i-th component.
960 * Atomic, transform feedback or block member offset.
965 * Highest element accessed with a constant expression array index
967 * Not used for non-array variables. -1 is never accessed.
969 int max_array_access
;
972 * Transform feedback buffer.
977 * Transform feedback stride.
982 * Allow (only) ir_variable direct access private members.
984 friend class ir_variable
;
988 * Value assigned in the initializer of a variable declared "const"
990 ir_constant
*constant_value
;
993 * Constant expression assigned in the initializer of the variable
996 * This field and \c ::constant_value are distinct. Even if the two fields
997 * refer to constants with the same value, they must point to separate
1000 ir_constant
*constant_initializer
;
1003 static const char *const warn_extension_table
[];
1007 * For variables which satisfy the is_interface_instance() predicate,
1008 * this points to an array of integers such that if the ith member of
1009 * the interface block is an array, max_ifc_array_access[i] is the
1010 * maximum array element of that member that has been accessed. If the
1011 * ith member of the interface block is not an array,
1012 * max_ifc_array_access[i] is unused.
1014 * For variables whose type is not an interface block, this pointer is
1017 int *max_ifc_array_access
;
1020 * Built-in state that backs this uniform
1022 * Once set at variable creation, \c state_slots must remain invariant.
1024 * If the variable is not a uniform, \c _num_state_slots will be zero
1025 * and \c state_slots will be \c NULL.
1027 ir_state_slot
*state_slots
;
1031 * For variables that are in an interface block or are an instance of an
1032 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
1034 * \sa ir_variable::location
1036 const glsl_type
*interface_type
;
1039 * Name used for anonymous compiler temporaries
1041 static const char tmp_name
[];
1045 * Should the construct keep names for ir_var_temporary variables?
1047 * When this global is false, names passed to the constructor for
1048 * \c ir_var_temporary variables will be dropped. Instead, the variable will
1049 * be named "compiler_temp". This name will be in static storage.
1052 * \b NEVER change the mode of an \c ir_var_temporary.
1055 * This variable is \b not thread-safe. It is global, \b not
1056 * per-context. It begins life false. A context can, at some point, make
1057 * it true. From that point on, it will be true forever. This should be
1058 * okay since it will only be set true while debugging.
1060 static bool temporaries_allocate_names
;
1064 * A function that returns whether a built-in function is available in the
1065 * current shading language (based on version, ES or desktop, and extensions).
1067 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
1069 #define MAKE_INTRINSIC_FOR_TYPE(op, t) \
1070 ir_intrinsic_generic_ ## op - ir_intrinsic_generic_load + ir_intrinsic_ ## t ## _ ## load
1072 #define MAP_INTRINSIC_TO_TYPE(i, t) \
1073 ir_intrinsic_id(int(i) - int(ir_intrinsic_generic_load) + int(ir_intrinsic_ ## t ## _ ## load))
1075 enum ir_intrinsic_id
{
1076 ir_intrinsic_invalid
= 0,
1079 * \name Generic intrinsics
1081 * Each of these intrinsics has a specific version for shared variables and
1085 ir_intrinsic_generic_load
,
1086 ir_intrinsic_generic_store
,
1087 ir_intrinsic_generic_atomic_add
,
1088 ir_intrinsic_generic_atomic_and
,
1089 ir_intrinsic_generic_atomic_or
,
1090 ir_intrinsic_generic_atomic_xor
,
1091 ir_intrinsic_generic_atomic_min
,
1092 ir_intrinsic_generic_atomic_max
,
1093 ir_intrinsic_generic_atomic_exchange
,
1094 ir_intrinsic_generic_atomic_comp_swap
,
1097 ir_intrinsic_atomic_counter_read
,
1098 ir_intrinsic_atomic_counter_increment
,
1099 ir_intrinsic_atomic_counter_predecrement
,
1100 ir_intrinsic_atomic_counter_add
,
1101 ir_intrinsic_atomic_counter_and
,
1102 ir_intrinsic_atomic_counter_or
,
1103 ir_intrinsic_atomic_counter_xor
,
1104 ir_intrinsic_atomic_counter_min
,
1105 ir_intrinsic_atomic_counter_max
,
1106 ir_intrinsic_atomic_counter_exchange
,
1107 ir_intrinsic_atomic_counter_comp_swap
,
1109 ir_intrinsic_image_load
,
1110 ir_intrinsic_image_store
,
1111 ir_intrinsic_image_atomic_add
,
1112 ir_intrinsic_image_atomic_and
,
1113 ir_intrinsic_image_atomic_or
,
1114 ir_intrinsic_image_atomic_xor
,
1115 ir_intrinsic_image_atomic_min
,
1116 ir_intrinsic_image_atomic_max
,
1117 ir_intrinsic_image_atomic_exchange
,
1118 ir_intrinsic_image_atomic_comp_swap
,
1119 ir_intrinsic_image_size
,
1120 ir_intrinsic_image_samples
,
1121 ir_intrinsic_image_atomic_inc_wrap
,
1122 ir_intrinsic_image_atomic_dec_wrap
,
1124 ir_intrinsic_ssbo_load
,
1125 ir_intrinsic_ssbo_store
= MAKE_INTRINSIC_FOR_TYPE(store
, ssbo
),
1126 ir_intrinsic_ssbo_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, ssbo
),
1127 ir_intrinsic_ssbo_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, ssbo
),
1128 ir_intrinsic_ssbo_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, ssbo
),
1129 ir_intrinsic_ssbo_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, ssbo
),
1130 ir_intrinsic_ssbo_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, ssbo
),
1131 ir_intrinsic_ssbo_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, ssbo
),
1132 ir_intrinsic_ssbo_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, ssbo
),
1133 ir_intrinsic_ssbo_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, ssbo
),
1135 ir_intrinsic_memory_barrier
,
1136 ir_intrinsic_shader_clock
,
1137 ir_intrinsic_group_memory_barrier
,
1138 ir_intrinsic_memory_barrier_atomic_counter
,
1139 ir_intrinsic_memory_barrier_buffer
,
1140 ir_intrinsic_memory_barrier_image
,
1141 ir_intrinsic_memory_barrier_shared
,
1142 ir_intrinsic_begin_invocation_interlock
,
1143 ir_intrinsic_end_invocation_interlock
,
1145 ir_intrinsic_vote_all
,
1146 ir_intrinsic_vote_any
,
1147 ir_intrinsic_vote_eq
,
1148 ir_intrinsic_ballot
,
1149 ir_intrinsic_read_invocation
,
1150 ir_intrinsic_read_first_invocation
,
1152 ir_intrinsic_helper_invocation
,
1154 ir_intrinsic_shared_load
,
1155 ir_intrinsic_shared_store
= MAKE_INTRINSIC_FOR_TYPE(store
, shared
),
1156 ir_intrinsic_shared_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, shared
),
1157 ir_intrinsic_shared_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, shared
),
1158 ir_intrinsic_shared_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, shared
),
1159 ir_intrinsic_shared_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, shared
),
1160 ir_intrinsic_shared_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, shared
),
1161 ir_intrinsic_shared_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, shared
),
1162 ir_intrinsic_shared_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, shared
),
1163 ir_intrinsic_shared_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, shared
),
1168 * The representation of a function instance; may be the full definition or
1169 * simply a prototype.
1171 class ir_function_signature
: public ir_instruction
{
1172 /* An ir_function_signature will be part of the list of signatures in
1176 ir_function_signature(const glsl_type
*return_type
,
1177 builtin_available_predicate builtin_avail
= NULL
);
1179 virtual ir_function_signature
*clone(void *mem_ctx
,
1180 struct hash_table
*ht
) const;
1181 ir_function_signature
*clone_prototype(void *mem_ctx
,
1182 struct hash_table
*ht
) const;
1184 virtual void accept(ir_visitor
*v
)
1189 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1192 * Attempt to evaluate this function as a constant expression,
1193 * given a list of the actual parameters and the variable context.
1194 * Returns NULL for non-built-ins.
1196 ir_constant
*constant_expression_value(void *mem_ctx
,
1197 exec_list
*actual_parameters
,
1198 struct hash_table
*variable_context
);
1201 * Get the name of the function for which this is a signature
1203 const char *function_name() const;
1206 * Get a handle to the function for which this is a signature
1208 * There is no setter function, this function returns a \c const pointer,
1209 * and \c ir_function_signature::_function is private for a reason. The
1210 * only way to make a connection between a function and function signature
1211 * is via \c ir_function::add_signature. This helps ensure that certain
1212 * invariants (i.e., a function signature is in the list of signatures for
1213 * its \c _function) are met.
1215 * \sa ir_function::add_signature
1217 inline const class ir_function
*function() const
1219 return this->_function
;
1223 * Check whether the qualifiers match between this signature's parameters
1224 * and the supplied parameter list. If not, returns the name of the first
1225 * parameter with mismatched qualifiers (for use in error messages).
1227 const char *qualifiers_match(exec_list
*params
);
1230 * Replace the current parameter list with the given one. This is useful
1231 * if the current information came from a prototype, and either has invalid
1232 * or missing parameter names.
1234 void replace_parameters(exec_list
*new_params
);
1237 * Function return type.
1239 * \note The precision qualifier is stored separately in return_precision.
1241 const struct glsl_type
*return_type
;
1244 * List of ir_variable of function parameters.
1246 * This represents the storage. The paramaters passed in a particular
1247 * call will be in ir_call::actual_paramaters.
1249 struct exec_list parameters
;
1251 /** Whether or not this function has a body (which may be empty). */
1252 unsigned is_defined
:1;
1255 * Precision qualifier for the return type.
1257 * See the comment for ir_variable_data::precision for more details.
1259 unsigned return_precision
:2;
1261 /** Whether or not this function signature is a built-in. */
1262 bool is_builtin() const;
1265 * Whether or not this function is an intrinsic to be implemented
1268 inline bool is_intrinsic() const
1270 return intrinsic_id
!= ir_intrinsic_invalid
;
1273 /** Indentifier for this intrinsic. */
1274 enum ir_intrinsic_id intrinsic_id
;
1276 /** Whether or not a built-in is available for this shader. */
1277 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1279 /** Body of instructions in the function. */
1280 struct exec_list body
;
1284 * A function pointer to a predicate that answers whether a built-in
1285 * function is available in the current shader. NULL if not a built-in.
1287 builtin_available_predicate builtin_avail
;
1289 /** Function of which this signature is one overload. */
1290 class ir_function
*_function
;
1292 /** Function signature of which this one is a prototype clone */
1293 const ir_function_signature
*origin
;
1295 friend class ir_function
;
1298 * Helper function to run a list of instructions for constant
1299 * expression evaluation.
1301 * The hash table represents the values of the visible variables.
1302 * There are no scoping issues because the table is indexed on
1303 * ir_variable pointers, not variable names.
1305 * Returns false if the expression is not constant, true otherwise,
1306 * and the value in *result if result is non-NULL.
1308 bool constant_expression_evaluate_expression_list(void *mem_ctx
,
1309 const struct exec_list
&body
,
1310 struct hash_table
*variable_context
,
1311 ir_constant
**result
);
1316 * Header for tracking multiple overloaded functions with the same name.
1317 * Contains a list of ir_function_signatures representing each of the
1320 class ir_function
: public ir_instruction
{
1322 ir_function(const char *name
);
1324 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1326 virtual void accept(ir_visitor
*v
)
1331 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1333 void add_signature(ir_function_signature
*sig
)
1335 sig
->_function
= this;
1336 this->signatures
.push_tail(sig
);
1340 * Find a signature that matches a set of actual parameters, taking implicit
1341 * conversions into account. Also flags whether the match was exact.
1343 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1344 const exec_list
*actual_param
,
1345 bool allow_builtins
,
1346 bool *match_is_exact
);
1349 * Find a signature that matches a set of actual parameters, taking implicit
1350 * conversions into account.
1352 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1353 const exec_list
*actual_param
,
1354 bool allow_builtins
);
1357 * Find a signature that exactly matches a set of actual parameters without
1358 * any implicit type conversions.
1360 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1361 const exec_list
*actual_ps
);
1364 * Name of the function.
1368 /** Whether or not this function has a signature that isn't a built-in. */
1369 bool has_user_signature();
1372 * List of ir_function_signature for each overloaded function with this name.
1374 struct exec_list signatures
;
1377 * is this function a subroutine type declaration
1378 * e.g. subroutine void type1(float arg1);
1383 * is this function associated to a subroutine type
1384 * e.g. subroutine (type1, type2) function_name { function_body };
1385 * would have num_subroutine_types 2,
1386 * and pointers to the type1 and type2 types.
1388 int num_subroutine_types
;
1389 const struct glsl_type
**subroutine_types
;
1391 int subroutine_index
;
1394 inline const char *ir_function_signature::function_name() const
1396 return this->_function
->name
;
1402 * IR instruction representing high-level if-statements
1404 class ir_if
: public ir_instruction
{
1406 ir_if(ir_rvalue
*condition
)
1407 : ir_instruction(ir_type_if
), condition(condition
)
1411 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1413 virtual void accept(ir_visitor
*v
)
1418 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1420 ir_rvalue
*condition
;
1421 /** List of ir_instruction for the body of the then branch */
1422 exec_list then_instructions
;
1423 /** List of ir_instruction for the body of the else branch */
1424 exec_list else_instructions
;
1429 * IR instruction representing a high-level loop structure.
1431 class ir_loop
: public ir_instruction
{
1435 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1437 virtual void accept(ir_visitor
*v
)
1442 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1444 /** List of ir_instruction that make up the body of the loop. */
1445 exec_list body_instructions
;
1449 class ir_assignment
: public ir_instruction
{
1451 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1454 * Construct an assignment with an explicit write mask
1457 * Since a write mask is supplied, the LHS must already be a bare
1458 * \c ir_dereference. The cannot be any swizzles in the LHS.
1460 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1461 unsigned write_mask
);
1463 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1465 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1466 struct hash_table
*variable_context
= NULL
);
1468 virtual void accept(ir_visitor
*v
)
1473 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1476 * Get a whole variable written by an assignment
1478 * If the LHS of the assignment writes a whole variable, the variable is
1479 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1482 * - Assigning to a scalar
1483 * - Assigning to all components of a vector
1484 * - Whole array (or matrix) assignment
1485 * - Whole structure assignment
1487 ir_variable
*whole_variable_written();
1490 * Set the LHS of an assignment
1492 void set_lhs(ir_rvalue
*lhs
);
1495 * Left-hand side of the assignment.
1497 * This should be treated as read only. If you need to set the LHS of an
1498 * assignment, use \c ir_assignment::set_lhs.
1500 ir_dereference
*lhs
;
1503 * Value being assigned
1508 * Optional condition for the assignment.
1510 ir_rvalue
*condition
;
1514 * Component mask written
1516 * For non-vector types in the LHS, this field will be zero. For vector
1517 * types, a bit will be set for each component that is written. Note that
1518 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1520 * A partially-set write mask means that each enabled channel gets
1521 * the value from a consecutive channel of the rhs. For example,
1522 * to write just .xyw of gl_FrontColor with color:
1524 * (assign (constant bool (1)) (xyw)
1525 * (var_ref gl_FragColor)
1526 * (swiz xyw (var_ref color)))
1528 unsigned write_mask
:4;
1531 #include "ir_expression_operation.h"
1533 extern const char *const ir_expression_operation_strings
[ir_last_opcode
+ 1];
1534 extern const char *const ir_expression_operation_enum_strings
[ir_last_opcode
+ 1];
1536 class ir_expression
: public ir_rvalue
{
1538 ir_expression(int op
, const struct glsl_type
*type
,
1539 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1540 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1543 * Constructor for unary operation expressions
1545 ir_expression(int op
, ir_rvalue
*);
1548 * Constructor for binary operation expressions
1550 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1553 * Constructor for ternary operation expressions
1555 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1557 virtual bool equals(const ir_instruction
*ir
,
1558 enum ir_node_type ignore
= ir_type_unset
) const;
1560 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1563 * Attempt to constant-fold the expression
1565 * The "variable_context" hash table links ir_variable * to ir_constant *
1566 * that represent the variables' values. \c NULL represents an empty
1569 * If the expression cannot be constant folded, this method will return
1572 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1573 struct hash_table
*variable_context
= NULL
);
1576 * This is only here for ir_reader to used for testing purposes please use
1577 * the precomputed num_operands field if you need the number of operands.
1579 static unsigned get_num_operands(ir_expression_operation
);
1582 * Return whether the expression operates on vectors horizontally.
1584 bool is_horizontal() const
1586 return operation
== ir_binop_all_equal
||
1587 operation
== ir_binop_any_nequal
||
1588 operation
== ir_binop_dot
||
1589 operation
== ir_binop_vector_extract
||
1590 operation
== ir_triop_vector_insert
||
1591 operation
== ir_binop_ubo_load
||
1592 operation
== ir_quadop_vector
;
1596 * Do a reverse-lookup to translate the given string into an operator.
1598 static ir_expression_operation
get_operator(const char *);
1600 virtual void accept(ir_visitor
*v
)
1605 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1607 virtual ir_variable
*variable_referenced() const;
1610 * Determine the number of operands used by an expression
1612 void init_num_operands()
1614 if (operation
== ir_quadop_vector
) {
1615 num_operands
= this->type
->vector_elements
;
1617 num_operands
= get_num_operands(operation
);
1621 ir_expression_operation operation
;
1622 ir_rvalue
*operands
[4];
1623 uint8_t num_operands
;
1628 * HIR instruction representing a high-level function call, containing a list
1629 * of parameters and returning a value in the supplied temporary.
1631 class ir_call
: public ir_instruction
{
1633 ir_call(ir_function_signature
*callee
,
1634 ir_dereference_variable
*return_deref
,
1635 exec_list
*actual_parameters
)
1636 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(NULL
), array_idx(NULL
)
1638 assert(callee
->return_type
!= NULL
);
1639 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1642 ir_call(ir_function_signature
*callee
,
1643 ir_dereference_variable
*return_deref
,
1644 exec_list
*actual_parameters
,
1645 ir_variable
*var
, ir_rvalue
*array_idx
)
1646 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(var
), array_idx(array_idx
)
1648 assert(callee
->return_type
!= NULL
);
1649 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1652 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1654 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1655 struct hash_table
*variable_context
= NULL
);
1657 virtual void accept(ir_visitor
*v
)
1662 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1665 * Get the name of the function being called.
1667 const char *callee_name() const
1669 return callee
->function_name();
1673 * Generates an inline version of the function before @ir,
1674 * storing the return value in return_deref.
1676 void generate_inline(ir_instruction
*ir
);
1679 * Storage for the function's return value.
1680 * This must be NULL if the return type is void.
1682 ir_dereference_variable
*return_deref
;
1685 * The specific function signature being called.
1687 ir_function_signature
*callee
;
1689 /* List of ir_rvalue of paramaters passed in this call. */
1690 exec_list actual_parameters
;
1693 * ARB_shader_subroutine support -
1694 * the subroutine uniform variable and array index
1695 * rvalue to be used in the lowering pass later.
1697 ir_variable
*sub_var
;
1698 ir_rvalue
*array_idx
;
1703 * \name Jump-like IR instructions.
1705 * These include \c break, \c continue, \c return, and \c discard.
1708 class ir_jump
: public ir_instruction
{
1710 ir_jump(enum ir_node_type t
)
1716 class ir_return
: public ir_jump
{
1719 : ir_jump(ir_type_return
), value(NULL
)
1723 ir_return(ir_rvalue
*value
)
1724 : ir_jump(ir_type_return
), value(value
)
1728 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1730 ir_rvalue
*get_value() const
1735 virtual void accept(ir_visitor
*v
)
1740 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1747 * Jump instructions used inside loops
1749 * These include \c break and \c continue. The \c break within a loop is
1750 * different from the \c break within a switch-statement.
1752 * \sa ir_switch_jump
1754 class ir_loop_jump
: public ir_jump
{
1761 ir_loop_jump(jump_mode mode
)
1762 : ir_jump(ir_type_loop_jump
)
1767 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1769 virtual void accept(ir_visitor
*v
)
1774 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1776 bool is_break() const
1778 return mode
== jump_break
;
1781 bool is_continue() const
1783 return mode
== jump_continue
;
1786 /** Mode selector for the jump instruction. */
1787 enum jump_mode mode
;
1791 * IR instruction representing discard statements.
1793 class ir_discard
: public ir_jump
{
1796 : ir_jump(ir_type_discard
)
1798 this->condition
= NULL
;
1801 ir_discard(ir_rvalue
*cond
)
1802 : ir_jump(ir_type_discard
)
1804 this->condition
= cond
;
1807 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1809 virtual void accept(ir_visitor
*v
)
1814 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1816 ir_rvalue
*condition
;
1822 * IR instruction representing demote statements from
1823 * GL_EXT_demote_to_helper_invocation.
1825 class ir_demote
: public ir_instruction
{
1828 : ir_instruction(ir_type_demote
)
1832 virtual ir_demote
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1834 virtual void accept(ir_visitor
*v
)
1839 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1844 * Texture sampling opcodes used in ir_texture
1846 enum ir_texture_opcode
{
1847 ir_tex
, /**< Regular texture look-up */
1848 ir_txb
, /**< Texture look-up with LOD bias */
1849 ir_txl
, /**< Texture look-up with explicit LOD */
1850 ir_txd
, /**< Texture look-up with partial derivatvies */
1851 ir_txf
, /**< Texel fetch with explicit LOD */
1852 ir_txf_ms
, /**< Multisample texture fetch */
1853 ir_txs
, /**< Texture size */
1854 ir_lod
, /**< Texture lod query */
1855 ir_tg4
, /**< Texture gather */
1856 ir_query_levels
, /**< Texture levels query */
1857 ir_texture_samples
, /**< Texture samples query */
1858 ir_samples_identical
, /**< Query whether all samples are definitely identical. */
1863 * IR instruction to sample a texture
1865 * The specific form of the IR instruction depends on the \c mode value
1866 * selected from \c ir_texture_opcodes. In the printed IR, these will
1869 * Texel offset (0 or an expression)
1870 * | Projection divisor
1871 * | | Shadow comparator
1874 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1875 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1876 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1877 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1878 * (txf <type> <sampler> <coordinate> 0 <lod>)
1880 * <type> <sampler> <coordinate> <sample_index>)
1881 * (txs <type> <sampler> <lod>)
1882 * (lod <type> <sampler> <coordinate>)
1883 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1884 * (query_levels <type> <sampler>)
1885 * (samples_identical <sampler> <coordinate>)
1887 class ir_texture
: public ir_rvalue
{
1889 ir_texture(enum ir_texture_opcode op
)
1890 : ir_rvalue(ir_type_texture
),
1891 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1892 shadow_comparator(NULL
), offset(NULL
)
1894 memset(&lod_info
, 0, sizeof(lod_info
));
1897 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1899 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1900 struct hash_table
*variable_context
= NULL
);
1902 virtual void accept(ir_visitor
*v
)
1907 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1909 virtual bool equals(const ir_instruction
*ir
,
1910 enum ir_node_type ignore
= ir_type_unset
) const;
1913 * Return a string representing the ir_texture_opcode.
1915 const char *opcode_string();
1917 /** Set the sampler and type. */
1918 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1921 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1923 static ir_texture_opcode
get_opcode(const char *);
1925 enum ir_texture_opcode op
;
1927 /** Sampler to use for the texture access. */
1928 ir_dereference
*sampler
;
1930 /** Texture coordinate to sample */
1931 ir_rvalue
*coordinate
;
1934 * Value used for projective divide.
1936 * If there is no projective divide (the common case), this will be
1937 * \c NULL. Optimization passes should check for this to point to a constant
1938 * of 1.0 and replace that with \c NULL.
1940 ir_rvalue
*projector
;
1943 * Coordinate used for comparison on shadow look-ups.
1945 * If there is no shadow comparison, this will be \c NULL. For the
1946 * \c ir_txf opcode, this *must* be \c NULL.
1948 ir_rvalue
*shadow_comparator
;
1950 /** Texel offset. */
1954 ir_rvalue
*lod
; /**< Floating point LOD */
1955 ir_rvalue
*bias
; /**< Floating point LOD bias */
1956 ir_rvalue
*sample_index
; /**< MSAA sample index */
1957 ir_rvalue
*component
; /**< Gather component selector */
1959 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1960 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1966 struct ir_swizzle_mask
{
1973 * Number of components in the swizzle.
1975 unsigned num_components
:3;
1978 * Does the swizzle contain duplicate components?
1980 * L-value swizzles cannot contain duplicate components.
1982 unsigned has_duplicates
:1;
1986 class ir_swizzle
: public ir_rvalue
{
1988 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
1991 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
1993 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
1995 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
1997 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1998 struct hash_table
*variable_context
= NULL
);
2001 * Construct an ir_swizzle from the textual representation. Can fail.
2003 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
2005 virtual void accept(ir_visitor
*v
)
2010 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2012 virtual bool equals(const ir_instruction
*ir
,
2013 enum ir_node_type ignore
= ir_type_unset
) const;
2015 bool is_lvalue(const struct _mesa_glsl_parse_state
*state
) const
2017 return val
->is_lvalue(state
) && !mask
.has_duplicates
;
2021 * Get the variable that is ultimately referenced by an r-value
2023 virtual ir_variable
*variable_referenced() const;
2026 ir_swizzle_mask mask
;
2030 * Initialize the mask component of a swizzle
2032 * This is used by the \c ir_swizzle constructors.
2034 void init_mask(const unsigned *components
, unsigned count
);
2038 class ir_dereference
: public ir_rvalue
{
2040 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2042 bool is_lvalue(const struct _mesa_glsl_parse_state
*state
) const;
2045 * Get the variable that is ultimately referenced by an r-value
2047 virtual ir_variable
*variable_referenced() const = 0;
2050 ir_dereference(enum ir_node_type t
)
2057 class ir_dereference_variable
: public ir_dereference
{
2059 ir_dereference_variable(ir_variable
*var
);
2061 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2062 struct hash_table
*) const;
2064 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2065 struct hash_table
*variable_context
= NULL
);
2067 virtual bool equals(const ir_instruction
*ir
,
2068 enum ir_node_type ignore
= ir_type_unset
) const;
2071 * Get the variable that is ultimately referenced by an r-value
2073 virtual ir_variable
*variable_referenced() const
2078 virtual ir_variable
*whole_variable_referenced()
2080 /* ir_dereference_variable objects always dereference the entire
2081 * variable. However, if this dereference is dereferenced by anything
2082 * else, the complete deferefernce chain is not a whole-variable
2083 * dereference. This method should only be called on the top most
2084 * ir_rvalue in a dereference chain.
2089 virtual void accept(ir_visitor
*v
)
2094 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2097 * Object being dereferenced.
2103 class ir_dereference_array
: public ir_dereference
{
2105 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2107 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2109 virtual ir_dereference_array
*clone(void *mem_ctx
,
2110 struct hash_table
*) const;
2112 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2113 struct hash_table
*variable_context
= NULL
);
2115 virtual bool equals(const ir_instruction
*ir
,
2116 enum ir_node_type ignore
= ir_type_unset
) const;
2119 * Get the variable that is ultimately referenced by an r-value
2121 virtual ir_variable
*variable_referenced() const
2123 return this->array
->variable_referenced();
2126 virtual void accept(ir_visitor
*v
)
2131 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2134 ir_rvalue
*array_index
;
2137 void set_array(ir_rvalue
*value
);
2141 class ir_dereference_record
: public ir_dereference
{
2143 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2145 ir_dereference_record(ir_variable
*var
, const char *field
);
2147 virtual ir_dereference_record
*clone(void *mem_ctx
,
2148 struct hash_table
*) const;
2150 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2151 struct hash_table
*variable_context
= NULL
);
2154 * Get the variable that is ultimately referenced by an r-value
2156 virtual ir_variable
*variable_referenced() const
2158 return this->record
->variable_referenced();
2161 virtual void accept(ir_visitor
*v
)
2166 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2174 * Data stored in an ir_constant
2176 union ir_constant_data
{
2187 class ir_constant
: public ir_rvalue
{
2189 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2190 ir_constant(bool b
, unsigned vector_elements
=1);
2191 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2192 ir_constant(int i
, unsigned vector_elements
=1);
2193 ir_constant(float f
, unsigned vector_elements
=1);
2194 ir_constant(double d
, unsigned vector_elements
=1);
2195 ir_constant(uint64_t u64
, unsigned vector_elements
=1);
2196 ir_constant(int64_t i64
, unsigned vector_elements
=1);
2199 * Construct an ir_constant from a list of ir_constant values
2201 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2204 * Construct an ir_constant from a scalar component of another ir_constant
2206 * The new \c ir_constant inherits the type of the component from the
2210 * In the case of a matrix constant, the new constant is a scalar, \b not
2213 ir_constant(const ir_constant
*c
, unsigned i
);
2216 * Return a new ir_constant of the specified type containing all zeros.
2218 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2220 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2222 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2223 struct hash_table
*variable_context
= NULL
);
2225 virtual void accept(ir_visitor
*v
)
2230 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2232 virtual bool equals(const ir_instruction
*ir
,
2233 enum ir_node_type ignore
= ir_type_unset
) const;
2236 * Get a particular component of a constant as a specific type
2238 * This is useful, for example, to get a value from an integer constant
2239 * as a float or bool. This appears frequently when constructors are
2240 * called with all constant parameters.
2243 bool get_bool_component(unsigned i
) const;
2244 float get_float_component(unsigned i
) const;
2245 double get_double_component(unsigned i
) const;
2246 int get_int_component(unsigned i
) const;
2247 unsigned get_uint_component(unsigned i
) const;
2248 int64_t get_int64_component(unsigned i
) const;
2249 uint64_t get_uint64_component(unsigned i
) const;
2252 ir_constant
*get_array_element(unsigned i
) const;
2254 ir_constant
*get_record_field(int idx
);
2257 * Copy the values on another constant at a given offset.
2259 * The offset is ignored for array or struct copies, it's only for
2260 * scalars or vectors into vectors or matrices.
2262 * With identical types on both sides and zero offset it's clone()
2263 * without creating a new object.
2266 void copy_offset(ir_constant
*src
, int offset
);
2269 * Copy the values on another constant at a given offset and
2270 * following an assign-like mask.
2272 * The mask is ignored for scalars.
2274 * Note that this function only handles what assign can handle,
2275 * i.e. at most a vector as source and a column of a matrix as
2279 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2282 * Determine whether a constant has the same value as another constant
2284 * \sa ir_constant::is_zero, ir_constant::is_one,
2285 * ir_constant::is_negative_one
2287 bool has_value(const ir_constant
*) const;
2290 * Return true if this ir_constant represents the given value.
2292 * For vectors, this checks that each component is the given value.
2294 virtual bool is_value(float f
, int i
) const;
2295 virtual bool is_zero() const;
2296 virtual bool is_one() const;
2297 virtual bool is_negative_one() const;
2300 * Return true for constants that could be stored as 16-bit unsigned values.
2302 * Note that this will return true even for signed integer ir_constants, as
2303 * long as the value is non-negative and fits in 16-bits.
2305 virtual bool is_uint16_constant() const;
2308 * Value of the constant.
2310 * The field used to back the values supplied by the constant is determined
2311 * by the type associated with the \c ir_instruction. Constants may be
2312 * scalars, vectors, or matrices.
2314 union ir_constant_data value
;
2316 /* Array elements and structure fields */
2317 ir_constant
**const_elements
;
2321 * Parameterless constructor only used by the clone method
2327 * IR instruction to emit a vertex in a geometry shader.
2329 class ir_emit_vertex
: public ir_instruction
{
2331 ir_emit_vertex(ir_rvalue
*stream
)
2332 : ir_instruction(ir_type_emit_vertex
),
2338 virtual void accept(ir_visitor
*v
)
2343 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2345 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2348 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2350 int stream_id() const
2352 return stream
->as_constant()->value
.i
[0];
2359 * IR instruction to complete the current primitive and start a new one in a
2362 class ir_end_primitive
: public ir_instruction
{
2364 ir_end_primitive(ir_rvalue
*stream
)
2365 : ir_instruction(ir_type_end_primitive
),
2371 virtual void accept(ir_visitor
*v
)
2376 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2378 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2381 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2383 int stream_id() const
2385 return stream
->as_constant()->value
.i
[0];
2392 * IR instruction for tessellation control and compute shader barrier.
2394 class ir_barrier
: public ir_instruction
{
2397 : ir_instruction(ir_type_barrier
)
2401 virtual void accept(ir_visitor
*v
)
2406 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2408 return new(mem_ctx
) ir_barrier();
2411 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2417 * Apply a visitor to each IR node in a list
2420 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2423 * Validate invariants on each IR node in a list
2425 void validate_ir_tree(exec_list
*instructions
);
2427 struct _mesa_glsl_parse_state
;
2428 struct gl_shader_program
;
2431 * Detect whether an unlinked shader contains static recursion
2433 * If the list of instructions is determined to contain static recursion,
2434 * \c _mesa_glsl_error will be called to emit error messages for each function
2435 * that is in the recursion cycle.
2438 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2439 exec_list
*instructions
);
2442 * Detect whether a linked shader contains static recursion
2444 * If the list of instructions is determined to contain static recursion,
2445 * \c link_error_printf will be called to emit error messages for each function
2446 * that is in the recursion cycle. In addition,
2447 * \c gl_shader_program::LinkStatus will be set to false.
2450 detect_recursion_linked(struct gl_shader_program
*prog
,
2451 exec_list
*instructions
);
2454 * Make a clone of each IR instruction in a list
2456 * \param in List of IR instructions that are to be cloned
2457 * \param out List to hold the cloned instructions
2460 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2463 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2464 struct _mesa_glsl_parse_state
*state
);
2467 reparent_ir(exec_list
*list
, void *mem_ctx
);
2470 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2471 gl_shader_stage shader_stage
);
2474 prototype_string(const glsl_type
*return_type
, const char *name
,
2475 exec_list
*parameters
);
2478 mode_string(const ir_variable
*var
);
2481 * Built-in / reserved GL variables names start with "gl_"
2484 is_gl_identifier(const char *s
)
2486 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2490 #endif /* __cplusplus */
2492 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2493 struct _mesa_glsl_parse_state
*state
);
2496 fprint_ir(FILE *f
, const void *instruction
);
2498 extern const struct gl_builtin_uniform_desc
*
2499 _mesa_glsl_get_builtin_uniform_desc(const char *name
);
2506 vertices_per_prim(GLenum prim
);