3 * Copyright © 2010 Intel Corporation
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
32 #include "util/ralloc.h"
33 #include "compiler/glsl_types.h"
35 #include "ir_visitor.h"
36 #include "ir_hierarchical_visitor.h"
37 #include "main/mtypes.h"
42 * \defgroup IR Intermediate representation nodes
50 * Each concrete class derived from \c ir_instruction has a value in this
51 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
52 * by the constructor. While using type tags is not very C++, it is extremely
53 * convenient. For example, during debugging you can simply inspect
54 * \c ir_instruction::ir_type to find out the actual type of the object.
56 * In addition, it is possible to use a switch-statement based on \c
57 * \c ir_instruction::ir_type to select different behavior for different object
58 * types. For functions that have only slight differences for several object
59 * types, this allows writing very straightforward, readable code.
62 ir_type_dereference_array
,
63 ir_type_dereference_record
,
64 ir_type_dereference_variable
,
73 ir_type_function_signature
,
80 ir_type_end_primitive
,
82 ir_type_max
, /**< maximum ir_type enum number, for validation */
83 ir_type_unset
= ir_type_max
88 * Base class of all IR instructions
90 class ir_instruction
: public exec_node
{
92 enum ir_node_type ir_type
;
95 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
96 * there's a virtual destructor present. Because we almost
97 * universally use ralloc for our memory management of
98 * ir_instructions, the destructor doesn't need to do any work.
100 virtual ~ir_instruction()
104 /** ir_print_visitor helper for debugging. */
105 void print(void) const;
106 void fprint(FILE *f
) const;
108 virtual void accept(ir_visitor
*) = 0;
109 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*) = 0;
110 virtual ir_instruction
*clone(void *mem_ctx
,
111 struct hash_table
*ht
) const = 0;
113 bool is_rvalue() const
115 return ir_type
== ir_type_dereference_array
||
116 ir_type
== ir_type_dereference_record
||
117 ir_type
== ir_type_dereference_variable
||
118 ir_type
== ir_type_constant
||
119 ir_type
== ir_type_expression
||
120 ir_type
== ir_type_swizzle
||
121 ir_type
== ir_type_texture
;
124 bool is_dereference() const
126 return ir_type
== ir_type_dereference_array
||
127 ir_type
== ir_type_dereference_record
||
128 ir_type
== ir_type_dereference_variable
;
133 return ir_type
== ir_type_loop_jump
||
134 ir_type
== ir_type_return
||
135 ir_type
== ir_type_discard
;
139 * \name IR instruction downcast functions
141 * These functions either cast the object to a derived class or return
142 * \c NULL if the object's type does not match the specified derived class.
143 * Additional downcast functions will be added as needed.
146 #define AS_BASE(TYPE) \
147 class ir_##TYPE *as_##TYPE() \
149 assume(this != NULL); \
150 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
152 const class ir_##TYPE *as_##TYPE() const \
154 assume(this != NULL); \
155 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
163 #define AS_CHILD(TYPE) \
164 class ir_##TYPE * as_##TYPE() \
166 assume(this != NULL); \
167 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
169 const class ir_##TYPE * as_##TYPE() const \
171 assume(this != NULL); \
172 return ir_type == ir_type_##TYPE ? (const ir_##TYPE *) this : NULL; \
176 AS_CHILD(dereference_array
)
177 AS_CHILD(dereference_variable
)
178 AS_CHILD(dereference_record
)
193 * IR equality method: Return true if the referenced instruction would
194 * return the same value as this one.
196 * This intended to be used for CSE and algebraic optimizations, on rvalues
197 * in particular. No support for other instruction types (assignments,
198 * jumps, calls, etc.) is planned.
200 virtual bool equals(const ir_instruction
*ir
,
201 enum ir_node_type ignore
= ir_type_unset
) const;
204 ir_instruction(enum ir_node_type t
)
212 assert(!"Should not get here.");
218 * The base class for all "values"/expression trees.
220 class ir_rvalue
: public ir_instruction
{
222 const struct glsl_type
*type
;
224 virtual ir_rvalue
*clone(void *mem_ctx
, struct hash_table
*) const;
226 virtual void accept(ir_visitor
*v
)
231 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
233 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
235 ir_rvalue
*as_rvalue_to_saturate();
237 virtual bool is_lvalue() const
243 * Get the variable that is ultimately referenced by an r-value
245 virtual ir_variable
*variable_referenced() const
252 * If an r-value is a reference to a whole variable, get that variable
255 * Pointer to a variable that is completely dereferenced by the r-value. If
256 * the r-value is not a dereference or the dereference does not access the
257 * entire variable (i.e., it's just one array element, struct field), \c NULL
260 virtual ir_variable
*whole_variable_referenced()
266 * Determine if an r-value has the value zero
268 * The base implementation of this function always returns \c false. The
269 * \c ir_constant class over-rides this function to return \c true \b only
270 * for vector and scalar types that have all elements set to the value
271 * zero (or \c false for booleans).
273 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
275 virtual bool is_zero() const;
278 * Determine if an r-value has the value one
280 * The base implementation of this function always returns \c false. The
281 * \c ir_constant class over-rides this function to return \c true \b only
282 * for vector and scalar types that have all elements set to the value
283 * one (or \c true for booleans).
285 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
287 virtual bool is_one() const;
290 * Determine if an r-value has the value negative one
292 * The base implementation of this function always returns \c false. The
293 * \c ir_constant class over-rides this function to return \c true \b only
294 * for vector and scalar types that have all elements set to the value
295 * negative one. For boolean types, the result is always \c false.
297 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
299 virtual bool is_negative_one() const;
302 * Determine if an r-value is an unsigned integer constant which can be
305 * \sa ir_constant::is_uint16_constant.
307 virtual bool is_uint16_constant() const { return false; }
310 * Return a generic value of error_type.
312 * Allocation will be performed with 'mem_ctx' as ralloc owner.
314 static ir_rvalue
*error_value(void *mem_ctx
);
317 ir_rvalue(enum ir_node_type t
);
322 * Variable storage classes
324 enum ir_variable_mode
{
325 ir_var_auto
= 0, /**< Function local variables and globals. */
326 ir_var_uniform
, /**< Variable declared as a uniform. */
327 ir_var_shader_storage
, /**< Variable declared as an ssbo. */
328 ir_var_shader_shared
, /**< Variable declared as shared. */
333 ir_var_function_inout
,
334 ir_var_const_in
, /**< "in" param that must be a constant expression */
335 ir_var_system_value
, /**< Ex: front-face, instance-id, etc. */
336 ir_var_temporary
, /**< Temporary variable generated during compilation. */
337 ir_var_mode_count
/**< Number of variable modes */
341 * Enum keeping track of how a variable was declared. For error checking of
342 * the gl_PerVertex redeclaration rules.
344 enum ir_var_declaration_type
{
346 * Normal declaration (for most variables, this means an explicit
347 * declaration. Exception: temporaries are always implicitly declared, but
348 * they still use ir_var_declared_normally).
350 * Note: an ir_variable that represents a named interface block uses
351 * ir_var_declared_normally.
353 ir_var_declared_normally
= 0,
356 * Variable was explicitly declared (or re-declared) in an unnamed
359 ir_var_declared_in_block
,
362 * Variable is an implicitly declared built-in that has not been explicitly
363 * re-declared by the shader.
365 ir_var_declared_implicitly
,
368 * Variable is implicitly generated by the compiler and should not be
369 * visible via the API.
375 * \brief Layout qualifiers for gl_FragDepth.
377 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
378 * with a layout qualifier.
380 enum ir_depth_layout
{
381 ir_depth_layout_none
, /**< No depth layout is specified. */
383 ir_depth_layout_greater
,
384 ir_depth_layout_less
,
385 ir_depth_layout_unchanged
389 * \brief Convert depth layout qualifier to string.
392 depth_layout_string(ir_depth_layout layout
);
395 * Description of built-in state associated with a uniform
397 * \sa ir_variable::state_slots
399 struct ir_state_slot
{
406 * Get the string value for an interpolation qualifier
408 * \return The string that would be used in a shader to specify \c
409 * mode will be returned.
411 * This function is used to generate error messages of the form "shader
412 * uses %s interpolation qualifier", so in the case where there is no
413 * interpolation qualifier, it returns "no".
415 * This function should only be used on a shader input or output variable.
417 const char *interpolation_string(unsigned interpolation
);
420 class ir_variable
: public ir_instruction
{
422 ir_variable(const struct glsl_type
*, const char *, ir_variable_mode
);
424 virtual ir_variable
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
426 virtual void accept(ir_visitor
*v
)
431 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
435 * Determine whether or not a variable is part of a uniform or
436 * shader storage block.
438 inline bool is_in_buffer_block() const
440 return (this->data
.mode
== ir_var_uniform
||
441 this->data
.mode
== ir_var_shader_storage
) &&
442 this->interface_type
!= NULL
;
446 * Determine whether or not a variable is part of a shader storage block.
448 inline bool is_in_shader_storage_block() const
450 return this->data
.mode
== ir_var_shader_storage
&&
451 this->interface_type
!= NULL
;
455 * Determine whether or not a variable is the declaration of an interface
458 * For the first declaration below, there will be an \c ir_variable named
459 * "instance" whose type and whose instance_type will be the same
460 * \cglsl_type. For the second declaration, there will be an \c ir_variable
461 * named "f" whose type is float and whose instance_type is B2.
463 * "instance" is an interface instance variable, but "f" is not.
473 inline bool is_interface_instance() const
475 return this->type
->without_array() == this->interface_type
;
479 * Set this->interface_type on a newly created variable.
481 void init_interface_type(const struct glsl_type
*type
)
483 assert(this->interface_type
== NULL
);
484 this->interface_type
= type
;
485 if (this->is_interface_instance()) {
486 this->u
.max_ifc_array_access
=
487 ralloc_array(this, int, type
->length
);
488 for (unsigned i
= 0; i
< type
->length
; i
++) {
489 this->u
.max_ifc_array_access
[i
] = -1;
495 * Change this->interface_type on a variable that previously had a
496 * different, but compatible, interface_type. This is used during linking
497 * to set the size of arrays in interface blocks.
499 void change_interface_type(const struct glsl_type
*type
)
501 if (this->u
.max_ifc_array_access
!= NULL
) {
502 /* max_ifc_array_access has already been allocated, so make sure the
503 * new interface has the same number of fields as the old one.
505 assert(this->interface_type
->length
== type
->length
);
507 this->interface_type
= type
;
511 * Change this->interface_type on a variable that previously had a
512 * different, and incompatible, interface_type. This is used during
513 * compilation to handle redeclaration of the built-in gl_PerVertex
516 void reinit_interface_type(const struct glsl_type
*type
)
518 if (this->u
.max_ifc_array_access
!= NULL
) {
520 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
521 * it defines have been accessed yet; so it's safe to throw away the
522 * old max_ifc_array_access pointer, since all of its values are
525 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
526 assert(this->u
.max_ifc_array_access
[i
] == -1);
528 ralloc_free(this->u
.max_ifc_array_access
);
529 this->u
.max_ifc_array_access
= NULL
;
531 this->interface_type
= NULL
;
532 init_interface_type(type
);
535 const glsl_type
*get_interface_type() const
537 return this->interface_type
;
540 enum glsl_interface_packing
get_interface_type_packing() const
542 return this->interface_type
->get_interface_packing();
545 * Get the max_ifc_array_access pointer
547 * A "set" function is not needed because the array is dynmically allocated
550 inline int *get_max_ifc_array_access()
552 assert(this->data
._num_state_slots
== 0);
553 return this->u
.max_ifc_array_access
;
556 inline unsigned get_num_state_slots() const
558 assert(!this->is_interface_instance()
559 || this->data
._num_state_slots
== 0);
560 return this->data
._num_state_slots
;
563 inline void set_num_state_slots(unsigned n
)
565 assert(!this->is_interface_instance()
567 this->data
._num_state_slots
= n
;
570 inline ir_state_slot
*get_state_slots()
572 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
575 inline const ir_state_slot
*get_state_slots() const
577 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
580 inline ir_state_slot
*allocate_state_slots(unsigned n
)
582 assert(!this->is_interface_instance());
584 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
585 this->data
._num_state_slots
= 0;
587 if (this->u
.state_slots
!= NULL
)
588 this->data
._num_state_slots
= n
;
590 return this->u
.state_slots
;
593 inline bool is_interpolation_flat() const
595 return this->data
.interpolation
== INTERP_MODE_FLAT
||
596 this->type
->contains_integer() ||
597 this->type
->contains_double();
600 inline bool is_name_ralloced() const
602 return this->name
!= ir_variable::tmp_name
&&
603 this->name
!= this->name_storage
;
607 * Enable emitting extension warnings for this variable
609 void enable_extension_warning(const char *extension
);
612 * Get the extension warning string for this variable
614 * If warnings are not enabled, \c NULL is returned.
616 const char *get_extension_warning() const;
619 * Declared type of the variable
621 const struct glsl_type
*type
;
624 * Declared name of the variable
630 * If the name length fits into name_storage, it's used, otherwise
631 * the name is ralloc'd. shader-db mining showed that 70% of variables
632 * fit here. This is a win over ralloc where only ralloc_header has
633 * 20 bytes on 64-bit (28 bytes with DEBUG), and we can also skip malloc.
635 char name_storage
[16];
638 struct ir_variable_data
{
641 * Is the variable read-only?
643 * This is set for variables declared as \c const, shader inputs,
646 unsigned read_only
:1;
650 unsigned invariant
:1;
654 * Has this variable been used for reading or writing?
656 * Several GLSL semantic checks require knowledge of whether or not a
657 * variable has been used. For example, it is an error to redeclare a
658 * variable as invariant after it has been used.
660 * This is only maintained in the ast_to_hir.cpp path, not in
661 * Mesa's fixed function or ARB program paths.
666 * Has this variable been statically assigned?
668 * This answers whether the variable was assigned in any path of
669 * the shader during ast_to_hir. This doesn't answer whether it is
670 * still written after dead code removal, nor is it maintained in
671 * non-ast_to_hir.cpp (GLSL parsing) paths.
676 * When separate shader programs are enabled, only input/outputs between
677 * the stages of a multi-stage separate program can be safely removed
678 * from the shader interface. Other input/outputs must remains active.
680 unsigned always_active_io
:1;
683 * Enum indicating how the variable was declared. See
684 * ir_var_declaration_type.
686 * This is used to detect certain kinds of illegal variable redeclarations.
688 unsigned how_declared
:2;
691 * Storage class of the variable.
693 * \sa ir_variable_mode
698 * Interpolation mode for shader inputs / outputs
700 * \sa glsl_interp_mode
702 unsigned interpolation
:2;
705 * \name ARB_fragment_coord_conventions
708 unsigned origin_upper_left
:1;
709 unsigned pixel_center_integer
:1;
713 * Was the location explicitly set in the shader?
715 * If the location is explicitly set in the shader, it \b cannot be changed
716 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
719 unsigned explicit_location
:1;
720 unsigned explicit_index
:1;
723 * Was an initial binding explicitly set in the shader?
725 * If so, constant_value contains an integer ir_constant representing the
726 * initial binding point.
728 unsigned explicit_binding
:1;
731 * Was an initial component explicitly set in the shader?
733 unsigned explicit_component
:1;
736 * Does this variable have an initializer?
738 * This is used by the linker to cross-validiate initializers of global
741 unsigned has_initializer
:1;
744 * Is this variable a generic output or input that has not yet been matched
745 * up to a variable in another stage of the pipeline?
747 * This is used by the linker as scratch storage while assigning locations
748 * to generic inputs and outputs.
750 unsigned is_unmatched_generic_inout
:1;
753 * Is this varying used only by transform feedback?
755 * This is used by the linker to decide if its safe to pack the varying.
757 unsigned is_xfb_only
:1;
760 * Was a transfor feedback buffer set in the shader?
762 unsigned explicit_xfb_buffer
:1;
765 * Was a transfor feedback offset set in the shader?
767 unsigned explicit_xfb_offset
:1;
770 * Was a transfor feedback stride set in the shader?
772 unsigned explicit_xfb_stride
:1;
775 * If non-zero, then this variable may be packed along with other variables
776 * into a single varying slot, so this offset should be applied when
777 * accessing components. For example, an offset of 1 means that the x
778 * component of this variable is actually stored in component y of the
779 * location specified by \c location.
781 unsigned location_frac
:2;
784 * Layout of the matrix. Uses glsl_matrix_layout values.
786 unsigned matrix_layout
:2;
789 * Non-zero if this variable was created by lowering a named interface
792 unsigned from_named_ifc_block
:1;
795 * Non-zero if the variable must be a shader input. This is useful for
796 * constraints on function parameters.
798 unsigned must_be_shader_input
:1;
801 * Output index for dual source blending.
804 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
810 * Precision qualifier.
812 * In desktop GLSL we do not care about precision qualifiers at all, in
813 * fact, the spec says that precision qualifiers are ignored.
815 * To make things easy, we make it so that this field is always
816 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
817 * have the same precision value and the checks we add in the compiler
818 * for this field will never break a desktop shader compile.
820 unsigned precision
:2;
823 * \brief Layout qualifier for gl_FragDepth.
825 * This is not equal to \c ir_depth_layout_none if and only if this
826 * variable is \c gl_FragDepth and a layout qualifier is specified.
828 ir_depth_layout depth_layout
:3;
831 * ARB_shader_image_load_store qualifiers.
833 unsigned image_read_only
:1; /**< "readonly" qualifier. */
834 unsigned image_write_only
:1; /**< "writeonly" qualifier. */
835 unsigned image_coherent
:1;
836 unsigned image_volatile
:1;
837 unsigned image_restrict
:1;
840 * ARB_shader_storage_buffer_object
842 unsigned from_ssbo_unsized_array
:1; /**< unsized array buffer variable. */
844 unsigned implicit_sized_array
:1;
847 * Is this a non-patch TCS output / TES input array that was implicitly
848 * sized to gl_MaxPatchVertices?
850 unsigned tess_varying_implicit_sized_array
:1;
853 * Whether this is a fragment shader output implicitly initialized with
854 * the previous contents of the specified render target at the
855 * framebuffer location corresponding to this shader invocation.
857 unsigned fb_fetch_output
:1;
860 * Emit a warning if this variable is accessed.
863 uint8_t warn_extension_index
;
866 /** Image internal format if specified explicitly, otherwise GL_NONE. */
867 uint16_t image_format
;
871 * Number of state slots used
874 * This could be stored in as few as 7-bits, if necessary. If it is made
875 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
878 uint16_t _num_state_slots
;
882 * Initial binding point for a sampler, atomic, or UBO.
884 * For array types, this represents the binding point for the first element.
889 * Storage location of the base of this variable
891 * The precise meaning of this field depends on the nature of the variable.
893 * - Vertex shader input: one of the values from \c gl_vert_attrib.
894 * - Vertex shader output: one of the values from \c gl_varying_slot.
895 * - Geometry shader input: one of the values from \c gl_varying_slot.
896 * - Geometry shader output: one of the values from \c gl_varying_slot.
897 * - Fragment shader input: one of the values from \c gl_varying_slot.
898 * - Fragment shader output: one of the values from \c gl_frag_result.
899 * - Uniforms: Per-stage uniform slot number for default uniform block.
900 * - Uniforms: Index within the uniform block definition for UBO members.
901 * - Non-UBO Uniforms: explicit location until linking then reused to
902 * store uniform slot number.
903 * - Other: This field is not currently used.
905 * If the variable is a uniform, shader input, or shader output, and the
906 * slot has not been assigned, the value will be -1.
911 * for glsl->tgsi/mesa IR we need to store the index into the
912 * parameters for uniforms, initially the code overloaded location
913 * but this causes problems with indirect samplers and AoA.
914 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
919 * Vertex stream output identifier.
924 * Atomic, transform feedback or block member offset.
929 * Highest element accessed with a constant expression array index
931 * Not used for non-array variables. -1 is never accessed.
933 int max_array_access
;
936 * Transform feedback buffer.
941 * Transform feedback stride.
946 * Allow (only) ir_variable direct access private members.
948 friend class ir_variable
;
952 * Value assigned in the initializer of a variable declared "const"
954 ir_constant
*constant_value
;
957 * Constant expression assigned in the initializer of the variable
960 * This field and \c ::constant_value are distinct. Even if the two fields
961 * refer to constants with the same value, they must point to separate
964 ir_constant
*constant_initializer
;
967 static const char *const warn_extension_table
[];
971 * For variables which satisfy the is_interface_instance() predicate,
972 * this points to an array of integers such that if the ith member of
973 * the interface block is an array, max_ifc_array_access[i] is the
974 * maximum array element of that member that has been accessed. If the
975 * ith member of the interface block is not an array,
976 * max_ifc_array_access[i] is unused.
978 * For variables whose type is not an interface block, this pointer is
981 int *max_ifc_array_access
;
984 * Built-in state that backs this uniform
986 * Once set at variable creation, \c state_slots must remain invariant.
988 * If the variable is not a uniform, \c _num_state_slots will be zero
989 * and \c state_slots will be \c NULL.
991 ir_state_slot
*state_slots
;
995 * For variables that are in an interface block or are an instance of an
996 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
998 * \sa ir_variable::location
1000 const glsl_type
*interface_type
;
1003 * Name used for anonymous compiler temporaries
1005 static const char tmp_name
[];
1009 * Should the construct keep names for ir_var_temporary variables?
1011 * When this global is false, names passed to the constructor for
1012 * \c ir_var_temporary variables will be dropped. Instead, the variable will
1013 * be named "compiler_temp". This name will be in static storage.
1016 * \b NEVER change the mode of an \c ir_var_temporary.
1019 * This variable is \b not thread-safe. It is global, \b not
1020 * per-context. It begins life false. A context can, at some point, make
1021 * it true. From that point on, it will be true forever. This should be
1022 * okay since it will only be set true while debugging.
1024 static bool temporaries_allocate_names
;
1028 * A function that returns whether a built-in function is available in the
1029 * current shading language (based on version, ES or desktop, and extensions).
1031 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
1033 #define MAKE_INTRINSIC_FOR_TYPE(op, t) \
1034 ir_intrinsic_generic_ ## op - ir_intrinsic_generic_load + ir_intrinsic_ ## t ## _ ## load
1036 #define MAP_INTRINSIC_TO_TYPE(i, t) \
1037 ir_intrinsic_id(int(i) - int(ir_intrinsic_generic_load) + int(ir_intrinsic_ ## t ## _ ## load))
1039 enum ir_intrinsic_id
{
1040 ir_intrinsic_invalid
= 0,
1043 * \name Generic intrinsics
1045 * Each of these intrinsics has a specific version for shared variables and
1049 ir_intrinsic_generic_load
,
1050 ir_intrinsic_generic_store
,
1051 ir_intrinsic_generic_atomic_add
,
1052 ir_intrinsic_generic_atomic_and
,
1053 ir_intrinsic_generic_atomic_or
,
1054 ir_intrinsic_generic_atomic_xor
,
1055 ir_intrinsic_generic_atomic_min
,
1056 ir_intrinsic_generic_atomic_max
,
1057 ir_intrinsic_generic_atomic_exchange
,
1058 ir_intrinsic_generic_atomic_comp_swap
,
1061 ir_intrinsic_atomic_counter_read
,
1062 ir_intrinsic_atomic_counter_increment
,
1063 ir_intrinsic_atomic_counter_predecrement
,
1064 ir_intrinsic_atomic_counter_add
,
1065 ir_intrinsic_atomic_counter_and
,
1066 ir_intrinsic_atomic_counter_or
,
1067 ir_intrinsic_atomic_counter_xor
,
1068 ir_intrinsic_atomic_counter_min
,
1069 ir_intrinsic_atomic_counter_max
,
1070 ir_intrinsic_atomic_counter_exchange
,
1071 ir_intrinsic_atomic_counter_comp_swap
,
1073 ir_intrinsic_image_load
,
1074 ir_intrinsic_image_store
,
1075 ir_intrinsic_image_atomic_add
,
1076 ir_intrinsic_image_atomic_and
,
1077 ir_intrinsic_image_atomic_or
,
1078 ir_intrinsic_image_atomic_xor
,
1079 ir_intrinsic_image_atomic_min
,
1080 ir_intrinsic_image_atomic_max
,
1081 ir_intrinsic_image_atomic_exchange
,
1082 ir_intrinsic_image_atomic_comp_swap
,
1083 ir_intrinsic_image_size
,
1084 ir_intrinsic_image_samples
,
1086 ir_intrinsic_ssbo_load
,
1087 ir_intrinsic_ssbo_store
= MAKE_INTRINSIC_FOR_TYPE(store
, ssbo
),
1088 ir_intrinsic_ssbo_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, ssbo
),
1089 ir_intrinsic_ssbo_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, ssbo
),
1090 ir_intrinsic_ssbo_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, ssbo
),
1091 ir_intrinsic_ssbo_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, ssbo
),
1092 ir_intrinsic_ssbo_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, ssbo
),
1093 ir_intrinsic_ssbo_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, ssbo
),
1094 ir_intrinsic_ssbo_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, ssbo
),
1095 ir_intrinsic_ssbo_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, ssbo
),
1097 ir_intrinsic_memory_barrier
,
1098 ir_intrinsic_shader_clock
,
1099 ir_intrinsic_group_memory_barrier
,
1100 ir_intrinsic_memory_barrier_atomic_counter
,
1101 ir_intrinsic_memory_barrier_buffer
,
1102 ir_intrinsic_memory_barrier_image
,
1103 ir_intrinsic_memory_barrier_shared
,
1105 ir_intrinsic_shared_load
,
1106 ir_intrinsic_shared_store
= MAKE_INTRINSIC_FOR_TYPE(store
, shared
),
1107 ir_intrinsic_shared_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, shared
),
1108 ir_intrinsic_shared_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, shared
),
1109 ir_intrinsic_shared_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, shared
),
1110 ir_intrinsic_shared_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, shared
),
1111 ir_intrinsic_shared_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, shared
),
1112 ir_intrinsic_shared_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, shared
),
1113 ir_intrinsic_shared_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, shared
),
1114 ir_intrinsic_shared_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, shared
),
1119 * The representation of a function instance; may be the full definition or
1120 * simply a prototype.
1122 class ir_function_signature
: public ir_instruction
{
1123 /* An ir_function_signature will be part of the list of signatures in
1127 ir_function_signature(const glsl_type
*return_type
,
1128 builtin_available_predicate builtin_avail
= NULL
);
1130 virtual ir_function_signature
*clone(void *mem_ctx
,
1131 struct hash_table
*ht
) const;
1132 ir_function_signature
*clone_prototype(void *mem_ctx
,
1133 struct hash_table
*ht
) const;
1135 virtual void accept(ir_visitor
*v
)
1140 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1143 * Attempt to evaluate this function as a constant expression,
1144 * given a list of the actual parameters and the variable context.
1145 * Returns NULL for non-built-ins.
1147 ir_constant
*constant_expression_value(exec_list
*actual_parameters
, struct hash_table
*variable_context
);
1150 * Get the name of the function for which this is a signature
1152 const char *function_name() const;
1155 * Get a handle to the function for which this is a signature
1157 * There is no setter function, this function returns a \c const pointer,
1158 * and \c ir_function_signature::_function is private for a reason. The
1159 * only way to make a connection between a function and function signature
1160 * is via \c ir_function::add_signature. This helps ensure that certain
1161 * invariants (i.e., a function signature is in the list of signatures for
1162 * its \c _function) are met.
1164 * \sa ir_function::add_signature
1166 inline const class ir_function
*function() const
1168 return this->_function
;
1172 * Check whether the qualifiers match between this signature's parameters
1173 * and the supplied parameter list. If not, returns the name of the first
1174 * parameter with mismatched qualifiers (for use in error messages).
1176 const char *qualifiers_match(exec_list
*params
);
1179 * Replace the current parameter list with the given one. This is useful
1180 * if the current information came from a prototype, and either has invalid
1181 * or missing parameter names.
1183 void replace_parameters(exec_list
*new_params
);
1186 * Function return type.
1188 * \note This discards the optional precision qualifier.
1190 const struct glsl_type
*return_type
;
1193 * List of ir_variable of function parameters.
1195 * This represents the storage. The paramaters passed in a particular
1196 * call will be in ir_call::actual_paramaters.
1198 struct exec_list parameters
;
1200 /** Whether or not this function has a body (which may be empty). */
1201 unsigned is_defined
:1;
1203 /** Whether or not this function signature is a built-in. */
1204 bool is_builtin() const;
1207 * Whether or not this function is an intrinsic to be implemented
1210 inline bool is_intrinsic() const
1212 return intrinsic_id
!= ir_intrinsic_invalid
;
1215 /** Indentifier for this intrinsic. */
1216 enum ir_intrinsic_id intrinsic_id
;
1218 /** Whether or not a built-in is available for this shader. */
1219 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1221 /** Body of instructions in the function. */
1222 struct exec_list body
;
1226 * A function pointer to a predicate that answers whether a built-in
1227 * function is available in the current shader. NULL if not a built-in.
1229 builtin_available_predicate builtin_avail
;
1231 /** Function of which this signature is one overload. */
1232 class ir_function
*_function
;
1234 /** Function signature of which this one is a prototype clone */
1235 const ir_function_signature
*origin
;
1237 friend class ir_function
;
1240 * Helper function to run a list of instructions for constant
1241 * expression evaluation.
1243 * The hash table represents the values of the visible variables.
1244 * There are no scoping issues because the table is indexed on
1245 * ir_variable pointers, not variable names.
1247 * Returns false if the expression is not constant, true otherwise,
1248 * and the value in *result if result is non-NULL.
1250 bool constant_expression_evaluate_expression_list(const struct exec_list
&body
,
1251 struct hash_table
*variable_context
,
1252 ir_constant
**result
);
1257 * Header for tracking multiple overloaded functions with the same name.
1258 * Contains a list of ir_function_signatures representing each of the
1261 class ir_function
: public ir_instruction
{
1263 ir_function(const char *name
);
1265 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1267 virtual void accept(ir_visitor
*v
)
1272 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1274 void add_signature(ir_function_signature
*sig
)
1276 sig
->_function
= this;
1277 this->signatures
.push_tail(sig
);
1281 * Find a signature that matches a set of actual parameters, taking implicit
1282 * conversions into account. Also flags whether the match was exact.
1284 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1285 const exec_list
*actual_param
,
1286 bool allow_builtins
,
1287 bool *match_is_exact
);
1290 * Find a signature that matches a set of actual parameters, taking implicit
1291 * conversions into account.
1293 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1294 const exec_list
*actual_param
,
1295 bool allow_builtins
);
1298 * Find a signature that exactly matches a set of actual parameters without
1299 * any implicit type conversions.
1301 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1302 const exec_list
*actual_ps
);
1305 * Name of the function.
1309 /** Whether or not this function has a signature that isn't a built-in. */
1310 bool has_user_signature();
1313 * List of ir_function_signature for each overloaded function with this name.
1315 struct exec_list signatures
;
1318 * is this function a subroutine type declaration
1319 * e.g. subroutine void type1(float arg1);
1324 * is this function associated to a subroutine type
1325 * e.g. subroutine (type1, type2) function_name { function_body };
1326 * would have num_subroutine_types 2,
1327 * and pointers to the type1 and type2 types.
1329 int num_subroutine_types
;
1330 const struct glsl_type
**subroutine_types
;
1332 int subroutine_index
;
1335 inline const char *ir_function_signature::function_name() const
1337 return this->_function
->name
;
1343 * IR instruction representing high-level if-statements
1345 class ir_if
: public ir_instruction
{
1347 ir_if(ir_rvalue
*condition
)
1348 : ir_instruction(ir_type_if
), condition(condition
)
1352 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1354 virtual void accept(ir_visitor
*v
)
1359 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1361 ir_rvalue
*condition
;
1362 /** List of ir_instruction for the body of the then branch */
1363 exec_list then_instructions
;
1364 /** List of ir_instruction for the body of the else branch */
1365 exec_list else_instructions
;
1370 * IR instruction representing a high-level loop structure.
1372 class ir_loop
: public ir_instruction
{
1376 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1378 virtual void accept(ir_visitor
*v
)
1383 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1385 /** List of ir_instruction that make up the body of the loop. */
1386 exec_list body_instructions
;
1390 class ir_assignment
: public ir_instruction
{
1392 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1395 * Construct an assignment with an explicit write mask
1398 * Since a write mask is supplied, the LHS must already be a bare
1399 * \c ir_dereference. The cannot be any swizzles in the LHS.
1401 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1402 unsigned write_mask
);
1404 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1406 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1408 virtual void accept(ir_visitor
*v
)
1413 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1416 * Get a whole variable written by an assignment
1418 * If the LHS of the assignment writes a whole variable, the variable is
1419 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1422 * - Assigning to a scalar
1423 * - Assigning to all components of a vector
1424 * - Whole array (or matrix) assignment
1425 * - Whole structure assignment
1427 ir_variable
*whole_variable_written();
1430 * Set the LHS of an assignment
1432 void set_lhs(ir_rvalue
*lhs
);
1435 * Left-hand side of the assignment.
1437 * This should be treated as read only. If you need to set the LHS of an
1438 * assignment, use \c ir_assignment::set_lhs.
1440 ir_dereference
*lhs
;
1443 * Value being assigned
1448 * Optional condition for the assignment.
1450 ir_rvalue
*condition
;
1454 * Component mask written
1456 * For non-vector types in the LHS, this field will be zero. For vector
1457 * types, a bit will be set for each component that is written. Note that
1458 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1460 * A partially-set write mask means that each enabled channel gets
1461 * the value from a consecutive channel of the rhs. For example,
1462 * to write just .xyw of gl_FrontColor with color:
1464 * (assign (constant bool (1)) (xyw)
1465 * (var_ref gl_FragColor)
1466 * (swiz xyw (var_ref color)))
1468 unsigned write_mask
:4;
1471 #include "ir_expression_operation.h"
1473 extern const char *const ir_expression_operation_strings
[ir_last_opcode
+ 1];
1475 class ir_expression
: public ir_rvalue
{
1477 ir_expression(int op
, const struct glsl_type
*type
,
1478 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1479 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1482 * Constructor for unary operation expressions
1484 ir_expression(int op
, ir_rvalue
*);
1487 * Constructor for binary operation expressions
1489 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1492 * Constructor for ternary operation expressions
1494 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1496 virtual bool equals(const ir_instruction
*ir
,
1497 enum ir_node_type ignore
= ir_type_unset
) const;
1499 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1502 * Attempt to constant-fold the expression
1504 * The "variable_context" hash table links ir_variable * to ir_constant *
1505 * that represent the variables' values. \c NULL represents an empty
1508 * If the expression cannot be constant folded, this method will return
1511 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1514 * Determine the number of operands used by an expression
1516 static unsigned int get_num_operands(ir_expression_operation
);
1519 * Determine the number of operands used by an expression
1521 unsigned int get_num_operands() const
1523 return (this->operation
== ir_quadop_vector
)
1524 ? this->type
->vector_elements
: get_num_operands(operation
);
1528 * Return whether the expression operates on vectors horizontally.
1530 bool is_horizontal() const
1532 return operation
== ir_binop_all_equal
||
1533 operation
== ir_binop_any_nequal
||
1534 operation
== ir_binop_dot
||
1535 operation
== ir_binop_vector_extract
||
1536 operation
== ir_triop_vector_insert
||
1537 operation
== ir_binop_ubo_load
||
1538 operation
== ir_quadop_vector
;
1542 * Do a reverse-lookup to translate the given string into an operator.
1544 static ir_expression_operation
get_operator(const char *);
1546 virtual void accept(ir_visitor
*v
)
1551 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1553 virtual ir_variable
*variable_referenced() const;
1555 ir_expression_operation operation
;
1556 ir_rvalue
*operands
[4];
1561 * HIR instruction representing a high-level function call, containing a list
1562 * of parameters and returning a value in the supplied temporary.
1564 class ir_call
: public ir_instruction
{
1566 ir_call(ir_function_signature
*callee
,
1567 ir_dereference_variable
*return_deref
,
1568 exec_list
*actual_parameters
)
1569 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(NULL
), array_idx(NULL
)
1571 assert(callee
->return_type
!= NULL
);
1572 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1573 this->use_builtin
= callee
->is_builtin();
1576 ir_call(ir_function_signature
*callee
,
1577 ir_dereference_variable
*return_deref
,
1578 exec_list
*actual_parameters
,
1579 ir_variable
*var
, ir_rvalue
*array_idx
)
1580 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(var
), array_idx(array_idx
)
1582 assert(callee
->return_type
!= NULL
);
1583 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1584 this->use_builtin
= callee
->is_builtin();
1587 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1589 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1591 virtual void accept(ir_visitor
*v
)
1596 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1599 * Get the name of the function being called.
1601 const char *callee_name() const
1603 return callee
->function_name();
1607 * Generates an inline version of the function before @ir,
1608 * storing the return value in return_deref.
1610 void generate_inline(ir_instruction
*ir
);
1613 * Storage for the function's return value.
1614 * This must be NULL if the return type is void.
1616 ir_dereference_variable
*return_deref
;
1619 * The specific function signature being called.
1621 ir_function_signature
*callee
;
1623 /* List of ir_rvalue of paramaters passed in this call. */
1624 exec_list actual_parameters
;
1626 /** Should this call only bind to a built-in function? */
1630 * ARB_shader_subroutine support -
1631 * the subroutine uniform variable and array index
1632 * rvalue to be used in the lowering pass later.
1634 ir_variable
*sub_var
;
1635 ir_rvalue
*array_idx
;
1640 * \name Jump-like IR instructions.
1642 * These include \c break, \c continue, \c return, and \c discard.
1645 class ir_jump
: public ir_instruction
{
1647 ir_jump(enum ir_node_type t
)
1653 class ir_return
: public ir_jump
{
1656 : ir_jump(ir_type_return
), value(NULL
)
1660 ir_return(ir_rvalue
*value
)
1661 : ir_jump(ir_type_return
), value(value
)
1665 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1667 ir_rvalue
*get_value() const
1672 virtual void accept(ir_visitor
*v
)
1677 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1684 * Jump instructions used inside loops
1686 * These include \c break and \c continue. The \c break within a loop is
1687 * different from the \c break within a switch-statement.
1689 * \sa ir_switch_jump
1691 class ir_loop_jump
: public ir_jump
{
1698 ir_loop_jump(jump_mode mode
)
1699 : ir_jump(ir_type_loop_jump
)
1704 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1706 virtual void accept(ir_visitor
*v
)
1711 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1713 bool is_break() const
1715 return mode
== jump_break
;
1718 bool is_continue() const
1720 return mode
== jump_continue
;
1723 /** Mode selector for the jump instruction. */
1724 enum jump_mode mode
;
1728 * IR instruction representing discard statements.
1730 class ir_discard
: public ir_jump
{
1733 : ir_jump(ir_type_discard
)
1735 this->condition
= NULL
;
1738 ir_discard(ir_rvalue
*cond
)
1739 : ir_jump(ir_type_discard
)
1741 this->condition
= cond
;
1744 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1746 virtual void accept(ir_visitor
*v
)
1751 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1753 ir_rvalue
*condition
;
1759 * Texture sampling opcodes used in ir_texture
1761 enum ir_texture_opcode
{
1762 ir_tex
, /**< Regular texture look-up */
1763 ir_txb
, /**< Texture look-up with LOD bias */
1764 ir_txl
, /**< Texture look-up with explicit LOD */
1765 ir_txd
, /**< Texture look-up with partial derivatvies */
1766 ir_txf
, /**< Texel fetch with explicit LOD */
1767 ir_txf_ms
, /**< Multisample texture fetch */
1768 ir_txs
, /**< Texture size */
1769 ir_lod
, /**< Texture lod query */
1770 ir_tg4
, /**< Texture gather */
1771 ir_query_levels
, /**< Texture levels query */
1772 ir_texture_samples
, /**< Texture samples query */
1773 ir_samples_identical
, /**< Query whether all samples are definitely identical. */
1778 * IR instruction to sample a texture
1780 * The specific form of the IR instruction depends on the \c mode value
1781 * selected from \c ir_texture_opcodes. In the printed IR, these will
1784 * Texel offset (0 or an expression)
1785 * | Projection divisor
1786 * | | Shadow comparitor
1789 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1790 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1791 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1792 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1793 * (txf <type> <sampler> <coordinate> 0 <lod>)
1795 * <type> <sampler> <coordinate> <sample_index>)
1796 * (txs <type> <sampler> <lod>)
1797 * (lod <type> <sampler> <coordinate>)
1798 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1799 * (query_levels <type> <sampler>)
1800 * (samples_identical <sampler> <coordinate>)
1802 class ir_texture
: public ir_rvalue
{
1804 ir_texture(enum ir_texture_opcode op
)
1805 : ir_rvalue(ir_type_texture
),
1806 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1807 shadow_comparitor(NULL
), offset(NULL
)
1809 memset(&lod_info
, 0, sizeof(lod_info
));
1812 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1814 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1816 virtual void accept(ir_visitor
*v
)
1821 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1823 virtual bool equals(const ir_instruction
*ir
,
1824 enum ir_node_type ignore
= ir_type_unset
) const;
1827 * Return a string representing the ir_texture_opcode.
1829 const char *opcode_string();
1831 /** Set the sampler and type. */
1832 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1835 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1837 static ir_texture_opcode
get_opcode(const char *);
1839 enum ir_texture_opcode op
;
1841 /** Sampler to use for the texture access. */
1842 ir_dereference
*sampler
;
1844 /** Texture coordinate to sample */
1845 ir_rvalue
*coordinate
;
1848 * Value used for projective divide.
1850 * If there is no projective divide (the common case), this will be
1851 * \c NULL. Optimization passes should check for this to point to a constant
1852 * of 1.0 and replace that with \c NULL.
1854 ir_rvalue
*projector
;
1857 * Coordinate used for comparison on shadow look-ups.
1859 * If there is no shadow comparison, this will be \c NULL. For the
1860 * \c ir_txf opcode, this *must* be \c NULL.
1862 ir_rvalue
*shadow_comparitor
;
1864 /** Texel offset. */
1868 ir_rvalue
*lod
; /**< Floating point LOD */
1869 ir_rvalue
*bias
; /**< Floating point LOD bias */
1870 ir_rvalue
*sample_index
; /**< MSAA sample index */
1871 ir_rvalue
*component
; /**< Gather component selector */
1873 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1874 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1880 struct ir_swizzle_mask
{
1887 * Number of components in the swizzle.
1889 unsigned num_components
:3;
1892 * Does the swizzle contain duplicate components?
1894 * L-value swizzles cannot contain duplicate components.
1896 unsigned has_duplicates
:1;
1900 class ir_swizzle
: public ir_rvalue
{
1902 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
1905 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
1907 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
1909 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
1911 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1914 * Construct an ir_swizzle from the textual representation. Can fail.
1916 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
1918 virtual void accept(ir_visitor
*v
)
1923 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1925 virtual bool equals(const ir_instruction
*ir
,
1926 enum ir_node_type ignore
= ir_type_unset
) const;
1928 bool is_lvalue() const
1930 return val
->is_lvalue() && !mask
.has_duplicates
;
1934 * Get the variable that is ultimately referenced by an r-value
1936 virtual ir_variable
*variable_referenced() const;
1939 ir_swizzle_mask mask
;
1943 * Initialize the mask component of a swizzle
1945 * This is used by the \c ir_swizzle constructors.
1947 void init_mask(const unsigned *components
, unsigned count
);
1951 class ir_dereference
: public ir_rvalue
{
1953 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
1955 bool is_lvalue() const;
1958 * Get the variable that is ultimately referenced by an r-value
1960 virtual ir_variable
*variable_referenced() const = 0;
1963 ir_dereference(enum ir_node_type t
)
1970 class ir_dereference_variable
: public ir_dereference
{
1972 ir_dereference_variable(ir_variable
*var
);
1974 virtual ir_dereference_variable
*clone(void *mem_ctx
,
1975 struct hash_table
*) const;
1977 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
1979 virtual bool equals(const ir_instruction
*ir
,
1980 enum ir_node_type ignore
= ir_type_unset
) const;
1983 * Get the variable that is ultimately referenced by an r-value
1985 virtual ir_variable
*variable_referenced() const
1990 virtual ir_variable
*whole_variable_referenced()
1992 /* ir_dereference_variable objects always dereference the entire
1993 * variable. However, if this dereference is dereferenced by anything
1994 * else, the complete deferefernce chain is not a whole-variable
1995 * dereference. This method should only be called on the top most
1996 * ir_rvalue in a dereference chain.
2001 virtual void accept(ir_visitor
*v
)
2006 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2009 * Object being dereferenced.
2015 class ir_dereference_array
: public ir_dereference
{
2017 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2019 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2021 virtual ir_dereference_array
*clone(void *mem_ctx
,
2022 struct hash_table
*) const;
2024 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2026 virtual bool equals(const ir_instruction
*ir
,
2027 enum ir_node_type ignore
= ir_type_unset
) const;
2030 * Get the variable that is ultimately referenced by an r-value
2032 virtual ir_variable
*variable_referenced() const
2034 return this->array
->variable_referenced();
2037 virtual void accept(ir_visitor
*v
)
2042 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2045 ir_rvalue
*array_index
;
2048 void set_array(ir_rvalue
*value
);
2052 class ir_dereference_record
: public ir_dereference
{
2054 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2056 ir_dereference_record(ir_variable
*var
, const char *field
);
2058 virtual ir_dereference_record
*clone(void *mem_ctx
,
2059 struct hash_table
*) const;
2061 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2064 * Get the variable that is ultimately referenced by an r-value
2066 virtual ir_variable
*variable_referenced() const
2068 return this->record
->variable_referenced();
2071 virtual void accept(ir_visitor
*v
)
2076 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2084 * Data stored in an ir_constant
2086 union ir_constant_data
{
2095 class ir_constant
: public ir_rvalue
{
2097 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2098 ir_constant(bool b
, unsigned vector_elements
=1);
2099 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2100 ir_constant(int i
, unsigned vector_elements
=1);
2101 ir_constant(float f
, unsigned vector_elements
=1);
2102 ir_constant(double d
, unsigned vector_elements
=1);
2105 * Construct an ir_constant from a list of ir_constant values
2107 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2110 * Construct an ir_constant from a scalar component of another ir_constant
2112 * The new \c ir_constant inherits the type of the component from the
2116 * In the case of a matrix constant, the new constant is a scalar, \b not
2119 ir_constant(const ir_constant
*c
, unsigned i
);
2122 * Return a new ir_constant of the specified type containing all zeros.
2124 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2126 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2128 virtual ir_constant
*constant_expression_value(struct hash_table
*variable_context
= NULL
);
2130 virtual void accept(ir_visitor
*v
)
2135 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2137 virtual bool equals(const ir_instruction
*ir
,
2138 enum ir_node_type ignore
= ir_type_unset
) const;
2141 * Get a particular component of a constant as a specific type
2143 * This is useful, for example, to get a value from an integer constant
2144 * as a float or bool. This appears frequently when constructors are
2145 * called with all constant parameters.
2148 bool get_bool_component(unsigned i
) const;
2149 float get_float_component(unsigned i
) const;
2150 double get_double_component(unsigned i
) const;
2151 int get_int_component(unsigned i
) const;
2152 unsigned get_uint_component(unsigned i
) const;
2155 ir_constant
*get_array_element(unsigned i
) const;
2157 ir_constant
*get_record_field(const char *name
);
2160 * Copy the values on another constant at a given offset.
2162 * The offset is ignored for array or struct copies, it's only for
2163 * scalars or vectors into vectors or matrices.
2165 * With identical types on both sides and zero offset it's clone()
2166 * without creating a new object.
2169 void copy_offset(ir_constant
*src
, int offset
);
2172 * Copy the values on another constant at a given offset and
2173 * following an assign-like mask.
2175 * The mask is ignored for scalars.
2177 * Note that this function only handles what assign can handle,
2178 * i.e. at most a vector as source and a column of a matrix as
2182 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2185 * Determine whether a constant has the same value as another constant
2187 * \sa ir_constant::is_zero, ir_constant::is_one,
2188 * ir_constant::is_negative_one
2190 bool has_value(const ir_constant
*) const;
2193 * Return true if this ir_constant represents the given value.
2195 * For vectors, this checks that each component is the given value.
2197 virtual bool is_value(float f
, int i
) const;
2198 virtual bool is_zero() const;
2199 virtual bool is_one() const;
2200 virtual bool is_negative_one() const;
2203 * Return true for constants that could be stored as 16-bit unsigned values.
2205 * Note that this will return true even for signed integer ir_constants, as
2206 * long as the value is non-negative and fits in 16-bits.
2208 virtual bool is_uint16_constant() const;
2211 * Value of the constant.
2213 * The field used to back the values supplied by the constant is determined
2214 * by the type associated with the \c ir_instruction. Constants may be
2215 * scalars, vectors, or matrices.
2217 union ir_constant_data value
;
2219 /* Array elements */
2220 ir_constant
**array_elements
;
2222 /* Structure fields */
2223 exec_list components
;
2227 * Parameterless constructor only used by the clone method
2233 * IR instruction to emit a vertex in a geometry shader.
2235 class ir_emit_vertex
: public ir_instruction
{
2237 ir_emit_vertex(ir_rvalue
*stream
)
2238 : ir_instruction(ir_type_emit_vertex
),
2244 virtual void accept(ir_visitor
*v
)
2249 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2251 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2254 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2256 int stream_id() const
2258 return stream
->as_constant()->value
.i
[0];
2265 * IR instruction to complete the current primitive and start a new one in a
2268 class ir_end_primitive
: public ir_instruction
{
2270 ir_end_primitive(ir_rvalue
*stream
)
2271 : ir_instruction(ir_type_end_primitive
),
2277 virtual void accept(ir_visitor
*v
)
2282 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2284 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2287 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2289 int stream_id() const
2291 return stream
->as_constant()->value
.i
[0];
2298 * IR instruction for tessellation control and compute shader barrier.
2300 class ir_barrier
: public ir_instruction
{
2303 : ir_instruction(ir_type_barrier
)
2307 virtual void accept(ir_visitor
*v
)
2312 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2314 return new(mem_ctx
) ir_barrier();
2317 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2323 * Apply a visitor to each IR node in a list
2326 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2329 * Validate invariants on each IR node in a list
2331 void validate_ir_tree(exec_list
*instructions
);
2333 struct _mesa_glsl_parse_state
;
2334 struct gl_shader_program
;
2337 * Detect whether an unlinked shader contains static recursion
2339 * If the list of instructions is determined to contain static recursion,
2340 * \c _mesa_glsl_error will be called to emit error messages for each function
2341 * that is in the recursion cycle.
2344 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2345 exec_list
*instructions
);
2348 * Detect whether a linked shader contains static recursion
2350 * If the list of instructions is determined to contain static recursion,
2351 * \c link_error_printf will be called to emit error messages for each function
2352 * that is in the recursion cycle. In addition,
2353 * \c gl_shader_program::LinkStatus will be set to false.
2356 detect_recursion_linked(struct gl_shader_program
*prog
,
2357 exec_list
*instructions
);
2360 * Make a clone of each IR instruction in a list
2362 * \param in List of IR instructions that are to be cloned
2363 * \param out List to hold the cloned instructions
2366 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2369 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2370 struct _mesa_glsl_parse_state
*state
);
2373 _mesa_glsl_initialize_derived_variables(struct gl_context
*ctx
,
2377 _mesa_glsl_initialize_builtin_functions();
2379 extern ir_function_signature
*
2380 _mesa_glsl_find_builtin_function(_mesa_glsl_parse_state
*state
,
2381 const char *name
, exec_list
*actual_parameters
);
2383 extern ir_function
*
2384 _mesa_glsl_find_builtin_function_by_name(const char *name
);
2387 _mesa_glsl_get_builtin_function_shader(void);
2389 extern ir_function_signature
*
2390 _mesa_get_main_function_signature(glsl_symbol_table
*symbols
);
2393 _mesa_glsl_release_builtin_functions(void);
2396 reparent_ir(exec_list
*list
, void *mem_ctx
);
2399 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2400 gl_shader_stage shader_stage
);
2403 prototype_string(const glsl_type
*return_type
, const char *name
,
2404 exec_list
*parameters
);
2407 mode_string(const ir_variable
*var
);
2410 * Built-in / reserved GL variables names start with "gl_"
2413 is_gl_identifier(const char *s
)
2415 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2419 #endif /* __cplusplus */
2421 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2422 struct _mesa_glsl_parse_state
*state
);
2425 fprint_ir(FILE *f
, const void *instruction
);
2427 extern const struct gl_builtin_uniform_desc
*
2428 _mesa_glsl_get_builtin_uniform_desc(const char *name
);
2435 vertices_per_prim(GLenum prim
);