3 * Copyright © 2010 Intel Corporation
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
31 #include "util/ralloc.h"
32 #include "compiler/glsl_types.h"
34 #include "ir_visitor.h"
35 #include "ir_hierarchical_visitor.h"
40 * \defgroup IR Intermediate representation nodes
48 * Each concrete class derived from \c ir_instruction has a value in this
49 * enumerant. The value for the type is stored in \c ir_instruction::ir_type
50 * by the constructor. While using type tags is not very C++, it is extremely
51 * convenient. For example, during debugging you can simply inspect
52 * \c ir_instruction::ir_type to find out the actual type of the object.
54 * In addition, it is possible to use a switch-statement based on \c
55 * \c ir_instruction::ir_type to select different behavior for different object
56 * types. For functions that have only slight differences for several object
57 * types, this allows writing very straightforward, readable code.
60 ir_type_dereference_array
,
61 ir_type_dereference_record
,
62 ir_type_dereference_variable
,
71 ir_type_function_signature
,
79 ir_type_end_primitive
,
81 ir_type_max
, /**< maximum ir_type enum number, for validation */
82 ir_type_unset
= ir_type_max
87 * Base class of all IR instructions
89 class ir_instruction
: public exec_node
{
91 enum ir_node_type ir_type
;
94 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
95 * there's a virtual destructor present. Because we almost
96 * universally use ralloc for our memory management of
97 * ir_instructions, the destructor doesn't need to do any work.
99 virtual ~ir_instruction()
103 /** ir_print_visitor helper for debugging. */
104 void print(void) const;
105 void fprint(FILE *f
) const;
107 virtual void accept(ir_visitor
*) = 0;
108 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*) = 0;
109 virtual ir_instruction
*clone(void *mem_ctx
,
110 struct hash_table
*ht
) const = 0;
112 bool is_rvalue() const
114 return ir_type
== ir_type_dereference_array
||
115 ir_type
== ir_type_dereference_record
||
116 ir_type
== ir_type_dereference_variable
||
117 ir_type
== ir_type_constant
||
118 ir_type
== ir_type_expression
||
119 ir_type
== ir_type_swizzle
||
120 ir_type
== ir_type_texture
;
123 bool is_dereference() const
125 return ir_type
== ir_type_dereference_array
||
126 ir_type
== ir_type_dereference_record
||
127 ir_type
== ir_type_dereference_variable
;
132 return ir_type
== ir_type_loop_jump
||
133 ir_type
== ir_type_return
||
134 ir_type
== ir_type_discard
;
138 * \name IR instruction downcast functions
140 * These functions either cast the object to a derived class or return
141 * \c NULL if the object's type does not match the specified derived class.
142 * Additional downcast functions will be added as needed.
145 #define AS_BASE(TYPE) \
146 class ir_##TYPE *as_##TYPE() \
148 assume(this != NULL); \
149 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
151 const class ir_##TYPE *as_##TYPE() const \
153 assume(this != NULL); \
154 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
162 #define AS_CHILD(TYPE) \
163 class ir_##TYPE * as_##TYPE() \
165 assume(this != NULL); \
166 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
168 const class ir_##TYPE * as_##TYPE() const \
170 assume(this != NULL); \
171 return ir_type == ir_type_##TYPE ? (const ir_##TYPE *) this : NULL; \
175 AS_CHILD(dereference_array
)
176 AS_CHILD(dereference_variable
)
177 AS_CHILD(dereference_record
)
192 * IR equality method: Return true if the referenced instruction would
193 * return the same value as this one.
195 * This intended to be used for CSE and algebraic optimizations, on rvalues
196 * in particular. No support for other instruction types (assignments,
197 * jumps, calls, etc.) is planned.
199 virtual bool equals(const ir_instruction
*ir
,
200 enum ir_node_type ignore
= ir_type_unset
) const;
203 ir_instruction(enum ir_node_type t
)
211 assert(!"Should not get here.");
217 * The base class for all "values"/expression trees.
219 class ir_rvalue
: public ir_instruction
{
221 const struct glsl_type
*type
;
223 virtual ir_rvalue
*clone(void *mem_ctx
, struct hash_table
*) const;
225 virtual void accept(ir_visitor
*v
)
230 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
232 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
233 struct hash_table
*variable_context
= NULL
);
235 ir_rvalue
*as_rvalue_to_saturate();
237 virtual bool is_lvalue(const struct _mesa_glsl_parse_state
* = NULL
) 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
{
400 gl_state_index16 tokens
[STATE_LENGTH
];
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 * \c glsl_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 * Return whether this variable contains a bindless sampler/image.
481 inline bool contains_bindless() const
483 if (!this->type
->contains_sampler() && !this->type
->contains_image())
486 return this->data
.bindless
|| this->data
.mode
!= ir_var_uniform
;
490 * Set this->interface_type on a newly created variable.
492 void init_interface_type(const struct glsl_type
*type
)
494 assert(this->interface_type
== NULL
);
495 this->interface_type
= type
;
496 if (this->is_interface_instance()) {
497 this->u
.max_ifc_array_access
=
498 ralloc_array(this, int, type
->length
);
499 for (unsigned i
= 0; i
< type
->length
; i
++) {
500 this->u
.max_ifc_array_access
[i
] = -1;
506 * Change this->interface_type on a variable that previously had a
507 * different, but compatible, interface_type. This is used during linking
508 * to set the size of arrays in interface blocks.
510 void change_interface_type(const struct glsl_type
*type
)
512 if (this->u
.max_ifc_array_access
!= NULL
) {
513 /* max_ifc_array_access has already been allocated, so make sure the
514 * new interface has the same number of fields as the old one.
516 assert(this->interface_type
->length
== type
->length
);
518 this->interface_type
= type
;
522 * Change this->interface_type on a variable that previously had a
523 * different, and incompatible, interface_type. This is used during
524 * compilation to handle redeclaration of the built-in gl_PerVertex
527 void reinit_interface_type(const struct glsl_type
*type
)
529 if (this->u
.max_ifc_array_access
!= NULL
) {
531 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
532 * it defines have been accessed yet; so it's safe to throw away the
533 * old max_ifc_array_access pointer, since all of its values are
536 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
537 assert(this->u
.max_ifc_array_access
[i
] == -1);
539 ralloc_free(this->u
.max_ifc_array_access
);
540 this->u
.max_ifc_array_access
= NULL
;
542 this->interface_type
= NULL
;
543 init_interface_type(type
);
546 const glsl_type
*get_interface_type() const
548 return this->interface_type
;
551 enum glsl_interface_packing
get_interface_type_packing() const
553 return this->interface_type
->get_interface_packing();
556 * Get the max_ifc_array_access pointer
558 * A "set" function is not needed because the array is dynmically allocated
561 inline int *get_max_ifc_array_access()
563 assert(this->data
._num_state_slots
== 0);
564 return this->u
.max_ifc_array_access
;
567 inline unsigned get_num_state_slots() const
569 assert(!this->is_interface_instance()
570 || this->data
._num_state_slots
== 0);
571 return this->data
._num_state_slots
;
574 inline void set_num_state_slots(unsigned n
)
576 assert(!this->is_interface_instance()
578 this->data
._num_state_slots
= n
;
581 inline ir_state_slot
*get_state_slots()
583 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
586 inline const ir_state_slot
*get_state_slots() const
588 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
591 inline ir_state_slot
*allocate_state_slots(unsigned n
)
593 assert(!this->is_interface_instance());
595 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
596 this->data
._num_state_slots
= 0;
598 if (this->u
.state_slots
!= NULL
)
599 this->data
._num_state_slots
= n
;
601 return this->u
.state_slots
;
604 inline bool is_interpolation_flat() const
606 return this->data
.interpolation
== INTERP_MODE_FLAT
||
607 this->type
->contains_integer() ||
608 this->type
->contains_double();
611 inline bool is_name_ralloced() const
613 return this->name
!= ir_variable::tmp_name
&&
614 this->name
!= this->name_storage
;
618 * Enable emitting extension warnings for this variable
620 void enable_extension_warning(const char *extension
);
623 * Get the extension warning string for this variable
625 * If warnings are not enabled, \c NULL is returned.
627 const char *get_extension_warning() const;
630 * Declared type of the variable
632 const struct glsl_type
*type
;
635 * Declared name of the variable
641 * If the name length fits into name_storage, it's used, otherwise
642 * the name is ralloc'd. shader-db mining showed that 70% of variables
643 * fit here. This is a win over ralloc where only ralloc_header has
644 * 20 bytes on 64-bit (28 bytes with DEBUG), and we can also skip malloc.
646 char name_storage
[16];
649 struct ir_variable_data
{
652 * Is the variable read-only?
654 * This is set for variables declared as \c const, shader inputs,
657 unsigned read_only
:1;
662 * Was an 'invariant' qualifier explicitly set in the shader?
664 * This is used to cross validate qualifiers.
666 unsigned explicit_invariant
:1;
668 * Is the variable invariant?
670 * It can happen either by having the 'invariant' qualifier
671 * explicitly set in the shader or by being used in calculations
672 * of other invariant variables.
674 unsigned invariant
:1;
678 * Has this variable been used for reading or writing?
680 * Several GLSL semantic checks require knowledge of whether or not a
681 * variable has been used. For example, it is an error to redeclare a
682 * variable as invariant after it has been used.
684 * This is maintained in the ast_to_hir.cpp path and during linking,
685 * but not in Mesa's fixed function or ARB program paths.
690 * Has this variable been statically assigned?
692 * This answers whether the variable was assigned in any path of
693 * the shader during ast_to_hir. This doesn't answer whether it is
694 * still written after dead code removal, nor is it maintained in
695 * non-ast_to_hir.cpp (GLSL parsing) paths.
700 * When separate shader programs are enabled, only input/outputs between
701 * the stages of a multi-stage separate program can be safely removed
702 * from the shader interface. Other input/outputs must remains active.
704 unsigned always_active_io
:1;
707 * Enum indicating how the variable was declared. See
708 * ir_var_declaration_type.
710 * This is used to detect certain kinds of illegal variable redeclarations.
712 unsigned how_declared
:2;
715 * Storage class of the variable.
717 * \sa ir_variable_mode
722 * Interpolation mode for shader inputs / outputs
724 * \sa glsl_interp_mode
726 unsigned interpolation
:2;
729 * Was the location explicitly set in the shader?
731 * If the location is explicitly set in the shader, it \b cannot be changed
732 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
735 unsigned explicit_location
:1;
736 unsigned explicit_index
:1;
739 * Was an initial binding explicitly set in the shader?
741 * If so, constant_value contains an integer ir_constant representing the
742 * initial binding point.
744 unsigned explicit_binding
:1;
747 * Was an initial component explicitly set in the shader?
749 unsigned explicit_component
:1;
752 * Does this variable have an initializer?
754 * This is used by the linker to cross-validiate initializers of global
757 unsigned has_initializer
:1;
760 * Is this variable a generic output or input that has not yet been matched
761 * up to a variable in another stage of the pipeline?
763 * This is used by the linker as scratch storage while assigning locations
764 * to generic inputs and outputs.
766 unsigned is_unmatched_generic_inout
:1;
769 * Is this varying used only by transform feedback?
771 * This is used by the linker to decide if its safe to pack the varying.
773 unsigned is_xfb_only
:1;
776 * Was a transform feedback buffer set in the shader?
778 unsigned explicit_xfb_buffer
:1;
781 * Was a transform feedback offset set in the shader?
783 unsigned explicit_xfb_offset
:1;
786 * Was a transform feedback stride set in the shader?
788 unsigned explicit_xfb_stride
:1;
791 * If non-zero, then this variable may be packed along with other variables
792 * into a single varying slot, so this offset should be applied when
793 * accessing components. For example, an offset of 1 means that the x
794 * component of this variable is actually stored in component y of the
795 * location specified by \c location.
797 unsigned location_frac
:2;
800 * Layout of the matrix. Uses glsl_matrix_layout values.
802 unsigned matrix_layout
:2;
805 * Non-zero if this variable was created by lowering a named interface
808 unsigned from_named_ifc_block
:1;
811 * Non-zero if the variable must be a shader input. This is useful for
812 * constraints on function parameters.
814 unsigned must_be_shader_input
:1;
817 * Output index for dual source blending.
820 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
826 * Precision qualifier.
828 * In desktop GLSL we do not care about precision qualifiers at all, in
829 * fact, the spec says that precision qualifiers are ignored.
831 * To make things easy, we make it so that this field is always
832 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
833 * have the same precision value and the checks we add in the compiler
834 * for this field will never break a desktop shader compile.
836 unsigned precision
:2;
839 * \brief Layout qualifier for gl_FragDepth.
841 * This is not equal to \c ir_depth_layout_none if and only if this
842 * variable is \c gl_FragDepth and a layout qualifier is specified.
844 ir_depth_layout depth_layout
:3;
849 unsigned memory_read_only
:1; /**< "readonly" qualifier. */
850 unsigned memory_write_only
:1; /**< "writeonly" qualifier. */
851 unsigned memory_coherent
:1;
852 unsigned memory_volatile
:1;
853 unsigned memory_restrict
:1;
856 * ARB_shader_storage_buffer_object
858 unsigned from_ssbo_unsized_array
:1; /**< unsized array buffer variable. */
860 unsigned implicit_sized_array
:1;
863 * Whether this is a fragment shader output implicitly initialized with
864 * the previous contents of the specified render target at the
865 * framebuffer location corresponding to this shader invocation.
867 unsigned fb_fetch_output
:1;
870 * Non-zero if this variable is considered bindless as defined by
871 * ARB_bindless_texture.
876 * Non-zero if this variable is considered bound as defined by
877 * ARB_bindless_texture.
882 * Emit a warning if this variable is accessed.
885 uint8_t warn_extension_index
;
888 /** Image internal format if specified explicitly, otherwise GL_NONE. */
889 uint16_t image_format
;
893 * Number of state slots used
896 * This could be stored in as few as 7-bits, if necessary. If it is made
897 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
900 uint16_t _num_state_slots
;
904 * Initial binding point for a sampler, atomic, or UBO.
906 * For array types, this represents the binding point for the first element.
911 * Storage location of the base of this variable
913 * The precise meaning of this field depends on the nature of the variable.
915 * - Vertex shader input: one of the values from \c gl_vert_attrib.
916 * - Vertex shader output: one of the values from \c gl_varying_slot.
917 * - Geometry shader input: one of the values from \c gl_varying_slot.
918 * - Geometry shader output: one of the values from \c gl_varying_slot.
919 * - Fragment shader input: one of the values from \c gl_varying_slot.
920 * - Fragment shader output: one of the values from \c gl_frag_result.
921 * - Uniforms: Per-stage uniform slot number for default uniform block.
922 * - Uniforms: Index within the uniform block definition for UBO members.
923 * - Non-UBO Uniforms: explicit location until linking then reused to
924 * store uniform slot number.
925 * - Other: This field is not currently used.
927 * If the variable is a uniform, shader input, or shader output, and the
928 * slot has not been assigned, the value will be -1.
933 * for glsl->tgsi/mesa IR we need to store the index into the
934 * parameters for uniforms, initially the code overloaded location
935 * but this causes problems with indirect samplers and AoA.
936 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
941 * Vertex stream output identifier.
943 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
944 * stream of the i-th component.
949 * Atomic, transform feedback or block member offset.
954 * Highest element accessed with a constant expression array index
956 * Not used for non-array variables. -1 is never accessed.
958 int max_array_access
;
961 * Transform feedback buffer.
966 * Transform feedback stride.
971 * Allow (only) ir_variable direct access private members.
973 friend class ir_variable
;
977 * Value assigned in the initializer of a variable declared "const"
979 ir_constant
*constant_value
;
982 * Constant expression assigned in the initializer of the variable
985 * This field and \c ::constant_value are distinct. Even if the two fields
986 * refer to constants with the same value, they must point to separate
989 ir_constant
*constant_initializer
;
992 static const char *const warn_extension_table
[];
996 * For variables which satisfy the is_interface_instance() predicate,
997 * this points to an array of integers such that if the ith member of
998 * the interface block is an array, max_ifc_array_access[i] is the
999 * maximum array element of that member that has been accessed. If the
1000 * ith member of the interface block is not an array,
1001 * max_ifc_array_access[i] is unused.
1003 * For variables whose type is not an interface block, this pointer is
1006 int *max_ifc_array_access
;
1009 * Built-in state that backs this uniform
1011 * Once set at variable creation, \c state_slots must remain invariant.
1013 * If the variable is not a uniform, \c _num_state_slots will be zero
1014 * and \c state_slots will be \c NULL.
1016 ir_state_slot
*state_slots
;
1020 * For variables that are in an interface block or are an instance of an
1021 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
1023 * \sa ir_variable::location
1025 const glsl_type
*interface_type
;
1028 * Name used for anonymous compiler temporaries
1030 static const char tmp_name
[];
1034 * Should the construct keep names for ir_var_temporary variables?
1036 * When this global is false, names passed to the constructor for
1037 * \c ir_var_temporary variables will be dropped. Instead, the variable will
1038 * be named "compiler_temp". This name will be in static storage.
1041 * \b NEVER change the mode of an \c ir_var_temporary.
1044 * This variable is \b not thread-safe. It is global, \b not
1045 * per-context. It begins life false. A context can, at some point, make
1046 * it true. From that point on, it will be true forever. This should be
1047 * okay since it will only be set true while debugging.
1049 static bool temporaries_allocate_names
;
1053 * A function that returns whether a built-in function is available in the
1054 * current shading language (based on version, ES or desktop, and extensions).
1056 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
1058 #define MAKE_INTRINSIC_FOR_TYPE(op, t) \
1059 ir_intrinsic_generic_ ## op - ir_intrinsic_generic_load + ir_intrinsic_ ## t ## _ ## load
1061 #define MAP_INTRINSIC_TO_TYPE(i, t) \
1062 ir_intrinsic_id(int(i) - int(ir_intrinsic_generic_load) + int(ir_intrinsic_ ## t ## _ ## load))
1064 enum ir_intrinsic_id
{
1065 ir_intrinsic_invalid
= 0,
1068 * \name Generic intrinsics
1070 * Each of these intrinsics has a specific version for shared variables and
1074 ir_intrinsic_generic_load
,
1075 ir_intrinsic_generic_store
,
1076 ir_intrinsic_generic_atomic_add
,
1077 ir_intrinsic_generic_atomic_and
,
1078 ir_intrinsic_generic_atomic_or
,
1079 ir_intrinsic_generic_atomic_xor
,
1080 ir_intrinsic_generic_atomic_min
,
1081 ir_intrinsic_generic_atomic_max
,
1082 ir_intrinsic_generic_atomic_exchange
,
1083 ir_intrinsic_generic_atomic_comp_swap
,
1086 ir_intrinsic_atomic_counter_read
,
1087 ir_intrinsic_atomic_counter_increment
,
1088 ir_intrinsic_atomic_counter_predecrement
,
1089 ir_intrinsic_atomic_counter_add
,
1090 ir_intrinsic_atomic_counter_and
,
1091 ir_intrinsic_atomic_counter_or
,
1092 ir_intrinsic_atomic_counter_xor
,
1093 ir_intrinsic_atomic_counter_min
,
1094 ir_intrinsic_atomic_counter_max
,
1095 ir_intrinsic_atomic_counter_exchange
,
1096 ir_intrinsic_atomic_counter_comp_swap
,
1098 ir_intrinsic_image_load
,
1099 ir_intrinsic_image_store
,
1100 ir_intrinsic_image_atomic_add
,
1101 ir_intrinsic_image_atomic_and
,
1102 ir_intrinsic_image_atomic_or
,
1103 ir_intrinsic_image_atomic_xor
,
1104 ir_intrinsic_image_atomic_min
,
1105 ir_intrinsic_image_atomic_max
,
1106 ir_intrinsic_image_atomic_exchange
,
1107 ir_intrinsic_image_atomic_comp_swap
,
1108 ir_intrinsic_image_size
,
1109 ir_intrinsic_image_samples
,
1110 ir_intrinsic_image_atomic_inc_wrap
,
1111 ir_intrinsic_image_atomic_dec_wrap
,
1113 ir_intrinsic_ssbo_load
,
1114 ir_intrinsic_ssbo_store
= MAKE_INTRINSIC_FOR_TYPE(store
, ssbo
),
1115 ir_intrinsic_ssbo_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, ssbo
),
1116 ir_intrinsic_ssbo_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, ssbo
),
1117 ir_intrinsic_ssbo_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, ssbo
),
1118 ir_intrinsic_ssbo_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, ssbo
),
1119 ir_intrinsic_ssbo_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, ssbo
),
1120 ir_intrinsic_ssbo_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, ssbo
),
1121 ir_intrinsic_ssbo_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, ssbo
),
1122 ir_intrinsic_ssbo_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, ssbo
),
1124 ir_intrinsic_memory_barrier
,
1125 ir_intrinsic_shader_clock
,
1126 ir_intrinsic_group_memory_barrier
,
1127 ir_intrinsic_memory_barrier_atomic_counter
,
1128 ir_intrinsic_memory_barrier_buffer
,
1129 ir_intrinsic_memory_barrier_image
,
1130 ir_intrinsic_memory_barrier_shared
,
1131 ir_intrinsic_begin_invocation_interlock
,
1132 ir_intrinsic_end_invocation_interlock
,
1134 ir_intrinsic_vote_all
,
1135 ir_intrinsic_vote_any
,
1136 ir_intrinsic_vote_eq
,
1137 ir_intrinsic_ballot
,
1138 ir_intrinsic_read_invocation
,
1139 ir_intrinsic_read_first_invocation
,
1141 ir_intrinsic_shared_load
,
1142 ir_intrinsic_shared_store
= MAKE_INTRINSIC_FOR_TYPE(store
, shared
),
1143 ir_intrinsic_shared_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, shared
),
1144 ir_intrinsic_shared_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, shared
),
1145 ir_intrinsic_shared_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, shared
),
1146 ir_intrinsic_shared_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, shared
),
1147 ir_intrinsic_shared_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, shared
),
1148 ir_intrinsic_shared_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, shared
),
1149 ir_intrinsic_shared_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, shared
),
1150 ir_intrinsic_shared_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, shared
),
1155 * The representation of a function instance; may be the full definition or
1156 * simply a prototype.
1158 class ir_function_signature
: public ir_instruction
{
1159 /* An ir_function_signature will be part of the list of signatures in
1163 ir_function_signature(const glsl_type
*return_type
,
1164 builtin_available_predicate builtin_avail
= NULL
);
1166 virtual ir_function_signature
*clone(void *mem_ctx
,
1167 struct hash_table
*ht
) const;
1168 ir_function_signature
*clone_prototype(void *mem_ctx
,
1169 struct hash_table
*ht
) const;
1171 virtual void accept(ir_visitor
*v
)
1176 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1179 * Attempt to evaluate this function as a constant expression,
1180 * given a list of the actual parameters and the variable context.
1181 * Returns NULL for non-built-ins.
1183 ir_constant
*constant_expression_value(void *mem_ctx
,
1184 exec_list
*actual_parameters
,
1185 struct hash_table
*variable_context
);
1188 * Get the name of the function for which this is a signature
1190 const char *function_name() const;
1193 * Get a handle to the function for which this is a signature
1195 * There is no setter function, this function returns a \c const pointer,
1196 * and \c ir_function_signature::_function is private for a reason. The
1197 * only way to make a connection between a function and function signature
1198 * is via \c ir_function::add_signature. This helps ensure that certain
1199 * invariants (i.e., a function signature is in the list of signatures for
1200 * its \c _function) are met.
1202 * \sa ir_function::add_signature
1204 inline const class ir_function
*function() const
1206 return this->_function
;
1210 * Check whether the qualifiers match between this signature's parameters
1211 * and the supplied parameter list. If not, returns the name of the first
1212 * parameter with mismatched qualifiers (for use in error messages).
1214 const char *qualifiers_match(exec_list
*params
);
1217 * Replace the current parameter list with the given one. This is useful
1218 * if the current information came from a prototype, and either has invalid
1219 * or missing parameter names.
1221 void replace_parameters(exec_list
*new_params
);
1224 * Function return type.
1226 * \note The precision qualifier is stored separately in return_precision.
1228 const struct glsl_type
*return_type
;
1231 * List of ir_variable of function parameters.
1233 * This represents the storage. The paramaters passed in a particular
1234 * call will be in ir_call::actual_paramaters.
1236 struct exec_list parameters
;
1238 /** Whether or not this function has a body (which may be empty). */
1239 unsigned is_defined
:1;
1242 * Precision qualifier for the return type.
1244 * See the comment for ir_variable_data::precision for more details.
1246 unsigned return_precision
:2;
1248 /** Whether or not this function signature is a built-in. */
1249 bool is_builtin() const;
1252 * Whether or not this function is an intrinsic to be implemented
1255 inline bool is_intrinsic() const
1257 return intrinsic_id
!= ir_intrinsic_invalid
;
1260 /** Indentifier for this intrinsic. */
1261 enum ir_intrinsic_id intrinsic_id
;
1263 /** Whether or not a built-in is available for this shader. */
1264 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1266 /** Body of instructions in the function. */
1267 struct exec_list body
;
1271 * A function pointer to a predicate that answers whether a built-in
1272 * function is available in the current shader. NULL if not a built-in.
1274 builtin_available_predicate builtin_avail
;
1276 /** Function of which this signature is one overload. */
1277 class ir_function
*_function
;
1279 /** Function signature of which this one is a prototype clone */
1280 const ir_function_signature
*origin
;
1282 friend class ir_function
;
1285 * Helper function to run a list of instructions for constant
1286 * expression evaluation.
1288 * The hash table represents the values of the visible variables.
1289 * There are no scoping issues because the table is indexed on
1290 * ir_variable pointers, not variable names.
1292 * Returns false if the expression is not constant, true otherwise,
1293 * and the value in *result if result is non-NULL.
1295 bool constant_expression_evaluate_expression_list(void *mem_ctx
,
1296 const struct exec_list
&body
,
1297 struct hash_table
*variable_context
,
1298 ir_constant
**result
);
1303 * Header for tracking multiple overloaded functions with the same name.
1304 * Contains a list of ir_function_signatures representing each of the
1307 class ir_function
: public ir_instruction
{
1309 ir_function(const char *name
);
1311 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1313 virtual void accept(ir_visitor
*v
)
1318 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1320 void add_signature(ir_function_signature
*sig
)
1322 sig
->_function
= this;
1323 this->signatures
.push_tail(sig
);
1327 * Find a signature that matches a set of actual parameters, taking implicit
1328 * conversions into account. Also flags whether the match was exact.
1330 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1331 const exec_list
*actual_param
,
1332 bool allow_builtins
,
1333 bool *match_is_exact
);
1336 * Find a signature that matches a set of actual parameters, taking implicit
1337 * conversions into account.
1339 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1340 const exec_list
*actual_param
,
1341 bool allow_builtins
);
1344 * Find a signature that exactly matches a set of actual parameters without
1345 * any implicit type conversions.
1347 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1348 const exec_list
*actual_ps
);
1351 * Name of the function.
1355 /** Whether or not this function has a signature that isn't a built-in. */
1356 bool has_user_signature();
1359 * List of ir_function_signature for each overloaded function with this name.
1361 struct exec_list signatures
;
1364 * is this function a subroutine type declaration
1365 * e.g. subroutine void type1(float arg1);
1370 * is this function associated to a subroutine type
1371 * e.g. subroutine (type1, type2) function_name { function_body };
1372 * would have num_subroutine_types 2,
1373 * and pointers to the type1 and type2 types.
1375 int num_subroutine_types
;
1376 const struct glsl_type
**subroutine_types
;
1378 int subroutine_index
;
1381 inline const char *ir_function_signature::function_name() const
1383 return this->_function
->name
;
1389 * IR instruction representing high-level if-statements
1391 class ir_if
: public ir_instruction
{
1393 ir_if(ir_rvalue
*condition
)
1394 : ir_instruction(ir_type_if
), condition(condition
)
1398 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1400 virtual void accept(ir_visitor
*v
)
1405 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1407 ir_rvalue
*condition
;
1408 /** List of ir_instruction for the body of the then branch */
1409 exec_list then_instructions
;
1410 /** List of ir_instruction for the body of the else branch */
1411 exec_list else_instructions
;
1416 * IR instruction representing a high-level loop structure.
1418 class ir_loop
: public ir_instruction
{
1422 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1424 virtual void accept(ir_visitor
*v
)
1429 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1431 /** List of ir_instruction that make up the body of the loop. */
1432 exec_list body_instructions
;
1436 class ir_assignment
: public ir_instruction
{
1438 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1441 * Construct an assignment with an explicit write mask
1444 * Since a write mask is supplied, the LHS must already be a bare
1445 * \c ir_dereference. The cannot be any swizzles in the LHS.
1447 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1448 unsigned write_mask
);
1450 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1452 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1453 struct hash_table
*variable_context
= NULL
);
1455 virtual void accept(ir_visitor
*v
)
1460 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1463 * Get a whole variable written by an assignment
1465 * If the LHS of the assignment writes a whole variable, the variable is
1466 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1469 * - Assigning to a scalar
1470 * - Assigning to all components of a vector
1471 * - Whole array (or matrix) assignment
1472 * - Whole structure assignment
1474 ir_variable
*whole_variable_written();
1477 * Set the LHS of an assignment
1479 void set_lhs(ir_rvalue
*lhs
);
1482 * Left-hand side of the assignment.
1484 * This should be treated as read only. If you need to set the LHS of an
1485 * assignment, use \c ir_assignment::set_lhs.
1487 ir_dereference
*lhs
;
1490 * Value being assigned
1495 * Optional condition for the assignment.
1497 ir_rvalue
*condition
;
1501 * Component mask written
1503 * For non-vector types in the LHS, this field will be zero. For vector
1504 * types, a bit will be set for each component that is written. Note that
1505 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1507 * A partially-set write mask means that each enabled channel gets
1508 * the value from a consecutive channel of the rhs. For example,
1509 * to write just .xyw of gl_FrontColor with color:
1511 * (assign (constant bool (1)) (xyw)
1512 * (var_ref gl_FragColor)
1513 * (swiz xyw (var_ref color)))
1515 unsigned write_mask
:4;
1518 #include "ir_expression_operation.h"
1520 extern const char *const ir_expression_operation_strings
[ir_last_opcode
+ 1];
1521 extern const char *const ir_expression_operation_enum_strings
[ir_last_opcode
+ 1];
1523 class ir_expression
: public ir_rvalue
{
1525 ir_expression(int op
, const struct glsl_type
*type
,
1526 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1527 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1530 * Constructor for unary operation expressions
1532 ir_expression(int op
, ir_rvalue
*);
1535 * Constructor for binary operation expressions
1537 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1540 * Constructor for ternary operation expressions
1542 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1544 virtual bool equals(const ir_instruction
*ir
,
1545 enum ir_node_type ignore
= ir_type_unset
) const;
1547 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1550 * Attempt to constant-fold the expression
1552 * The "variable_context" hash table links ir_variable * to ir_constant *
1553 * that represent the variables' values. \c NULL represents an empty
1556 * If the expression cannot be constant folded, this method will return
1559 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1560 struct hash_table
*variable_context
= NULL
);
1563 * This is only here for ir_reader to used for testing purposes please use
1564 * the precomputed num_operands field if you need the number of operands.
1566 static unsigned get_num_operands(ir_expression_operation
);
1569 * Return whether the expression operates on vectors horizontally.
1571 bool is_horizontal() const
1573 return operation
== ir_binop_all_equal
||
1574 operation
== ir_binop_any_nequal
||
1575 operation
== ir_binop_dot
||
1576 operation
== ir_binop_vector_extract
||
1577 operation
== ir_triop_vector_insert
||
1578 operation
== ir_binop_ubo_load
||
1579 operation
== ir_quadop_vector
;
1583 * Do a reverse-lookup to translate the given string into an operator.
1585 static ir_expression_operation
get_operator(const char *);
1587 virtual void accept(ir_visitor
*v
)
1592 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1594 virtual ir_variable
*variable_referenced() const;
1597 * Determine the number of operands used by an expression
1599 void init_num_operands()
1601 if (operation
== ir_quadop_vector
) {
1602 num_operands
= this->type
->vector_elements
;
1604 num_operands
= get_num_operands(operation
);
1608 ir_expression_operation operation
;
1609 ir_rvalue
*operands
[4];
1610 uint8_t num_operands
;
1615 * HIR instruction representing a high-level function call, containing a list
1616 * of parameters and returning a value in the supplied temporary.
1618 class ir_call
: public ir_instruction
{
1620 ir_call(ir_function_signature
*callee
,
1621 ir_dereference_variable
*return_deref
,
1622 exec_list
*actual_parameters
)
1623 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(NULL
), array_idx(NULL
)
1625 assert(callee
->return_type
!= NULL
);
1626 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1629 ir_call(ir_function_signature
*callee
,
1630 ir_dereference_variable
*return_deref
,
1631 exec_list
*actual_parameters
,
1632 ir_variable
*var
, ir_rvalue
*array_idx
)
1633 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(var
), array_idx(array_idx
)
1635 assert(callee
->return_type
!= NULL
);
1636 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1639 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1641 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1642 struct hash_table
*variable_context
= NULL
);
1644 virtual void accept(ir_visitor
*v
)
1649 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1652 * Get the name of the function being called.
1654 const char *callee_name() const
1656 return callee
->function_name();
1660 * Generates an inline version of the function before @ir,
1661 * storing the return value in return_deref.
1663 void generate_inline(ir_instruction
*ir
);
1666 * Storage for the function's return value.
1667 * This must be NULL if the return type is void.
1669 ir_dereference_variable
*return_deref
;
1672 * The specific function signature being called.
1674 ir_function_signature
*callee
;
1676 /* List of ir_rvalue of paramaters passed in this call. */
1677 exec_list actual_parameters
;
1680 * ARB_shader_subroutine support -
1681 * the subroutine uniform variable and array index
1682 * rvalue to be used in the lowering pass later.
1684 ir_variable
*sub_var
;
1685 ir_rvalue
*array_idx
;
1690 * \name Jump-like IR instructions.
1692 * These include \c break, \c continue, \c return, and \c discard.
1695 class ir_jump
: public ir_instruction
{
1697 ir_jump(enum ir_node_type t
)
1703 class ir_return
: public ir_jump
{
1706 : ir_jump(ir_type_return
), value(NULL
)
1710 ir_return(ir_rvalue
*value
)
1711 : ir_jump(ir_type_return
), value(value
)
1715 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1717 ir_rvalue
*get_value() const
1722 virtual void accept(ir_visitor
*v
)
1727 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1734 * Jump instructions used inside loops
1736 * These include \c break and \c continue. The \c break within a loop is
1737 * different from the \c break within a switch-statement.
1739 * \sa ir_switch_jump
1741 class ir_loop_jump
: public ir_jump
{
1748 ir_loop_jump(jump_mode mode
)
1749 : ir_jump(ir_type_loop_jump
)
1754 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1756 virtual void accept(ir_visitor
*v
)
1761 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1763 bool is_break() const
1765 return mode
== jump_break
;
1768 bool is_continue() const
1770 return mode
== jump_continue
;
1773 /** Mode selector for the jump instruction. */
1774 enum jump_mode mode
;
1778 * IR instruction representing discard statements.
1780 class ir_discard
: public ir_jump
{
1783 : ir_jump(ir_type_discard
)
1785 this->condition
= NULL
;
1788 ir_discard(ir_rvalue
*cond
)
1789 : ir_jump(ir_type_discard
)
1791 this->condition
= cond
;
1794 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1796 virtual void accept(ir_visitor
*v
)
1801 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1803 ir_rvalue
*condition
;
1809 * IR instruction representing demote statements from
1810 * GL_EXT_demote_to_helper_invocation.
1812 class ir_demote
: public ir_instruction
{
1815 : ir_instruction(ir_type_demote
)
1819 virtual ir_demote
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1821 virtual void accept(ir_visitor
*v
)
1826 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1831 * Texture sampling opcodes used in ir_texture
1833 enum ir_texture_opcode
{
1834 ir_tex
, /**< Regular texture look-up */
1835 ir_txb
, /**< Texture look-up with LOD bias */
1836 ir_txl
, /**< Texture look-up with explicit LOD */
1837 ir_txd
, /**< Texture look-up with partial derivatvies */
1838 ir_txf
, /**< Texel fetch with explicit LOD */
1839 ir_txf_ms
, /**< Multisample texture fetch */
1840 ir_txs
, /**< Texture size */
1841 ir_lod
, /**< Texture lod query */
1842 ir_tg4
, /**< Texture gather */
1843 ir_query_levels
, /**< Texture levels query */
1844 ir_texture_samples
, /**< Texture samples query */
1845 ir_samples_identical
, /**< Query whether all samples are definitely identical. */
1850 * IR instruction to sample a texture
1852 * The specific form of the IR instruction depends on the \c mode value
1853 * selected from \c ir_texture_opcodes. In the printed IR, these will
1856 * Texel offset (0 or an expression)
1857 * | Projection divisor
1858 * | | Shadow comparator
1861 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1862 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1863 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1864 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1865 * (txf <type> <sampler> <coordinate> 0 <lod>)
1867 * <type> <sampler> <coordinate> <sample_index>)
1868 * (txs <type> <sampler> <lod>)
1869 * (lod <type> <sampler> <coordinate>)
1870 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1871 * (query_levels <type> <sampler>)
1872 * (samples_identical <sampler> <coordinate>)
1874 class ir_texture
: public ir_rvalue
{
1876 ir_texture(enum ir_texture_opcode op
)
1877 : ir_rvalue(ir_type_texture
),
1878 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1879 shadow_comparator(NULL
), offset(NULL
)
1881 memset(&lod_info
, 0, sizeof(lod_info
));
1884 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1886 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1887 struct hash_table
*variable_context
= NULL
);
1889 virtual void accept(ir_visitor
*v
)
1894 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1896 virtual bool equals(const ir_instruction
*ir
,
1897 enum ir_node_type ignore
= ir_type_unset
) const;
1900 * Return a string representing the ir_texture_opcode.
1902 const char *opcode_string();
1904 /** Set the sampler and type. */
1905 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1908 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1910 static ir_texture_opcode
get_opcode(const char *);
1912 enum ir_texture_opcode op
;
1914 /** Sampler to use for the texture access. */
1915 ir_dereference
*sampler
;
1917 /** Texture coordinate to sample */
1918 ir_rvalue
*coordinate
;
1921 * Value used for projective divide.
1923 * If there is no projective divide (the common case), this will be
1924 * \c NULL. Optimization passes should check for this to point to a constant
1925 * of 1.0 and replace that with \c NULL.
1927 ir_rvalue
*projector
;
1930 * Coordinate used for comparison on shadow look-ups.
1932 * If there is no shadow comparison, this will be \c NULL. For the
1933 * \c ir_txf opcode, this *must* be \c NULL.
1935 ir_rvalue
*shadow_comparator
;
1937 /** Texel offset. */
1941 ir_rvalue
*lod
; /**< Floating point LOD */
1942 ir_rvalue
*bias
; /**< Floating point LOD bias */
1943 ir_rvalue
*sample_index
; /**< MSAA sample index */
1944 ir_rvalue
*component
; /**< Gather component selector */
1946 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1947 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1953 struct ir_swizzle_mask
{
1960 * Number of components in the swizzle.
1962 unsigned num_components
:3;
1965 * Does the swizzle contain duplicate components?
1967 * L-value swizzles cannot contain duplicate components.
1969 unsigned has_duplicates
:1;
1973 class ir_swizzle
: public ir_rvalue
{
1975 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
1978 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
1980 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
1982 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
1984 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1985 struct hash_table
*variable_context
= NULL
);
1988 * Construct an ir_swizzle from the textual representation. Can fail.
1990 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
1992 virtual void accept(ir_visitor
*v
)
1997 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1999 virtual bool equals(const ir_instruction
*ir
,
2000 enum ir_node_type ignore
= ir_type_unset
) const;
2002 bool is_lvalue(const struct _mesa_glsl_parse_state
*state
) const
2004 return val
->is_lvalue(state
) && !mask
.has_duplicates
;
2008 * Get the variable that is ultimately referenced by an r-value
2010 virtual ir_variable
*variable_referenced() const;
2013 ir_swizzle_mask mask
;
2017 * Initialize the mask component of a swizzle
2019 * This is used by the \c ir_swizzle constructors.
2021 void init_mask(const unsigned *components
, unsigned count
);
2025 class ir_dereference
: public ir_rvalue
{
2027 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2029 bool is_lvalue(const struct _mesa_glsl_parse_state
*state
) const;
2032 * Get the variable that is ultimately referenced by an r-value
2034 virtual ir_variable
*variable_referenced() const = 0;
2037 ir_dereference(enum ir_node_type t
)
2044 class ir_dereference_variable
: public ir_dereference
{
2046 ir_dereference_variable(ir_variable
*var
);
2048 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2049 struct hash_table
*) const;
2051 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2052 struct hash_table
*variable_context
= NULL
);
2054 virtual bool equals(const ir_instruction
*ir
,
2055 enum ir_node_type ignore
= ir_type_unset
) const;
2058 * Get the variable that is ultimately referenced by an r-value
2060 virtual ir_variable
*variable_referenced() const
2065 virtual ir_variable
*whole_variable_referenced()
2067 /* ir_dereference_variable objects always dereference the entire
2068 * variable. However, if this dereference is dereferenced by anything
2069 * else, the complete deferefernce chain is not a whole-variable
2070 * dereference. This method should only be called on the top most
2071 * ir_rvalue in a dereference chain.
2076 virtual void accept(ir_visitor
*v
)
2081 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2084 * Object being dereferenced.
2090 class ir_dereference_array
: public ir_dereference
{
2092 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2094 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2096 virtual ir_dereference_array
*clone(void *mem_ctx
,
2097 struct hash_table
*) const;
2099 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2100 struct hash_table
*variable_context
= NULL
);
2102 virtual bool equals(const ir_instruction
*ir
,
2103 enum ir_node_type ignore
= ir_type_unset
) const;
2106 * Get the variable that is ultimately referenced by an r-value
2108 virtual ir_variable
*variable_referenced() const
2110 return this->array
->variable_referenced();
2113 virtual void accept(ir_visitor
*v
)
2118 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2121 ir_rvalue
*array_index
;
2124 void set_array(ir_rvalue
*value
);
2128 class ir_dereference_record
: public ir_dereference
{
2130 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2132 ir_dereference_record(ir_variable
*var
, const char *field
);
2134 virtual ir_dereference_record
*clone(void *mem_ctx
,
2135 struct hash_table
*) const;
2137 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2138 struct hash_table
*variable_context
= NULL
);
2141 * Get the variable that is ultimately referenced by an r-value
2143 virtual ir_variable
*variable_referenced() const
2145 return this->record
->variable_referenced();
2148 virtual void accept(ir_visitor
*v
)
2153 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2161 * Data stored in an ir_constant
2163 union ir_constant_data
{
2174 class ir_constant
: public ir_rvalue
{
2176 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2177 ir_constant(bool b
, unsigned vector_elements
=1);
2178 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2179 ir_constant(int i
, unsigned vector_elements
=1);
2180 ir_constant(float f
, unsigned vector_elements
=1);
2181 ir_constant(double d
, unsigned vector_elements
=1);
2182 ir_constant(uint64_t u64
, unsigned vector_elements
=1);
2183 ir_constant(int64_t i64
, unsigned vector_elements
=1);
2186 * Construct an ir_constant from a list of ir_constant values
2188 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2191 * Construct an ir_constant from a scalar component of another ir_constant
2193 * The new \c ir_constant inherits the type of the component from the
2197 * In the case of a matrix constant, the new constant is a scalar, \b not
2200 ir_constant(const ir_constant
*c
, unsigned i
);
2203 * Return a new ir_constant of the specified type containing all zeros.
2205 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2207 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2209 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2210 struct hash_table
*variable_context
= NULL
);
2212 virtual void accept(ir_visitor
*v
)
2217 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2219 virtual bool equals(const ir_instruction
*ir
,
2220 enum ir_node_type ignore
= ir_type_unset
) const;
2223 * Get a particular component of a constant as a specific type
2225 * This is useful, for example, to get a value from an integer constant
2226 * as a float or bool. This appears frequently when constructors are
2227 * called with all constant parameters.
2230 bool get_bool_component(unsigned i
) const;
2231 float get_float_component(unsigned i
) const;
2232 double get_double_component(unsigned i
) const;
2233 int get_int_component(unsigned i
) const;
2234 unsigned get_uint_component(unsigned i
) const;
2235 int64_t get_int64_component(unsigned i
) const;
2236 uint64_t get_uint64_component(unsigned i
) const;
2239 ir_constant
*get_array_element(unsigned i
) const;
2241 ir_constant
*get_record_field(int idx
);
2244 * Copy the values on another constant at a given offset.
2246 * The offset is ignored for array or struct copies, it's only for
2247 * scalars or vectors into vectors or matrices.
2249 * With identical types on both sides and zero offset it's clone()
2250 * without creating a new object.
2253 void copy_offset(ir_constant
*src
, int offset
);
2256 * Copy the values on another constant at a given offset and
2257 * following an assign-like mask.
2259 * The mask is ignored for scalars.
2261 * Note that this function only handles what assign can handle,
2262 * i.e. at most a vector as source and a column of a matrix as
2266 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2269 * Determine whether a constant has the same value as another constant
2271 * \sa ir_constant::is_zero, ir_constant::is_one,
2272 * ir_constant::is_negative_one
2274 bool has_value(const ir_constant
*) const;
2277 * Return true if this ir_constant represents the given value.
2279 * For vectors, this checks that each component is the given value.
2281 virtual bool is_value(float f
, int i
) const;
2282 virtual bool is_zero() const;
2283 virtual bool is_one() const;
2284 virtual bool is_negative_one() const;
2287 * Return true for constants that could be stored as 16-bit unsigned values.
2289 * Note that this will return true even for signed integer ir_constants, as
2290 * long as the value is non-negative and fits in 16-bits.
2292 virtual bool is_uint16_constant() const;
2295 * Value of the constant.
2297 * The field used to back the values supplied by the constant is determined
2298 * by the type associated with the \c ir_instruction. Constants may be
2299 * scalars, vectors, or matrices.
2301 union ir_constant_data value
;
2303 /* Array elements and structure fields */
2304 ir_constant
**const_elements
;
2308 * Parameterless constructor only used by the clone method
2314 * IR instruction to emit a vertex in a geometry shader.
2316 class ir_emit_vertex
: public ir_instruction
{
2318 ir_emit_vertex(ir_rvalue
*stream
)
2319 : ir_instruction(ir_type_emit_vertex
),
2325 virtual void accept(ir_visitor
*v
)
2330 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2332 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2335 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2337 int stream_id() const
2339 return stream
->as_constant()->value
.i
[0];
2346 * IR instruction to complete the current primitive and start a new one in a
2349 class ir_end_primitive
: public ir_instruction
{
2351 ir_end_primitive(ir_rvalue
*stream
)
2352 : ir_instruction(ir_type_end_primitive
),
2358 virtual void accept(ir_visitor
*v
)
2363 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2365 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2368 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2370 int stream_id() const
2372 return stream
->as_constant()->value
.i
[0];
2379 * IR instruction for tessellation control and compute shader barrier.
2381 class ir_barrier
: public ir_instruction
{
2384 : ir_instruction(ir_type_barrier
)
2388 virtual void accept(ir_visitor
*v
)
2393 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2395 return new(mem_ctx
) ir_barrier();
2398 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2404 * Apply a visitor to each IR node in a list
2407 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2410 * Validate invariants on each IR node in a list
2412 void validate_ir_tree(exec_list
*instructions
);
2414 struct _mesa_glsl_parse_state
;
2415 struct gl_shader_program
;
2418 * Detect whether an unlinked shader contains static recursion
2420 * If the list of instructions is determined to contain static recursion,
2421 * \c _mesa_glsl_error will be called to emit error messages for each function
2422 * that is in the recursion cycle.
2425 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2426 exec_list
*instructions
);
2429 * Detect whether a linked shader contains static recursion
2431 * If the list of instructions is determined to contain static recursion,
2432 * \c link_error_printf will be called to emit error messages for each function
2433 * that is in the recursion cycle. In addition,
2434 * \c gl_shader_program::LinkStatus will be set to false.
2437 detect_recursion_linked(struct gl_shader_program
*prog
,
2438 exec_list
*instructions
);
2441 * Make a clone of each IR instruction in a list
2443 * \param in List of IR instructions that are to be cloned
2444 * \param out List to hold the cloned instructions
2447 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2450 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2451 struct _mesa_glsl_parse_state
*state
);
2454 reparent_ir(exec_list
*list
, void *mem_ctx
);
2457 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2458 gl_shader_stage shader_stage
);
2461 prototype_string(const glsl_type
*return_type
, const char *name
,
2462 exec_list
*parameters
);
2465 mode_string(const ir_variable
*var
);
2468 * Built-in / reserved GL variables names start with "gl_"
2471 is_gl_identifier(const char *s
)
2473 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2477 #endif /* __cplusplus */
2479 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2480 struct _mesa_glsl_parse_state
*state
);
2483 fprint_ir(FILE *f
, const void *instruction
);
2485 extern const struct gl_builtin_uniform_desc
*
2486 _mesa_glsl_get_builtin_uniform_desc(const char *name
);
2493 vertices_per_prim(GLenum prim
);