3 * Copyright © 2010 Intel Corporation
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
31 #include "util/ralloc.h"
32 #include "util/format/u_format.h"
33 #include "util/half_float.h"
34 #include "compiler/glsl_types.h"
36 #include "ir_visitor.h"
37 #include "ir_hierarchical_visitor.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
,
81 ir_type_end_primitive
,
83 ir_type_max
, /**< maximum ir_type enum number, for validation */
84 ir_type_unset
= ir_type_max
89 * Base class of all IR instructions
91 class ir_instruction
: public exec_node
{
93 enum ir_node_type ir_type
;
96 * GCC 4.7+ and clang warn when deleting an ir_instruction unless
97 * there's a virtual destructor present. Because we almost
98 * universally use ralloc for our memory management of
99 * ir_instructions, the destructor doesn't need to do any work.
101 virtual ~ir_instruction()
105 /** ir_print_visitor helper for debugging. */
106 void print(void) const;
107 void fprint(FILE *f
) const;
109 virtual void accept(ir_visitor
*) = 0;
110 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*) = 0;
111 virtual ir_instruction
*clone(void *mem_ctx
,
112 struct hash_table
*ht
) const = 0;
114 bool is_rvalue() const
116 return ir_type
== ir_type_dereference_array
||
117 ir_type
== ir_type_dereference_record
||
118 ir_type
== ir_type_dereference_variable
||
119 ir_type
== ir_type_constant
||
120 ir_type
== ir_type_expression
||
121 ir_type
== ir_type_swizzle
||
122 ir_type
== ir_type_texture
;
125 bool is_dereference() const
127 return ir_type
== ir_type_dereference_array
||
128 ir_type
== ir_type_dereference_record
||
129 ir_type
== ir_type_dereference_variable
;
134 return ir_type
== ir_type_loop_jump
||
135 ir_type
== ir_type_return
||
136 ir_type
== ir_type_discard
;
140 * \name IR instruction downcast functions
142 * These functions either cast the object to a derived class or return
143 * \c NULL if the object's type does not match the specified derived class.
144 * Additional downcast functions will be added as needed.
147 #define AS_BASE(TYPE) \
148 class ir_##TYPE *as_##TYPE() \
150 assume(this != NULL); \
151 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
153 const class ir_##TYPE *as_##TYPE() const \
155 assume(this != NULL); \
156 return is_##TYPE() ? (ir_##TYPE *) this : NULL; \
164 #define AS_CHILD(TYPE) \
165 class ir_##TYPE * as_##TYPE() \
167 assume(this != NULL); \
168 return ir_type == ir_type_##TYPE ? (ir_##TYPE *) this : NULL; \
170 const class ir_##TYPE * as_##TYPE() const \
172 assume(this != NULL); \
173 return ir_type == ir_type_##TYPE ? (const ir_##TYPE *) this : NULL; \
177 AS_CHILD(dereference_array
)
178 AS_CHILD(dereference_variable
)
179 AS_CHILD(dereference_record
)
194 * IR equality method: Return true if the referenced instruction would
195 * return the same value as this one.
197 * This intended to be used for CSE and algebraic optimizations, on rvalues
198 * in particular. No support for other instruction types (assignments,
199 * jumps, calls, etc.) is planned.
201 virtual bool equals(const ir_instruction
*ir
,
202 enum ir_node_type ignore
= ir_type_unset
) const;
205 ir_instruction(enum ir_node_type t
)
213 assert(!"Should not get here.");
219 * The base class for all "values"/expression trees.
221 class ir_rvalue
: public ir_instruction
{
223 const struct glsl_type
*type
;
225 virtual ir_rvalue
*clone(void *mem_ctx
, struct hash_table
*) const;
227 virtual void accept(ir_visitor
*v
)
232 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
234 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
235 struct hash_table
*variable_context
= NULL
);
237 ir_rvalue
*as_rvalue_to_saturate();
239 virtual bool is_lvalue(const struct _mesa_glsl_parse_state
* = NULL
) const
245 * Get the variable that is ultimately referenced by an r-value
247 virtual ir_variable
*variable_referenced() const
254 * If an r-value is a reference to a whole variable, get that variable
257 * Pointer to a variable that is completely dereferenced by the r-value. If
258 * the r-value is not a dereference or the dereference does not access the
259 * entire variable (i.e., it's just one array element, struct field), \c NULL
262 virtual ir_variable
*whole_variable_referenced()
268 * Determine if an r-value has the value zero
270 * The base implementation of this function always returns \c false. The
271 * \c ir_constant class over-rides this function to return \c true \b only
272 * for vector and scalar types that have all elements set to the value
273 * zero (or \c false for booleans).
275 * \sa ir_constant::has_value, ir_rvalue::is_one, ir_rvalue::is_negative_one
277 virtual bool is_zero() const;
280 * Determine if an r-value has the value one
282 * The base implementation of this function always returns \c false. The
283 * \c ir_constant class over-rides this function to return \c true \b only
284 * for vector and scalar types that have all elements set to the value
285 * one (or \c true for booleans).
287 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_negative_one
289 virtual bool is_one() const;
292 * Determine if an r-value has the value negative one
294 * The base implementation of this function always returns \c false. The
295 * \c ir_constant class over-rides this function to return \c true \b only
296 * for vector and scalar types that have all elements set to the value
297 * negative one. For boolean types, the result is always \c false.
299 * \sa ir_constant::has_value, ir_rvalue::is_zero, ir_rvalue::is_one
301 virtual bool is_negative_one() const;
304 * Determine if an r-value is an unsigned integer constant which can be
307 * \sa ir_constant::is_uint16_constant.
309 virtual bool is_uint16_constant() const { return false; }
312 * Return a generic value of error_type.
314 * Allocation will be performed with 'mem_ctx' as ralloc owner.
316 static ir_rvalue
*error_value(void *mem_ctx
);
319 ir_rvalue(enum ir_node_type t
);
324 * Variable storage classes
326 enum ir_variable_mode
{
327 ir_var_auto
= 0, /**< Function local variables and globals. */
328 ir_var_uniform
, /**< Variable declared as a uniform. */
329 ir_var_shader_storage
, /**< Variable declared as an ssbo. */
330 ir_var_shader_shared
, /**< Variable declared as shared. */
335 ir_var_function_inout
,
336 ir_var_const_in
, /**< "in" param that must be a constant expression */
337 ir_var_system_value
, /**< Ex: front-face, instance-id, etc. */
338 ir_var_temporary
, /**< Temporary variable generated during compilation. */
339 ir_var_mode_count
/**< Number of variable modes */
343 * Enum keeping track of how a variable was declared. For error checking of
344 * the gl_PerVertex redeclaration rules.
346 enum ir_var_declaration_type
{
348 * Normal declaration (for most variables, this means an explicit
349 * declaration. Exception: temporaries are always implicitly declared, but
350 * they still use ir_var_declared_normally).
352 * Note: an ir_variable that represents a named interface block uses
353 * ir_var_declared_normally.
355 ir_var_declared_normally
= 0,
358 * Variable was explicitly declared (or re-declared) in an unnamed
361 ir_var_declared_in_block
,
364 * Variable is an implicitly declared built-in that has not been explicitly
365 * re-declared by the shader.
367 ir_var_declared_implicitly
,
370 * Variable is implicitly generated by the compiler and should not be
371 * visible via the API.
377 * \brief Layout qualifiers for gl_FragDepth.
379 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
380 * with a layout qualifier.
382 enum ir_depth_layout
{
383 ir_depth_layout_none
, /**< No depth layout is specified. */
385 ir_depth_layout_greater
,
386 ir_depth_layout_less
,
387 ir_depth_layout_unchanged
391 * \brief Convert depth layout qualifier to string.
394 depth_layout_string(ir_depth_layout layout
);
397 * Description of built-in state associated with a uniform
399 * \sa ir_variable::state_slots
401 struct ir_state_slot
{
402 gl_state_index16 tokens
[STATE_LENGTH
];
408 * Get the string value for an interpolation qualifier
410 * \return The string that would be used in a shader to specify \c
411 * mode will be returned.
413 * This function is used to generate error messages of the form "shader
414 * uses %s interpolation qualifier", so in the case where there is no
415 * interpolation qualifier, it returns "no".
417 * This function should only be used on a shader input or output variable.
419 const char *interpolation_string(unsigned interpolation
);
422 class ir_variable
: public ir_instruction
{
424 ir_variable(const struct glsl_type
*, const char *, ir_variable_mode
);
426 virtual ir_variable
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
428 virtual void accept(ir_visitor
*v
)
433 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
437 * Determine whether or not a variable is part of a uniform or
438 * shader storage block.
440 inline bool is_in_buffer_block() const
442 return (this->data
.mode
== ir_var_uniform
||
443 this->data
.mode
== ir_var_shader_storage
) &&
444 this->interface_type
!= NULL
;
448 * Determine whether or not a variable is part of a shader storage block.
450 inline bool is_in_shader_storage_block() const
452 return this->data
.mode
== ir_var_shader_storage
&&
453 this->interface_type
!= NULL
;
457 * Determine whether or not a variable is the declaration of an interface
460 * For the first declaration below, there will be an \c ir_variable named
461 * "instance" whose type and whose instance_type will be the same
462 * \c glsl_type. For the second declaration, there will be an \c ir_variable
463 * named "f" whose type is float and whose instance_type is B2.
465 * "instance" is an interface instance variable, but "f" is not.
475 inline bool is_interface_instance() const
477 return this->type
->without_array() == this->interface_type
;
481 * Return whether this variable contains a bindless sampler/image.
483 inline bool contains_bindless() const
485 if (!this->type
->contains_sampler() && !this->type
->contains_image())
488 return this->data
.bindless
|| this->data
.mode
!= ir_var_uniform
;
492 * Set this->interface_type on a newly created variable.
494 void init_interface_type(const struct glsl_type
*type
)
496 assert(this->interface_type
== NULL
);
497 this->interface_type
= type
;
498 if (this->is_interface_instance()) {
499 this->u
.max_ifc_array_access
=
500 ralloc_array(this, int, type
->length
);
501 for (unsigned i
= 0; i
< type
->length
; i
++) {
502 this->u
.max_ifc_array_access
[i
] = -1;
508 * Change this->interface_type on a variable that previously had a
509 * different, but compatible, interface_type. This is used during linking
510 * to set the size of arrays in interface blocks.
512 void change_interface_type(const struct glsl_type
*type
)
514 if (this->u
.max_ifc_array_access
!= NULL
) {
515 /* max_ifc_array_access has already been allocated, so make sure the
516 * new interface has the same number of fields as the old one.
518 assert(this->interface_type
->length
== type
->length
);
520 this->interface_type
= type
;
524 * Change this->interface_type on a variable that previously had a
525 * different, and incompatible, interface_type. This is used during
526 * compilation to handle redeclaration of the built-in gl_PerVertex
529 void reinit_interface_type(const struct glsl_type
*type
)
531 if (this->u
.max_ifc_array_access
!= NULL
) {
533 /* Redeclaring gl_PerVertex is only allowed if none of the built-ins
534 * it defines have been accessed yet; so it's safe to throw away the
535 * old max_ifc_array_access pointer, since all of its values are
538 for (unsigned i
= 0; i
< this->interface_type
->length
; i
++)
539 assert(this->u
.max_ifc_array_access
[i
] == -1);
541 ralloc_free(this->u
.max_ifc_array_access
);
542 this->u
.max_ifc_array_access
= NULL
;
544 this->interface_type
= NULL
;
545 init_interface_type(type
);
548 const glsl_type
*get_interface_type() const
550 return this->interface_type
;
553 enum glsl_interface_packing
get_interface_type_packing() const
555 return this->interface_type
->get_interface_packing();
558 * Get the max_ifc_array_access pointer
560 * A "set" function is not needed because the array is dynmically allocated
563 inline int *get_max_ifc_array_access()
565 assert(this->data
._num_state_slots
== 0);
566 return this->u
.max_ifc_array_access
;
569 inline unsigned get_num_state_slots() const
571 assert(!this->is_interface_instance()
572 || this->data
._num_state_slots
== 0);
573 return this->data
._num_state_slots
;
576 inline void set_num_state_slots(unsigned n
)
578 assert(!this->is_interface_instance()
580 this->data
._num_state_slots
= n
;
583 inline ir_state_slot
*get_state_slots()
585 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
588 inline const ir_state_slot
*get_state_slots() const
590 return this->is_interface_instance() ? NULL
: this->u
.state_slots
;
593 inline ir_state_slot
*allocate_state_slots(unsigned n
)
595 assert(!this->is_interface_instance());
597 this->u
.state_slots
= ralloc_array(this, ir_state_slot
, n
);
598 this->data
._num_state_slots
= 0;
600 if (this->u
.state_slots
!= NULL
)
601 this->data
._num_state_slots
= n
;
603 return this->u
.state_slots
;
606 inline bool is_interpolation_flat() const
608 return this->data
.interpolation
== INTERP_MODE_FLAT
||
609 this->type
->contains_integer() ||
610 this->type
->contains_double();
613 inline bool is_name_ralloced() const
615 return this->name
!= ir_variable::tmp_name
&&
616 this->name
!= this->name_storage
;
620 * Enable emitting extension warnings for this variable
622 void enable_extension_warning(const char *extension
);
625 * Get the extension warning string for this variable
627 * If warnings are not enabled, \c NULL is returned.
629 const char *get_extension_warning() const;
632 * Declared type of the variable
634 const struct glsl_type
*type
;
637 * Declared name of the variable
643 * If the name length fits into name_storage, it's used, otherwise
644 * the name is ralloc'd. shader-db mining showed that 70% of variables
645 * fit here. This is a win over ralloc where only ralloc_header has
646 * 20 bytes on 64-bit (28 bytes with DEBUG), and we can also skip malloc.
648 char name_storage
[16];
651 struct ir_variable_data
{
654 * Is the variable read-only?
656 * This is set for variables declared as \c const, shader inputs,
659 unsigned read_only
:1;
664 * Was an 'invariant' qualifier explicitly set in the shader?
666 * This is used to cross validate qualifiers.
668 unsigned explicit_invariant
:1;
670 * Is the variable invariant?
672 * It can happen either by having the 'invariant' qualifier
673 * explicitly set in the shader or by being used in calculations
674 * of other invariant variables.
676 unsigned invariant
:1;
680 * Has this variable been used for reading or writing?
682 * Several GLSL semantic checks require knowledge of whether or not a
683 * variable has been used. For example, it is an error to redeclare a
684 * variable as invariant after it has been used.
686 * This is maintained in the ast_to_hir.cpp path and during linking,
687 * but not in Mesa's fixed function or ARB program paths.
692 * Has this variable been statically assigned?
694 * This answers whether the variable was assigned in any path of
695 * the shader during ast_to_hir. This doesn't answer whether it is
696 * still written after dead code removal, nor is it maintained in
697 * non-ast_to_hir.cpp (GLSL parsing) paths.
702 * When separate shader programs are enabled, only input/outputs between
703 * the stages of a multi-stage separate program can be safely removed
704 * from the shader interface. Other input/outputs must remains active.
706 unsigned always_active_io
:1;
709 * Enum indicating how the variable was declared. See
710 * ir_var_declaration_type.
712 * This is used to detect certain kinds of illegal variable redeclarations.
714 unsigned how_declared
:2;
717 * Storage class of the variable.
719 * \sa ir_variable_mode
724 * Interpolation mode for shader inputs / outputs
726 * \sa glsl_interp_mode
728 unsigned interpolation
:2;
731 * Was the location explicitly set in the shader?
733 * If the location is explicitly set in the shader, it \b cannot be changed
734 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
737 unsigned explicit_location
:1;
738 unsigned explicit_index
:1;
741 * Was an initial binding explicitly set in the shader?
743 * If so, constant_value contains an integer ir_constant representing the
744 * initial binding point.
746 unsigned explicit_binding
:1;
749 * Was an initial component explicitly set in the shader?
751 unsigned explicit_component
:1;
754 * Does this variable have an initializer?
756 * This is used by the linker to cross-validiate initializers of global
759 unsigned has_initializer
:1;
762 * Is the initializer created by the compiler (glsl_zero_init)
764 unsigned is_implicit_initializer
:1;
767 * Is this variable a generic output or input that has not yet been matched
768 * up to a variable in another stage of the pipeline?
770 * This is used by the linker as scratch storage while assigning locations
771 * to generic inputs and outputs.
773 unsigned is_unmatched_generic_inout
:1;
776 * Is this varying used by transform feedback?
778 * This is used by the linker to decide if it's safe to pack the varying.
783 * Is this varying used only by transform feedback?
785 * This is used by the linker to decide if its safe to pack the varying.
787 unsigned is_xfb_only
:1;
790 * Was a transform feedback buffer set in the shader?
792 unsigned explicit_xfb_buffer
:1;
795 * Was a transform feedback offset set in the shader?
797 unsigned explicit_xfb_offset
:1;
800 * Was a transform feedback stride set in the shader?
802 unsigned explicit_xfb_stride
:1;
805 * If non-zero, then this variable may be packed along with other variables
806 * into a single varying slot, so this offset should be applied when
807 * accessing components. For example, an offset of 1 means that the x
808 * component of this variable is actually stored in component y of the
809 * location specified by \c location.
811 unsigned location_frac
:2;
814 * Layout of the matrix. Uses glsl_matrix_layout values.
816 unsigned matrix_layout
:2;
819 * Non-zero if this variable was created by lowering a named interface
822 unsigned from_named_ifc_block
:1;
825 * Non-zero if the variable must be a shader input. This is useful for
826 * constraints on function parameters.
828 unsigned must_be_shader_input
:1;
831 * Output index for dual source blending.
834 * The GLSL spec only allows the values 0 or 1 for the index in \b dual
840 * Precision qualifier.
842 * In desktop GLSL we do not care about precision qualifiers at all, in
843 * fact, the spec says that precision qualifiers are ignored.
845 * To make things easy, we make it so that this field is always
846 * GLSL_PRECISION_NONE on desktop shaders. This way all the variables
847 * have the same precision value and the checks we add in the compiler
848 * for this field will never break a desktop shader compile.
850 unsigned precision
:2;
853 * \brief Layout qualifier for gl_FragDepth.
855 * This is not equal to \c ir_depth_layout_none if and only if this
856 * variable is \c gl_FragDepth and a layout qualifier is specified.
858 ir_depth_layout depth_layout
:3;
863 unsigned memory_read_only
:1; /**< "readonly" qualifier. */
864 unsigned memory_write_only
:1; /**< "writeonly" qualifier. */
865 unsigned memory_coherent
:1;
866 unsigned memory_volatile
:1;
867 unsigned memory_restrict
:1;
870 * ARB_shader_storage_buffer_object
872 unsigned from_ssbo_unsized_array
:1; /**< unsized array buffer variable. */
874 unsigned implicit_sized_array
:1;
877 * Whether this is a fragment shader output implicitly initialized with
878 * the previous contents of the specified render target at the
879 * framebuffer location corresponding to this shader invocation.
881 unsigned fb_fetch_output
:1;
884 * Non-zero if this variable is considered bindless as defined by
885 * ARB_bindless_texture.
890 * Non-zero if this variable is considered bound as defined by
891 * ARB_bindless_texture.
896 * Emit a warning if this variable is accessed.
899 uint8_t warn_extension_index
;
903 * Image internal format if specified explicitly, otherwise
906 enum pipe_format image_format
;
910 * Number of state slots used
913 * This could be stored in as few as 7-bits, if necessary. If it is made
914 * smaller, add an assertion to \c ir_variable::allocate_state_slots to
917 uint16_t _num_state_slots
;
921 * Initial binding point for a sampler, atomic, or UBO.
923 * For array types, this represents the binding point for the first element.
928 * Storage location of the base of this variable
930 * The precise meaning of this field depends on the nature of the variable.
932 * - Vertex shader input: one of the values from \c gl_vert_attrib.
933 * - Vertex shader output: one of the values from \c gl_varying_slot.
934 * - Geometry shader input: one of the values from \c gl_varying_slot.
935 * - Geometry shader output: one of the values from \c gl_varying_slot.
936 * - Fragment shader input: one of the values from \c gl_varying_slot.
937 * - Fragment shader output: one of the values from \c gl_frag_result.
938 * - Uniforms: Per-stage uniform slot number for default uniform block.
939 * - Uniforms: Index within the uniform block definition for UBO members.
940 * - Non-UBO Uniforms: explicit location until linking then reused to
941 * store uniform slot number.
942 * - Other: This field is not currently used.
944 * If the variable is a uniform, shader input, or shader output, and the
945 * slot has not been assigned, the value will be -1.
950 * for glsl->tgsi/mesa IR we need to store the index into the
951 * parameters for uniforms, initially the code overloaded location
952 * but this causes problems with indirect samplers and AoA.
953 * This is assigned in _mesa_generate_parameters_list_for_uniforms.
958 * Vertex stream output identifier.
960 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
961 * stream of the i-th component.
966 * Atomic, transform feedback or block member offset.
971 * Highest element accessed with a constant expression array index
973 * Not used for non-array variables. -1 is never accessed.
975 int max_array_access
;
978 * Transform feedback buffer.
983 * Transform feedback stride.
988 * Allow (only) ir_variable direct access private members.
990 friend class ir_variable
;
994 * Value assigned in the initializer of a variable declared "const"
996 ir_constant
*constant_value
;
999 * Constant expression assigned in the initializer of the variable
1002 * This field and \c ::constant_value are distinct. Even if the two fields
1003 * refer to constants with the same value, they must point to separate
1006 ir_constant
*constant_initializer
;
1009 static const char *const warn_extension_table
[];
1013 * For variables which satisfy the is_interface_instance() predicate,
1014 * this points to an array of integers such that if the ith member of
1015 * the interface block is an array, max_ifc_array_access[i] is the
1016 * maximum array element of that member that has been accessed. If the
1017 * ith member of the interface block is not an array,
1018 * max_ifc_array_access[i] is unused.
1020 * For variables whose type is not an interface block, this pointer is
1023 int *max_ifc_array_access
;
1026 * Built-in state that backs this uniform
1028 * Once set at variable creation, \c state_slots must remain invariant.
1030 * If the variable is not a uniform, \c _num_state_slots will be zero
1031 * and \c state_slots will be \c NULL.
1033 ir_state_slot
*state_slots
;
1037 * For variables that are in an interface block or are an instance of an
1038 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
1040 * \sa ir_variable::location
1042 const glsl_type
*interface_type
;
1045 * Name used for anonymous compiler temporaries
1047 static const char tmp_name
[];
1051 * Should the construct keep names for ir_var_temporary variables?
1053 * When this global is false, names passed to the constructor for
1054 * \c ir_var_temporary variables will be dropped. Instead, the variable will
1055 * be named "compiler_temp". This name will be in static storage.
1058 * \b NEVER change the mode of an \c ir_var_temporary.
1061 * This variable is \b not thread-safe. It is global, \b not
1062 * per-context. It begins life false. A context can, at some point, make
1063 * it true. From that point on, it will be true forever. This should be
1064 * okay since it will only be set true while debugging.
1066 static bool temporaries_allocate_names
;
1070 * A function that returns whether a built-in function is available in the
1071 * current shading language (based on version, ES or desktop, and extensions).
1073 typedef bool (*builtin_available_predicate
)(const _mesa_glsl_parse_state
*);
1075 #define MAKE_INTRINSIC_FOR_TYPE(op, t) \
1076 ir_intrinsic_generic_ ## op - ir_intrinsic_generic_load + ir_intrinsic_ ## t ## _ ## load
1078 #define MAP_INTRINSIC_TO_TYPE(i, t) \
1079 ir_intrinsic_id(int(i) - int(ir_intrinsic_generic_load) + int(ir_intrinsic_ ## t ## _ ## load))
1081 enum ir_intrinsic_id
{
1082 ir_intrinsic_invalid
= 0,
1085 * \name Generic intrinsics
1087 * Each of these intrinsics has a specific version for shared variables and
1091 ir_intrinsic_generic_load
,
1092 ir_intrinsic_generic_store
,
1093 ir_intrinsic_generic_atomic_add
,
1094 ir_intrinsic_generic_atomic_and
,
1095 ir_intrinsic_generic_atomic_or
,
1096 ir_intrinsic_generic_atomic_xor
,
1097 ir_intrinsic_generic_atomic_min
,
1098 ir_intrinsic_generic_atomic_max
,
1099 ir_intrinsic_generic_atomic_exchange
,
1100 ir_intrinsic_generic_atomic_comp_swap
,
1103 ir_intrinsic_atomic_counter_read
,
1104 ir_intrinsic_atomic_counter_increment
,
1105 ir_intrinsic_atomic_counter_predecrement
,
1106 ir_intrinsic_atomic_counter_add
,
1107 ir_intrinsic_atomic_counter_and
,
1108 ir_intrinsic_atomic_counter_or
,
1109 ir_intrinsic_atomic_counter_xor
,
1110 ir_intrinsic_atomic_counter_min
,
1111 ir_intrinsic_atomic_counter_max
,
1112 ir_intrinsic_atomic_counter_exchange
,
1113 ir_intrinsic_atomic_counter_comp_swap
,
1115 ir_intrinsic_image_load
,
1116 ir_intrinsic_image_store
,
1117 ir_intrinsic_image_atomic_add
,
1118 ir_intrinsic_image_atomic_and
,
1119 ir_intrinsic_image_atomic_or
,
1120 ir_intrinsic_image_atomic_xor
,
1121 ir_intrinsic_image_atomic_min
,
1122 ir_intrinsic_image_atomic_max
,
1123 ir_intrinsic_image_atomic_exchange
,
1124 ir_intrinsic_image_atomic_comp_swap
,
1125 ir_intrinsic_image_size
,
1126 ir_intrinsic_image_samples
,
1127 ir_intrinsic_image_atomic_inc_wrap
,
1128 ir_intrinsic_image_atomic_dec_wrap
,
1130 ir_intrinsic_ssbo_load
,
1131 ir_intrinsic_ssbo_store
= MAKE_INTRINSIC_FOR_TYPE(store
, ssbo
),
1132 ir_intrinsic_ssbo_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, ssbo
),
1133 ir_intrinsic_ssbo_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, ssbo
),
1134 ir_intrinsic_ssbo_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, ssbo
),
1135 ir_intrinsic_ssbo_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, ssbo
),
1136 ir_intrinsic_ssbo_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, ssbo
),
1137 ir_intrinsic_ssbo_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, ssbo
),
1138 ir_intrinsic_ssbo_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, ssbo
),
1139 ir_intrinsic_ssbo_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, ssbo
),
1141 ir_intrinsic_memory_barrier
,
1142 ir_intrinsic_shader_clock
,
1143 ir_intrinsic_group_memory_barrier
,
1144 ir_intrinsic_memory_barrier_atomic_counter
,
1145 ir_intrinsic_memory_barrier_buffer
,
1146 ir_intrinsic_memory_barrier_image
,
1147 ir_intrinsic_memory_barrier_shared
,
1148 ir_intrinsic_begin_invocation_interlock
,
1149 ir_intrinsic_end_invocation_interlock
,
1151 ir_intrinsic_vote_all
,
1152 ir_intrinsic_vote_any
,
1153 ir_intrinsic_vote_eq
,
1154 ir_intrinsic_ballot
,
1155 ir_intrinsic_read_invocation
,
1156 ir_intrinsic_read_first_invocation
,
1158 ir_intrinsic_helper_invocation
,
1160 ir_intrinsic_shared_load
,
1161 ir_intrinsic_shared_store
= MAKE_INTRINSIC_FOR_TYPE(store
, shared
),
1162 ir_intrinsic_shared_atomic_add
= MAKE_INTRINSIC_FOR_TYPE(atomic_add
, shared
),
1163 ir_intrinsic_shared_atomic_and
= MAKE_INTRINSIC_FOR_TYPE(atomic_and
, shared
),
1164 ir_intrinsic_shared_atomic_or
= MAKE_INTRINSIC_FOR_TYPE(atomic_or
, shared
),
1165 ir_intrinsic_shared_atomic_xor
= MAKE_INTRINSIC_FOR_TYPE(atomic_xor
, shared
),
1166 ir_intrinsic_shared_atomic_min
= MAKE_INTRINSIC_FOR_TYPE(atomic_min
, shared
),
1167 ir_intrinsic_shared_atomic_max
= MAKE_INTRINSIC_FOR_TYPE(atomic_max
, shared
),
1168 ir_intrinsic_shared_atomic_exchange
= MAKE_INTRINSIC_FOR_TYPE(atomic_exchange
, shared
),
1169 ir_intrinsic_shared_atomic_comp_swap
= MAKE_INTRINSIC_FOR_TYPE(atomic_comp_swap
, shared
),
1174 * The representation of a function instance; may be the full definition or
1175 * simply a prototype.
1177 class ir_function_signature
: public ir_instruction
{
1178 /* An ir_function_signature will be part of the list of signatures in
1182 ir_function_signature(const glsl_type
*return_type
,
1183 builtin_available_predicate builtin_avail
= NULL
);
1185 virtual ir_function_signature
*clone(void *mem_ctx
,
1186 struct hash_table
*ht
) const;
1187 ir_function_signature
*clone_prototype(void *mem_ctx
,
1188 struct hash_table
*ht
) const;
1190 virtual void accept(ir_visitor
*v
)
1195 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1198 * Attempt to evaluate this function as a constant expression,
1199 * given a list of the actual parameters and the variable context.
1200 * Returns NULL for non-built-ins.
1202 ir_constant
*constant_expression_value(void *mem_ctx
,
1203 exec_list
*actual_parameters
,
1204 struct hash_table
*variable_context
);
1207 * Get the name of the function for which this is a signature
1209 const char *function_name() const;
1212 * Get a handle to the function for which this is a signature
1214 * There is no setter function, this function returns a \c const pointer,
1215 * and \c ir_function_signature::_function is private for a reason. The
1216 * only way to make a connection between a function and function signature
1217 * is via \c ir_function::add_signature. This helps ensure that certain
1218 * invariants (i.e., a function signature is in the list of signatures for
1219 * its \c _function) are met.
1221 * \sa ir_function::add_signature
1223 inline const class ir_function
*function() const
1225 return this->_function
;
1229 * Check whether the qualifiers match between this signature's parameters
1230 * and the supplied parameter list. If not, returns the name of the first
1231 * parameter with mismatched qualifiers (for use in error messages).
1233 const char *qualifiers_match(exec_list
*params
);
1236 * Replace the current parameter list with the given one. This is useful
1237 * if the current information came from a prototype, and either has invalid
1238 * or missing parameter names.
1240 void replace_parameters(exec_list
*new_params
);
1243 * Function return type.
1245 * \note The precision qualifier is stored separately in return_precision.
1247 const struct glsl_type
*return_type
;
1250 * List of ir_variable of function parameters.
1252 * This represents the storage. The paramaters passed in a particular
1253 * call will be in ir_call::actual_paramaters.
1255 struct exec_list parameters
;
1257 /** Whether or not this function has a body (which may be empty). */
1258 unsigned is_defined
:1;
1261 * Precision qualifier for the return type.
1263 * See the comment for ir_variable_data::precision for more details.
1265 unsigned return_precision
:2;
1267 /** Whether or not this function signature is a built-in. */
1268 bool is_builtin() const;
1271 * Whether or not this function is an intrinsic to be implemented
1274 inline bool is_intrinsic() const
1276 return intrinsic_id
!= ir_intrinsic_invalid
;
1279 /** Indentifier for this intrinsic. */
1280 enum ir_intrinsic_id intrinsic_id
;
1282 /** Whether or not a built-in is available for this shader. */
1283 bool is_builtin_available(const _mesa_glsl_parse_state
*state
) const;
1285 /** Body of instructions in the function. */
1286 struct exec_list body
;
1290 * A function pointer to a predicate that answers whether a built-in
1291 * function is available in the current shader. NULL if not a built-in.
1293 builtin_available_predicate builtin_avail
;
1295 /** Function of which this signature is one overload. */
1296 class ir_function
*_function
;
1298 /** Function signature of which this one is a prototype clone */
1299 const ir_function_signature
*origin
;
1301 friend class ir_function
;
1304 * Helper function to run a list of instructions for constant
1305 * expression evaluation.
1307 * The hash table represents the values of the visible variables.
1308 * There are no scoping issues because the table is indexed on
1309 * ir_variable pointers, not variable names.
1311 * Returns false if the expression is not constant, true otherwise,
1312 * and the value in *result if result is non-NULL.
1314 bool constant_expression_evaluate_expression_list(void *mem_ctx
,
1315 const struct exec_list
&body
,
1316 struct hash_table
*variable_context
,
1317 ir_constant
**result
);
1322 * Header for tracking multiple overloaded functions with the same name.
1323 * Contains a list of ir_function_signatures representing each of the
1326 class ir_function
: public ir_instruction
{
1328 ir_function(const char *name
);
1330 virtual ir_function
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1332 virtual void accept(ir_visitor
*v
)
1337 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1339 void add_signature(ir_function_signature
*sig
)
1341 sig
->_function
= this;
1342 this->signatures
.push_tail(sig
);
1346 * Find a signature that matches a set of actual parameters, taking implicit
1347 * conversions into account. Also flags whether the match was exact.
1349 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1350 const exec_list
*actual_param
,
1351 bool allow_builtins
,
1352 bool *match_is_exact
);
1355 * Find a signature that matches a set of actual parameters, taking implicit
1356 * conversions into account.
1358 ir_function_signature
*matching_signature(_mesa_glsl_parse_state
*state
,
1359 const exec_list
*actual_param
,
1360 bool allow_builtins
);
1363 * Find a signature that exactly matches a set of actual parameters without
1364 * any implicit type conversions.
1366 ir_function_signature
*exact_matching_signature(_mesa_glsl_parse_state
*state
,
1367 const exec_list
*actual_ps
);
1370 * Name of the function.
1374 /** Whether or not this function has a signature that isn't a built-in. */
1375 bool has_user_signature();
1378 * List of ir_function_signature for each overloaded function with this name.
1380 struct exec_list signatures
;
1383 * is this function a subroutine type declaration
1384 * e.g. subroutine void type1(float arg1);
1389 * is this function associated to a subroutine type
1390 * e.g. subroutine (type1, type2) function_name { function_body };
1391 * would have num_subroutine_types 2,
1392 * and pointers to the type1 and type2 types.
1394 int num_subroutine_types
;
1395 const struct glsl_type
**subroutine_types
;
1397 int subroutine_index
;
1400 inline const char *ir_function_signature::function_name() const
1402 return this->_function
->name
;
1408 * IR instruction representing high-level if-statements
1410 class ir_if
: public ir_instruction
{
1412 ir_if(ir_rvalue
*condition
)
1413 : ir_instruction(ir_type_if
), condition(condition
)
1417 virtual ir_if
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1419 virtual void accept(ir_visitor
*v
)
1424 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1426 ir_rvalue
*condition
;
1427 /** List of ir_instruction for the body of the then branch */
1428 exec_list then_instructions
;
1429 /** List of ir_instruction for the body of the else branch */
1430 exec_list else_instructions
;
1435 * IR instruction representing a high-level loop structure.
1437 class ir_loop
: public ir_instruction
{
1441 virtual ir_loop
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1443 virtual void accept(ir_visitor
*v
)
1448 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1450 /** List of ir_instruction that make up the body of the loop. */
1451 exec_list body_instructions
;
1455 class ir_assignment
: public ir_instruction
{
1457 ir_assignment(ir_rvalue
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
= NULL
);
1460 * Construct an assignment with an explicit write mask
1463 * Since a write mask is supplied, the LHS must already be a bare
1464 * \c ir_dereference. The cannot be any swizzles in the LHS.
1466 ir_assignment(ir_dereference
*lhs
, ir_rvalue
*rhs
, ir_rvalue
*condition
,
1467 unsigned write_mask
);
1469 virtual ir_assignment
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1471 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1472 struct hash_table
*variable_context
= NULL
);
1474 virtual void accept(ir_visitor
*v
)
1479 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1482 * Get a whole variable written by an assignment
1484 * If the LHS of the assignment writes a whole variable, the variable is
1485 * returned. Otherwise \c NULL is returned. Examples of whole-variable
1488 * - Assigning to a scalar
1489 * - Assigning to all components of a vector
1490 * - Whole array (or matrix) assignment
1491 * - Whole structure assignment
1493 ir_variable
*whole_variable_written();
1496 * Set the LHS of an assignment
1498 void set_lhs(ir_rvalue
*lhs
);
1501 * Left-hand side of the assignment.
1503 * This should be treated as read only. If you need to set the LHS of an
1504 * assignment, use \c ir_assignment::set_lhs.
1506 ir_dereference
*lhs
;
1509 * Value being assigned
1514 * Optional condition for the assignment.
1516 ir_rvalue
*condition
;
1520 * Component mask written
1522 * For non-vector types in the LHS, this field will be zero. For vector
1523 * types, a bit will be set for each component that is written. Note that
1524 * for \c vec2 and \c vec3 types only the lower bits will ever be set.
1526 * A partially-set write mask means that each enabled channel gets
1527 * the value from a consecutive channel of the rhs. For example,
1528 * to write just .xyw of gl_FrontColor with color:
1530 * (assign (constant bool (1)) (xyw)
1531 * (var_ref gl_FragColor)
1532 * (swiz xyw (var_ref color)))
1534 unsigned write_mask
:4;
1537 #include "ir_expression_operation.h"
1539 extern const char *const ir_expression_operation_strings
[ir_last_opcode
+ 1];
1540 extern const char *const ir_expression_operation_enum_strings
[ir_last_opcode
+ 1];
1542 class ir_expression
: public ir_rvalue
{
1544 ir_expression(int op
, const struct glsl_type
*type
,
1545 ir_rvalue
*op0
, ir_rvalue
*op1
= NULL
,
1546 ir_rvalue
*op2
= NULL
, ir_rvalue
*op3
= NULL
);
1549 * Constructor for unary operation expressions
1551 ir_expression(int op
, ir_rvalue
*);
1554 * Constructor for binary operation expressions
1556 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
);
1559 * Constructor for ternary operation expressions
1561 ir_expression(int op
, ir_rvalue
*op0
, ir_rvalue
*op1
, ir_rvalue
*op2
);
1563 virtual bool equals(const ir_instruction
*ir
,
1564 enum ir_node_type ignore
= ir_type_unset
) const;
1566 virtual ir_expression
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1569 * Attempt to constant-fold the expression
1571 * The "variable_context" hash table links ir_variable * to ir_constant *
1572 * that represent the variables' values. \c NULL represents an empty
1575 * If the expression cannot be constant folded, this method will return
1578 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1579 struct hash_table
*variable_context
= NULL
);
1582 * This is only here for ir_reader to used for testing purposes please use
1583 * the precomputed num_operands field if you need the number of operands.
1585 static unsigned get_num_operands(ir_expression_operation
);
1588 * Return whether the expression operates on vectors horizontally.
1590 bool is_horizontal() const
1592 return operation
== ir_binop_all_equal
||
1593 operation
== ir_binop_any_nequal
||
1594 operation
== ir_binop_dot
||
1595 operation
== ir_binop_vector_extract
||
1596 operation
== ir_triop_vector_insert
||
1597 operation
== ir_binop_ubo_load
||
1598 operation
== ir_quadop_vector
;
1602 * Do a reverse-lookup to translate the given string into an operator.
1604 static ir_expression_operation
get_operator(const char *);
1606 virtual void accept(ir_visitor
*v
)
1611 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1613 virtual ir_variable
*variable_referenced() const;
1616 * Determine the number of operands used by an expression
1618 void init_num_operands()
1620 if (operation
== ir_quadop_vector
) {
1621 num_operands
= this->type
->vector_elements
;
1623 num_operands
= get_num_operands(operation
);
1627 ir_expression_operation operation
;
1628 ir_rvalue
*operands
[4];
1629 uint8_t num_operands
;
1634 * HIR instruction representing a high-level function call, containing a list
1635 * of parameters and returning a value in the supplied temporary.
1637 class ir_call
: public ir_instruction
{
1639 ir_call(ir_function_signature
*callee
,
1640 ir_dereference_variable
*return_deref
,
1641 exec_list
*actual_parameters
)
1642 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(NULL
), array_idx(NULL
)
1644 assert(callee
->return_type
!= NULL
);
1645 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1648 ir_call(ir_function_signature
*callee
,
1649 ir_dereference_variable
*return_deref
,
1650 exec_list
*actual_parameters
,
1651 ir_variable
*var
, ir_rvalue
*array_idx
)
1652 : ir_instruction(ir_type_call
), return_deref(return_deref
), callee(callee
), sub_var(var
), array_idx(array_idx
)
1654 assert(callee
->return_type
!= NULL
);
1655 actual_parameters
->move_nodes_to(& this->actual_parameters
);
1658 virtual ir_call
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1660 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1661 struct hash_table
*variable_context
= NULL
);
1663 virtual void accept(ir_visitor
*v
)
1668 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1671 * Get the name of the function being called.
1673 const char *callee_name() const
1675 return callee
->function_name();
1679 * Generates an inline version of the function before @ir,
1680 * storing the return value in return_deref.
1682 void generate_inline(ir_instruction
*ir
);
1685 * Storage for the function's return value.
1686 * This must be NULL if the return type is void.
1688 ir_dereference_variable
*return_deref
;
1691 * The specific function signature being called.
1693 ir_function_signature
*callee
;
1695 /* List of ir_rvalue of paramaters passed in this call. */
1696 exec_list actual_parameters
;
1699 * ARB_shader_subroutine support -
1700 * the subroutine uniform variable and array index
1701 * rvalue to be used in the lowering pass later.
1703 ir_variable
*sub_var
;
1704 ir_rvalue
*array_idx
;
1709 * \name Jump-like IR instructions.
1711 * These include \c break, \c continue, \c return, and \c discard.
1714 class ir_jump
: public ir_instruction
{
1716 ir_jump(enum ir_node_type t
)
1722 class ir_return
: public ir_jump
{
1725 : ir_jump(ir_type_return
), value(NULL
)
1729 ir_return(ir_rvalue
*value
)
1730 : ir_jump(ir_type_return
), value(value
)
1734 virtual ir_return
*clone(void *mem_ctx
, struct hash_table
*) const;
1736 ir_rvalue
*get_value() const
1741 virtual void accept(ir_visitor
*v
)
1746 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1753 * Jump instructions used inside loops
1755 * These include \c break and \c continue. The \c break within a loop is
1756 * different from the \c break within a switch-statement.
1758 * \sa ir_switch_jump
1760 class ir_loop_jump
: public ir_jump
{
1767 ir_loop_jump(jump_mode mode
)
1768 : ir_jump(ir_type_loop_jump
)
1773 virtual ir_loop_jump
*clone(void *mem_ctx
, struct hash_table
*) const;
1775 virtual void accept(ir_visitor
*v
)
1780 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1782 bool is_break() const
1784 return mode
== jump_break
;
1787 bool is_continue() const
1789 return mode
== jump_continue
;
1792 /** Mode selector for the jump instruction. */
1793 enum jump_mode mode
;
1797 * IR instruction representing discard statements.
1799 class ir_discard
: public ir_jump
{
1802 : ir_jump(ir_type_discard
)
1804 this->condition
= NULL
;
1807 ir_discard(ir_rvalue
*cond
)
1808 : ir_jump(ir_type_discard
)
1810 this->condition
= cond
;
1813 virtual ir_discard
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1815 virtual void accept(ir_visitor
*v
)
1820 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1822 ir_rvalue
*condition
;
1828 * IR instruction representing demote statements from
1829 * GL_EXT_demote_to_helper_invocation.
1831 class ir_demote
: public ir_instruction
{
1834 : ir_instruction(ir_type_demote
)
1838 virtual ir_demote
*clone(void *mem_ctx
, struct hash_table
*ht
) const;
1840 virtual void accept(ir_visitor
*v
)
1845 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1850 * Texture sampling opcodes used in ir_texture
1852 enum ir_texture_opcode
{
1853 ir_tex
, /**< Regular texture look-up */
1854 ir_txb
, /**< Texture look-up with LOD bias */
1855 ir_txl
, /**< Texture look-up with explicit LOD */
1856 ir_txd
, /**< Texture look-up with partial derivatvies */
1857 ir_txf
, /**< Texel fetch with explicit LOD */
1858 ir_txf_ms
, /**< Multisample texture fetch */
1859 ir_txs
, /**< Texture size */
1860 ir_lod
, /**< Texture lod query */
1861 ir_tg4
, /**< Texture gather */
1862 ir_query_levels
, /**< Texture levels query */
1863 ir_texture_samples
, /**< Texture samples query */
1864 ir_samples_identical
, /**< Query whether all samples are definitely identical. */
1869 * IR instruction to sample a texture
1871 * The specific form of the IR instruction depends on the \c mode value
1872 * selected from \c ir_texture_opcodes. In the printed IR, these will
1875 * Texel offset (0 or an expression)
1876 * | Projection divisor
1877 * | | Shadow comparator
1880 * (tex <type> <sampler> <coordinate> 0 1 ( ))
1881 * (txb <type> <sampler> <coordinate> 0 1 ( ) <bias>)
1882 * (txl <type> <sampler> <coordinate> 0 1 ( ) <lod>)
1883 * (txd <type> <sampler> <coordinate> 0 1 ( ) (dPdx dPdy))
1884 * (txf <type> <sampler> <coordinate> 0 <lod>)
1886 * <type> <sampler> <coordinate> <sample_index>)
1887 * (txs <type> <sampler> <lod>)
1888 * (lod <type> <sampler> <coordinate>)
1889 * (tg4 <type> <sampler> <coordinate> <offset> <component>)
1890 * (query_levels <type> <sampler>)
1891 * (samples_identical <sampler> <coordinate>)
1893 class ir_texture
: public ir_rvalue
{
1895 ir_texture(enum ir_texture_opcode op
)
1896 : ir_rvalue(ir_type_texture
),
1897 op(op
), sampler(NULL
), coordinate(NULL
), projector(NULL
),
1898 shadow_comparator(NULL
), offset(NULL
)
1900 memset(&lod_info
, 0, sizeof(lod_info
));
1903 virtual ir_texture
*clone(void *mem_ctx
, struct hash_table
*) const;
1905 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
1906 struct hash_table
*variable_context
= NULL
);
1908 virtual void accept(ir_visitor
*v
)
1913 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
1915 virtual bool equals(const ir_instruction
*ir
,
1916 enum ir_node_type ignore
= ir_type_unset
) const;
1919 * Return a string representing the ir_texture_opcode.
1921 const char *opcode_string();
1923 /** Set the sampler and type. */
1924 void set_sampler(ir_dereference
*sampler
, const glsl_type
*type
);
1927 * Do a reverse-lookup to translate a string into an ir_texture_opcode.
1929 static ir_texture_opcode
get_opcode(const char *);
1931 enum ir_texture_opcode op
;
1933 /** Sampler to use for the texture access. */
1934 ir_dereference
*sampler
;
1936 /** Texture coordinate to sample */
1937 ir_rvalue
*coordinate
;
1940 * Value used for projective divide.
1942 * If there is no projective divide (the common case), this will be
1943 * \c NULL. Optimization passes should check for this to point to a constant
1944 * of 1.0 and replace that with \c NULL.
1946 ir_rvalue
*projector
;
1949 * Coordinate used for comparison on shadow look-ups.
1951 * If there is no shadow comparison, this will be \c NULL. For the
1952 * \c ir_txf opcode, this *must* be \c NULL.
1954 ir_rvalue
*shadow_comparator
;
1956 /** Texel offset. */
1960 ir_rvalue
*lod
; /**< Floating point LOD */
1961 ir_rvalue
*bias
; /**< Floating point LOD bias */
1962 ir_rvalue
*sample_index
; /**< MSAA sample index */
1963 ir_rvalue
*component
; /**< Gather component selector */
1965 ir_rvalue
*dPdx
; /**< Partial derivative of coordinate wrt X */
1966 ir_rvalue
*dPdy
; /**< Partial derivative of coordinate wrt Y */
1972 struct ir_swizzle_mask
{
1979 * Number of components in the swizzle.
1981 unsigned num_components
:3;
1984 * Does the swizzle contain duplicate components?
1986 * L-value swizzles cannot contain duplicate components.
1988 unsigned has_duplicates
:1;
1992 class ir_swizzle
: public ir_rvalue
{
1994 ir_swizzle(ir_rvalue
*, unsigned x
, unsigned y
, unsigned z
, unsigned w
,
1997 ir_swizzle(ir_rvalue
*val
, const unsigned *components
, unsigned count
);
1999 ir_swizzle(ir_rvalue
*val
, ir_swizzle_mask mask
);
2001 virtual ir_swizzle
*clone(void *mem_ctx
, struct hash_table
*) const;
2003 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2004 struct hash_table
*variable_context
= NULL
);
2007 * Construct an ir_swizzle from the textual representation. Can fail.
2009 static ir_swizzle
*create(ir_rvalue
*, const char *, unsigned vector_length
);
2011 virtual void accept(ir_visitor
*v
)
2016 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2018 virtual bool equals(const ir_instruction
*ir
,
2019 enum ir_node_type ignore
= ir_type_unset
) const;
2021 bool is_lvalue(const struct _mesa_glsl_parse_state
*state
) const
2023 return val
->is_lvalue(state
) && !mask
.has_duplicates
;
2027 * Get the variable that is ultimately referenced by an r-value
2029 virtual ir_variable
*variable_referenced() const;
2032 ir_swizzle_mask mask
;
2036 * Initialize the mask component of a swizzle
2038 * This is used by the \c ir_swizzle constructors.
2040 void init_mask(const unsigned *components
, unsigned count
);
2044 class ir_dereference
: public ir_rvalue
{
2046 virtual ir_dereference
*clone(void *mem_ctx
, struct hash_table
*) const = 0;
2048 bool is_lvalue(const struct _mesa_glsl_parse_state
*state
) const;
2051 * Get the variable that is ultimately referenced by an r-value
2053 virtual ir_variable
*variable_referenced() const = 0;
2056 * Get the precision. This can either come from the eventual variable that
2057 * is dereferenced, or from a record member.
2059 virtual int precision() const = 0;
2062 ir_dereference(enum ir_node_type t
)
2069 class ir_dereference_variable
: public ir_dereference
{
2071 ir_dereference_variable(ir_variable
*var
);
2073 virtual ir_dereference_variable
*clone(void *mem_ctx
,
2074 struct hash_table
*) const;
2076 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2077 struct hash_table
*variable_context
= NULL
);
2079 virtual bool equals(const ir_instruction
*ir
,
2080 enum ir_node_type ignore
= ir_type_unset
) const;
2083 * Get the variable that is ultimately referenced by an r-value
2085 virtual ir_variable
*variable_referenced() const
2090 virtual int precision() const
2092 return this->var
->data
.precision
;
2095 virtual ir_variable
*whole_variable_referenced()
2097 /* ir_dereference_variable objects always dereference the entire
2098 * variable. However, if this dereference is dereferenced by anything
2099 * else, the complete deferefernce chain is not a whole-variable
2100 * dereference. This method should only be called on the top most
2101 * ir_rvalue in a dereference chain.
2106 virtual void accept(ir_visitor
*v
)
2111 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2114 * Object being dereferenced.
2120 class ir_dereference_array
: public ir_dereference
{
2122 ir_dereference_array(ir_rvalue
*value
, ir_rvalue
*array_index
);
2124 ir_dereference_array(ir_variable
*var
, ir_rvalue
*array_index
);
2126 virtual ir_dereference_array
*clone(void *mem_ctx
,
2127 struct hash_table
*) const;
2129 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2130 struct hash_table
*variable_context
= NULL
);
2132 virtual bool equals(const ir_instruction
*ir
,
2133 enum ir_node_type ignore
= ir_type_unset
) const;
2136 * Get the variable that is ultimately referenced by an r-value
2138 virtual ir_variable
*variable_referenced() const
2140 return this->array
->variable_referenced();
2143 virtual int precision() const
2145 ir_dereference
*deref
= this->array
->as_dereference();
2148 return GLSL_PRECISION_NONE
;
2150 return deref
->precision();
2153 virtual void accept(ir_visitor
*v
)
2158 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2161 ir_rvalue
*array_index
;
2164 void set_array(ir_rvalue
*value
);
2168 class ir_dereference_record
: public ir_dereference
{
2170 ir_dereference_record(ir_rvalue
*value
, const char *field
);
2172 ir_dereference_record(ir_variable
*var
, const char *field
);
2174 virtual ir_dereference_record
*clone(void *mem_ctx
,
2175 struct hash_table
*) const;
2177 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2178 struct hash_table
*variable_context
= NULL
);
2181 * Get the variable that is ultimately referenced by an r-value
2183 virtual ir_variable
*variable_referenced() const
2185 return this->record
->variable_referenced();
2188 virtual int precision() const
2190 glsl_struct_field
*field
= record
->type
->fields
.structure
+ field_idx
;
2192 return field
->precision
;
2195 virtual void accept(ir_visitor
*v
)
2200 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2208 * Data stored in an ir_constant
2210 union ir_constant_data
{
2224 class ir_constant
: public ir_rvalue
{
2226 ir_constant(const struct glsl_type
*type
, const ir_constant_data
*data
);
2227 ir_constant(bool b
, unsigned vector_elements
=1);
2228 ir_constant(int16_t i16
, unsigned vector_elements
=1);
2229 ir_constant(uint16_t u16
, unsigned vector_elements
=1);
2230 ir_constant(unsigned int u
, unsigned vector_elements
=1);
2231 ir_constant(int i
, unsigned vector_elements
=1);
2232 ir_constant(float16_t f16
, unsigned vector_elements
=1);
2233 ir_constant(float f
, unsigned vector_elements
=1);
2234 ir_constant(double d
, unsigned vector_elements
=1);
2235 ir_constant(uint64_t u64
, unsigned vector_elements
=1);
2236 ir_constant(int64_t i64
, unsigned vector_elements
=1);
2239 * Construct an ir_constant from a list of ir_constant values
2241 ir_constant(const struct glsl_type
*type
, exec_list
*values
);
2244 * Construct an ir_constant from a scalar component of another ir_constant
2246 * The new \c ir_constant inherits the type of the component from the
2250 * In the case of a matrix constant, the new constant is a scalar, \b not
2253 ir_constant(const ir_constant
*c
, unsigned i
);
2256 * Return a new ir_constant of the specified type containing all zeros.
2258 static ir_constant
*zero(void *mem_ctx
, const glsl_type
*type
);
2260 virtual ir_constant
*clone(void *mem_ctx
, struct hash_table
*) const;
2262 virtual ir_constant
*constant_expression_value(void *mem_ctx
,
2263 struct hash_table
*variable_context
= NULL
);
2265 virtual void accept(ir_visitor
*v
)
2270 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2272 virtual bool equals(const ir_instruction
*ir
,
2273 enum ir_node_type ignore
= ir_type_unset
) const;
2276 * Get a particular component of a constant as a specific type
2278 * This is useful, for example, to get a value from an integer constant
2279 * as a float or bool. This appears frequently when constructors are
2280 * called with all constant parameters.
2283 bool get_bool_component(unsigned i
) const;
2284 float get_float_component(unsigned i
) const;
2285 uint16_t get_float16_component(unsigned i
) const;
2286 double get_double_component(unsigned i
) const;
2287 int16_t get_int16_component(unsigned i
) const;
2288 uint16_t get_uint16_component(unsigned i
) const;
2289 int get_int_component(unsigned i
) const;
2290 unsigned get_uint_component(unsigned i
) const;
2291 int64_t get_int64_component(unsigned i
) const;
2292 uint64_t get_uint64_component(unsigned i
) const;
2295 ir_constant
*get_array_element(unsigned i
) const;
2297 ir_constant
*get_record_field(int idx
);
2300 * Copy the values on another constant at a given offset.
2302 * The offset is ignored for array or struct copies, it's only for
2303 * scalars or vectors into vectors or matrices.
2305 * With identical types on both sides and zero offset it's clone()
2306 * without creating a new object.
2309 void copy_offset(ir_constant
*src
, int offset
);
2312 * Copy the values on another constant at a given offset and
2313 * following an assign-like mask.
2315 * The mask is ignored for scalars.
2317 * Note that this function only handles what assign can handle,
2318 * i.e. at most a vector as source and a column of a matrix as
2322 void copy_masked_offset(ir_constant
*src
, int offset
, unsigned int mask
);
2325 * Determine whether a constant has the same value as another constant
2327 * \sa ir_constant::is_zero, ir_constant::is_one,
2328 * ir_constant::is_negative_one
2330 bool has_value(const ir_constant
*) const;
2333 * Return true if this ir_constant represents the given value.
2335 * For vectors, this checks that each component is the given value.
2337 virtual bool is_value(float f
, int i
) const;
2338 virtual bool is_zero() const;
2339 virtual bool is_one() const;
2340 virtual bool is_negative_one() const;
2343 * Return true for constants that could be stored as 16-bit unsigned values.
2345 * Note that this will return true even for signed integer ir_constants, as
2346 * long as the value is non-negative and fits in 16-bits.
2348 virtual bool is_uint16_constant() const;
2351 * Value of the constant.
2353 * The field used to back the values supplied by the constant is determined
2354 * by the type associated with the \c ir_instruction. Constants may be
2355 * scalars, vectors, or matrices.
2357 union ir_constant_data value
;
2359 /* Array elements and structure fields */
2360 ir_constant
**const_elements
;
2364 * Parameterless constructor only used by the clone method
2370 * IR instruction to emit a vertex in a geometry shader.
2372 class ir_emit_vertex
: public ir_instruction
{
2374 ir_emit_vertex(ir_rvalue
*stream
)
2375 : ir_instruction(ir_type_emit_vertex
),
2381 virtual void accept(ir_visitor
*v
)
2386 virtual ir_emit_vertex
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2388 return new(mem_ctx
) ir_emit_vertex(this->stream
->clone(mem_ctx
, ht
));
2391 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2393 int stream_id() const
2395 return stream
->as_constant()->value
.i
[0];
2402 * IR instruction to complete the current primitive and start a new one in a
2405 class ir_end_primitive
: public ir_instruction
{
2407 ir_end_primitive(ir_rvalue
*stream
)
2408 : ir_instruction(ir_type_end_primitive
),
2414 virtual void accept(ir_visitor
*v
)
2419 virtual ir_end_primitive
*clone(void *mem_ctx
, struct hash_table
*ht
) const
2421 return new(mem_ctx
) ir_end_primitive(this->stream
->clone(mem_ctx
, ht
));
2424 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2426 int stream_id() const
2428 return stream
->as_constant()->value
.i
[0];
2435 * IR instruction for tessellation control and compute shader barrier.
2437 class ir_barrier
: public ir_instruction
{
2440 : ir_instruction(ir_type_barrier
)
2444 virtual void accept(ir_visitor
*v
)
2449 virtual ir_barrier
*clone(void *mem_ctx
, struct hash_table
*) const
2451 return new(mem_ctx
) ir_barrier();
2454 virtual ir_visitor_status
accept(ir_hierarchical_visitor
*);
2460 * Apply a visitor to each IR node in a list
2463 visit_exec_list(exec_list
*list
, ir_visitor
*visitor
);
2466 * Validate invariants on each IR node in a list
2468 void validate_ir_tree(exec_list
*instructions
);
2470 struct _mesa_glsl_parse_state
;
2471 struct gl_shader_program
;
2474 * Detect whether an unlinked shader contains static recursion
2476 * If the list of instructions is determined to contain static recursion,
2477 * \c _mesa_glsl_error will be called to emit error messages for each function
2478 * that is in the recursion cycle.
2481 detect_recursion_unlinked(struct _mesa_glsl_parse_state
*state
,
2482 exec_list
*instructions
);
2485 * Detect whether a linked shader contains static recursion
2487 * If the list of instructions is determined to contain static recursion,
2488 * \c link_error_printf will be called to emit error messages for each function
2489 * that is in the recursion cycle. In addition,
2490 * \c gl_shader_program::LinkStatus will be set to false.
2493 detect_recursion_linked(struct gl_shader_program
*prog
,
2494 exec_list
*instructions
);
2497 * Make a clone of each IR instruction in a list
2499 * \param in List of IR instructions that are to be cloned
2500 * \param out List to hold the cloned instructions
2503 clone_ir_list(void *mem_ctx
, exec_list
*out
, const exec_list
*in
);
2506 _mesa_glsl_initialize_variables(exec_list
*instructions
,
2507 struct _mesa_glsl_parse_state
*state
);
2510 reparent_ir(exec_list
*list
, void *mem_ctx
);
2513 do_set_program_inouts(exec_list
*instructions
, struct gl_program
*prog
,
2514 gl_shader_stage shader_stage
);
2517 prototype_string(const glsl_type
*return_type
, const char *name
,
2518 exec_list
*parameters
);
2521 mode_string(const ir_variable
*var
);
2524 * Built-in / reserved GL variables names start with "gl_"
2527 is_gl_identifier(const char *s
)
2529 return s
&& s
[0] == 'g' && s
[1] == 'l' && s
[2] == '_';
2533 #endif /* __cplusplus */
2535 extern void _mesa_print_ir(FILE *f
, struct exec_list
*instructions
,
2536 struct _mesa_glsl_parse_state
*state
);
2539 fprint_ir(FILE *f
, const void *instruction
);
2541 extern const struct gl_builtin_uniform_desc
*
2542 _mesa_glsl_get_builtin_uniform_desc(const char *name
);
2549 vertices_per_prim(GLenum prim
);