2 * Copyright © 2014 Connor Abbott
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Connor Abbott (cwabbott0@gmail.com)
30 #include "util/hash_table.h"
31 #include "compiler/glsl/list.h"
32 #include "GL/gl.h" /* GLenum */
33 #include "util/list.h"
34 #include "util/ralloc.h"
36 #include "util/bitset.h"
37 #include "util/macros.h"
38 #include "compiler/nir_types.h"
39 #include "compiler/shader_enums.h"
42 #include "nir_opcodes.h"
49 struct gl_shader_program
;
52 #define NIR_TRUE (~0u)
54 /** Defines a cast function
56 * This macro defines a cast function from in_type to out_type where
57 * out_type is some structure type that contains a field of type out_type.
59 * Note that you have to be a bit careful as the generated cast function
62 #define NIR_DEFINE_CAST(name, in_type, out_type, field) \
63 static inline out_type * \
64 name(const in_type *parent) \
66 return exec_node_data(out_type, parent, field); \
75 * Description of built-in state associated with a uniform
77 * \sa nir_variable::state_slots
85 nir_var_shader_in
= (1 << 0),
86 nir_var_shader_out
= (1 << 1),
87 nir_var_global
= (1 << 2),
88 nir_var_local
= (1 << 3),
89 nir_var_uniform
= (1 << 4),
90 nir_var_shader_storage
= (1 << 5),
91 nir_var_system_value
= (1 << 6),
92 nir_var_param
= (1 << 7),
93 nir_var_shared
= (1 << 8),
98 * Data stored in an nir_constant
100 union nir_constant_data
{
108 typedef struct nir_constant
{
110 * Value of the constant.
112 * The field used to back the values supplied by the constant is determined
113 * by the type associated with the \c nir_variable. Constants may be
114 * scalars, vectors, or matrices.
116 union nir_constant_data value
;
118 /* we could get this from the var->type but makes clone *much* easier to
119 * not have to care about the type.
121 unsigned num_elements
;
123 /* Array elements / Structure Fields */
124 struct nir_constant
**elements
;
128 * \brief Layout qualifiers for gl_FragDepth.
130 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
131 * with a layout qualifier.
134 nir_depth_layout_none
, /**< No depth layout is specified. */
135 nir_depth_layout_any
,
136 nir_depth_layout_greater
,
137 nir_depth_layout_less
,
138 nir_depth_layout_unchanged
142 * Either a uniform, global variable, shader input, or shader output. Based on
143 * ir_variable - it should be easy to translate between the two.
146 typedef struct nir_variable
{
147 struct exec_node node
;
150 * Declared type of the variable
152 const struct glsl_type
*type
;
155 * Declared name of the variable
159 struct nir_variable_data
{
161 * Storage class of the variable.
163 * \sa nir_variable_mode
165 nir_variable_mode mode
;
168 * Is the variable read-only?
170 * This is set for variables declared as \c const, shader inputs,
173 unsigned read_only
:1;
177 unsigned invariant
:1;
180 * Interpolation mode for shader inputs / outputs
182 * \sa glsl_interp_qualifier
184 unsigned interpolation
:2;
187 * \name ARB_fragment_coord_conventions
190 unsigned origin_upper_left
:1;
191 unsigned pixel_center_integer
:1;
195 * Was the location explicitly set in the shader?
197 * If the location is explicitly set in the shader, it \b cannot be changed
198 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
201 unsigned explicit_location
:1;
202 unsigned explicit_index
:1;
205 * Was an initial binding explicitly set in the shader?
207 * If so, constant_initializer contains an integer nir_constant
208 * representing the initial binding point.
210 unsigned explicit_binding
:1;
213 * Does this variable have an initializer?
215 * This is used by the linker to cross-validiate initializers of global
218 unsigned has_initializer
:1;
221 * If non-zero, then this variable may be packed along with other variables
222 * into a single varying slot, so this offset should be applied when
223 * accessing components. For example, an offset of 1 means that the x
224 * component of this variable is actually stored in component y of the
225 * location specified by \c location.
227 unsigned location_frac
:2;
230 * \brief Layout qualifier for gl_FragDepth.
232 * This is not equal to \c ir_depth_layout_none if and only if this
233 * variable is \c gl_FragDepth and a layout qualifier is specified.
235 nir_depth_layout depth_layout
;
238 * Storage location of the base of this variable
240 * The precise meaning of this field depends on the nature of the variable.
242 * - Vertex shader input: one of the values from \c gl_vert_attrib.
243 * - Vertex shader output: one of the values from \c gl_varying_slot.
244 * - Geometry shader input: one of the values from \c gl_varying_slot.
245 * - Geometry shader output: one of the values from \c gl_varying_slot.
246 * - Fragment shader input: one of the values from \c gl_varying_slot.
247 * - Fragment shader output: one of the values from \c gl_frag_result.
248 * - Uniforms: Per-stage uniform slot number for default uniform block.
249 * - Uniforms: Index within the uniform block definition for UBO members.
250 * - Non-UBO Uniforms: uniform slot number.
251 * - Other: This field is not currently used.
253 * If the variable is a uniform, shader input, or shader output, and the
254 * slot has not been assigned, the value will be -1.
259 * The actual location of the variable in the IR. Only valid for inputs
262 unsigned int driver_location
;
265 * output index for dual source blending.
270 * Descriptor set binding for sampler or UBO.
275 * Initial binding point for a sampler or UBO.
277 * For array types, this represents the binding point for the first element.
282 * Location an atomic counter is stored at.
287 * ARB_shader_image_load_store qualifiers.
290 bool read_only
; /**< "readonly" qualifier. */
291 bool write_only
; /**< "writeonly" qualifier. */
296 /** Image internal format if specified explicitly, otherwise GL_NONE. */
301 * Highest element accessed with a constant expression array index
303 * Not used for non-array variables.
305 unsigned max_array_access
;
310 * Built-in state that backs this uniform
312 * Once set at variable creation, \c state_slots must remain invariant.
313 * This is because, ideally, this array would be shared by all clones of
314 * this variable in the IR tree. In other words, we'd really like for it
315 * to be a fly-weight.
317 * If the variable is not a uniform, \c num_state_slots will be zero and
318 * \c state_slots will be \c NULL.
321 unsigned num_state_slots
; /**< Number of state slots used */
322 nir_state_slot
*state_slots
; /**< State descriptors. */
326 * Constant expression assigned in the initializer of the variable
328 nir_constant
*constant_initializer
;
331 * For variables that are in an interface block or are an instance of an
332 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
334 * \sa ir_variable::location
336 const struct glsl_type
*interface_type
;
339 #define nir_foreach_variable(var, var_list) \
340 foreach_list_typed(nir_variable, var, node, var_list)
342 #define nir_foreach_variable_safe(var, var_list) \
343 foreach_list_typed_safe(nir_variable, var, node, var_list)
346 nir_variable_is_global(const nir_variable
*var
)
348 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
351 typedef struct nir_register
{
352 struct exec_node node
;
354 unsigned num_components
; /** < number of vector components */
355 unsigned num_array_elems
; /** < size of array (0 for no array) */
357 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
360 /** generic register index. */
363 /** only for debug purposes, can be NULL */
366 /** whether this register is local (per-function) or global (per-shader) */
370 * If this flag is set to true, then accessing channels >= num_components
371 * is well-defined, and simply spills over to the next array element. This
372 * is useful for backends that can do per-component accessing, in
373 * particular scalar backends. By setting this flag and making
374 * num_components equal to 1, structures can be packed tightly into
375 * registers and then registers can be accessed per-component to get to
376 * each structure member, even if it crosses vec4 boundaries.
380 /** set of nir_src's where this register is used (read from) */
381 struct list_head uses
;
383 /** set of nir_dest's where this register is defined (written to) */
384 struct list_head defs
;
386 /** set of nir_if's where this register is used as a condition */
387 struct list_head if_uses
;
394 nir_instr_type_intrinsic
,
395 nir_instr_type_load_const
,
397 nir_instr_type_ssa_undef
,
399 nir_instr_type_parallel_copy
,
402 typedef struct nir_instr
{
403 struct exec_node node
;
405 struct nir_block
*block
;
407 /** generic instruction index. */
410 /* A temporary for optimization and analysis passes to use for storing
411 * flags. For instance, DCE uses this to store the "dead/live" info.
416 static inline nir_instr
*
417 nir_instr_next(nir_instr
*instr
)
419 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
420 if (exec_node_is_tail_sentinel(next
))
423 return exec_node_data(nir_instr
, next
, node
);
426 static inline nir_instr
*
427 nir_instr_prev(nir_instr
*instr
)
429 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
430 if (exec_node_is_head_sentinel(prev
))
433 return exec_node_data(nir_instr
, prev
, node
);
437 nir_instr_is_first(nir_instr
*instr
)
439 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
443 nir_instr_is_last(nir_instr
*instr
)
445 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
448 typedef struct nir_ssa_def
{
449 /** for debugging only, can be NULL */
452 /** generic SSA definition index. */
455 /** Index into the live_in and live_out bitfields */
458 nir_instr
*parent_instr
;
460 /** set of nir_instr's where this register is used (read from) */
461 struct list_head uses
;
463 /** set of nir_if's where this register is used as a condition */
464 struct list_head if_uses
;
466 uint8_t num_components
;
468 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
476 struct nir_src
*indirect
; /** < NULL for no indirect offset */
477 unsigned base_offset
;
479 /* TODO use-def chain goes here */
483 nir_instr
*parent_instr
;
484 struct list_head def_link
;
487 struct nir_src
*indirect
; /** < NULL for no indirect offset */
488 unsigned base_offset
;
490 /* TODO def-use chain goes here */
495 typedef struct nir_src
{
497 nir_instr
*parent_instr
;
498 struct nir_if
*parent_if
;
501 struct list_head use_link
;
511 static inline nir_src
514 nir_src src
= { { NULL
} };
518 #define NIR_SRC_INIT nir_src_init()
520 #define nir_foreach_use(src, reg_or_ssa_def) \
521 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
523 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
524 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
526 #define nir_foreach_if_use(src, reg_or_ssa_def) \
527 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
529 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
530 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
541 static inline nir_dest
544 nir_dest dest
= { { { NULL
} } };
548 #define NIR_DEST_INIT nir_dest_init()
550 #define nir_foreach_def(dest, reg) \
551 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
553 #define nir_foreach_def_safe(dest, reg) \
554 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
556 static inline nir_src
557 nir_src_for_ssa(nir_ssa_def
*def
)
559 nir_src src
= NIR_SRC_INIT
;
567 static inline nir_src
568 nir_src_for_reg(nir_register
*reg
)
570 nir_src src
= NIR_SRC_INIT
;
574 src
.reg
.indirect
= NULL
;
575 src
.reg
.base_offset
= 0;
580 static inline nir_dest
581 nir_dest_for_reg(nir_register
*reg
)
583 nir_dest dest
= NIR_DEST_INIT
;
590 static inline unsigned
591 nir_src_bit_size(nir_src src
)
593 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
596 static inline unsigned
597 nir_dest_bit_size(nir_dest dest
)
599 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
602 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
603 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
609 * \name input modifiers
613 * For inputs interpreted as floating point, flips the sign bit. For
614 * inputs interpreted as integers, performs the two's complement negation.
619 * Clears the sign bit for floating point values, and computes the integer
620 * absolute value for integers. Note that the negate modifier acts after
621 * the absolute value modifier, therefore if both are set then all inputs
622 * will become negative.
628 * For each input component, says which component of the register it is
629 * chosen from. Note that which elements of the swizzle are used and which
630 * are ignored are based on the write mask for most opcodes - for example,
631 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
632 * a swizzle of {2, x, 1, 0} where x means "don't care."
641 * \name saturate output modifier
643 * Only valid for opcodes that output floating-point numbers. Clamps the
644 * output to between 0.0 and 1.0 inclusive.
649 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
653 nir_type_invalid
= 0, /* Not a valid type */
658 nir_type_bool32
= 32 | nir_type_bool
,
659 nir_type_int8
= 8 | nir_type_int
,
660 nir_type_int16
= 16 | nir_type_int
,
661 nir_type_int32
= 32 | nir_type_int
,
662 nir_type_int64
= 64 | nir_type_int
,
663 nir_type_uint8
= 8 | nir_type_uint
,
664 nir_type_uint16
= 16 | nir_type_uint
,
665 nir_type_uint32
= 32 | nir_type_uint
,
666 nir_type_uint64
= 64 | nir_type_uint
,
667 nir_type_float16
= 16 | nir_type_float
,
668 nir_type_float32
= 32 | nir_type_float
,
669 nir_type_float64
= 64 | nir_type_float
,
672 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
673 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
675 static inline unsigned
676 nir_alu_type_get_type_size(nir_alu_type type
)
678 return type
& NIR_ALU_TYPE_SIZE_MASK
;
681 static inline unsigned
682 nir_alu_type_get_base_type(nir_alu_type type
)
684 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
688 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
689 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
690 } nir_op_algebraic_property
;
698 * The number of components in the output
700 * If non-zero, this is the size of the output and input sizes are
701 * explicitly given; swizzle and writemask are still in effect, but if
702 * the output component is masked out, then the input component may
705 * If zero, the opcode acts in the standard, per-component manner; the
706 * operation is performed on each component (except the ones that are
707 * masked out) with the input being taken from the input swizzle for
710 * The size of some of the inputs may be given (i.e. non-zero) even
711 * though output_size is zero; in that case, the inputs with a zero
712 * size act per-component, while the inputs with non-zero size don't.
714 unsigned output_size
;
717 * The type of vector that the instruction outputs. Note that the
718 * staurate modifier is only allowed on outputs with the float type.
721 nir_alu_type output_type
;
724 * The number of components in each input
726 unsigned input_sizes
[4];
729 * The type of vector that each input takes. Note that negate and
730 * absolute value are only allowed on inputs with int or float type and
731 * behave differently on the two.
733 nir_alu_type input_types
[4];
735 nir_op_algebraic_property algebraic_properties
;
738 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
740 typedef struct nir_alu_instr
{
744 /** Indicates that this ALU instruction generates an exact value
746 * This is kind of a mixture of GLSL "precise" and "invariant" and not
747 * really equivalent to either. This indicates that the value generated by
748 * this operation is high-precision and any code transformations that touch
749 * it must ensure that the resulting value is bit-for-bit identical to the
758 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
759 nir_alu_instr
*instr
);
760 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
761 nir_alu_instr
*instr
);
763 /* is this source channel used? */
765 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
767 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
768 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
770 return (instr
->dest
.write_mask
>> channel
) & 1;
774 * For instructions whose destinations are SSA, get the number of channels
777 static inline unsigned
778 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
780 assert(instr
->dest
.dest
.is_ssa
);
782 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
783 return nir_op_infos
[instr
->op
].input_sizes
[src
];
785 return instr
->dest
.dest
.ssa
.num_components
;
790 nir_deref_type_array
,
791 nir_deref_type_struct
794 typedef struct nir_deref
{
795 nir_deref_type deref_type
;
796 struct nir_deref
*child
;
797 const struct glsl_type
*type
;
806 /* This enum describes how the array is referenced. If the deref is
807 * direct then the base_offset is used. If the deref is indirect then
808 * offset is given by base_offset + indirect. If the deref is a wildcard
809 * then the deref refers to all of the elements of the array at the same
810 * time. Wildcard dereferences are only ever allowed in copy_var
811 * intrinsics and the source and destination derefs must have matching
815 nir_deref_array_type_direct
,
816 nir_deref_array_type_indirect
,
817 nir_deref_array_type_wildcard
,
818 } nir_deref_array_type
;
823 nir_deref_array_type deref_array_type
;
824 unsigned base_offset
;
834 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
835 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
836 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
838 /* Returns the last deref in the chain. */
839 static inline nir_deref
*
840 nir_deref_tail(nir_deref
*deref
)
843 deref
= deref
->child
;
851 nir_deref_var
**params
;
852 nir_deref_var
*return_deref
;
854 struct nir_function
*callee
;
857 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
858 num_variables, num_indices, idx0, idx1, idx2, flags) \
859 nir_intrinsic_##name,
861 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
864 #include "nir_intrinsics.h"
865 nir_num_intrinsics
= nir_last_intrinsic
+ 1
869 #undef LAST_INTRINSIC
871 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
873 /** Represents an intrinsic
875 * An intrinsic is an instruction type for handling things that are
876 * more-or-less regular operations but don't just consume and produce SSA
877 * values like ALU operations do. Intrinsics are not for things that have
878 * special semantic meaning such as phi nodes and parallel copies.
879 * Examples of intrinsics include variable load/store operations, system
880 * value loads, and the like. Even though texturing more-or-less falls
881 * under this category, texturing is its own instruction type because
882 * trying to represent texturing with intrinsics would lead to a
883 * combinatorial explosion of intrinsic opcodes.
885 * By having a single instruction type for handling a lot of different
886 * cases, optimization passes can look for intrinsics and, for the most
887 * part, completely ignore them. Each intrinsic type also has a few
888 * possible flags that govern whether or not they can be reordered or
889 * eliminated. That way passes like dead code elimination can still work
890 * on intrisics without understanding the meaning of each.
892 * Each intrinsic has some number of constant indices, some number of
893 * variables, and some number of sources. What these sources, variables,
894 * and indices mean depends on the intrinsic and is documented with the
895 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
896 * instructions are the only types of instruction that can operate on
902 nir_intrinsic_op intrinsic
;
906 /** number of components if this is a vectorized intrinsic
908 * Similarly to ALU operations, some intrinsics are vectorized.
909 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
910 * For vectorized intrinsics, the num_components field specifies the
911 * number of destination components and the number of source components
912 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
914 uint8_t num_components
;
916 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
918 nir_deref_var
*variables
[2];
921 } nir_intrinsic_instr
;
924 * \name NIR intrinsics semantic flags
926 * information about what the compiler can do with the intrinsics.
928 * \sa nir_intrinsic_info::flags
932 * whether the intrinsic can be safely eliminated if none of its output
933 * value is not being used.
935 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
938 * Whether the intrinsic can be reordered with respect to any other
939 * intrinsic, i.e. whether the only reordering dependencies of the
940 * intrinsic are due to the register reads/writes.
942 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
943 } nir_intrinsic_semantic_flag
;
946 * \name NIR intrinsics const-index flag
948 * Indicates the usage of a const_index slot.
950 * \sa nir_intrinsic_info::index_map
954 * Generally instructions that take a offset src argument, can encode
955 * a constant 'base' value which is added to the offset.
957 NIR_INTRINSIC_BASE
= 1,
960 * For store instructions, a writemask for the store.
962 NIR_INTRINSIC_WRMASK
= 2,
965 * The stream-id for GS emit_vertex/end_primitive intrinsics.
967 NIR_INTRINSIC_STREAM_ID
= 3,
970 * The clip-plane id for load_user_clip_plane intrinsic.
972 NIR_INTRINSIC_UCP_ID
= 4,
975 * The amount of data, starting from BASE, that this instruction may
976 * access. This is used to provide bounds if the offset is not constant.
978 NIR_INTRINSIC_RANGE
= 5,
981 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
983 NIR_INTRINSIC_DESC_SET
= 6,
986 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
988 NIR_INTRINSIC_BINDING
= 7,
993 NIR_INTRINSIC_COMPONENT
= 8,
995 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
997 } nir_intrinsic_index_flag
;
999 #define NIR_INTRINSIC_MAX_INPUTS 4
1004 unsigned num_srcs
; /** < number of register/SSA inputs */
1006 /** number of components of each input register
1008 * If this value is 0, the number of components is given by the
1009 * num_components field of nir_intrinsic_instr.
1011 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1015 /** number of components of the output register
1017 * If this value is 0, the number of components is given by the
1018 * num_components field of nir_intrinsic_instr.
1020 unsigned dest_components
;
1022 /** the number of inputs/outputs that are variables */
1023 unsigned num_variables
;
1025 /** the number of constant indices used by the intrinsic */
1026 unsigned num_indices
;
1028 /** indicates the usage of intr->const_index[n] */
1029 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1031 /** semantic flags for calls to this intrinsic */
1032 nir_intrinsic_semantic_flag flags
;
1033 } nir_intrinsic_info
;
1035 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1038 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1039 static inline type \
1040 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
1042 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1043 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1044 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1046 static inline void \
1047 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1049 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1050 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1051 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1054 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1055 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1056 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1057 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1058 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1059 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1060 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1061 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1064 * \group texture information
1066 * This gives semantic information about textures which is useful to the
1067 * frontend, the backend, and lowering passes, but not the optimizer.
1072 nir_tex_src_projector
,
1073 nir_tex_src_comparitor
, /* shadow comparitor */
1077 nir_tex_src_ms_index
, /* MSAA sample index */
1078 nir_tex_src_ms_mcs
, /* MSAA compression value */
1081 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1082 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1083 nir_tex_src_plane
, /* < selects plane for planar textures */
1084 nir_num_tex_src_types
1089 nir_tex_src_type src_type
;
1093 nir_texop_tex
, /**< Regular texture look-up */
1094 nir_texop_txb
, /**< Texture look-up with LOD bias */
1095 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1096 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1097 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1098 nir_texop_txf_ms
, /**< Multisample texture fetch */
1099 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1100 nir_texop_txs
, /**< Texture size */
1101 nir_texop_lod
, /**< Texture lod query */
1102 nir_texop_tg4
, /**< Texture gather */
1103 nir_texop_query_levels
, /**< Texture levels query */
1104 nir_texop_texture_samples
, /**< Texture samples query */
1105 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1113 enum glsl_sampler_dim sampler_dim
;
1114 nir_alu_type dest_type
;
1119 unsigned num_srcs
, coord_components
;
1120 bool is_array
, is_shadow
;
1123 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1124 * components or the new-style shadow that outputs 1 component.
1126 bool is_new_style_shadow
;
1128 /* gather component selector */
1129 unsigned component
: 2;
1131 /** The texture index
1133 * If this texture instruction has a nir_tex_src_texture_offset source,
1134 * then the texture index is given by texture_index + texture_offset.
1136 unsigned texture_index
;
1138 /** The size of the texture array or 0 if it's not an array */
1139 unsigned texture_array_size
;
1141 /** The texture deref
1143 * If this is null, use texture_index instead.
1145 nir_deref_var
*texture
;
1147 /** The sampler index
1149 * The following operations do not require a sampler and, as such, this
1150 * field should be ignored:
1152 * - nir_texop_txf_ms
1156 * - nir_texop_query_levels
1157 * - nir_texop_texture_samples
1158 * - nir_texop_samples_identical
1160 * If this texture instruction has a nir_tex_src_sampler_offset source,
1161 * then the sampler index is given by sampler_index + sampler_offset.
1163 unsigned sampler_index
;
1165 /** The sampler deref
1167 * If this is null, use sampler_index instead.
1169 nir_deref_var
*sampler
;
1172 static inline unsigned
1173 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1175 switch (instr
->op
) {
1176 case nir_texop_txs
: {
1178 switch (instr
->sampler_dim
) {
1179 case GLSL_SAMPLER_DIM_1D
:
1180 case GLSL_SAMPLER_DIM_BUF
:
1183 case GLSL_SAMPLER_DIM_2D
:
1184 case GLSL_SAMPLER_DIM_CUBE
:
1185 case GLSL_SAMPLER_DIM_MS
:
1186 case GLSL_SAMPLER_DIM_RECT
:
1187 case GLSL_SAMPLER_DIM_EXTERNAL
:
1190 case GLSL_SAMPLER_DIM_3D
:
1194 unreachable("not reached");
1196 if (instr
->is_array
)
1204 case nir_texop_texture_samples
:
1205 case nir_texop_query_levels
:
1206 case nir_texop_samples_identical
:
1210 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1217 /* Returns true if this texture operation queries something about the texture
1218 * rather than actually sampling it.
1221 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1223 switch (instr
->op
) {
1226 case nir_texop_texture_samples
:
1227 case nir_texop_query_levels
:
1228 case nir_texop_txf_ms_mcs
:
1235 case nir_texop_txf_ms
:
1239 unreachable("Invalid texture opcode");
1243 static inline unsigned
1244 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1246 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1247 return instr
->coord_components
;
1249 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1250 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1253 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1254 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1255 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1256 if (instr
->is_array
)
1257 return instr
->coord_components
- 1;
1259 return instr
->coord_components
;
1266 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1268 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1269 if (instr
->src
[i
].src_type
== type
)
1287 nir_const_value value
;
1290 } nir_load_const_instr
;
1303 /* creates a new SSA variable in an undefined state */
1308 } nir_ssa_undef_instr
;
1311 struct exec_node node
;
1313 /* The predecessor block corresponding to this source */
1314 struct nir_block
*pred
;
1319 #define nir_foreach_phi_src(phi_src, phi) \
1320 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1321 #define nir_foreach_phi_src_safe(phi_src, phi) \
1322 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1327 struct exec_list srcs
; /** < list of nir_phi_src */
1333 struct exec_node node
;
1336 } nir_parallel_copy_entry
;
1338 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1339 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1344 /* A list of nir_parallel_copy_entry's. The sources of all of the
1345 * entries are copied to the corresponding destinations "in parallel".
1346 * In other words, if we have two entries: a -> b and b -> a, the values
1349 struct exec_list entries
;
1350 } nir_parallel_copy_instr
;
1352 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1353 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1354 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1355 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1356 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1357 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1358 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1359 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1360 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1361 nir_parallel_copy_instr
, instr
)
1366 * Control flow consists of a tree of control flow nodes, which include
1367 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1368 * instructions that always run start-to-finish. Each basic block also keeps
1369 * track of its successors (blocks which may run immediately after the current
1370 * block) and predecessors (blocks which could have run immediately before the
1371 * current block). Each function also has a start block and an end block which
1372 * all return statements point to (which is always empty). Together, all the
1373 * blocks with their predecessors and successors make up the control flow
1374 * graph (CFG) of the function. There are helpers that modify the tree of
1375 * control flow nodes while modifying the CFG appropriately; these should be
1376 * used instead of modifying the tree directly.
1383 nir_cf_node_function
1386 typedef struct nir_cf_node
{
1387 struct exec_node node
;
1388 nir_cf_node_type type
;
1389 struct nir_cf_node
*parent
;
1392 typedef struct nir_block
{
1393 nir_cf_node cf_node
;
1395 struct exec_list instr_list
; /** < list of nir_instr */
1397 /** generic block index; generated by nir_index_blocks */
1401 * Each block can only have up to 2 successors, so we put them in a simple
1402 * array - no need for anything more complicated.
1404 struct nir_block
*successors
[2];
1406 /* Set of nir_block predecessors in the CFG */
1407 struct set
*predecessors
;
1410 * this node's immediate dominator in the dominance tree - set to NULL for
1413 struct nir_block
*imm_dom
;
1415 /* This node's children in the dominance tree */
1416 unsigned num_dom_children
;
1417 struct nir_block
**dom_children
;
1419 /* Set of nir_block's on the dominance frontier of this block */
1420 struct set
*dom_frontier
;
1423 * These two indices have the property that dom_{pre,post}_index for each
1424 * child of this block in the dominance tree will always be between
1425 * dom_pre_index and dom_post_index for this block, which makes testing if
1426 * a given block is dominated by another block an O(1) operation.
1428 unsigned dom_pre_index
, dom_post_index
;
1430 /* live in and out for this block; used for liveness analysis */
1431 BITSET_WORD
*live_in
;
1432 BITSET_WORD
*live_out
;
1435 static inline nir_instr
*
1436 nir_block_first_instr(nir_block
*block
)
1438 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1439 return exec_node_data(nir_instr
, head
, node
);
1442 static inline nir_instr
*
1443 nir_block_last_instr(nir_block
*block
)
1445 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1446 return exec_node_data(nir_instr
, tail
, node
);
1449 #define nir_foreach_instr(instr, block) \
1450 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1451 #define nir_foreach_instr_reverse(instr, block) \
1452 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1453 #define nir_foreach_instr_safe(instr, block) \
1454 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1455 #define nir_foreach_instr_reverse_safe(instr, block) \
1456 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1458 typedef struct nir_if
{
1459 nir_cf_node cf_node
;
1462 struct exec_list then_list
; /** < list of nir_cf_node */
1463 struct exec_list else_list
; /** < list of nir_cf_node */
1466 static inline nir_cf_node
*
1467 nir_if_first_then_node(nir_if
*if_stmt
)
1469 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1470 return exec_node_data(nir_cf_node
, head
, node
);
1473 static inline nir_cf_node
*
1474 nir_if_last_then_node(nir_if
*if_stmt
)
1476 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1477 return exec_node_data(nir_cf_node
, tail
, node
);
1480 static inline nir_cf_node
*
1481 nir_if_first_else_node(nir_if
*if_stmt
)
1483 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1484 return exec_node_data(nir_cf_node
, head
, node
);
1487 static inline nir_cf_node
*
1488 nir_if_last_else_node(nir_if
*if_stmt
)
1490 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1491 return exec_node_data(nir_cf_node
, tail
, node
);
1495 nir_cf_node cf_node
;
1497 struct exec_list body
; /** < list of nir_cf_node */
1500 static inline nir_cf_node
*
1501 nir_loop_first_cf_node(nir_loop
*loop
)
1503 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1506 static inline nir_cf_node
*
1507 nir_loop_last_cf_node(nir_loop
*loop
)
1509 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1513 * Various bits of metadata that can may be created or required by
1514 * optimization and analysis passes
1517 nir_metadata_none
= 0x0,
1518 nir_metadata_block_index
= 0x1,
1519 nir_metadata_dominance
= 0x2,
1520 nir_metadata_live_ssa_defs
= 0x4,
1521 nir_metadata_not_properly_reset
= 0x8,
1525 nir_cf_node cf_node
;
1527 /** pointer to the function of which this is an implementation */
1528 struct nir_function
*function
;
1530 struct exec_list body
; /** < list of nir_cf_node */
1532 nir_block
*end_block
;
1534 /** list for all local variables in the function */
1535 struct exec_list locals
;
1537 /** array of variables used as parameters */
1538 unsigned num_params
;
1539 nir_variable
**params
;
1541 /** variable used to hold the result of the function */
1542 nir_variable
*return_var
;
1544 /** list of local registers in the function */
1545 struct exec_list registers
;
1547 /** next available local register index */
1550 /** next available SSA value index */
1553 /* total number of basic blocks, only valid when block_index_dirty = false */
1554 unsigned num_blocks
;
1556 nir_metadata valid_metadata
;
1557 } nir_function_impl
;
1559 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1560 nir_start_block(nir_function_impl
*impl
)
1562 return (nir_block
*) impl
->body
.head
;
1565 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1566 nir_impl_last_block(nir_function_impl
*impl
)
1568 return (nir_block
*) impl
->body
.tail_pred
;
1571 static inline nir_cf_node
*
1572 nir_cf_node_next(nir_cf_node
*node
)
1574 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1575 if (exec_node_is_tail_sentinel(next
))
1578 return exec_node_data(nir_cf_node
, next
, node
);
1581 static inline nir_cf_node
*
1582 nir_cf_node_prev(nir_cf_node
*node
)
1584 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1585 if (exec_node_is_head_sentinel(prev
))
1588 return exec_node_data(nir_cf_node
, prev
, node
);
1592 nir_cf_node_is_first(const nir_cf_node
*node
)
1594 return exec_node_is_head_sentinel(node
->node
.prev
);
1598 nir_cf_node_is_last(const nir_cf_node
*node
)
1600 return exec_node_is_tail_sentinel(node
->node
.next
);
1603 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1604 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1605 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1606 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1611 nir_parameter_inout
,
1612 } nir_parameter_type
;
1615 nir_parameter_type param_type
;
1616 const struct glsl_type
*type
;
1619 typedef struct nir_function
{
1620 struct exec_node node
;
1623 struct nir_shader
*shader
;
1625 unsigned num_params
;
1626 nir_parameter
*params
;
1627 const struct glsl_type
*return_type
;
1629 /** The implementation of this function.
1631 * If the function is only declared and not implemented, this is NULL.
1633 nir_function_impl
*impl
;
1636 typedef struct nir_shader_compiler_options
{
1641 /** Lowers flrp when it does not support doubles */
1648 bool lower_bitfield_extract
;
1649 bool lower_bitfield_insert
;
1650 bool lower_uadd_carry
;
1651 bool lower_usub_borrow
;
1652 /** lowers fneg and ineg to fsub and isub. */
1654 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1657 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1660 /** enables rules to lower idiv by power-of-two: */
1663 /* Does the native fdot instruction replicate its result for four
1664 * components? If so, then opt_algebraic_late will turn all fdotN
1665 * instructions into fdot_replicatedN instructions.
1667 bool fdot_replicates
;
1669 /** lowers ffract to fsub+ffloor: */
1672 bool lower_pack_half_2x16
;
1673 bool lower_pack_unorm_2x16
;
1674 bool lower_pack_snorm_2x16
;
1675 bool lower_pack_unorm_4x8
;
1676 bool lower_pack_snorm_4x8
;
1677 bool lower_unpack_half_2x16
;
1678 bool lower_unpack_unorm_2x16
;
1679 bool lower_unpack_snorm_2x16
;
1680 bool lower_unpack_unorm_4x8
;
1681 bool lower_unpack_snorm_4x8
;
1683 bool lower_extract_byte
;
1684 bool lower_extract_word
;
1687 * Does the driver support real 32-bit integers? (Otherwise, integers
1688 * are simulated by floats.)
1690 bool native_integers
;
1692 /* Indicates that the driver only has zero-based vertex id */
1693 bool vertex_id_zero_based
;
1695 bool lower_cs_local_index_from_id
;
1696 } nir_shader_compiler_options
;
1698 typedef struct nir_shader_info
{
1701 /* Descriptive name provided by the client; may be NULL */
1704 /* Number of textures used by this shader */
1705 unsigned num_textures
;
1706 /* Number of uniform buffers used by this shader */
1708 /* Number of atomic buffers used by this shader */
1710 /* Number of shader storage buffers used by this shader */
1712 /* Number of images used by this shader */
1713 unsigned num_images
;
1715 /* Which inputs are actually read */
1716 uint64_t inputs_read
;
1717 /* Which inputs are actually read and are double */
1718 uint64_t double_inputs_read
;
1719 /* Which outputs are actually written */
1720 uint64_t outputs_written
;
1721 /* Which system values are actually read */
1722 uint64_t system_values_read
;
1724 /* Which patch inputs are actually read */
1725 uint32_t patch_inputs_read
;
1726 /* Which patch outputs are actually written */
1727 uint32_t patch_outputs_written
;
1729 /* Whether or not this shader ever uses textureGather() */
1730 bool uses_texture_gather
;
1732 /* Whether or not this shader uses the gl_ClipDistance output */
1733 bool uses_clip_distance_out
;
1735 /* Whether or not separate shader objects were used */
1736 bool separate_shader
;
1738 /** Was this shader linked with any transform feedback varyings? */
1739 bool has_transform_feedback_varyings
;
1743 /** The number of vertices recieves per input primitive */
1744 unsigned vertices_in
;
1746 /** The output primitive type (GL enum value) */
1747 unsigned output_primitive
;
1749 /** The maximum number of vertices the geometry shader might write. */
1750 unsigned vertices_out
;
1752 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1753 unsigned invocations
;
1755 /** Whether or not this shader uses EndPrimitive */
1756 bool uses_end_primitive
;
1758 /** Whether or not this shader uses non-zero streams */
1766 * Whether any inputs are declared with the "sample" qualifier.
1768 bool uses_sample_qualifier
;
1771 * Whether early fragment tests are enabled as defined by
1772 * ARB_shader_image_load_store.
1774 bool early_fragment_tests
;
1776 /** gl_FragDepth layout for ARB_conservative_depth. */
1777 enum gl_frag_depth_layout depth_layout
;
1781 unsigned local_size
[3];
1785 /** The number of vertices in the TCS output patch. */
1786 unsigned vertices_out
;
1791 typedef struct nir_shader
{
1792 /** list of uniforms (nir_variable) */
1793 struct exec_list uniforms
;
1795 /** list of inputs (nir_variable) */
1796 struct exec_list inputs
;
1798 /** list of outputs (nir_variable) */
1799 struct exec_list outputs
;
1801 /** list of shared compute variables (nir_variable) */
1802 struct exec_list shared
;
1804 /** Set of driver-specific options for the shader.
1806 * The memory for the options is expected to be kept in a single static
1807 * copy by the driver.
1809 const struct nir_shader_compiler_options
*options
;
1811 /** Various bits of compile-time information about a given shader */
1812 struct nir_shader_info info
;
1814 /** list of global variables in the shader (nir_variable) */
1815 struct exec_list globals
;
1817 /** list of system value variables in the shader (nir_variable) */
1818 struct exec_list system_values
;
1820 struct exec_list functions
; /** < list of nir_function */
1822 /** list of global register in the shader */
1823 struct exec_list registers
;
1825 /** next available global register index */
1829 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1832 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1834 /** The shader stage, such as MESA_SHADER_VERTEX. */
1835 gl_shader_stage stage
;
1838 static inline nir_function
*
1839 nir_shader_get_entrypoint(nir_shader
*shader
)
1841 assert(exec_list_length(&shader
->functions
) == 1);
1842 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1843 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1844 assert(func
->return_type
== glsl_void_type());
1845 assert(func
->num_params
== 0);
1849 #define nir_foreach_function(func, shader) \
1850 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1852 nir_shader
*nir_shader_create(void *mem_ctx
,
1853 gl_shader_stage stage
,
1854 const nir_shader_compiler_options
*options
);
1856 /** creates a register, including assigning it an index and adding it to the list */
1857 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1859 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1861 void nir_reg_remove(nir_register
*reg
);
1863 /** Adds a variable to the appropreate list in nir_shader */
1864 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1867 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1869 assert(var
->data
.mode
== nir_var_local
);
1870 exec_list_push_tail(&impl
->locals
, &var
->node
);
1873 /** creates a variable, sets a few defaults, and adds it to the list */
1874 nir_variable
*nir_variable_create(nir_shader
*shader
,
1875 nir_variable_mode mode
,
1876 const struct glsl_type
*type
,
1878 /** creates a local variable and adds it to the list */
1879 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1880 const struct glsl_type
*type
,
1883 /** creates a function and adds it to the shader's list of functions */
1884 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1886 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1887 /** creates a function_impl that isn't tied to any particular function */
1888 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1890 nir_block
*nir_block_create(nir_shader
*shader
);
1891 nir_if
*nir_if_create(nir_shader
*shader
);
1892 nir_loop
*nir_loop_create(nir_shader
*shader
);
1894 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1896 /** requests that the given pieces of metadata be generated */
1897 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1898 /** dirties all but the preserved metadata */
1899 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1901 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1902 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1904 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1906 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1907 unsigned num_components
,
1910 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1911 nir_intrinsic_op op
);
1913 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1914 nir_function
*callee
);
1916 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1918 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1920 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1922 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1923 unsigned num_components
,
1926 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1927 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1928 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1930 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1932 nir_load_const_instr
*
1933 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1936 * NIR Cursors and Instruction Insertion API
1939 * A tiny struct representing a point to insert/extract instructions or
1940 * control flow nodes. Helps reduce the combinatorial explosion of possible
1941 * points to insert/extract.
1943 * \sa nir_control_flow.h
1946 nir_cursor_before_block
,
1947 nir_cursor_after_block
,
1948 nir_cursor_before_instr
,
1949 nir_cursor_after_instr
,
1950 } nir_cursor_option
;
1953 nir_cursor_option option
;
1960 static inline nir_block
*
1961 nir_cursor_current_block(nir_cursor cursor
)
1963 if (cursor
.option
== nir_cursor_before_instr
||
1964 cursor
.option
== nir_cursor_after_instr
) {
1965 return cursor
.instr
->block
;
1967 return cursor
.block
;
1971 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
1973 static inline nir_cursor
1974 nir_before_block(nir_block
*block
)
1977 cursor
.option
= nir_cursor_before_block
;
1978 cursor
.block
= block
;
1982 static inline nir_cursor
1983 nir_after_block(nir_block
*block
)
1986 cursor
.option
= nir_cursor_after_block
;
1987 cursor
.block
= block
;
1991 static inline nir_cursor
1992 nir_before_instr(nir_instr
*instr
)
1995 cursor
.option
= nir_cursor_before_instr
;
1996 cursor
.instr
= instr
;
2000 static inline nir_cursor
2001 nir_after_instr(nir_instr
*instr
)
2004 cursor
.option
= nir_cursor_after_instr
;
2005 cursor
.instr
= instr
;
2009 static inline nir_cursor
2010 nir_after_block_before_jump(nir_block
*block
)
2012 nir_instr
*last_instr
= nir_block_last_instr(block
);
2013 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2014 return nir_before_instr(last_instr
);
2016 return nir_after_block(block
);
2020 static inline nir_cursor
2021 nir_before_cf_node(nir_cf_node
*node
)
2023 if (node
->type
== nir_cf_node_block
)
2024 return nir_before_block(nir_cf_node_as_block(node
));
2026 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2029 static inline nir_cursor
2030 nir_after_cf_node(nir_cf_node
*node
)
2032 if (node
->type
== nir_cf_node_block
)
2033 return nir_after_block(nir_cf_node_as_block(node
));
2035 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2038 static inline nir_cursor
2039 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2041 if (node
->type
== nir_cf_node_block
)
2042 return nir_after_block(nir_cf_node_as_block(node
));
2044 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2045 assert(block
->cf_node
.type
== nir_cf_node_block
);
2047 nir_foreach_instr(instr
, block
) {
2048 if (instr
->type
!= nir_instr_type_phi
)
2049 return nir_before_instr(instr
);
2051 return nir_after_block(block
);
2054 static inline nir_cursor
2055 nir_before_cf_list(struct exec_list
*cf_list
)
2057 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2058 exec_list_get_head(cf_list
), node
);
2059 return nir_before_cf_node(first_node
);
2062 static inline nir_cursor
2063 nir_after_cf_list(struct exec_list
*cf_list
)
2065 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2066 exec_list_get_tail(cf_list
), node
);
2067 return nir_after_cf_node(last_node
);
2071 * Insert a NIR instruction at the given cursor.
2073 * Note: This does not update the cursor.
2075 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2078 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2080 nir_instr_insert(nir_before_instr(instr
), before
);
2084 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2086 nir_instr_insert(nir_after_instr(instr
), after
);
2090 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2092 nir_instr_insert(nir_before_block(block
), before
);
2096 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2098 nir_instr_insert(nir_after_block(block
), after
);
2102 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2104 nir_instr_insert(nir_before_cf_node(node
), before
);
2108 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2110 nir_instr_insert(nir_after_cf_node(node
), after
);
2114 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2116 nir_instr_insert(nir_before_cf_list(list
), before
);
2120 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2122 nir_instr_insert(nir_after_cf_list(list
), after
);
2125 void nir_instr_remove(nir_instr
*instr
);
2129 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2130 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2131 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2132 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2134 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2135 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2137 nir_const_value
*nir_src_as_const_value(nir_src src
);
2138 bool nir_src_is_dynamically_uniform(nir_src src
);
2139 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2140 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2141 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2142 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2143 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2146 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2147 unsigned num_components
, unsigned bit_size
,
2149 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2150 unsigned num_components
, unsigned bit_size
,
2152 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2153 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2154 nir_instr
*after_me
);
2156 uint8_t nir_ssa_def_components_read(nir_ssa_def
*def
);
2159 * finds the next basic block in source-code order, returns NULL if there is
2163 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2165 /* Performs the opposite of nir_block_cf_tree_next() */
2167 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2169 /* Gets the first block in a CF node in source-code order */
2171 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2173 /* Gets the last block in a CF node in source-code order */
2175 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2177 /* Gets the next block after a CF node in source-code order */
2179 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2181 /* Macros for loops that visit blocks in source-code order */
2183 #define nir_foreach_block(block, impl) \
2184 for (nir_block *block = nir_start_block(impl); block != NULL; \
2185 block = nir_block_cf_tree_next(block))
2187 #define nir_foreach_block_safe(block, impl) \
2188 for (nir_block *block = nir_start_block(impl), \
2189 *next = nir_block_cf_tree_next(block); \
2191 block = next, next = nir_block_cf_tree_next(block))
2193 #define nir_foreach_block_reverse(block, impl) \
2194 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2195 block = nir_block_cf_tree_prev(block))
2197 #define nir_foreach_block_reverse_safe(block, impl) \
2198 for (nir_block *block = nir_impl_last_block(impl), \
2199 *prev = nir_block_cf_tree_prev(block); \
2201 block = prev, prev = nir_block_cf_tree_prev(block))
2203 #define nir_foreach_block_in_cf_node(block, node) \
2204 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2205 block != nir_cf_node_cf_tree_next(node); \
2206 block = nir_block_cf_tree_next(block))
2208 /* If the following CF node is an if, this function returns that if.
2209 * Otherwise, it returns NULL.
2211 nir_if
*nir_block_get_following_if(nir_block
*block
);
2213 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2215 void nir_index_local_regs(nir_function_impl
*impl
);
2216 void nir_index_global_regs(nir_shader
*shader
);
2217 void nir_index_ssa_defs(nir_function_impl
*impl
);
2218 unsigned nir_index_instrs(nir_function_impl
*impl
);
2220 void nir_index_blocks(nir_function_impl
*impl
);
2222 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2223 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2224 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2226 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2227 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2228 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2229 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2232 void nir_validate_shader(nir_shader
*shader
);
2233 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2234 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2236 #include "util/debug.h"
2238 should_clone_nir(void)
2240 static int should_clone
= -1;
2241 if (should_clone
< 0)
2242 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2244 return should_clone
;
2247 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2248 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2249 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2250 static inline bool should_clone_nir(void) { return false; }
2253 #define _PASS(nir, do_pass) do { \
2255 nir_validate_shader(nir); \
2256 if (should_clone_nir()) { \
2257 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2263 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2264 nir_metadata_set_validation_flag(nir); \
2265 if (pass(nir, ##__VA_ARGS__)) { \
2267 nir_metadata_check_validation_flag(nir); \
2271 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2272 pass(nir, ##__VA_ARGS__); \
2275 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2276 void nir_calc_dominance(nir_shader
*shader
);
2278 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2279 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2281 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2282 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2284 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2285 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2287 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2288 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2290 int nir_gs_count_vertices(const nir_shader
*shader
);
2292 bool nir_split_var_copies(nir_shader
*shader
);
2294 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2295 bool nir_lower_returns(nir_shader
*shader
);
2297 bool nir_inline_functions(nir_shader
*shader
);
2299 bool nir_propagate_invariant(nir_shader
*shader
);
2301 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2302 void nir_lower_var_copies(nir_shader
*shader
);
2304 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2306 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2308 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2310 void nir_lower_io_to_temporaries(nir_shader
*shader
, nir_function
*entrypoint
,
2311 bool outputs
, bool inputs
);
2313 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2315 void nir_assign_var_locations(struct exec_list
*var_list
,
2317 int (*type_size
)(const struct glsl_type
*));
2319 void nir_lower_io(nir_shader
*shader
,
2320 nir_variable_mode modes
,
2321 int (*type_size
)(const struct glsl_type
*));
2322 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2323 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2325 void nir_lower_io_types(nir_shader
*shader
);
2326 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2328 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2330 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2331 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2332 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2333 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2335 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2337 void nir_lower_samplers(nir_shader
*shader
,
2338 const struct gl_shader_program
*shader_program
);
2340 bool nir_lower_system_values(nir_shader
*shader
);
2342 typedef struct nir_lower_tex_options
{
2344 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2345 * sampler types a texture projector is lowered.
2350 * If true, lower rect textures to 2D, using txs to fetch the
2351 * texture dimensions and dividing the texture coords by the
2352 * texture dims to normalize.
2357 * If true, convert yuv to rgb.
2359 unsigned lower_y_uv_external
;
2360 unsigned lower_y_u_v_external
;
2361 unsigned lower_yx_xuxv_external
;
2364 * To emulate certain texture wrap modes, this can be used
2365 * to saturate the specified tex coord to [0.0, 1.0]. The
2366 * bits are according to sampler #, ie. if, for example:
2368 * (conf->saturate_s & (1 << n))
2370 * is true, then the s coord for sampler n is saturated.
2372 * Note that clamping must happen *after* projector lowering
2373 * so any projected texture sample instruction with a clamped
2374 * coordinate gets automatically lowered, regardless of the
2375 * 'lower_txp' setting.
2377 unsigned saturate_s
;
2378 unsigned saturate_t
;
2379 unsigned saturate_r
;
2381 /* Bitmask of textures that need swizzling.
2383 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2384 * swizzles[texture_index] is applied to the result of the texturing
2387 unsigned swizzle_result
;
2389 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2390 * while 4 and 5 represent 0 and 1 respectively.
2392 uint8_t swizzles
[32][4];
2395 * Bitmap of textures that need srgb to linear conversion. If
2396 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2397 * of the texture are lowered to linear.
2399 unsigned lower_srgb
;
2400 } nir_lower_tex_options
;
2402 bool nir_lower_tex(nir_shader
*shader
,
2403 const nir_lower_tex_options
*options
);
2405 bool nir_lower_idiv(nir_shader
*shader
);
2407 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2408 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2410 void nir_lower_two_sided_color(nir_shader
*shader
);
2412 void nir_lower_clamp_color_outputs(nir_shader
*shader
);
2414 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2416 typedef struct nir_lower_wpos_ytransform_options
{
2417 int state_tokens
[5];
2418 bool fs_coord_origin_upper_left
:1;
2419 bool fs_coord_origin_lower_left
:1;
2420 bool fs_coord_pixel_center_integer
:1;
2421 bool fs_coord_pixel_center_half_integer
:1;
2422 } nir_lower_wpos_ytransform_options
;
2424 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2425 const nir_lower_wpos_ytransform_options
*options
);
2426 bool nir_lower_wpos_center(nir_shader
*shader
);
2428 typedef struct nir_lower_drawpixels_options
{
2429 int texcoord_state_tokens
[5];
2430 int scale_state_tokens
[5];
2431 int bias_state_tokens
[5];
2432 unsigned drawpix_sampler
;
2433 unsigned pixelmap_sampler
;
2435 bool scale_and_bias
:1;
2436 } nir_lower_drawpixels_options
;
2438 void nir_lower_drawpixels(nir_shader
*shader
,
2439 const nir_lower_drawpixels_options
*options
);
2441 typedef struct nir_lower_bitmap_options
{
2444 } nir_lower_bitmap_options
;
2446 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2448 void nir_lower_atomics(nir_shader
*shader
,
2449 const struct gl_shader_program
*shader_program
);
2450 void nir_lower_to_source_mods(nir_shader
*shader
);
2452 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2455 nir_lower_drcp
= (1 << 0),
2456 nir_lower_dsqrt
= (1 << 1),
2457 nir_lower_drsq
= (1 << 2),
2458 nir_lower_dtrunc
= (1 << 3),
2459 nir_lower_dfloor
= (1 << 4),
2460 nir_lower_dceil
= (1 << 5),
2461 nir_lower_dfract
= (1 << 6),
2462 nir_lower_dround_even
= (1 << 7),
2463 nir_lower_dmod
= (1 << 8)
2464 } nir_lower_doubles_options
;
2466 void nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2467 void nir_lower_double_pack(nir_shader
*shader
);
2469 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2471 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2472 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2474 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2475 void nir_convert_to_ssa(nir_shader
*shader
);
2477 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2478 bool nir_repair_ssa(nir_shader
*shader
);
2480 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2481 * registers. If false, convert all values (even those not involved in a phi
2482 * node) to registers.
2484 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2486 bool nir_opt_algebraic(nir_shader
*shader
);
2487 bool nir_opt_algebraic_late(nir_shader
*shader
);
2488 bool nir_opt_constant_folding(nir_shader
*shader
);
2490 bool nir_opt_global_to_local(nir_shader
*shader
);
2492 bool nir_copy_prop(nir_shader
*shader
);
2494 bool nir_opt_cse(nir_shader
*shader
);
2496 bool nir_opt_dce(nir_shader
*shader
);
2498 bool nir_opt_dead_cf(nir_shader
*shader
);
2500 void nir_opt_gcm(nir_shader
*shader
);
2502 bool nir_opt_peephole_select(nir_shader
*shader
);
2504 bool nir_opt_remove_phis(nir_shader
*shader
);
2506 bool nir_opt_undef(nir_shader
*shader
);
2508 void nir_sweep(nir_shader
*shader
);
2510 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2511 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);