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 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,
990 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
992 } nir_intrinsic_index_flag
;
994 #define NIR_INTRINSIC_MAX_INPUTS 4
999 unsigned num_srcs
; /** < number of register/SSA inputs */
1001 /** number of components of each input register
1003 * If this value is 0, the number of components is given by the
1004 * num_components field of nir_intrinsic_instr.
1006 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1010 /** number of components of the output register
1012 * If this value is 0, the number of components is given by the
1013 * num_components field of nir_intrinsic_instr.
1015 unsigned dest_components
;
1017 /** the number of inputs/outputs that are variables */
1018 unsigned num_variables
;
1020 /** the number of constant indices used by the intrinsic */
1021 unsigned num_indices
;
1023 /** indicates the usage of intr->const_index[n] */
1024 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1026 /** semantic flags for calls to this intrinsic */
1027 nir_intrinsic_semantic_flag flags
;
1028 } nir_intrinsic_info
;
1030 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1033 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1034 static inline type \
1035 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
1037 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1038 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1039 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1041 static inline void \
1042 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1044 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1045 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1046 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1049 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1050 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1051 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1052 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1053 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1054 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1055 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1058 * \group texture information
1060 * This gives semantic information about textures which is useful to the
1061 * frontend, the backend, and lowering passes, but not the optimizer.
1066 nir_tex_src_projector
,
1067 nir_tex_src_comparitor
, /* shadow comparitor */
1071 nir_tex_src_ms_index
, /* MSAA sample index */
1072 nir_tex_src_ms_mcs
, /* MSAA compression value */
1075 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1076 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1077 nir_num_tex_src_types
1082 nir_tex_src_type src_type
;
1086 nir_texop_tex
, /**< Regular texture look-up */
1087 nir_texop_txb
, /**< Texture look-up with LOD bias */
1088 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1089 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1090 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1091 nir_texop_txf_ms
, /**< Multisample texture fetch */
1092 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1093 nir_texop_txs
, /**< Texture size */
1094 nir_texop_lod
, /**< Texture lod query */
1095 nir_texop_tg4
, /**< Texture gather */
1096 nir_texop_query_levels
, /**< Texture levels query */
1097 nir_texop_texture_samples
, /**< Texture samples query */
1098 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1106 enum glsl_sampler_dim sampler_dim
;
1107 nir_alu_type dest_type
;
1112 unsigned num_srcs
, coord_components
;
1113 bool is_array
, is_shadow
;
1116 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1117 * components or the new-style shadow that outputs 1 component.
1119 bool is_new_style_shadow
;
1121 /* gather component selector */
1122 unsigned component
: 2;
1124 /** The texture index
1126 * If this texture instruction has a nir_tex_src_texture_offset source,
1127 * then the texture index is given by texture_index + texture_offset.
1129 unsigned texture_index
;
1131 /** The size of the texture array or 0 if it's not an array */
1132 unsigned texture_array_size
;
1134 /** The texture deref
1136 * If this is null, use texture_index instead.
1138 nir_deref_var
*texture
;
1140 /** The sampler index
1142 * The following operations do not require a sampler and, as such, this
1143 * field should be ignored:
1145 * - nir_texop_txf_ms
1149 * - nir_texop_query_levels
1150 * - nir_texop_texture_samples
1151 * - nir_texop_samples_identical
1153 * If this texture instruction has a nir_tex_src_sampler_offset source,
1154 * then the sampler index is given by sampler_index + sampler_offset.
1156 unsigned sampler_index
;
1158 /** The sampler deref
1160 * If this is null, use sampler_index instead.
1162 nir_deref_var
*sampler
;
1165 static inline unsigned
1166 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1168 switch (instr
->op
) {
1169 case nir_texop_txs
: {
1171 switch (instr
->sampler_dim
) {
1172 case GLSL_SAMPLER_DIM_1D
:
1173 case GLSL_SAMPLER_DIM_BUF
:
1176 case GLSL_SAMPLER_DIM_2D
:
1177 case GLSL_SAMPLER_DIM_CUBE
:
1178 case GLSL_SAMPLER_DIM_MS
:
1179 case GLSL_SAMPLER_DIM_RECT
:
1180 case GLSL_SAMPLER_DIM_EXTERNAL
:
1183 case GLSL_SAMPLER_DIM_3D
:
1187 unreachable("not reached");
1189 if (instr
->is_array
)
1197 case nir_texop_texture_samples
:
1198 case nir_texop_query_levels
:
1199 case nir_texop_samples_identical
:
1203 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1210 /* Returns true if this texture operation queries something about the texture
1211 * rather than actually sampling it.
1214 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1216 switch (instr
->op
) {
1219 case nir_texop_texture_samples
:
1220 case nir_texop_query_levels
:
1221 case nir_texop_txf_ms_mcs
:
1228 case nir_texop_txf_ms
:
1232 unreachable("Invalid texture opcode");
1236 static inline unsigned
1237 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1239 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1240 return instr
->coord_components
;
1242 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1243 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1246 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1247 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1248 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1249 if (instr
->is_array
)
1250 return instr
->coord_components
- 1;
1252 return instr
->coord_components
;
1259 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1261 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1262 if (instr
->src
[i
].src_type
== type
)
1282 nir_const_value value
;
1285 } nir_load_const_instr
;
1298 /* creates a new SSA variable in an undefined state */
1303 } nir_ssa_undef_instr
;
1306 struct exec_node node
;
1308 /* The predecessor block corresponding to this source */
1309 struct nir_block
*pred
;
1314 #define nir_foreach_phi_src(phi_src, phi) \
1315 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1316 #define nir_foreach_phi_src_safe(phi_src, phi) \
1317 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1322 struct exec_list srcs
; /** < list of nir_phi_src */
1328 struct exec_node node
;
1331 } nir_parallel_copy_entry
;
1333 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1334 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1339 /* A list of nir_parallel_copy_entry's. The sources of all of the
1340 * entries are copied to the corresponding destinations "in parallel".
1341 * In other words, if we have two entries: a -> b and b -> a, the values
1344 struct exec_list entries
;
1345 } nir_parallel_copy_instr
;
1347 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1348 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1349 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1350 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1351 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1352 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1353 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1354 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1355 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1356 nir_parallel_copy_instr
, instr
)
1361 * Control flow consists of a tree of control flow nodes, which include
1362 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1363 * instructions that always run start-to-finish. Each basic block also keeps
1364 * track of its successors (blocks which may run immediately after the current
1365 * block) and predecessors (blocks which could have run immediately before the
1366 * current block). Each function also has a start block and an end block which
1367 * all return statements point to (which is always empty). Together, all the
1368 * blocks with their predecessors and successors make up the control flow
1369 * graph (CFG) of the function. There are helpers that modify the tree of
1370 * control flow nodes while modifying the CFG appropriately; these should be
1371 * used instead of modifying the tree directly.
1378 nir_cf_node_function
1381 typedef struct nir_cf_node
{
1382 struct exec_node node
;
1383 nir_cf_node_type type
;
1384 struct nir_cf_node
*parent
;
1387 typedef struct nir_block
{
1388 nir_cf_node cf_node
;
1390 struct exec_list instr_list
; /** < list of nir_instr */
1392 /** generic block index; generated by nir_index_blocks */
1396 * Each block can only have up to 2 successors, so we put them in a simple
1397 * array - no need for anything more complicated.
1399 struct nir_block
*successors
[2];
1401 /* Set of nir_block predecessors in the CFG */
1402 struct set
*predecessors
;
1405 * this node's immediate dominator in the dominance tree - set to NULL for
1408 struct nir_block
*imm_dom
;
1410 /* This node's children in the dominance tree */
1411 unsigned num_dom_children
;
1412 struct nir_block
**dom_children
;
1414 /* Set of nir_block's on the dominance frontier of this block */
1415 struct set
*dom_frontier
;
1418 * These two indices have the property that dom_{pre,post}_index for each
1419 * child of this block in the dominance tree will always be between
1420 * dom_pre_index and dom_post_index for this block, which makes testing if
1421 * a given block is dominated by another block an O(1) operation.
1423 unsigned dom_pre_index
, dom_post_index
;
1425 /* live in and out for this block; used for liveness analysis */
1426 BITSET_WORD
*live_in
;
1427 BITSET_WORD
*live_out
;
1430 static inline nir_instr
*
1431 nir_block_first_instr(nir_block
*block
)
1433 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1434 return exec_node_data(nir_instr
, head
, node
);
1437 static inline nir_instr
*
1438 nir_block_last_instr(nir_block
*block
)
1440 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1441 return exec_node_data(nir_instr
, tail
, node
);
1444 #define nir_foreach_instr(instr, block) \
1445 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1446 #define nir_foreach_instr_reverse(instr, block) \
1447 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1448 #define nir_foreach_instr_safe(instr, block) \
1449 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1450 #define nir_foreach_instr_reverse_safe(instr, block) \
1451 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1453 typedef struct nir_if
{
1454 nir_cf_node cf_node
;
1457 struct exec_list then_list
; /** < list of nir_cf_node */
1458 struct exec_list else_list
; /** < list of nir_cf_node */
1461 static inline nir_cf_node
*
1462 nir_if_first_then_node(nir_if
*if_stmt
)
1464 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1465 return exec_node_data(nir_cf_node
, head
, node
);
1468 static inline nir_cf_node
*
1469 nir_if_last_then_node(nir_if
*if_stmt
)
1471 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1472 return exec_node_data(nir_cf_node
, tail
, node
);
1475 static inline nir_cf_node
*
1476 nir_if_first_else_node(nir_if
*if_stmt
)
1478 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1479 return exec_node_data(nir_cf_node
, head
, node
);
1482 static inline nir_cf_node
*
1483 nir_if_last_else_node(nir_if
*if_stmt
)
1485 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1486 return exec_node_data(nir_cf_node
, tail
, node
);
1490 nir_cf_node cf_node
;
1492 struct exec_list body
; /** < list of nir_cf_node */
1495 static inline nir_cf_node
*
1496 nir_loop_first_cf_node(nir_loop
*loop
)
1498 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1501 static inline nir_cf_node
*
1502 nir_loop_last_cf_node(nir_loop
*loop
)
1504 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1508 * Various bits of metadata that can may be created or required by
1509 * optimization and analysis passes
1512 nir_metadata_none
= 0x0,
1513 nir_metadata_block_index
= 0x1,
1514 nir_metadata_dominance
= 0x2,
1515 nir_metadata_live_ssa_defs
= 0x4,
1516 nir_metadata_not_properly_reset
= 0x8,
1520 nir_cf_node cf_node
;
1522 /** pointer to the function of which this is an implementation */
1523 struct nir_function
*function
;
1525 struct exec_list body
; /** < list of nir_cf_node */
1527 nir_block
*end_block
;
1529 /** list for all local variables in the function */
1530 struct exec_list locals
;
1532 /** array of variables used as parameters */
1533 unsigned num_params
;
1534 nir_variable
**params
;
1536 /** variable used to hold the result of the function */
1537 nir_variable
*return_var
;
1539 /** list of local registers in the function */
1540 struct exec_list registers
;
1542 /** next available local register index */
1545 /** next available SSA value index */
1548 /* total number of basic blocks, only valid when block_index_dirty = false */
1549 unsigned num_blocks
;
1551 nir_metadata valid_metadata
;
1552 } nir_function_impl
;
1554 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1555 nir_start_block(nir_function_impl
*impl
)
1557 return (nir_block
*) impl
->body
.head
;
1560 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1561 nir_impl_last_block(nir_function_impl
*impl
)
1563 return (nir_block
*) impl
->body
.tail_pred
;
1566 static inline nir_cf_node
*
1567 nir_cf_node_next(nir_cf_node
*node
)
1569 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1570 if (exec_node_is_tail_sentinel(next
))
1573 return exec_node_data(nir_cf_node
, next
, node
);
1576 static inline nir_cf_node
*
1577 nir_cf_node_prev(nir_cf_node
*node
)
1579 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1580 if (exec_node_is_head_sentinel(prev
))
1583 return exec_node_data(nir_cf_node
, prev
, node
);
1587 nir_cf_node_is_first(const nir_cf_node
*node
)
1589 return exec_node_is_head_sentinel(node
->node
.prev
);
1593 nir_cf_node_is_last(const nir_cf_node
*node
)
1595 return exec_node_is_tail_sentinel(node
->node
.next
);
1598 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1599 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1600 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1601 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1606 nir_parameter_inout
,
1607 } nir_parameter_type
;
1610 nir_parameter_type param_type
;
1611 const struct glsl_type
*type
;
1614 typedef struct nir_function
{
1615 struct exec_node node
;
1618 struct nir_shader
*shader
;
1620 unsigned num_params
;
1621 nir_parameter
*params
;
1622 const struct glsl_type
*return_type
;
1624 /** The implementation of this function.
1626 * If the function is only declared and not implemented, this is NULL.
1628 nir_function_impl
*impl
;
1631 typedef struct nir_shader_compiler_options
{
1636 /** Lowers flrp when it does not support doubles */
1643 bool lower_bitfield_extract
;
1644 bool lower_bitfield_insert
;
1645 bool lower_uadd_carry
;
1646 bool lower_usub_borrow
;
1647 /** lowers fneg and ineg to fsub and isub. */
1649 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1652 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1655 /* Does the native fdot instruction replicate its result for four
1656 * components? If so, then opt_algebraic_late will turn all fdotN
1657 * instructions into fdot_replicatedN instructions.
1659 bool fdot_replicates
;
1661 /** lowers ffract to fsub+ffloor: */
1664 bool lower_pack_half_2x16
;
1665 bool lower_pack_unorm_2x16
;
1666 bool lower_pack_snorm_2x16
;
1667 bool lower_pack_unorm_4x8
;
1668 bool lower_pack_snorm_4x8
;
1669 bool lower_unpack_half_2x16
;
1670 bool lower_unpack_unorm_2x16
;
1671 bool lower_unpack_snorm_2x16
;
1672 bool lower_unpack_unorm_4x8
;
1673 bool lower_unpack_snorm_4x8
;
1675 bool lower_extract_byte
;
1676 bool lower_extract_word
;
1679 * Does the driver support real 32-bit integers? (Otherwise, integers
1680 * are simulated by floats.)
1682 bool native_integers
;
1684 /* Indicates that the driver only has zero-based vertex id */
1685 bool vertex_id_zero_based
;
1686 } nir_shader_compiler_options
;
1688 typedef struct nir_shader_info
{
1691 /* Descriptive name provided by the client; may be NULL */
1694 /* Number of textures used by this shader */
1695 unsigned num_textures
;
1696 /* Number of uniform buffers used by this shader */
1698 /* Number of atomic buffers used by this shader */
1700 /* Number of shader storage buffers used by this shader */
1702 /* Number of images used by this shader */
1703 unsigned num_images
;
1705 /* Which inputs are actually read */
1706 uint64_t inputs_read
;
1707 /* Which inputs are actually read and are double */
1708 uint64_t double_inputs_read
;
1709 /* Which outputs are actually written */
1710 uint64_t outputs_written
;
1711 /* Which system values are actually read */
1712 uint64_t system_values_read
;
1714 /* Which patch inputs are actually read */
1715 uint32_t patch_inputs_read
;
1716 /* Which patch outputs are actually written */
1717 uint32_t patch_outputs_written
;
1719 /* Whether or not this shader ever uses textureGather() */
1720 bool uses_texture_gather
;
1722 /** Whether or not this shader uses nir_intrinsic_interp_var_at_offset */
1723 bool uses_interp_var_at_offset
;
1725 /* Whether or not this shader uses the gl_ClipDistance output */
1726 bool uses_clip_distance_out
;
1728 /* Whether or not separate shader objects were used */
1729 bool separate_shader
;
1731 /** Was this shader linked with any transform feedback varyings? */
1732 bool has_transform_feedback_varyings
;
1736 /** The number of vertices recieves per input primitive */
1737 unsigned vertices_in
;
1739 /** The output primitive type (GL enum value) */
1740 unsigned output_primitive
;
1742 /** The maximum number of vertices the geometry shader might write. */
1743 unsigned vertices_out
;
1745 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1746 unsigned invocations
;
1748 /** Whether or not this shader uses EndPrimitive */
1749 bool uses_end_primitive
;
1751 /** Whether or not this shader uses non-zero streams */
1759 * Whether any inputs are declared with the "sample" qualifier.
1761 bool uses_sample_qualifier
;
1764 * Whether early fragment tests are enabled as defined by
1765 * ARB_shader_image_load_store.
1767 bool early_fragment_tests
;
1769 /** gl_FragDepth layout for ARB_conservative_depth. */
1770 enum gl_frag_depth_layout depth_layout
;
1774 unsigned local_size
[3];
1778 /** The number of vertices in the TCS output patch. */
1779 unsigned vertices_out
;
1784 typedef struct nir_shader
{
1785 /** list of uniforms (nir_variable) */
1786 struct exec_list uniforms
;
1788 /** list of inputs (nir_variable) */
1789 struct exec_list inputs
;
1791 /** list of outputs (nir_variable) */
1792 struct exec_list outputs
;
1794 /** list of shared compute variables (nir_variable) */
1795 struct exec_list shared
;
1797 /** Set of driver-specific options for the shader.
1799 * The memory for the options is expected to be kept in a single static
1800 * copy by the driver.
1802 const struct nir_shader_compiler_options
*options
;
1804 /** Various bits of compile-time information about a given shader */
1805 struct nir_shader_info info
;
1807 /** list of global variables in the shader (nir_variable) */
1808 struct exec_list globals
;
1810 /** list of system value variables in the shader (nir_variable) */
1811 struct exec_list system_values
;
1813 struct exec_list functions
; /** < list of nir_function */
1815 /** list of global register in the shader */
1816 struct exec_list registers
;
1818 /** next available global register index */
1822 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1825 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1827 /** The shader stage, such as MESA_SHADER_VERTEX. */
1828 gl_shader_stage stage
;
1831 static inline nir_function
*
1832 nir_shader_get_entrypoint(nir_shader
*shader
)
1834 assert(exec_list_length(&shader
->functions
) == 1);
1835 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1836 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1837 assert(func
->return_type
== glsl_void_type());
1838 assert(func
->num_params
== 0);
1842 #define nir_foreach_function(func, shader) \
1843 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1845 nir_shader
*nir_shader_create(void *mem_ctx
,
1846 gl_shader_stage stage
,
1847 const nir_shader_compiler_options
*options
);
1849 /** creates a register, including assigning it an index and adding it to the list */
1850 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1852 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1854 void nir_reg_remove(nir_register
*reg
);
1856 /** Adds a variable to the appropreate list in nir_shader */
1857 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1860 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1862 assert(var
->data
.mode
== nir_var_local
);
1863 exec_list_push_tail(&impl
->locals
, &var
->node
);
1866 /** creates a variable, sets a few defaults, and adds it to the list */
1867 nir_variable
*nir_variable_create(nir_shader
*shader
,
1868 nir_variable_mode mode
,
1869 const struct glsl_type
*type
,
1871 /** creates a local variable and adds it to the list */
1872 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1873 const struct glsl_type
*type
,
1876 /** creates a function and adds it to the shader's list of functions */
1877 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1879 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1880 /** creates a function_impl that isn't tied to any particular function */
1881 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1883 nir_block
*nir_block_create(nir_shader
*shader
);
1884 nir_if
*nir_if_create(nir_shader
*shader
);
1885 nir_loop
*nir_loop_create(nir_shader
*shader
);
1887 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1889 /** requests that the given pieces of metadata be generated */
1890 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1891 /** dirties all but the preserved metadata */
1892 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1894 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1895 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1897 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1899 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1900 unsigned num_components
,
1903 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1904 nir_intrinsic_op op
);
1906 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1907 nir_function
*callee
);
1909 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1911 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1913 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1915 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1916 unsigned num_components
,
1919 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1920 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1921 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1923 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1925 nir_load_const_instr
*
1926 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1929 * NIR Cursors and Instruction Insertion API
1932 * A tiny struct representing a point to insert/extract instructions or
1933 * control flow nodes. Helps reduce the combinatorial explosion of possible
1934 * points to insert/extract.
1936 * \sa nir_control_flow.h
1939 nir_cursor_before_block
,
1940 nir_cursor_after_block
,
1941 nir_cursor_before_instr
,
1942 nir_cursor_after_instr
,
1943 } nir_cursor_option
;
1946 nir_cursor_option option
;
1953 static inline nir_block
*
1954 nir_cursor_current_block(nir_cursor cursor
)
1956 if (cursor
.option
== nir_cursor_before_instr
||
1957 cursor
.option
== nir_cursor_after_instr
) {
1958 return cursor
.instr
->block
;
1960 return cursor
.block
;
1964 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
1966 static inline nir_cursor
1967 nir_before_block(nir_block
*block
)
1970 cursor
.option
= nir_cursor_before_block
;
1971 cursor
.block
= block
;
1975 static inline nir_cursor
1976 nir_after_block(nir_block
*block
)
1979 cursor
.option
= nir_cursor_after_block
;
1980 cursor
.block
= block
;
1984 static inline nir_cursor
1985 nir_before_instr(nir_instr
*instr
)
1988 cursor
.option
= nir_cursor_before_instr
;
1989 cursor
.instr
= instr
;
1993 static inline nir_cursor
1994 nir_after_instr(nir_instr
*instr
)
1997 cursor
.option
= nir_cursor_after_instr
;
1998 cursor
.instr
= instr
;
2002 static inline nir_cursor
2003 nir_after_block_before_jump(nir_block
*block
)
2005 nir_instr
*last_instr
= nir_block_last_instr(block
);
2006 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2007 return nir_before_instr(last_instr
);
2009 return nir_after_block(block
);
2013 static inline nir_cursor
2014 nir_before_cf_node(nir_cf_node
*node
)
2016 if (node
->type
== nir_cf_node_block
)
2017 return nir_before_block(nir_cf_node_as_block(node
));
2019 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2022 static inline nir_cursor
2023 nir_after_cf_node(nir_cf_node
*node
)
2025 if (node
->type
== nir_cf_node_block
)
2026 return nir_after_block(nir_cf_node_as_block(node
));
2028 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2031 static inline nir_cursor
2032 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2034 if (node
->type
== nir_cf_node_block
)
2035 return nir_after_block(nir_cf_node_as_block(node
));
2037 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2038 assert(block
->cf_node
.type
== nir_cf_node_block
);
2040 nir_foreach_instr(instr
, block
) {
2041 if (instr
->type
!= nir_instr_type_phi
)
2042 return nir_before_instr(instr
);
2044 return nir_after_block(block
);
2047 static inline nir_cursor
2048 nir_before_cf_list(struct exec_list
*cf_list
)
2050 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2051 exec_list_get_head(cf_list
), node
);
2052 return nir_before_cf_node(first_node
);
2055 static inline nir_cursor
2056 nir_after_cf_list(struct exec_list
*cf_list
)
2058 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2059 exec_list_get_tail(cf_list
), node
);
2060 return nir_after_cf_node(last_node
);
2064 * Insert a NIR instruction at the given cursor.
2066 * Note: This does not update the cursor.
2068 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2071 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2073 nir_instr_insert(nir_before_instr(instr
), before
);
2077 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2079 nir_instr_insert(nir_after_instr(instr
), after
);
2083 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2085 nir_instr_insert(nir_before_block(block
), before
);
2089 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2091 nir_instr_insert(nir_after_block(block
), after
);
2095 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2097 nir_instr_insert(nir_before_cf_node(node
), before
);
2101 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2103 nir_instr_insert(nir_after_cf_node(node
), after
);
2107 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2109 nir_instr_insert(nir_before_cf_list(list
), before
);
2113 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2115 nir_instr_insert(nir_after_cf_list(list
), after
);
2118 void nir_instr_remove(nir_instr
*instr
);
2122 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2123 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2124 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2125 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2127 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2128 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2130 nir_const_value
*nir_src_as_const_value(nir_src src
);
2131 bool nir_src_is_dynamically_uniform(nir_src src
);
2132 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2133 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2134 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2135 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2136 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2139 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2140 unsigned num_components
, unsigned bit_size
,
2142 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2143 unsigned num_components
, unsigned bit_size
,
2145 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2146 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2147 nir_instr
*after_me
);
2149 uint8_t nir_ssa_def_components_read(nir_ssa_def
*def
);
2152 * finds the next basic block in source-code order, returns NULL if there is
2156 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2158 /* Performs the opposite of nir_block_cf_tree_next() */
2160 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2162 /* Gets the first block in a CF node in source-code order */
2164 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2166 /* Gets the last block in a CF node in source-code order */
2168 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2170 /* Gets the next block after a CF node in source-code order */
2172 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2174 /* Macros for loops that visit blocks in source-code order */
2176 #define nir_foreach_block(block, impl) \
2177 for (nir_block *block = nir_start_block(impl); block != NULL; \
2178 block = nir_block_cf_tree_next(block))
2180 #define nir_foreach_block_safe(block, impl) \
2181 for (nir_block *block = nir_start_block(impl), \
2182 *next = nir_block_cf_tree_next(block); \
2184 block = next, next = nir_block_cf_tree_next(block))
2186 #define nir_foreach_block_reverse(block, impl) \
2187 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2188 block = nir_block_cf_tree_prev(block))
2190 #define nir_foreach_block_reverse_safe(block, impl) \
2191 for (nir_block *block = nir_impl_last_block(impl), \
2192 *prev = nir_block_cf_tree_prev(block); \
2194 block = prev, prev = nir_block_cf_tree_prev(block))
2196 #define nir_foreach_block_in_cf_node(block, node) \
2197 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2198 block != nir_cf_node_cf_tree_next(node); \
2199 block = nir_block_cf_tree_next(block))
2201 /* If the following CF node is an if, this function returns that if.
2202 * Otherwise, it returns NULL.
2204 nir_if
*nir_block_get_following_if(nir_block
*block
);
2206 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2208 void nir_index_local_regs(nir_function_impl
*impl
);
2209 void nir_index_global_regs(nir_shader
*shader
);
2210 void nir_index_ssa_defs(nir_function_impl
*impl
);
2211 unsigned nir_index_instrs(nir_function_impl
*impl
);
2213 void nir_index_blocks(nir_function_impl
*impl
);
2215 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2216 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2217 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2219 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2220 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2221 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2222 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2225 void nir_validate_shader(nir_shader
*shader
);
2226 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2227 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2229 #include "util/debug.h"
2231 should_clone_nir(void)
2233 static int should_clone
= -1;
2234 if (should_clone
< 0)
2235 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2237 return should_clone
;
2240 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2241 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2242 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2243 static inline bool should_clone_nir(void) { return false; }
2246 #define _PASS(nir, do_pass) do { \
2248 nir_validate_shader(nir); \
2249 if (should_clone_nir()) { \
2250 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2256 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2257 nir_metadata_set_validation_flag(nir); \
2258 if (pass(nir, ##__VA_ARGS__)) { \
2260 nir_metadata_check_validation_flag(nir); \
2264 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2265 pass(nir, ##__VA_ARGS__); \
2268 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2269 void nir_calc_dominance(nir_shader
*shader
);
2271 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2272 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2274 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2275 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2277 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2278 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2280 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2281 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2283 int nir_gs_count_vertices(const nir_shader
*shader
);
2285 bool nir_split_var_copies(nir_shader
*shader
);
2287 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2288 bool nir_lower_returns(nir_shader
*shader
);
2290 bool nir_inline_functions(nir_shader
*shader
);
2292 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2293 void nir_lower_var_copies(nir_shader
*shader
);
2295 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2297 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2299 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2301 void nir_lower_io_to_temporaries(nir_shader
*shader
, nir_function
*entrypoint
,
2302 bool outputs
, bool inputs
);
2304 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2306 void nir_assign_var_locations(struct exec_list
*var_list
,
2308 int (*type_size
)(const struct glsl_type
*));
2310 void nir_lower_io(nir_shader
*shader
,
2311 nir_variable_mode modes
,
2312 int (*type_size
)(const struct glsl_type
*));
2313 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2314 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2316 void nir_lower_io_types(nir_shader
*shader
);
2317 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2319 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2321 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2322 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2323 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2324 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2326 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2328 void nir_lower_samplers(nir_shader
*shader
,
2329 const struct gl_shader_program
*shader_program
);
2331 bool nir_lower_system_values(nir_shader
*shader
);
2333 typedef struct nir_lower_tex_options
{
2335 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2336 * sampler types a texture projector is lowered.
2341 * If true, lower rect textures to 2D, using txs to fetch the
2342 * texture dimensions and dividing the texture coords by the
2343 * texture dims to normalize.
2348 * To emulate certain texture wrap modes, this can be used
2349 * to saturate the specified tex coord to [0.0, 1.0]. The
2350 * bits are according to sampler #, ie. if, for example:
2352 * (conf->saturate_s & (1 << n))
2354 * is true, then the s coord for sampler n is saturated.
2356 * Note that clamping must happen *after* projector lowering
2357 * so any projected texture sample instruction with a clamped
2358 * coordinate gets automatically lowered, regardless of the
2359 * 'lower_txp' setting.
2361 unsigned saturate_s
;
2362 unsigned saturate_t
;
2363 unsigned saturate_r
;
2365 /* Bitmask of textures that need swizzling.
2367 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2368 * swizzles[texture_index] is applied to the result of the texturing
2371 unsigned swizzle_result
;
2373 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2374 * while 4 and 5 represent 0 and 1 respectively.
2376 uint8_t swizzles
[32][4];
2379 * Bitmap of textures that need srgb to linear conversion. If
2380 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2381 * of the texture are lowered to linear.
2383 unsigned lower_srgb
;
2384 } nir_lower_tex_options
;
2386 bool nir_lower_tex(nir_shader
*shader
,
2387 const nir_lower_tex_options
*options
);
2389 bool nir_lower_idiv(nir_shader
*shader
);
2391 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2392 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2394 void nir_lower_two_sided_color(nir_shader
*shader
);
2396 void nir_lower_clamp_color_outputs(nir_shader
*shader
);
2398 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2400 typedef struct nir_lower_wpos_ytransform_options
{
2401 int state_tokens
[5];
2402 bool fs_coord_origin_upper_left
:1;
2403 bool fs_coord_origin_lower_left
:1;
2404 bool fs_coord_pixel_center_integer
:1;
2405 bool fs_coord_pixel_center_half_integer
:1;
2406 } nir_lower_wpos_ytransform_options
;
2408 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2409 const nir_lower_wpos_ytransform_options
*options
);
2411 typedef struct nir_lower_drawpixels_options
{
2412 int texcoord_state_tokens
[5];
2413 int scale_state_tokens
[5];
2414 int bias_state_tokens
[5];
2415 unsigned drawpix_sampler
;
2416 unsigned pixelmap_sampler
;
2418 bool scale_and_bias
:1;
2419 } nir_lower_drawpixels_options
;
2421 void nir_lower_drawpixels(nir_shader
*shader
,
2422 const nir_lower_drawpixels_options
*options
);
2424 typedef struct nir_lower_bitmap_options
{
2427 } nir_lower_bitmap_options
;
2429 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2431 void nir_lower_atomics(nir_shader
*shader
,
2432 const struct gl_shader_program
*shader_program
);
2433 void nir_lower_to_source_mods(nir_shader
*shader
);
2435 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2438 nir_lower_drcp
= (1 << 0),
2439 nir_lower_dsqrt
= (1 << 1),
2440 nir_lower_drsq
= (1 << 2),
2441 nir_lower_dtrunc
= (1 << 3),
2442 nir_lower_dfloor
= (1 << 4),
2443 nir_lower_dceil
= (1 << 5),
2444 nir_lower_dfract
= (1 << 6),
2445 nir_lower_dround_even
= (1 << 7),
2446 nir_lower_dmod
= (1 << 8)
2447 } nir_lower_doubles_options
;
2449 void nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2450 void nir_lower_double_pack(nir_shader
*shader
);
2452 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2454 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2455 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2457 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2458 void nir_convert_to_ssa(nir_shader
*shader
);
2460 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2461 bool nir_repair_ssa(nir_shader
*shader
);
2463 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2464 * registers. If false, convert all values (even those not involved in a phi
2465 * node) to registers.
2467 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2469 bool nir_opt_algebraic(nir_shader
*shader
);
2470 bool nir_opt_algebraic_late(nir_shader
*shader
);
2471 bool nir_opt_constant_folding(nir_shader
*shader
);
2473 bool nir_opt_global_to_local(nir_shader
*shader
);
2475 bool nir_copy_prop(nir_shader
*shader
);
2477 bool nir_opt_cse(nir_shader
*shader
);
2479 bool nir_opt_dce(nir_shader
*shader
);
2481 bool nir_opt_dead_cf(nir_shader
*shader
);
2483 void nir_opt_gcm(nir_shader
*shader
);
2485 bool nir_opt_peephole_select(nir_shader
*shader
);
2487 bool nir_opt_remove_phis(nir_shader
*shader
);
2489 bool nir_opt_undef(nir_shader
*shader
);
2491 void nir_sweep(nir_shader
*shader
);
2493 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2494 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);