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"
40 #include "compiler/shader_info.h"
43 #include "nir_opcodes.h"
50 struct gl_shader_program
;
53 #define NIR_TRUE (~0u)
55 /** Defines a cast function
57 * This macro defines a cast function from in_type to out_type where
58 * out_type is some structure type that contains a field of type out_type.
60 * Note that you have to be a bit careful as the generated cast function
63 #define NIR_DEFINE_CAST(name, in_type, out_type, field, \
64 type_field, type_value) \
65 static inline out_type * \
66 name(const in_type *parent) \
68 assert(parent && parent->type_field == type_value); \
69 return exec_node_data(out_type, parent, field); \
78 * Description of built-in state associated with a uniform
80 * \sa nir_variable::state_slots
88 nir_var_shader_in
= (1 << 0),
89 nir_var_shader_out
= (1 << 1),
90 nir_var_global
= (1 << 2),
91 nir_var_local
= (1 << 3),
92 nir_var_uniform
= (1 << 4),
93 nir_var_shader_storage
= (1 << 5),
94 nir_var_system_value
= (1 << 6),
95 nir_var_param
= (1 << 7),
96 nir_var_shared
= (1 << 8),
101 * Data stored in an nir_constant
103 union nir_constant_data
{
111 typedef struct nir_constant
{
113 * Value of the constant.
115 * The field used to back the values supplied by the constant is determined
116 * by the type associated with the \c nir_variable. Constants may be
117 * scalars, vectors, or matrices.
119 union nir_constant_data value
;
121 /* we could get this from the var->type but makes clone *much* easier to
122 * not have to care about the type.
124 unsigned num_elements
;
126 /* Array elements / Structure Fields */
127 struct nir_constant
**elements
;
131 * \brief Layout qualifiers for gl_FragDepth.
133 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
134 * with a layout qualifier.
137 nir_depth_layout_none
, /**< No depth layout is specified. */
138 nir_depth_layout_any
,
139 nir_depth_layout_greater
,
140 nir_depth_layout_less
,
141 nir_depth_layout_unchanged
145 * Either a uniform, global variable, shader input, or shader output. Based on
146 * ir_variable - it should be easy to translate between the two.
149 typedef struct nir_variable
{
150 struct exec_node node
;
153 * Declared type of the variable
155 const struct glsl_type
*type
;
158 * Declared name of the variable
162 struct nir_variable_data
{
164 * Storage class of the variable.
166 * \sa nir_variable_mode
168 nir_variable_mode mode
;
171 * Is the variable read-only?
173 * This is set for variables declared as \c const, shader inputs,
176 unsigned read_only
:1;
180 unsigned invariant
:1;
183 * Interpolation mode for shader inputs / outputs
185 * \sa glsl_interp_mode
187 unsigned interpolation
:2;
190 * \name ARB_fragment_coord_conventions
193 unsigned origin_upper_left
:1;
194 unsigned pixel_center_integer
:1;
198 * Was the location explicitly set in the shader?
200 * If the location is explicitly set in the shader, it \b cannot be changed
201 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
204 unsigned explicit_location
:1;
205 unsigned explicit_index
:1;
208 * Was an initial binding explicitly set in the shader?
210 * If so, constant_initializer contains an integer nir_constant
211 * representing the initial binding point.
213 unsigned explicit_binding
:1;
216 * Does this variable have an initializer?
218 * This is used by the linker to cross-validiate initializers of global
221 unsigned has_initializer
:1;
224 * If non-zero, then this variable may be packed along with other variables
225 * into a single varying slot, so this offset should be applied when
226 * accessing components. For example, an offset of 1 means that the x
227 * component of this variable is actually stored in component y of the
228 * location specified by \c location.
230 unsigned location_frac
:2;
233 * Whether this is a fragment shader output implicitly initialized with
234 * the previous contents of the specified render target at the
235 * framebuffer location corresponding to this shader invocation.
237 unsigned fb_fetch_output
:1;
240 * \brief Layout qualifier for gl_FragDepth.
242 * This is not equal to \c ir_depth_layout_none if and only if this
243 * variable is \c gl_FragDepth and a layout qualifier is specified.
245 nir_depth_layout depth_layout
;
248 * Storage location of the base of this variable
250 * The precise meaning of this field depends on the nature of the variable.
252 * - Vertex shader input: one of the values from \c gl_vert_attrib.
253 * - Vertex shader output: one of the values from \c gl_varying_slot.
254 * - Geometry shader input: one of the values from \c gl_varying_slot.
255 * - Geometry shader output: one of the values from \c gl_varying_slot.
256 * - Fragment shader input: one of the values from \c gl_varying_slot.
257 * - Fragment shader output: one of the values from \c gl_frag_result.
258 * - Uniforms: Per-stage uniform slot number for default uniform block.
259 * - Uniforms: Index within the uniform block definition for UBO members.
260 * - Non-UBO Uniforms: uniform slot number.
261 * - Other: This field is not currently used.
263 * If the variable is a uniform, shader input, or shader output, and the
264 * slot has not been assigned, the value will be -1.
269 * The actual location of the variable in the IR. Only valid for inputs
272 unsigned int driver_location
;
275 * output index for dual source blending.
280 * Descriptor set binding for sampler or UBO.
285 * Initial binding point for a sampler or UBO.
287 * For array types, this represents the binding point for the first element.
292 * Location an atomic counter is stored at.
297 * ARB_shader_image_load_store qualifiers.
300 bool read_only
; /**< "readonly" qualifier. */
301 bool write_only
; /**< "writeonly" qualifier. */
306 /** Image internal format if specified explicitly, otherwise GL_NONE. */
311 * Highest element accessed with a constant expression array index
313 * Not used for non-array variables.
315 unsigned max_array_access
;
320 * Built-in state that backs this uniform
322 * Once set at variable creation, \c state_slots must remain invariant.
323 * This is because, ideally, this array would be shared by all clones of
324 * this variable in the IR tree. In other words, we'd really like for it
325 * to be a fly-weight.
327 * If the variable is not a uniform, \c num_state_slots will be zero and
328 * \c state_slots will be \c NULL.
331 unsigned num_state_slots
; /**< Number of state slots used */
332 nir_state_slot
*state_slots
; /**< State descriptors. */
336 * Constant expression assigned in the initializer of the variable
338 nir_constant
*constant_initializer
;
341 * For variables that are in an interface block or are an instance of an
342 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
344 * \sa ir_variable::location
346 const struct glsl_type
*interface_type
;
349 #define nir_foreach_variable(var, var_list) \
350 foreach_list_typed(nir_variable, var, node, var_list)
352 #define nir_foreach_variable_safe(var, var_list) \
353 foreach_list_typed_safe(nir_variable, var, node, var_list)
356 nir_variable_is_global(const nir_variable
*var
)
358 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
361 typedef struct nir_register
{
362 struct exec_node node
;
364 unsigned num_components
; /** < number of vector components */
365 unsigned num_array_elems
; /** < size of array (0 for no array) */
367 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
370 /** generic register index. */
373 /** only for debug purposes, can be NULL */
376 /** whether this register is local (per-function) or global (per-shader) */
380 * If this flag is set to true, then accessing channels >= num_components
381 * is well-defined, and simply spills over to the next array element. This
382 * is useful for backends that can do per-component accessing, in
383 * particular scalar backends. By setting this flag and making
384 * num_components equal to 1, structures can be packed tightly into
385 * registers and then registers can be accessed per-component to get to
386 * each structure member, even if it crosses vec4 boundaries.
390 /** set of nir_src's where this register is used (read from) */
391 struct list_head uses
;
393 /** set of nir_dest's where this register is defined (written to) */
394 struct list_head defs
;
396 /** set of nir_if's where this register is used as a condition */
397 struct list_head if_uses
;
404 nir_instr_type_intrinsic
,
405 nir_instr_type_load_const
,
407 nir_instr_type_ssa_undef
,
409 nir_instr_type_parallel_copy
,
412 typedef struct nir_instr
{
413 struct exec_node node
;
415 struct nir_block
*block
;
417 /** generic instruction index. */
420 /* A temporary for optimization and analysis passes to use for storing
421 * flags. For instance, DCE uses this to store the "dead/live" info.
426 static inline nir_instr
*
427 nir_instr_next(nir_instr
*instr
)
429 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
430 if (exec_node_is_tail_sentinel(next
))
433 return exec_node_data(nir_instr
, next
, node
);
436 static inline nir_instr
*
437 nir_instr_prev(nir_instr
*instr
)
439 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
440 if (exec_node_is_head_sentinel(prev
))
443 return exec_node_data(nir_instr
, prev
, node
);
447 nir_instr_is_first(nir_instr
*instr
)
449 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
453 nir_instr_is_last(nir_instr
*instr
)
455 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
458 typedef struct nir_ssa_def
{
459 /** for debugging only, can be NULL */
462 /** generic SSA definition index. */
465 /** Index into the live_in and live_out bitfields */
468 nir_instr
*parent_instr
;
470 /** set of nir_instr's where this register is used (read from) */
471 struct list_head uses
;
473 /** set of nir_if's where this register is used as a condition */
474 struct list_head if_uses
;
476 uint8_t num_components
;
478 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
486 struct nir_src
*indirect
; /** < NULL for no indirect offset */
487 unsigned base_offset
;
489 /* TODO use-def chain goes here */
493 nir_instr
*parent_instr
;
494 struct list_head def_link
;
497 struct nir_src
*indirect
; /** < NULL for no indirect offset */
498 unsigned base_offset
;
500 /* TODO def-use chain goes here */
505 typedef struct nir_src
{
507 nir_instr
*parent_instr
;
508 struct nir_if
*parent_if
;
511 struct list_head use_link
;
521 static inline nir_src
524 nir_src src
= { { NULL
} };
528 #define NIR_SRC_INIT nir_src_init()
530 #define nir_foreach_use(src, reg_or_ssa_def) \
531 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
533 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
534 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
536 #define nir_foreach_if_use(src, reg_or_ssa_def) \
537 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
539 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
540 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
551 static inline nir_dest
554 nir_dest dest
= { { { NULL
} } };
558 #define NIR_DEST_INIT nir_dest_init()
560 #define nir_foreach_def(dest, reg) \
561 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
563 #define nir_foreach_def_safe(dest, reg) \
564 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
566 static inline nir_src
567 nir_src_for_ssa(nir_ssa_def
*def
)
569 nir_src src
= NIR_SRC_INIT
;
577 static inline nir_src
578 nir_src_for_reg(nir_register
*reg
)
580 nir_src src
= NIR_SRC_INIT
;
584 src
.reg
.indirect
= NULL
;
585 src
.reg
.base_offset
= 0;
590 static inline nir_dest
591 nir_dest_for_reg(nir_register
*reg
)
593 nir_dest dest
= NIR_DEST_INIT
;
600 static inline unsigned
601 nir_src_bit_size(nir_src src
)
603 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
606 static inline unsigned
607 nir_dest_bit_size(nir_dest dest
)
609 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
612 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
613 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
619 * \name input modifiers
623 * For inputs interpreted as floating point, flips the sign bit. For
624 * inputs interpreted as integers, performs the two's complement negation.
629 * Clears the sign bit for floating point values, and computes the integer
630 * absolute value for integers. Note that the negate modifier acts after
631 * the absolute value modifier, therefore if both are set then all inputs
632 * will become negative.
638 * For each input component, says which component of the register it is
639 * chosen from. Note that which elements of the swizzle are used and which
640 * are ignored are based on the write mask for most opcodes - for example,
641 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
642 * a swizzle of {2, x, 1, 0} where x means "don't care."
651 * \name saturate output modifier
653 * Only valid for opcodes that output floating-point numbers. Clamps the
654 * output to between 0.0 and 1.0 inclusive.
659 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
663 nir_type_invalid
= 0, /* Not a valid type */
668 nir_type_bool32
= 32 | nir_type_bool
,
669 nir_type_int8
= 8 | nir_type_int
,
670 nir_type_int16
= 16 | nir_type_int
,
671 nir_type_int32
= 32 | nir_type_int
,
672 nir_type_int64
= 64 | nir_type_int
,
673 nir_type_uint8
= 8 | nir_type_uint
,
674 nir_type_uint16
= 16 | nir_type_uint
,
675 nir_type_uint32
= 32 | nir_type_uint
,
676 nir_type_uint64
= 64 | nir_type_uint
,
677 nir_type_float16
= 16 | nir_type_float
,
678 nir_type_float32
= 32 | nir_type_float
,
679 nir_type_float64
= 64 | nir_type_float
,
682 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
683 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
685 static inline unsigned
686 nir_alu_type_get_type_size(nir_alu_type type
)
688 return type
& NIR_ALU_TYPE_SIZE_MASK
;
691 static inline unsigned
692 nir_alu_type_get_base_type(nir_alu_type type
)
694 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
698 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
699 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
700 } nir_op_algebraic_property
;
708 * The number of components in the output
710 * If non-zero, this is the size of the output and input sizes are
711 * explicitly given; swizzle and writemask are still in effect, but if
712 * the output component is masked out, then the input component may
715 * If zero, the opcode acts in the standard, per-component manner; the
716 * operation is performed on each component (except the ones that are
717 * masked out) with the input being taken from the input swizzle for
720 * The size of some of the inputs may be given (i.e. non-zero) even
721 * though output_size is zero; in that case, the inputs with a zero
722 * size act per-component, while the inputs with non-zero size don't.
724 unsigned output_size
;
727 * The type of vector that the instruction outputs. Note that the
728 * staurate modifier is only allowed on outputs with the float type.
731 nir_alu_type output_type
;
734 * The number of components in each input
736 unsigned input_sizes
[4];
739 * The type of vector that each input takes. Note that negate and
740 * absolute value are only allowed on inputs with int or float type and
741 * behave differently on the two.
743 nir_alu_type input_types
[4];
745 nir_op_algebraic_property algebraic_properties
;
748 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
750 typedef struct nir_alu_instr
{
754 /** Indicates that this ALU instruction generates an exact value
756 * This is kind of a mixture of GLSL "precise" and "invariant" and not
757 * really equivalent to either. This indicates that the value generated by
758 * this operation is high-precision and any code transformations that touch
759 * it must ensure that the resulting value is bit-for-bit identical to the
768 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
769 nir_alu_instr
*instr
);
770 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
771 nir_alu_instr
*instr
);
773 /* is this source channel used? */
775 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
777 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
778 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
780 return (instr
->dest
.write_mask
>> channel
) & 1;
784 * For instructions whose destinations are SSA, get the number of channels
787 static inline unsigned
788 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
790 assert(instr
->dest
.dest
.is_ssa
);
792 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
793 return nir_op_infos
[instr
->op
].input_sizes
[src
];
795 return instr
->dest
.dest
.ssa
.num_components
;
798 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
799 unsigned src1
, unsigned src2
);
803 nir_deref_type_array
,
804 nir_deref_type_struct
807 typedef struct nir_deref
{
808 nir_deref_type deref_type
;
809 struct nir_deref
*child
;
810 const struct glsl_type
*type
;
819 /* This enum describes how the array is referenced. If the deref is
820 * direct then the base_offset is used. If the deref is indirect then
821 * offset is given by base_offset + indirect. If the deref is a wildcard
822 * then the deref refers to all of the elements of the array at the same
823 * time. Wildcard dereferences are only ever allowed in copy_var
824 * intrinsics and the source and destination derefs must have matching
828 nir_deref_array_type_direct
,
829 nir_deref_array_type_indirect
,
830 nir_deref_array_type_wildcard
,
831 } nir_deref_array_type
;
836 nir_deref_array_type deref_array_type
;
837 unsigned base_offset
;
847 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
,
848 deref_type
, nir_deref_type_var
)
849 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
,
850 deref_type
, nir_deref_type_array
)
851 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
,
852 deref_type
, nir_deref_type_struct
)
854 /* Returns the last deref in the chain. */
855 static inline nir_deref
*
856 nir_deref_tail(nir_deref
*deref
)
859 deref
= deref
->child
;
867 nir_deref_var
**params
;
868 nir_deref_var
*return_deref
;
870 struct nir_function
*callee
;
873 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
874 num_variables, num_indices, idx0, idx1, idx2, flags) \
875 nir_intrinsic_##name,
877 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
880 #include "nir_intrinsics.h"
881 nir_num_intrinsics
= nir_last_intrinsic
+ 1
884 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
886 /** Represents an intrinsic
888 * An intrinsic is an instruction type for handling things that are
889 * more-or-less regular operations but don't just consume and produce SSA
890 * values like ALU operations do. Intrinsics are not for things that have
891 * special semantic meaning such as phi nodes and parallel copies.
892 * Examples of intrinsics include variable load/store operations, system
893 * value loads, and the like. Even though texturing more-or-less falls
894 * under this category, texturing is its own instruction type because
895 * trying to represent texturing with intrinsics would lead to a
896 * combinatorial explosion of intrinsic opcodes.
898 * By having a single instruction type for handling a lot of different
899 * cases, optimization passes can look for intrinsics and, for the most
900 * part, completely ignore them. Each intrinsic type also has a few
901 * possible flags that govern whether or not they can be reordered or
902 * eliminated. That way passes like dead code elimination can still work
903 * on intrisics without understanding the meaning of each.
905 * Each intrinsic has some number of constant indices, some number of
906 * variables, and some number of sources. What these sources, variables,
907 * and indices mean depends on the intrinsic and is documented with the
908 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
909 * instructions are the only types of instruction that can operate on
915 nir_intrinsic_op intrinsic
;
919 /** number of components if this is a vectorized intrinsic
921 * Similarly to ALU operations, some intrinsics are vectorized.
922 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
923 * For vectorized intrinsics, the num_components field specifies the
924 * number of destination components and the number of source components
925 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
927 uint8_t num_components
;
929 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
931 nir_deref_var
*variables
[2];
934 } nir_intrinsic_instr
;
937 * \name NIR intrinsics semantic flags
939 * information about what the compiler can do with the intrinsics.
941 * \sa nir_intrinsic_info::flags
945 * whether the intrinsic can be safely eliminated if none of its output
946 * value is not being used.
948 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
951 * Whether the intrinsic can be reordered with respect to any other
952 * intrinsic, i.e. whether the only reordering dependencies of the
953 * intrinsic are due to the register reads/writes.
955 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
956 } nir_intrinsic_semantic_flag
;
959 * \name NIR intrinsics const-index flag
961 * Indicates the usage of a const_index slot.
963 * \sa nir_intrinsic_info::index_map
967 * Generally instructions that take a offset src argument, can encode
968 * a constant 'base' value which is added to the offset.
970 NIR_INTRINSIC_BASE
= 1,
973 * For store instructions, a writemask for the store.
975 NIR_INTRINSIC_WRMASK
= 2,
978 * The stream-id for GS emit_vertex/end_primitive intrinsics.
980 NIR_INTRINSIC_STREAM_ID
= 3,
983 * The clip-plane id for load_user_clip_plane intrinsic.
985 NIR_INTRINSIC_UCP_ID
= 4,
988 * The amount of data, starting from BASE, that this instruction may
989 * access. This is used to provide bounds if the offset is not constant.
991 NIR_INTRINSIC_RANGE
= 5,
994 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
996 NIR_INTRINSIC_DESC_SET
= 6,
999 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1001 NIR_INTRINSIC_BINDING
= 7,
1006 NIR_INTRINSIC_COMPONENT
= 8,
1009 * Interpolation mode (only meaningful for FS inputs).
1011 NIR_INTRINSIC_INTERP_MODE
= 9,
1013 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1015 } nir_intrinsic_index_flag
;
1017 #define NIR_INTRINSIC_MAX_INPUTS 4
1022 unsigned num_srcs
; /** < number of register/SSA inputs */
1024 /** number of components of each input register
1026 * If this value is 0, the number of components is given by the
1027 * num_components field of nir_intrinsic_instr.
1029 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1033 /** number of components of the output register
1035 * If this value is 0, the number of components is given by the
1036 * num_components field of nir_intrinsic_instr.
1038 unsigned dest_components
;
1040 /** the number of inputs/outputs that are variables */
1041 unsigned num_variables
;
1043 /** the number of constant indices used by the intrinsic */
1044 unsigned num_indices
;
1046 /** indicates the usage of intr->const_index[n] */
1047 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1049 /** semantic flags for calls to this intrinsic */
1050 nir_intrinsic_semantic_flag flags
;
1051 } nir_intrinsic_info
;
1053 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1056 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1057 static inline type \
1058 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
1060 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1061 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1062 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1064 static inline void \
1065 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1067 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1068 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1069 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1072 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1073 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1074 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1075 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1076 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1077 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1078 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1079 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1080 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1083 * \group texture information
1085 * This gives semantic information about textures which is useful to the
1086 * frontend, the backend, and lowering passes, but not the optimizer.
1091 nir_tex_src_projector
,
1092 nir_tex_src_comparitor
, /* shadow comparitor */
1096 nir_tex_src_ms_index
, /* MSAA sample index */
1097 nir_tex_src_ms_mcs
, /* MSAA compression value */
1100 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1101 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1102 nir_tex_src_plane
, /* < selects plane for planar textures */
1103 nir_num_tex_src_types
1108 nir_tex_src_type src_type
;
1112 nir_texop_tex
, /**< Regular texture look-up */
1113 nir_texop_txb
, /**< Texture look-up with LOD bias */
1114 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1115 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1116 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1117 nir_texop_txf_ms
, /**< Multisample texture fetch */
1118 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1119 nir_texop_txs
, /**< Texture size */
1120 nir_texop_lod
, /**< Texture lod query */
1121 nir_texop_tg4
, /**< Texture gather */
1122 nir_texop_query_levels
, /**< Texture levels query */
1123 nir_texop_texture_samples
, /**< Texture samples query */
1124 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1132 enum glsl_sampler_dim sampler_dim
;
1133 nir_alu_type dest_type
;
1138 unsigned num_srcs
, coord_components
;
1139 bool is_array
, is_shadow
;
1142 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1143 * components or the new-style shadow that outputs 1 component.
1145 bool is_new_style_shadow
;
1147 /* gather component selector */
1148 unsigned component
: 2;
1150 /** The texture index
1152 * If this texture instruction has a nir_tex_src_texture_offset source,
1153 * then the texture index is given by texture_index + texture_offset.
1155 unsigned texture_index
;
1157 /** The size of the texture array or 0 if it's not an array */
1158 unsigned texture_array_size
;
1160 /** The texture deref
1162 * If this is null, use texture_index instead.
1164 nir_deref_var
*texture
;
1166 /** The sampler index
1168 * The following operations do not require a sampler and, as such, this
1169 * field should be ignored:
1171 * - nir_texop_txf_ms
1175 * - nir_texop_query_levels
1176 * - nir_texop_texture_samples
1177 * - nir_texop_samples_identical
1179 * If this texture instruction has a nir_tex_src_sampler_offset source,
1180 * then the sampler index is given by sampler_index + sampler_offset.
1182 unsigned sampler_index
;
1184 /** The sampler deref
1186 * If this is null, use sampler_index instead.
1188 nir_deref_var
*sampler
;
1191 static inline unsigned
1192 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1194 switch (instr
->op
) {
1195 case nir_texop_txs
: {
1197 switch (instr
->sampler_dim
) {
1198 case GLSL_SAMPLER_DIM_1D
:
1199 case GLSL_SAMPLER_DIM_BUF
:
1202 case GLSL_SAMPLER_DIM_2D
:
1203 case GLSL_SAMPLER_DIM_CUBE
:
1204 case GLSL_SAMPLER_DIM_MS
:
1205 case GLSL_SAMPLER_DIM_RECT
:
1206 case GLSL_SAMPLER_DIM_EXTERNAL
:
1207 case GLSL_SAMPLER_DIM_SUBPASS
:
1210 case GLSL_SAMPLER_DIM_3D
:
1214 unreachable("not reached");
1216 if (instr
->is_array
)
1224 case nir_texop_texture_samples
:
1225 case nir_texop_query_levels
:
1226 case nir_texop_samples_identical
:
1230 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1237 /* Returns true if this texture operation queries something about the texture
1238 * rather than actually sampling it.
1241 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1243 switch (instr
->op
) {
1246 case nir_texop_texture_samples
:
1247 case nir_texop_query_levels
:
1248 case nir_texop_txf_ms_mcs
:
1255 case nir_texop_txf_ms
:
1259 unreachable("Invalid texture opcode");
1263 static inline nir_alu_type
1264 nir_tex_instr_src_type(nir_tex_instr
*instr
, unsigned src
)
1266 switch (instr
->src
[src
].src_type
) {
1267 case nir_tex_src_coord
:
1268 switch (instr
->op
) {
1270 case nir_texop_txf_ms
:
1271 case nir_texop_txf_ms_mcs
:
1272 case nir_texop_samples_identical
:
1273 return nir_type_int
;
1276 return nir_type_float
;
1279 case nir_tex_src_lod
:
1280 switch (instr
->op
) {
1283 return nir_type_int
;
1286 return nir_type_float
;
1289 case nir_tex_src_projector
:
1290 case nir_tex_src_comparitor
:
1291 case nir_tex_src_bias
:
1292 case nir_tex_src_ddx
:
1293 case nir_tex_src_ddy
:
1294 return nir_type_float
;
1296 case nir_tex_src_offset
:
1297 case nir_tex_src_ms_index
:
1298 case nir_tex_src_texture_offset
:
1299 case nir_tex_src_sampler_offset
:
1300 return nir_type_int
;
1303 unreachable("Invalid texture source type");
1307 static inline unsigned
1308 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1310 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1311 return instr
->coord_components
;
1313 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1314 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1317 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1318 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1319 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1320 if (instr
->is_array
)
1321 return instr
->coord_components
- 1;
1323 return instr
->coord_components
;
1330 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1332 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1333 if (instr
->src
[i
].src_type
== type
)
1339 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1353 nir_const_value value
;
1356 } nir_load_const_instr
;
1369 /* creates a new SSA variable in an undefined state */
1374 } nir_ssa_undef_instr
;
1377 struct exec_node node
;
1379 /* The predecessor block corresponding to this source */
1380 struct nir_block
*pred
;
1385 #define nir_foreach_phi_src(phi_src, phi) \
1386 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1387 #define nir_foreach_phi_src_safe(phi_src, phi) \
1388 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1393 struct exec_list srcs
; /** < list of nir_phi_src */
1399 struct exec_node node
;
1402 } nir_parallel_copy_entry
;
1404 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1405 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1410 /* A list of nir_parallel_copy_entry's. The sources of all of the
1411 * entries are copied to the corresponding destinations "in parallel".
1412 * In other words, if we have two entries: a -> b and b -> a, the values
1415 struct exec_list entries
;
1416 } nir_parallel_copy_instr
;
1418 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1419 type
, nir_instr_type_alu
)
1420 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1421 type
, nir_instr_type_call
)
1422 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1423 type
, nir_instr_type_jump
)
1424 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1425 type
, nir_instr_type_tex
)
1426 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1427 type
, nir_instr_type_intrinsic
)
1428 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1429 type
, nir_instr_type_load_const
)
1430 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1431 type
, nir_instr_type_ssa_undef
)
1432 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1433 type
, nir_instr_type_phi
)
1434 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1435 nir_parallel_copy_instr
, instr
,
1436 type
, nir_instr_type_parallel_copy
)
1441 * Control flow consists of a tree of control flow nodes, which include
1442 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1443 * instructions that always run start-to-finish. Each basic block also keeps
1444 * track of its successors (blocks which may run immediately after the current
1445 * block) and predecessors (blocks which could have run immediately before the
1446 * current block). Each function also has a start block and an end block which
1447 * all return statements point to (which is always empty). Together, all the
1448 * blocks with their predecessors and successors make up the control flow
1449 * graph (CFG) of the function. There are helpers that modify the tree of
1450 * control flow nodes while modifying the CFG appropriately; these should be
1451 * used instead of modifying the tree directly.
1458 nir_cf_node_function
1461 typedef struct nir_cf_node
{
1462 struct exec_node node
;
1463 nir_cf_node_type type
;
1464 struct nir_cf_node
*parent
;
1467 typedef struct nir_block
{
1468 nir_cf_node cf_node
;
1470 struct exec_list instr_list
; /** < list of nir_instr */
1472 /** generic block index; generated by nir_index_blocks */
1476 * Each block can only have up to 2 successors, so we put them in a simple
1477 * array - no need for anything more complicated.
1479 struct nir_block
*successors
[2];
1481 /* Set of nir_block predecessors in the CFG */
1482 struct set
*predecessors
;
1485 * this node's immediate dominator in the dominance tree - set to NULL for
1488 struct nir_block
*imm_dom
;
1490 /* This node's children in the dominance tree */
1491 unsigned num_dom_children
;
1492 struct nir_block
**dom_children
;
1494 /* Set of nir_block's on the dominance frontier of this block */
1495 struct set
*dom_frontier
;
1498 * These two indices have the property that dom_{pre,post}_index for each
1499 * child of this block in the dominance tree will always be between
1500 * dom_pre_index and dom_post_index for this block, which makes testing if
1501 * a given block is dominated by another block an O(1) operation.
1503 unsigned dom_pre_index
, dom_post_index
;
1505 /* live in and out for this block; used for liveness analysis */
1506 BITSET_WORD
*live_in
;
1507 BITSET_WORD
*live_out
;
1510 static inline nir_instr
*
1511 nir_block_first_instr(nir_block
*block
)
1513 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1514 return exec_node_data(nir_instr
, head
, node
);
1517 static inline nir_instr
*
1518 nir_block_last_instr(nir_block
*block
)
1520 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1521 return exec_node_data(nir_instr
, tail
, node
);
1524 #define nir_foreach_instr(instr, block) \
1525 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1526 #define nir_foreach_instr_reverse(instr, block) \
1527 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1528 #define nir_foreach_instr_safe(instr, block) \
1529 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1530 #define nir_foreach_instr_reverse_safe(instr, block) \
1531 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1533 typedef struct nir_if
{
1534 nir_cf_node cf_node
;
1537 struct exec_list then_list
; /** < list of nir_cf_node */
1538 struct exec_list else_list
; /** < list of nir_cf_node */
1542 nir_cf_node cf_node
;
1544 struct exec_list body
; /** < list of nir_cf_node */
1548 * Various bits of metadata that can may be created or required by
1549 * optimization and analysis passes
1552 nir_metadata_none
= 0x0,
1553 nir_metadata_block_index
= 0x1,
1554 nir_metadata_dominance
= 0x2,
1555 nir_metadata_live_ssa_defs
= 0x4,
1556 nir_metadata_not_properly_reset
= 0x8,
1560 nir_cf_node cf_node
;
1562 /** pointer to the function of which this is an implementation */
1563 struct nir_function
*function
;
1565 struct exec_list body
; /** < list of nir_cf_node */
1567 nir_block
*end_block
;
1569 /** list for all local variables in the function */
1570 struct exec_list locals
;
1572 /** array of variables used as parameters */
1573 unsigned num_params
;
1574 nir_variable
**params
;
1576 /** variable used to hold the result of the function */
1577 nir_variable
*return_var
;
1579 /** list of local registers in the function */
1580 struct exec_list registers
;
1582 /** next available local register index */
1585 /** next available SSA value index */
1588 /* total number of basic blocks, only valid when block_index_dirty = false */
1589 unsigned num_blocks
;
1591 nir_metadata valid_metadata
;
1592 } nir_function_impl
;
1594 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1595 nir_start_block(nir_function_impl
*impl
)
1597 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1600 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1601 nir_impl_last_block(nir_function_impl
*impl
)
1603 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1606 static inline nir_cf_node
*
1607 nir_cf_node_next(nir_cf_node
*node
)
1609 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1610 if (exec_node_is_tail_sentinel(next
))
1613 return exec_node_data(nir_cf_node
, next
, node
);
1616 static inline nir_cf_node
*
1617 nir_cf_node_prev(nir_cf_node
*node
)
1619 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1620 if (exec_node_is_head_sentinel(prev
))
1623 return exec_node_data(nir_cf_node
, prev
, node
);
1627 nir_cf_node_is_first(const nir_cf_node
*node
)
1629 return exec_node_is_head_sentinel(node
->node
.prev
);
1633 nir_cf_node_is_last(const nir_cf_node
*node
)
1635 return exec_node_is_tail_sentinel(node
->node
.next
);
1638 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1639 type
, nir_cf_node_block
)
1640 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1641 type
, nir_cf_node_if
)
1642 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1643 type
, nir_cf_node_loop
)
1644 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1645 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1647 static inline nir_block
*
1648 nir_if_first_then_block(nir_if
*if_stmt
)
1650 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1651 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1654 static inline nir_block
*
1655 nir_if_last_then_block(nir_if
*if_stmt
)
1657 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1658 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1661 static inline nir_block
*
1662 nir_if_first_else_block(nir_if
*if_stmt
)
1664 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1665 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1668 static inline nir_block
*
1669 nir_if_last_else_block(nir_if
*if_stmt
)
1671 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1672 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1675 static inline nir_block
*
1676 nir_loop_first_block(nir_loop
*loop
)
1678 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1679 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1682 static inline nir_block
*
1683 nir_loop_last_block(nir_loop
*loop
)
1685 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1686 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1692 nir_parameter_inout
,
1693 } nir_parameter_type
;
1696 nir_parameter_type param_type
;
1697 const struct glsl_type
*type
;
1700 typedef struct nir_function
{
1701 struct exec_node node
;
1704 struct nir_shader
*shader
;
1706 unsigned num_params
;
1707 nir_parameter
*params
;
1708 const struct glsl_type
*return_type
;
1710 /** The implementation of this function.
1712 * If the function is only declared and not implemented, this is NULL.
1714 nir_function_impl
*impl
;
1717 typedef struct nir_shader_compiler_options
{
1722 /** Lowers flrp when it does not support doubles */
1729 bool lower_bitfield_extract
;
1730 bool lower_bitfield_insert
;
1731 bool lower_uadd_carry
;
1732 bool lower_usub_borrow
;
1733 /** lowers fneg and ineg to fsub and isub. */
1735 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1738 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1741 /** enables rules to lower idiv by power-of-two: */
1744 /* Does the native fdot instruction replicate its result for four
1745 * components? If so, then opt_algebraic_late will turn all fdotN
1746 * instructions into fdot_replicatedN instructions.
1748 bool fdot_replicates
;
1750 /** lowers ffract to fsub+ffloor: */
1753 bool lower_pack_half_2x16
;
1754 bool lower_pack_unorm_2x16
;
1755 bool lower_pack_snorm_2x16
;
1756 bool lower_pack_unorm_4x8
;
1757 bool lower_pack_snorm_4x8
;
1758 bool lower_unpack_half_2x16
;
1759 bool lower_unpack_unorm_2x16
;
1760 bool lower_unpack_snorm_2x16
;
1761 bool lower_unpack_unorm_4x8
;
1762 bool lower_unpack_snorm_4x8
;
1764 bool lower_extract_byte
;
1765 bool lower_extract_word
;
1768 * Does the driver support real 32-bit integers? (Otherwise, integers
1769 * are simulated by floats.)
1771 bool native_integers
;
1773 /* Indicates that the driver only has zero-based vertex id */
1774 bool vertex_id_zero_based
;
1776 bool lower_cs_local_index_from_id
;
1779 * Should nir_lower_io() create load_interpolated_input intrinsics?
1781 * If not, it generates regular load_input intrinsics and interpolation
1782 * information must be inferred from the list of input nir_variables.
1784 bool use_interpolated_input_intrinsics
;
1785 } nir_shader_compiler_options
;
1787 typedef struct nir_shader
{
1788 /** list of uniforms (nir_variable) */
1789 struct exec_list uniforms
;
1791 /** list of inputs (nir_variable) */
1792 struct exec_list inputs
;
1794 /** list of outputs (nir_variable) */
1795 struct exec_list outputs
;
1797 /** list of shared compute variables (nir_variable) */
1798 struct exec_list shared
;
1800 /** Set of driver-specific options for the shader.
1802 * The memory for the options is expected to be kept in a single static
1803 * copy by the driver.
1805 const struct nir_shader_compiler_options
*options
;
1807 /** Various bits of compile-time information about a given shader */
1808 struct shader_info
*info
;
1810 /** list of global variables in the shader (nir_variable) */
1811 struct exec_list globals
;
1813 /** list of system value variables in the shader (nir_variable) */
1814 struct exec_list system_values
;
1816 struct exec_list functions
; /** < list of nir_function */
1818 /** list of global register in the shader */
1819 struct exec_list registers
;
1821 /** next available global register index */
1825 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1828 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1830 /** The shader stage, such as MESA_SHADER_VERTEX. */
1831 gl_shader_stage stage
;
1834 static inline nir_function_impl
*
1835 nir_shader_get_entrypoint(nir_shader
*shader
)
1837 assert(exec_list_length(&shader
->functions
) == 1);
1838 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1839 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1840 assert(func
->return_type
== glsl_void_type());
1841 assert(func
->num_params
== 0);
1846 #define nir_foreach_function(func, shader) \
1847 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1849 nir_shader
*nir_shader_create(void *mem_ctx
,
1850 gl_shader_stage stage
,
1851 const nir_shader_compiler_options
*options
,
1854 /** creates a register, including assigning it an index and adding it to the list */
1855 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1857 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1859 void nir_reg_remove(nir_register
*reg
);
1861 /** Adds a variable to the appropreate list in nir_shader */
1862 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1865 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1867 assert(var
->data
.mode
== nir_var_local
);
1868 exec_list_push_tail(&impl
->locals
, &var
->node
);
1871 /** creates a variable, sets a few defaults, and adds it to the list */
1872 nir_variable
*nir_variable_create(nir_shader
*shader
,
1873 nir_variable_mode mode
,
1874 const struct glsl_type
*type
,
1876 /** creates a local variable and adds it to the list */
1877 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1878 const struct glsl_type
*type
,
1881 /** creates a function and adds it to the shader's list of functions */
1882 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1884 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1885 /** creates a function_impl that isn't tied to any particular function */
1886 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1888 nir_block
*nir_block_create(nir_shader
*shader
);
1889 nir_if
*nir_if_create(nir_shader
*shader
);
1890 nir_loop
*nir_loop_create(nir_shader
*shader
);
1892 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1894 /** requests that the given pieces of metadata be generated */
1895 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1896 /** dirties all but the preserved metadata */
1897 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1899 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1900 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1902 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1904 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1905 unsigned num_components
,
1908 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1909 nir_intrinsic_op op
);
1911 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1912 nir_function
*callee
);
1914 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1916 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1918 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1920 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1921 unsigned num_components
,
1924 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1925 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1926 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1928 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1930 typedef bool (*nir_deref_foreach_leaf_cb
)(nir_deref_var
*deref
, void *state
);
1931 bool nir_deref_foreach_leaf(nir_deref_var
*deref
,
1932 nir_deref_foreach_leaf_cb cb
, void *state
);
1934 nir_load_const_instr
*
1935 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1938 * NIR Cursors and Instruction Insertion API
1941 * A tiny struct representing a point to insert/extract instructions or
1942 * control flow nodes. Helps reduce the combinatorial explosion of possible
1943 * points to insert/extract.
1945 * \sa nir_control_flow.h
1948 nir_cursor_before_block
,
1949 nir_cursor_after_block
,
1950 nir_cursor_before_instr
,
1951 nir_cursor_after_instr
,
1952 } nir_cursor_option
;
1955 nir_cursor_option option
;
1962 static inline nir_block
*
1963 nir_cursor_current_block(nir_cursor cursor
)
1965 if (cursor
.option
== nir_cursor_before_instr
||
1966 cursor
.option
== nir_cursor_after_instr
) {
1967 return cursor
.instr
->block
;
1969 return cursor
.block
;
1973 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
1975 static inline nir_cursor
1976 nir_before_block(nir_block
*block
)
1979 cursor
.option
= nir_cursor_before_block
;
1980 cursor
.block
= block
;
1984 static inline nir_cursor
1985 nir_after_block(nir_block
*block
)
1988 cursor
.option
= nir_cursor_after_block
;
1989 cursor
.block
= block
;
1993 static inline nir_cursor
1994 nir_before_instr(nir_instr
*instr
)
1997 cursor
.option
= nir_cursor_before_instr
;
1998 cursor
.instr
= instr
;
2002 static inline nir_cursor
2003 nir_after_instr(nir_instr
*instr
)
2006 cursor
.option
= nir_cursor_after_instr
;
2007 cursor
.instr
= instr
;
2011 static inline nir_cursor
2012 nir_after_block_before_jump(nir_block
*block
)
2014 nir_instr
*last_instr
= nir_block_last_instr(block
);
2015 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2016 return nir_before_instr(last_instr
);
2018 return nir_after_block(block
);
2022 static inline nir_cursor
2023 nir_before_cf_node(nir_cf_node
*node
)
2025 if (node
->type
== nir_cf_node_block
)
2026 return nir_before_block(nir_cf_node_as_block(node
));
2028 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2031 static inline nir_cursor
2032 nir_after_cf_node(nir_cf_node
*node
)
2034 if (node
->type
== nir_cf_node_block
)
2035 return nir_after_block(nir_cf_node_as_block(node
));
2037 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2040 static inline nir_cursor
2041 nir_after_phis(nir_block
*block
)
2043 nir_foreach_instr(instr
, block
) {
2044 if (instr
->type
!= nir_instr_type_phi
)
2045 return nir_before_instr(instr
);
2047 return nir_after_block(block
);
2050 static inline nir_cursor
2051 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2053 if (node
->type
== nir_cf_node_block
)
2054 return nir_after_block(nir_cf_node_as_block(node
));
2056 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2058 return nir_after_phis(block
);
2061 static inline nir_cursor
2062 nir_before_cf_list(struct exec_list
*cf_list
)
2064 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2065 exec_list_get_head(cf_list
), node
);
2066 return nir_before_cf_node(first_node
);
2069 static inline nir_cursor
2070 nir_after_cf_list(struct exec_list
*cf_list
)
2072 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2073 exec_list_get_tail(cf_list
), node
);
2074 return nir_after_cf_node(last_node
);
2078 * Insert a NIR instruction at the given cursor.
2080 * Note: This does not update the cursor.
2082 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2085 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2087 nir_instr_insert(nir_before_instr(instr
), before
);
2091 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2093 nir_instr_insert(nir_after_instr(instr
), after
);
2097 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2099 nir_instr_insert(nir_before_block(block
), before
);
2103 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2105 nir_instr_insert(nir_after_block(block
), after
);
2109 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2111 nir_instr_insert(nir_before_cf_node(node
), before
);
2115 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2117 nir_instr_insert(nir_after_cf_node(node
), after
);
2121 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2123 nir_instr_insert(nir_before_cf_list(list
), before
);
2127 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2129 nir_instr_insert(nir_after_cf_list(list
), after
);
2132 void nir_instr_remove(nir_instr
*instr
);
2136 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2137 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2138 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2139 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2141 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2142 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2144 nir_const_value
*nir_src_as_const_value(nir_src src
);
2145 bool nir_src_is_dynamically_uniform(nir_src src
);
2146 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2147 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2148 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2149 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2150 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2153 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2154 unsigned num_components
, unsigned bit_size
,
2156 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2157 unsigned num_components
, unsigned bit_size
,
2159 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2160 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2161 nir_instr
*after_me
);
2163 uint8_t nir_ssa_def_components_read(nir_ssa_def
*def
);
2166 * finds the next basic block in source-code order, returns NULL if there is
2170 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2172 /* Performs the opposite of nir_block_cf_tree_next() */
2174 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2176 /* Gets the first block in a CF node in source-code order */
2178 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2180 /* Gets the last block in a CF node in source-code order */
2182 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2184 /* Gets the next block after a CF node in source-code order */
2186 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2188 /* Macros for loops that visit blocks in source-code order */
2190 #define nir_foreach_block(block, impl) \
2191 for (nir_block *block = nir_start_block(impl); block != NULL; \
2192 block = nir_block_cf_tree_next(block))
2194 #define nir_foreach_block_safe(block, impl) \
2195 for (nir_block *block = nir_start_block(impl), \
2196 *next = nir_block_cf_tree_next(block); \
2198 block = next, next = nir_block_cf_tree_next(block))
2200 #define nir_foreach_block_reverse(block, impl) \
2201 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2202 block = nir_block_cf_tree_prev(block))
2204 #define nir_foreach_block_reverse_safe(block, impl) \
2205 for (nir_block *block = nir_impl_last_block(impl), \
2206 *prev = nir_block_cf_tree_prev(block); \
2208 block = prev, prev = nir_block_cf_tree_prev(block))
2210 #define nir_foreach_block_in_cf_node(block, node) \
2211 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2212 block != nir_cf_node_cf_tree_next(node); \
2213 block = nir_block_cf_tree_next(block))
2215 /* If the following CF node is an if, this function returns that if.
2216 * Otherwise, it returns NULL.
2218 nir_if
*nir_block_get_following_if(nir_block
*block
);
2220 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2222 void nir_index_local_regs(nir_function_impl
*impl
);
2223 void nir_index_global_regs(nir_shader
*shader
);
2224 void nir_index_ssa_defs(nir_function_impl
*impl
);
2225 unsigned nir_index_instrs(nir_function_impl
*impl
);
2227 void nir_index_blocks(nir_function_impl
*impl
);
2229 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2230 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2231 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2233 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2234 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2235 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2236 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2239 void nir_validate_shader(nir_shader
*shader
);
2240 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2241 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2243 #include "util/debug.h"
2245 should_clone_nir(void)
2247 static int should_clone
= -1;
2248 if (should_clone
< 0)
2249 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2251 return should_clone
;
2254 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2255 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2256 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2257 static inline bool should_clone_nir(void) { return false; }
2260 #define _PASS(nir, do_pass) do { \
2262 nir_validate_shader(nir); \
2263 if (should_clone_nir()) { \
2264 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2270 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2271 nir_metadata_set_validation_flag(nir); \
2272 if (pass(nir, ##__VA_ARGS__)) { \
2274 nir_metadata_check_validation_flag(nir); \
2278 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2279 pass(nir, ##__VA_ARGS__); \
2282 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2283 void nir_calc_dominance(nir_shader
*shader
);
2285 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2286 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2288 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2289 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2291 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2292 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2294 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2295 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2297 int nir_gs_count_vertices(const nir_shader
*shader
);
2299 bool nir_split_var_copies(nir_shader
*shader
);
2301 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2302 bool nir_lower_returns(nir_shader
*shader
);
2304 bool nir_inline_functions(nir_shader
*shader
);
2306 bool nir_propagate_invariant(nir_shader
*shader
);
2308 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2309 void nir_lower_var_copies(nir_shader
*shader
);
2311 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2313 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2315 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2317 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2318 nir_function_impl
*entrypoint
,
2319 bool outputs
, bool inputs
);
2321 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2323 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2324 unsigned base_offset
,
2325 int (*type_size
)(const struct glsl_type
*));
2328 /* If set, this forces all non-flat fragment shader inputs to be
2329 * interpolated as if with the "sample" qualifier. This requires
2330 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2332 nir_lower_io_force_sample_interpolation
= (1 << 1),
2333 } nir_lower_io_options
;
2334 void nir_lower_io(nir_shader
*shader
,
2335 nir_variable_mode modes
,
2336 int (*type_size
)(const struct glsl_type
*),
2337 nir_lower_io_options
);
2338 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2339 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2341 void nir_lower_io_types(nir_shader
*shader
);
2342 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2344 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2346 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2347 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2348 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2349 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2351 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2352 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2354 void nir_lower_samplers(nir_shader
*shader
,
2355 const struct gl_shader_program
*shader_program
);
2357 bool nir_lower_system_values(nir_shader
*shader
);
2359 typedef struct nir_lower_tex_options
{
2361 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2362 * sampler types a texture projector is lowered.
2367 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2369 bool lower_txf_offset
;
2372 * If true, lower away nir_tex_src_offset for all rect textures.
2374 bool lower_rect_offset
;
2377 * If true, lower rect textures to 2D, using txs to fetch the
2378 * texture dimensions and dividing the texture coords by the
2379 * texture dims to normalize.
2384 * If true, convert yuv to rgb.
2386 unsigned lower_y_uv_external
;
2387 unsigned lower_y_u_v_external
;
2388 unsigned lower_yx_xuxv_external
;
2391 * To emulate certain texture wrap modes, this can be used
2392 * to saturate the specified tex coord to [0.0, 1.0]. The
2393 * bits are according to sampler #, ie. if, for example:
2395 * (conf->saturate_s & (1 << n))
2397 * is true, then the s coord for sampler n is saturated.
2399 * Note that clamping must happen *after* projector lowering
2400 * so any projected texture sample instruction with a clamped
2401 * coordinate gets automatically lowered, regardless of the
2402 * 'lower_txp' setting.
2404 unsigned saturate_s
;
2405 unsigned saturate_t
;
2406 unsigned saturate_r
;
2408 /* Bitmask of textures that need swizzling.
2410 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2411 * swizzles[texture_index] is applied to the result of the texturing
2414 unsigned swizzle_result
;
2416 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2417 * while 4 and 5 represent 0 and 1 respectively.
2419 uint8_t swizzles
[32][4];
2422 * Bitmap of textures that need srgb to linear conversion. If
2423 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2424 * of the texture are lowered to linear.
2426 unsigned lower_srgb
;
2427 } nir_lower_tex_options
;
2429 bool nir_lower_tex(nir_shader
*shader
,
2430 const nir_lower_tex_options
*options
);
2432 bool nir_lower_idiv(nir_shader
*shader
);
2434 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2435 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2437 void nir_lower_two_sided_color(nir_shader
*shader
);
2439 void nir_lower_clamp_color_outputs(nir_shader
*shader
);
2441 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2443 typedef struct nir_lower_wpos_ytransform_options
{
2444 int state_tokens
[5];
2445 bool fs_coord_origin_upper_left
:1;
2446 bool fs_coord_origin_lower_left
:1;
2447 bool fs_coord_pixel_center_integer
:1;
2448 bool fs_coord_pixel_center_half_integer
:1;
2449 } nir_lower_wpos_ytransform_options
;
2451 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2452 const nir_lower_wpos_ytransform_options
*options
);
2453 bool nir_lower_wpos_center(nir_shader
*shader
);
2455 typedef struct nir_lower_drawpixels_options
{
2456 int texcoord_state_tokens
[5];
2457 int scale_state_tokens
[5];
2458 int bias_state_tokens
[5];
2459 unsigned drawpix_sampler
;
2460 unsigned pixelmap_sampler
;
2462 bool scale_and_bias
:1;
2463 } nir_lower_drawpixels_options
;
2465 void nir_lower_drawpixels(nir_shader
*shader
,
2466 const nir_lower_drawpixels_options
*options
);
2468 typedef struct nir_lower_bitmap_options
{
2471 } nir_lower_bitmap_options
;
2473 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2475 void nir_lower_atomics(nir_shader
*shader
,
2476 const struct gl_shader_program
*shader_program
);
2477 void nir_lower_to_source_mods(nir_shader
*shader
);
2479 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2482 nir_lower_drcp
= (1 << 0),
2483 nir_lower_dsqrt
= (1 << 1),
2484 nir_lower_drsq
= (1 << 2),
2485 nir_lower_dtrunc
= (1 << 3),
2486 nir_lower_dfloor
= (1 << 4),
2487 nir_lower_dceil
= (1 << 5),
2488 nir_lower_dfract
= (1 << 6),
2489 nir_lower_dround_even
= (1 << 7),
2490 nir_lower_dmod
= (1 << 8)
2491 } nir_lower_doubles_options
;
2493 void nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2494 void nir_lower_double_pack(nir_shader
*shader
);
2496 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2498 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2499 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2501 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2502 void nir_convert_to_ssa(nir_shader
*shader
);
2504 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2505 bool nir_repair_ssa(nir_shader
*shader
);
2507 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2508 * registers. If false, convert all values (even those not involved in a phi
2509 * node) to registers.
2511 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2513 bool nir_opt_algebraic(nir_shader
*shader
);
2514 bool nir_opt_algebraic_late(nir_shader
*shader
);
2515 bool nir_opt_constant_folding(nir_shader
*shader
);
2517 bool nir_opt_global_to_local(nir_shader
*shader
);
2519 bool nir_copy_prop(nir_shader
*shader
);
2521 bool nir_opt_cse(nir_shader
*shader
);
2523 bool nir_opt_dce(nir_shader
*shader
);
2525 bool nir_opt_dead_cf(nir_shader
*shader
);
2527 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2529 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2531 bool nir_opt_remove_phis(nir_shader
*shader
);
2533 bool nir_opt_undef(nir_shader
*shader
);
2535 void nir_sweep(nir_shader
*shader
);
2537 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2538 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);