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 * If true, this variable represents an array of scalars that should
234 * be tightly packed. In other words, consecutive array elements
235 * should be stored one component apart, rather than one slot apart.
240 * Whether this is a fragment shader output implicitly initialized with
241 * the previous contents of the specified render target at the
242 * framebuffer location corresponding to this shader invocation.
244 unsigned fb_fetch_output
:1;
247 * \brief Layout qualifier for gl_FragDepth.
249 * This is not equal to \c ir_depth_layout_none if and only if this
250 * variable is \c gl_FragDepth and a layout qualifier is specified.
252 nir_depth_layout depth_layout
;
255 * Storage location of the base of this variable
257 * The precise meaning of this field depends on the nature of the variable.
259 * - Vertex shader input: one of the values from \c gl_vert_attrib.
260 * - Vertex shader output: one of the values from \c gl_varying_slot.
261 * - Geometry shader input: one of the values from \c gl_varying_slot.
262 * - Geometry shader output: one of the values from \c gl_varying_slot.
263 * - Fragment shader input: one of the values from \c gl_varying_slot.
264 * - Fragment shader output: one of the values from \c gl_frag_result.
265 * - Uniforms: Per-stage uniform slot number for default uniform block.
266 * - Uniforms: Index within the uniform block definition for UBO members.
267 * - Non-UBO Uniforms: uniform slot number.
268 * - Other: This field is not currently used.
270 * If the variable is a uniform, shader input, or shader output, and the
271 * slot has not been assigned, the value will be -1.
276 * The actual location of the variable in the IR. Only valid for inputs
279 unsigned int driver_location
;
282 * output index for dual source blending.
287 * Descriptor set binding for sampler or UBO.
292 * Initial binding point for a sampler or UBO.
294 * For array types, this represents the binding point for the first element.
299 * Location an atomic counter is stored at.
304 * ARB_shader_image_load_store qualifiers.
307 bool read_only
; /**< "readonly" qualifier. */
308 bool write_only
; /**< "writeonly" qualifier. */
313 /** Image internal format if specified explicitly, otherwise GL_NONE. */
318 * Highest element accessed with a constant expression array index
320 * Not used for non-array variables.
322 unsigned max_array_access
;
327 * Built-in state that backs this uniform
329 * Once set at variable creation, \c state_slots must remain invariant.
330 * This is because, ideally, this array would be shared by all clones of
331 * this variable in the IR tree. In other words, we'd really like for it
332 * to be a fly-weight.
334 * If the variable is not a uniform, \c num_state_slots will be zero and
335 * \c state_slots will be \c NULL.
338 unsigned num_state_slots
; /**< Number of state slots used */
339 nir_state_slot
*state_slots
; /**< State descriptors. */
343 * Constant expression assigned in the initializer of the variable
345 nir_constant
*constant_initializer
;
348 * For variables that are in an interface block or are an instance of an
349 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
351 * \sa ir_variable::location
353 const struct glsl_type
*interface_type
;
356 #define nir_foreach_variable(var, var_list) \
357 foreach_list_typed(nir_variable, var, node, var_list)
359 #define nir_foreach_variable_safe(var, var_list) \
360 foreach_list_typed_safe(nir_variable, var, node, var_list)
363 nir_variable_is_global(const nir_variable
*var
)
365 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
368 typedef struct nir_register
{
369 struct exec_node node
;
371 unsigned num_components
; /** < number of vector components */
372 unsigned num_array_elems
; /** < size of array (0 for no array) */
374 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
377 /** generic register index. */
380 /** only for debug purposes, can be NULL */
383 /** whether this register is local (per-function) or global (per-shader) */
387 * If this flag is set to true, then accessing channels >= num_components
388 * is well-defined, and simply spills over to the next array element. This
389 * is useful for backends that can do per-component accessing, in
390 * particular scalar backends. By setting this flag and making
391 * num_components equal to 1, structures can be packed tightly into
392 * registers and then registers can be accessed per-component to get to
393 * each structure member, even if it crosses vec4 boundaries.
397 /** set of nir_src's where this register is used (read from) */
398 struct list_head uses
;
400 /** set of nir_dest's where this register is defined (written to) */
401 struct list_head defs
;
403 /** set of nir_if's where this register is used as a condition */
404 struct list_head if_uses
;
411 nir_instr_type_intrinsic
,
412 nir_instr_type_load_const
,
414 nir_instr_type_ssa_undef
,
416 nir_instr_type_parallel_copy
,
419 typedef struct nir_instr
{
420 struct exec_node node
;
422 struct nir_block
*block
;
424 /** generic instruction index. */
427 /* A temporary for optimization and analysis passes to use for storing
428 * flags. For instance, DCE uses this to store the "dead/live" info.
433 static inline nir_instr
*
434 nir_instr_next(nir_instr
*instr
)
436 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
437 if (exec_node_is_tail_sentinel(next
))
440 return exec_node_data(nir_instr
, next
, node
);
443 static inline nir_instr
*
444 nir_instr_prev(nir_instr
*instr
)
446 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
447 if (exec_node_is_head_sentinel(prev
))
450 return exec_node_data(nir_instr
, prev
, node
);
454 nir_instr_is_first(nir_instr
*instr
)
456 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
460 nir_instr_is_last(nir_instr
*instr
)
462 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
465 typedef struct nir_ssa_def
{
466 /** for debugging only, can be NULL */
469 /** generic SSA definition index. */
472 /** Index into the live_in and live_out bitfields */
475 nir_instr
*parent_instr
;
477 /** set of nir_instr's where this register is used (read from) */
478 struct list_head uses
;
480 /** set of nir_if's where this register is used as a condition */
481 struct list_head if_uses
;
483 uint8_t num_components
;
485 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
493 struct nir_src
*indirect
; /** < NULL for no indirect offset */
494 unsigned base_offset
;
496 /* TODO use-def chain goes here */
500 nir_instr
*parent_instr
;
501 struct list_head def_link
;
504 struct nir_src
*indirect
; /** < NULL for no indirect offset */
505 unsigned base_offset
;
507 /* TODO def-use chain goes here */
512 typedef struct nir_src
{
514 nir_instr
*parent_instr
;
515 struct nir_if
*parent_if
;
518 struct list_head use_link
;
528 static inline nir_src
531 nir_src src
= { { NULL
} };
535 #define NIR_SRC_INIT nir_src_init()
537 #define nir_foreach_use(src, reg_or_ssa_def) \
538 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
540 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
541 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
543 #define nir_foreach_if_use(src, reg_or_ssa_def) \
544 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
546 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
547 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
558 static inline nir_dest
561 nir_dest dest
= { { { NULL
} } };
565 #define NIR_DEST_INIT nir_dest_init()
567 #define nir_foreach_def(dest, reg) \
568 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
570 #define nir_foreach_def_safe(dest, reg) \
571 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
573 static inline nir_src
574 nir_src_for_ssa(nir_ssa_def
*def
)
576 nir_src src
= NIR_SRC_INIT
;
584 static inline nir_src
585 nir_src_for_reg(nir_register
*reg
)
587 nir_src src
= NIR_SRC_INIT
;
591 src
.reg
.indirect
= NULL
;
592 src
.reg
.base_offset
= 0;
597 static inline nir_dest
598 nir_dest_for_reg(nir_register
*reg
)
600 nir_dest dest
= NIR_DEST_INIT
;
607 static inline unsigned
608 nir_src_bit_size(nir_src src
)
610 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
613 static inline unsigned
614 nir_dest_bit_size(nir_dest dest
)
616 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
619 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
620 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
626 * \name input modifiers
630 * For inputs interpreted as floating point, flips the sign bit. For
631 * inputs interpreted as integers, performs the two's complement negation.
636 * Clears the sign bit for floating point values, and computes the integer
637 * absolute value for integers. Note that the negate modifier acts after
638 * the absolute value modifier, therefore if both are set then all inputs
639 * will become negative.
645 * For each input component, says which component of the register it is
646 * chosen from. Note that which elements of the swizzle are used and which
647 * are ignored are based on the write mask for most opcodes - for example,
648 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
649 * a swizzle of {2, x, 1, 0} where x means "don't care."
658 * \name saturate output modifier
660 * Only valid for opcodes that output floating-point numbers. Clamps the
661 * output to between 0.0 and 1.0 inclusive.
666 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
670 nir_type_invalid
= 0, /* Not a valid type */
675 nir_type_bool32
= 32 | nir_type_bool
,
676 nir_type_int8
= 8 | nir_type_int
,
677 nir_type_int16
= 16 | nir_type_int
,
678 nir_type_int32
= 32 | nir_type_int
,
679 nir_type_int64
= 64 | nir_type_int
,
680 nir_type_uint8
= 8 | nir_type_uint
,
681 nir_type_uint16
= 16 | nir_type_uint
,
682 nir_type_uint32
= 32 | nir_type_uint
,
683 nir_type_uint64
= 64 | nir_type_uint
,
684 nir_type_float16
= 16 | nir_type_float
,
685 nir_type_float32
= 32 | nir_type_float
,
686 nir_type_float64
= 64 | nir_type_float
,
689 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
690 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
692 static inline unsigned
693 nir_alu_type_get_type_size(nir_alu_type type
)
695 return type
& NIR_ALU_TYPE_SIZE_MASK
;
698 static inline unsigned
699 nir_alu_type_get_base_type(nir_alu_type type
)
701 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
705 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
706 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
707 } nir_op_algebraic_property
;
715 * The number of components in the output
717 * If non-zero, this is the size of the output and input sizes are
718 * explicitly given; swizzle and writemask are still in effect, but if
719 * the output component is masked out, then the input component may
722 * If zero, the opcode acts in the standard, per-component manner; the
723 * operation is performed on each component (except the ones that are
724 * masked out) with the input being taken from the input swizzle for
727 * The size of some of the inputs may be given (i.e. non-zero) even
728 * though output_size is zero; in that case, the inputs with a zero
729 * size act per-component, while the inputs with non-zero size don't.
731 unsigned output_size
;
734 * The type of vector that the instruction outputs. Note that the
735 * staurate modifier is only allowed on outputs with the float type.
738 nir_alu_type output_type
;
741 * The number of components in each input
743 unsigned input_sizes
[4];
746 * The type of vector that each input takes. Note that negate and
747 * absolute value are only allowed on inputs with int or float type and
748 * behave differently on the two.
750 nir_alu_type input_types
[4];
752 nir_op_algebraic_property algebraic_properties
;
755 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
757 typedef struct nir_alu_instr
{
761 /** Indicates that this ALU instruction generates an exact value
763 * This is kind of a mixture of GLSL "precise" and "invariant" and not
764 * really equivalent to either. This indicates that the value generated by
765 * this operation is high-precision and any code transformations that touch
766 * it must ensure that the resulting value is bit-for-bit identical to the
775 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
776 nir_alu_instr
*instr
);
777 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
778 nir_alu_instr
*instr
);
780 /* is this source channel used? */
782 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
784 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
785 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
787 return (instr
->dest
.write_mask
>> channel
) & 1;
791 * For instructions whose destinations are SSA, get the number of channels
794 static inline unsigned
795 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
797 assert(instr
->dest
.dest
.is_ssa
);
799 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
800 return nir_op_infos
[instr
->op
].input_sizes
[src
];
802 return instr
->dest
.dest
.ssa
.num_components
;
805 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
806 unsigned src1
, unsigned src2
);
810 nir_deref_type_array
,
811 nir_deref_type_struct
814 typedef struct nir_deref
{
815 nir_deref_type deref_type
;
816 struct nir_deref
*child
;
817 const struct glsl_type
*type
;
826 /* This enum describes how the array is referenced. If the deref is
827 * direct then the base_offset is used. If the deref is indirect then
828 * offset is given by base_offset + indirect. If the deref is a wildcard
829 * then the deref refers to all of the elements of the array at the same
830 * time. Wildcard dereferences are only ever allowed in copy_var
831 * intrinsics and the source and destination derefs must have matching
835 nir_deref_array_type_direct
,
836 nir_deref_array_type_indirect
,
837 nir_deref_array_type_wildcard
,
838 } nir_deref_array_type
;
843 nir_deref_array_type deref_array_type
;
844 unsigned base_offset
;
854 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
,
855 deref_type
, nir_deref_type_var
)
856 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
,
857 deref_type
, nir_deref_type_array
)
858 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
,
859 deref_type
, nir_deref_type_struct
)
861 /* Returns the last deref in the chain. */
862 static inline nir_deref
*
863 nir_deref_tail(nir_deref
*deref
)
866 deref
= deref
->child
;
874 nir_deref_var
**params
;
875 nir_deref_var
*return_deref
;
877 struct nir_function
*callee
;
880 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
881 num_variables, num_indices, idx0, idx1, idx2, flags) \
882 nir_intrinsic_##name,
884 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
887 #include "nir_intrinsics.h"
888 nir_num_intrinsics
= nir_last_intrinsic
+ 1
891 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
893 /** Represents an intrinsic
895 * An intrinsic is an instruction type for handling things that are
896 * more-or-less regular operations but don't just consume and produce SSA
897 * values like ALU operations do. Intrinsics are not for things that have
898 * special semantic meaning such as phi nodes and parallel copies.
899 * Examples of intrinsics include variable load/store operations, system
900 * value loads, and the like. Even though texturing more-or-less falls
901 * under this category, texturing is its own instruction type because
902 * trying to represent texturing with intrinsics would lead to a
903 * combinatorial explosion of intrinsic opcodes.
905 * By having a single instruction type for handling a lot of different
906 * cases, optimization passes can look for intrinsics and, for the most
907 * part, completely ignore them. Each intrinsic type also has a few
908 * possible flags that govern whether or not they can be reordered or
909 * eliminated. That way passes like dead code elimination can still work
910 * on intrisics without understanding the meaning of each.
912 * Each intrinsic has some number of constant indices, some number of
913 * variables, and some number of sources. What these sources, variables,
914 * and indices mean depends on the intrinsic and is documented with the
915 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
916 * instructions are the only types of instruction that can operate on
922 nir_intrinsic_op intrinsic
;
926 /** number of components if this is a vectorized intrinsic
928 * Similarly to ALU operations, some intrinsics are vectorized.
929 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
930 * For vectorized intrinsics, the num_components field specifies the
931 * number of destination components and the number of source components
932 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
934 uint8_t num_components
;
936 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
938 nir_deref_var
*variables
[2];
941 } nir_intrinsic_instr
;
944 * \name NIR intrinsics semantic flags
946 * information about what the compiler can do with the intrinsics.
948 * \sa nir_intrinsic_info::flags
952 * whether the intrinsic can be safely eliminated if none of its output
953 * value is not being used.
955 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
958 * Whether the intrinsic can be reordered with respect to any other
959 * intrinsic, i.e. whether the only reordering dependencies of the
960 * intrinsic are due to the register reads/writes.
962 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
963 } nir_intrinsic_semantic_flag
;
966 * \name NIR intrinsics const-index flag
968 * Indicates the usage of a const_index slot.
970 * \sa nir_intrinsic_info::index_map
974 * Generally instructions that take a offset src argument, can encode
975 * a constant 'base' value which is added to the offset.
977 NIR_INTRINSIC_BASE
= 1,
980 * For store instructions, a writemask for the store.
982 NIR_INTRINSIC_WRMASK
= 2,
985 * The stream-id for GS emit_vertex/end_primitive intrinsics.
987 NIR_INTRINSIC_STREAM_ID
= 3,
990 * The clip-plane id for load_user_clip_plane intrinsic.
992 NIR_INTRINSIC_UCP_ID
= 4,
995 * The amount of data, starting from BASE, that this instruction may
996 * access. This is used to provide bounds if the offset is not constant.
998 NIR_INTRINSIC_RANGE
= 5,
1001 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1003 NIR_INTRINSIC_DESC_SET
= 6,
1006 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1008 NIR_INTRINSIC_BINDING
= 7,
1013 NIR_INTRINSIC_COMPONENT
= 8,
1016 * Interpolation mode (only meaningful for FS inputs).
1018 NIR_INTRINSIC_INTERP_MODE
= 9,
1020 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1022 } nir_intrinsic_index_flag
;
1024 #define NIR_INTRINSIC_MAX_INPUTS 4
1029 unsigned num_srcs
; /** < number of register/SSA inputs */
1031 /** number of components of each input register
1033 * If this value is 0, the number of components is given by the
1034 * num_components field of nir_intrinsic_instr.
1036 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1040 /** number of components of the output register
1042 * If this value is 0, the number of components is given by the
1043 * num_components field of nir_intrinsic_instr.
1045 unsigned dest_components
;
1047 /** the number of inputs/outputs that are variables */
1048 unsigned num_variables
;
1050 /** the number of constant indices used by the intrinsic */
1051 unsigned num_indices
;
1053 /** indicates the usage of intr->const_index[n] */
1054 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1056 /** semantic flags for calls to this intrinsic */
1057 nir_intrinsic_semantic_flag flags
;
1058 } nir_intrinsic_info
;
1060 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1063 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1064 static inline type \
1065 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
1067 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1068 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1069 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1071 static inline void \
1072 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1074 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1075 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1076 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1079 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1080 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1081 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1082 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1083 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1084 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1085 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1086 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1087 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1090 * \group texture information
1092 * This gives semantic information about textures which is useful to the
1093 * frontend, the backend, and lowering passes, but not the optimizer.
1098 nir_tex_src_projector
,
1099 nir_tex_src_comparitor
, /* shadow comparitor */
1103 nir_tex_src_ms_index
, /* MSAA sample index */
1104 nir_tex_src_ms_mcs
, /* MSAA compression value */
1107 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1108 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1109 nir_tex_src_plane
, /* < selects plane for planar textures */
1110 nir_num_tex_src_types
1115 nir_tex_src_type src_type
;
1119 nir_texop_tex
, /**< Regular texture look-up */
1120 nir_texop_txb
, /**< Texture look-up with LOD bias */
1121 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1122 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1123 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1124 nir_texop_txf_ms
, /**< Multisample texture fetch */
1125 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1126 nir_texop_txs
, /**< Texture size */
1127 nir_texop_lod
, /**< Texture lod query */
1128 nir_texop_tg4
, /**< Texture gather */
1129 nir_texop_query_levels
, /**< Texture levels query */
1130 nir_texop_texture_samples
, /**< Texture samples query */
1131 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1139 enum glsl_sampler_dim sampler_dim
;
1140 nir_alu_type dest_type
;
1145 unsigned num_srcs
, coord_components
;
1146 bool is_array
, is_shadow
;
1149 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1150 * components or the new-style shadow that outputs 1 component.
1152 bool is_new_style_shadow
;
1154 /* gather component selector */
1155 unsigned component
: 2;
1157 /** The texture index
1159 * If this texture instruction has a nir_tex_src_texture_offset source,
1160 * then the texture index is given by texture_index + texture_offset.
1162 unsigned texture_index
;
1164 /** The size of the texture array or 0 if it's not an array */
1165 unsigned texture_array_size
;
1167 /** The texture deref
1169 * If this is null, use texture_index instead.
1171 nir_deref_var
*texture
;
1173 /** The sampler index
1175 * The following operations do not require a sampler and, as such, this
1176 * field should be ignored:
1178 * - nir_texop_txf_ms
1182 * - nir_texop_query_levels
1183 * - nir_texop_texture_samples
1184 * - nir_texop_samples_identical
1186 * If this texture instruction has a nir_tex_src_sampler_offset source,
1187 * then the sampler index is given by sampler_index + sampler_offset.
1189 unsigned sampler_index
;
1191 /** The sampler deref
1193 * If this is null, use sampler_index instead.
1195 nir_deref_var
*sampler
;
1198 static inline unsigned
1199 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1201 switch (instr
->op
) {
1202 case nir_texop_txs
: {
1204 switch (instr
->sampler_dim
) {
1205 case GLSL_SAMPLER_DIM_1D
:
1206 case GLSL_SAMPLER_DIM_BUF
:
1209 case GLSL_SAMPLER_DIM_2D
:
1210 case GLSL_SAMPLER_DIM_CUBE
:
1211 case GLSL_SAMPLER_DIM_MS
:
1212 case GLSL_SAMPLER_DIM_RECT
:
1213 case GLSL_SAMPLER_DIM_EXTERNAL
:
1214 case GLSL_SAMPLER_DIM_SUBPASS
:
1217 case GLSL_SAMPLER_DIM_3D
:
1221 unreachable("not reached");
1223 if (instr
->is_array
)
1231 case nir_texop_texture_samples
:
1232 case nir_texop_query_levels
:
1233 case nir_texop_samples_identical
:
1237 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1244 /* Returns true if this texture operation queries something about the texture
1245 * rather than actually sampling it.
1248 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1250 switch (instr
->op
) {
1253 case nir_texop_texture_samples
:
1254 case nir_texop_query_levels
:
1255 case nir_texop_txf_ms_mcs
:
1262 case nir_texop_txf_ms
:
1266 unreachable("Invalid texture opcode");
1270 static inline nir_alu_type
1271 nir_tex_instr_src_type(nir_tex_instr
*instr
, unsigned src
)
1273 switch (instr
->src
[src
].src_type
) {
1274 case nir_tex_src_coord
:
1275 switch (instr
->op
) {
1277 case nir_texop_txf_ms
:
1278 case nir_texop_txf_ms_mcs
:
1279 case nir_texop_samples_identical
:
1280 return nir_type_int
;
1283 return nir_type_float
;
1286 case nir_tex_src_lod
:
1287 switch (instr
->op
) {
1290 return nir_type_int
;
1293 return nir_type_float
;
1296 case nir_tex_src_projector
:
1297 case nir_tex_src_comparitor
:
1298 case nir_tex_src_bias
:
1299 case nir_tex_src_ddx
:
1300 case nir_tex_src_ddy
:
1301 return nir_type_float
;
1303 case nir_tex_src_offset
:
1304 case nir_tex_src_ms_index
:
1305 case nir_tex_src_texture_offset
:
1306 case nir_tex_src_sampler_offset
:
1307 return nir_type_int
;
1310 unreachable("Invalid texture source type");
1314 static inline unsigned
1315 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1317 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1318 return instr
->coord_components
;
1320 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1321 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1324 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1325 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1326 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1327 if (instr
->is_array
)
1328 return instr
->coord_components
- 1;
1330 return instr
->coord_components
;
1337 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1339 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1340 if (instr
->src
[i
].src_type
== type
)
1346 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1360 nir_const_value value
;
1363 } nir_load_const_instr
;
1376 /* creates a new SSA variable in an undefined state */
1381 } nir_ssa_undef_instr
;
1384 struct exec_node node
;
1386 /* The predecessor block corresponding to this source */
1387 struct nir_block
*pred
;
1392 #define nir_foreach_phi_src(phi_src, phi) \
1393 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1394 #define nir_foreach_phi_src_safe(phi_src, phi) \
1395 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1400 struct exec_list srcs
; /** < list of nir_phi_src */
1406 struct exec_node node
;
1409 } nir_parallel_copy_entry
;
1411 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1412 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1417 /* A list of nir_parallel_copy_entry's. The sources of all of the
1418 * entries are copied to the corresponding destinations "in parallel".
1419 * In other words, if we have two entries: a -> b and b -> a, the values
1422 struct exec_list entries
;
1423 } nir_parallel_copy_instr
;
1425 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1426 type
, nir_instr_type_alu
)
1427 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1428 type
, nir_instr_type_call
)
1429 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1430 type
, nir_instr_type_jump
)
1431 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1432 type
, nir_instr_type_tex
)
1433 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1434 type
, nir_instr_type_intrinsic
)
1435 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1436 type
, nir_instr_type_load_const
)
1437 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1438 type
, nir_instr_type_ssa_undef
)
1439 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1440 type
, nir_instr_type_phi
)
1441 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1442 nir_parallel_copy_instr
, instr
,
1443 type
, nir_instr_type_parallel_copy
)
1448 * Control flow consists of a tree of control flow nodes, which include
1449 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1450 * instructions that always run start-to-finish. Each basic block also keeps
1451 * track of its successors (blocks which may run immediately after the current
1452 * block) and predecessors (blocks which could have run immediately before the
1453 * current block). Each function also has a start block and an end block which
1454 * all return statements point to (which is always empty). Together, all the
1455 * blocks with their predecessors and successors make up the control flow
1456 * graph (CFG) of the function. There are helpers that modify the tree of
1457 * control flow nodes while modifying the CFG appropriately; these should be
1458 * used instead of modifying the tree directly.
1465 nir_cf_node_function
1468 typedef struct nir_cf_node
{
1469 struct exec_node node
;
1470 nir_cf_node_type type
;
1471 struct nir_cf_node
*parent
;
1474 typedef struct nir_block
{
1475 nir_cf_node cf_node
;
1477 struct exec_list instr_list
; /** < list of nir_instr */
1479 /** generic block index; generated by nir_index_blocks */
1483 * Each block can only have up to 2 successors, so we put them in a simple
1484 * array - no need for anything more complicated.
1486 struct nir_block
*successors
[2];
1488 /* Set of nir_block predecessors in the CFG */
1489 struct set
*predecessors
;
1492 * this node's immediate dominator in the dominance tree - set to NULL for
1495 struct nir_block
*imm_dom
;
1497 /* This node's children in the dominance tree */
1498 unsigned num_dom_children
;
1499 struct nir_block
**dom_children
;
1501 /* Set of nir_block's on the dominance frontier of this block */
1502 struct set
*dom_frontier
;
1505 * These two indices have the property that dom_{pre,post}_index for each
1506 * child of this block in the dominance tree will always be between
1507 * dom_pre_index and dom_post_index for this block, which makes testing if
1508 * a given block is dominated by another block an O(1) operation.
1510 unsigned dom_pre_index
, dom_post_index
;
1512 /* live in and out for this block; used for liveness analysis */
1513 BITSET_WORD
*live_in
;
1514 BITSET_WORD
*live_out
;
1517 static inline nir_instr
*
1518 nir_block_first_instr(nir_block
*block
)
1520 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1521 return exec_node_data(nir_instr
, head
, node
);
1524 static inline nir_instr
*
1525 nir_block_last_instr(nir_block
*block
)
1527 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1528 return exec_node_data(nir_instr
, tail
, node
);
1531 #define nir_foreach_instr(instr, block) \
1532 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1533 #define nir_foreach_instr_reverse(instr, block) \
1534 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1535 #define nir_foreach_instr_safe(instr, block) \
1536 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1537 #define nir_foreach_instr_reverse_safe(instr, block) \
1538 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1540 typedef struct nir_if
{
1541 nir_cf_node cf_node
;
1544 struct exec_list then_list
; /** < list of nir_cf_node */
1545 struct exec_list else_list
; /** < list of nir_cf_node */
1549 nir_cf_node cf_node
;
1551 struct exec_list body
; /** < list of nir_cf_node */
1555 * Various bits of metadata that can may be created or required by
1556 * optimization and analysis passes
1559 nir_metadata_none
= 0x0,
1560 nir_metadata_block_index
= 0x1,
1561 nir_metadata_dominance
= 0x2,
1562 nir_metadata_live_ssa_defs
= 0x4,
1563 nir_metadata_not_properly_reset
= 0x8,
1567 nir_cf_node cf_node
;
1569 /** pointer to the function of which this is an implementation */
1570 struct nir_function
*function
;
1572 struct exec_list body
; /** < list of nir_cf_node */
1574 nir_block
*end_block
;
1576 /** list for all local variables in the function */
1577 struct exec_list locals
;
1579 /** array of variables used as parameters */
1580 unsigned num_params
;
1581 nir_variable
**params
;
1583 /** variable used to hold the result of the function */
1584 nir_variable
*return_var
;
1586 /** list of local registers in the function */
1587 struct exec_list registers
;
1589 /** next available local register index */
1592 /** next available SSA value index */
1595 /* total number of basic blocks, only valid when block_index_dirty = false */
1596 unsigned num_blocks
;
1598 nir_metadata valid_metadata
;
1599 } nir_function_impl
;
1601 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1602 nir_start_block(nir_function_impl
*impl
)
1604 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1607 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1608 nir_impl_last_block(nir_function_impl
*impl
)
1610 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1613 static inline nir_cf_node
*
1614 nir_cf_node_next(nir_cf_node
*node
)
1616 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1617 if (exec_node_is_tail_sentinel(next
))
1620 return exec_node_data(nir_cf_node
, next
, node
);
1623 static inline nir_cf_node
*
1624 nir_cf_node_prev(nir_cf_node
*node
)
1626 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1627 if (exec_node_is_head_sentinel(prev
))
1630 return exec_node_data(nir_cf_node
, prev
, node
);
1634 nir_cf_node_is_first(const nir_cf_node
*node
)
1636 return exec_node_is_head_sentinel(node
->node
.prev
);
1640 nir_cf_node_is_last(const nir_cf_node
*node
)
1642 return exec_node_is_tail_sentinel(node
->node
.next
);
1645 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1646 type
, nir_cf_node_block
)
1647 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1648 type
, nir_cf_node_if
)
1649 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1650 type
, nir_cf_node_loop
)
1651 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1652 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1654 static inline nir_block
*
1655 nir_if_first_then_block(nir_if
*if_stmt
)
1657 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1658 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1661 static inline nir_block
*
1662 nir_if_last_then_block(nir_if
*if_stmt
)
1664 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1665 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1668 static inline nir_block
*
1669 nir_if_first_else_block(nir_if
*if_stmt
)
1671 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1672 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1675 static inline nir_block
*
1676 nir_if_last_else_block(nir_if
*if_stmt
)
1678 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1679 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1682 static inline nir_block
*
1683 nir_loop_first_block(nir_loop
*loop
)
1685 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1686 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1689 static inline nir_block
*
1690 nir_loop_last_block(nir_loop
*loop
)
1692 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1693 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1699 nir_parameter_inout
,
1700 } nir_parameter_type
;
1703 nir_parameter_type param_type
;
1704 const struct glsl_type
*type
;
1707 typedef struct nir_function
{
1708 struct exec_node node
;
1711 struct nir_shader
*shader
;
1713 unsigned num_params
;
1714 nir_parameter
*params
;
1715 const struct glsl_type
*return_type
;
1717 /** The implementation of this function.
1719 * If the function is only declared and not implemented, this is NULL.
1721 nir_function_impl
*impl
;
1724 typedef struct nir_shader_compiler_options
{
1729 /** Lowers flrp when it does not support doubles */
1736 bool lower_bitfield_extract
;
1737 bool lower_bitfield_insert
;
1738 bool lower_uadd_carry
;
1739 bool lower_usub_borrow
;
1740 /** lowers fneg and ineg to fsub and isub. */
1742 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1745 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1748 /** enables rules to lower idiv by power-of-two: */
1751 /* Does the native fdot instruction replicate its result for four
1752 * components? If so, then opt_algebraic_late will turn all fdotN
1753 * instructions into fdot_replicatedN instructions.
1755 bool fdot_replicates
;
1757 /** lowers ffract to fsub+ffloor: */
1760 bool lower_pack_half_2x16
;
1761 bool lower_pack_unorm_2x16
;
1762 bool lower_pack_snorm_2x16
;
1763 bool lower_pack_unorm_4x8
;
1764 bool lower_pack_snorm_4x8
;
1765 bool lower_unpack_half_2x16
;
1766 bool lower_unpack_unorm_2x16
;
1767 bool lower_unpack_snorm_2x16
;
1768 bool lower_unpack_unorm_4x8
;
1769 bool lower_unpack_snorm_4x8
;
1771 bool lower_extract_byte
;
1772 bool lower_extract_word
;
1775 * Does the driver support real 32-bit integers? (Otherwise, integers
1776 * are simulated by floats.)
1778 bool native_integers
;
1780 /* Indicates that the driver only has zero-based vertex id */
1781 bool vertex_id_zero_based
;
1783 bool lower_cs_local_index_from_id
;
1786 * Should nir_lower_io() create load_interpolated_input intrinsics?
1788 * If not, it generates regular load_input intrinsics and interpolation
1789 * information must be inferred from the list of input nir_variables.
1791 bool use_interpolated_input_intrinsics
;
1792 } nir_shader_compiler_options
;
1794 typedef struct nir_shader
{
1795 /** list of uniforms (nir_variable) */
1796 struct exec_list uniforms
;
1798 /** list of inputs (nir_variable) */
1799 struct exec_list inputs
;
1801 /** list of outputs (nir_variable) */
1802 struct exec_list outputs
;
1804 /** list of shared compute variables (nir_variable) */
1805 struct exec_list shared
;
1807 /** Set of driver-specific options for the shader.
1809 * The memory for the options is expected to be kept in a single static
1810 * copy by the driver.
1812 const struct nir_shader_compiler_options
*options
;
1814 /** Various bits of compile-time information about a given shader */
1815 struct shader_info
*info
;
1817 /** list of global variables in the shader (nir_variable) */
1818 struct exec_list globals
;
1820 /** list of system value variables in the shader (nir_variable) */
1821 struct exec_list system_values
;
1823 struct exec_list functions
; /** < list of nir_function */
1825 /** list of global register in the shader */
1826 struct exec_list registers
;
1828 /** next available global register index */
1832 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1835 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1837 /** The shader stage, such as MESA_SHADER_VERTEX. */
1838 gl_shader_stage stage
;
1841 static inline nir_function_impl
*
1842 nir_shader_get_entrypoint(nir_shader
*shader
)
1844 assert(exec_list_length(&shader
->functions
) == 1);
1845 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1846 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1847 assert(func
->return_type
== glsl_void_type());
1848 assert(func
->num_params
== 0);
1853 #define nir_foreach_function(func, shader) \
1854 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1856 nir_shader
*nir_shader_create(void *mem_ctx
,
1857 gl_shader_stage stage
,
1858 const nir_shader_compiler_options
*options
,
1861 /** creates a register, including assigning it an index and adding it to the list */
1862 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1864 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1866 void nir_reg_remove(nir_register
*reg
);
1868 /** Adds a variable to the appropreate list in nir_shader */
1869 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1872 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1874 assert(var
->data
.mode
== nir_var_local
);
1875 exec_list_push_tail(&impl
->locals
, &var
->node
);
1878 /** creates a variable, sets a few defaults, and adds it to the list */
1879 nir_variable
*nir_variable_create(nir_shader
*shader
,
1880 nir_variable_mode mode
,
1881 const struct glsl_type
*type
,
1883 /** creates a local variable and adds it to the list */
1884 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1885 const struct glsl_type
*type
,
1888 /** creates a function and adds it to the shader's list of functions */
1889 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1891 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1892 /** creates a function_impl that isn't tied to any particular function */
1893 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1895 nir_block
*nir_block_create(nir_shader
*shader
);
1896 nir_if
*nir_if_create(nir_shader
*shader
);
1897 nir_loop
*nir_loop_create(nir_shader
*shader
);
1899 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1901 /** requests that the given pieces of metadata be generated */
1902 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1903 /** dirties all but the preserved metadata */
1904 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1906 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1907 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1909 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1911 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1912 unsigned num_components
,
1915 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1916 nir_intrinsic_op op
);
1918 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1919 nir_function
*callee
);
1921 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1923 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1925 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1927 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1928 unsigned num_components
,
1931 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1932 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1933 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1935 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1937 typedef bool (*nir_deref_foreach_leaf_cb
)(nir_deref_var
*deref
, void *state
);
1938 bool nir_deref_foreach_leaf(nir_deref_var
*deref
,
1939 nir_deref_foreach_leaf_cb cb
, void *state
);
1941 nir_load_const_instr
*
1942 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1945 * NIR Cursors and Instruction Insertion API
1948 * A tiny struct representing a point to insert/extract instructions or
1949 * control flow nodes. Helps reduce the combinatorial explosion of possible
1950 * points to insert/extract.
1952 * \sa nir_control_flow.h
1955 nir_cursor_before_block
,
1956 nir_cursor_after_block
,
1957 nir_cursor_before_instr
,
1958 nir_cursor_after_instr
,
1959 } nir_cursor_option
;
1962 nir_cursor_option option
;
1969 static inline nir_block
*
1970 nir_cursor_current_block(nir_cursor cursor
)
1972 if (cursor
.option
== nir_cursor_before_instr
||
1973 cursor
.option
== nir_cursor_after_instr
) {
1974 return cursor
.instr
->block
;
1976 return cursor
.block
;
1980 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
1982 static inline nir_cursor
1983 nir_before_block(nir_block
*block
)
1986 cursor
.option
= nir_cursor_before_block
;
1987 cursor
.block
= block
;
1991 static inline nir_cursor
1992 nir_after_block(nir_block
*block
)
1995 cursor
.option
= nir_cursor_after_block
;
1996 cursor
.block
= block
;
2000 static inline nir_cursor
2001 nir_before_instr(nir_instr
*instr
)
2004 cursor
.option
= nir_cursor_before_instr
;
2005 cursor
.instr
= instr
;
2009 static inline nir_cursor
2010 nir_after_instr(nir_instr
*instr
)
2013 cursor
.option
= nir_cursor_after_instr
;
2014 cursor
.instr
= instr
;
2018 static inline nir_cursor
2019 nir_after_block_before_jump(nir_block
*block
)
2021 nir_instr
*last_instr
= nir_block_last_instr(block
);
2022 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2023 return nir_before_instr(last_instr
);
2025 return nir_after_block(block
);
2029 static inline nir_cursor
2030 nir_before_cf_node(nir_cf_node
*node
)
2032 if (node
->type
== nir_cf_node_block
)
2033 return nir_before_block(nir_cf_node_as_block(node
));
2035 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2038 static inline nir_cursor
2039 nir_after_cf_node(nir_cf_node
*node
)
2041 if (node
->type
== nir_cf_node_block
)
2042 return nir_after_block(nir_cf_node_as_block(node
));
2044 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2047 static inline nir_cursor
2048 nir_after_phis(nir_block
*block
)
2050 nir_foreach_instr(instr
, block
) {
2051 if (instr
->type
!= nir_instr_type_phi
)
2052 return nir_before_instr(instr
);
2054 return nir_after_block(block
);
2057 static inline nir_cursor
2058 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2060 if (node
->type
== nir_cf_node_block
)
2061 return nir_after_block(nir_cf_node_as_block(node
));
2063 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2065 return nir_after_phis(block
);
2068 static inline nir_cursor
2069 nir_before_cf_list(struct exec_list
*cf_list
)
2071 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2072 exec_list_get_head(cf_list
), node
);
2073 return nir_before_cf_node(first_node
);
2076 static inline nir_cursor
2077 nir_after_cf_list(struct exec_list
*cf_list
)
2079 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2080 exec_list_get_tail(cf_list
), node
);
2081 return nir_after_cf_node(last_node
);
2085 * Insert a NIR instruction at the given cursor.
2087 * Note: This does not update the cursor.
2089 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2092 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2094 nir_instr_insert(nir_before_instr(instr
), before
);
2098 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2100 nir_instr_insert(nir_after_instr(instr
), after
);
2104 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2106 nir_instr_insert(nir_before_block(block
), before
);
2110 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2112 nir_instr_insert(nir_after_block(block
), after
);
2116 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2118 nir_instr_insert(nir_before_cf_node(node
), before
);
2122 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2124 nir_instr_insert(nir_after_cf_node(node
), after
);
2128 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2130 nir_instr_insert(nir_before_cf_list(list
), before
);
2134 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2136 nir_instr_insert(nir_after_cf_list(list
), after
);
2139 void nir_instr_remove(nir_instr
*instr
);
2143 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2144 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2145 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2146 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2148 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2149 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2151 nir_const_value
*nir_src_as_const_value(nir_src src
);
2152 bool nir_src_is_dynamically_uniform(nir_src src
);
2153 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2154 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2155 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2156 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2157 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2160 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2161 unsigned num_components
, unsigned bit_size
,
2163 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2164 unsigned num_components
, unsigned bit_size
,
2166 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2167 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2168 nir_instr
*after_me
);
2170 uint8_t nir_ssa_def_components_read(nir_ssa_def
*def
);
2173 * finds the next basic block in source-code order, returns NULL if there is
2177 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2179 /* Performs the opposite of nir_block_cf_tree_next() */
2181 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2183 /* Gets the first block in a CF node in source-code order */
2185 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2187 /* Gets the last block in a CF node in source-code order */
2189 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2191 /* Gets the next block after a CF node in source-code order */
2193 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2195 /* Macros for loops that visit blocks in source-code order */
2197 #define nir_foreach_block(block, impl) \
2198 for (nir_block *block = nir_start_block(impl); block != NULL; \
2199 block = nir_block_cf_tree_next(block))
2201 #define nir_foreach_block_safe(block, impl) \
2202 for (nir_block *block = nir_start_block(impl), \
2203 *next = nir_block_cf_tree_next(block); \
2205 block = next, next = nir_block_cf_tree_next(block))
2207 #define nir_foreach_block_reverse(block, impl) \
2208 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2209 block = nir_block_cf_tree_prev(block))
2211 #define nir_foreach_block_reverse_safe(block, impl) \
2212 for (nir_block *block = nir_impl_last_block(impl), \
2213 *prev = nir_block_cf_tree_prev(block); \
2215 block = prev, prev = nir_block_cf_tree_prev(block))
2217 #define nir_foreach_block_in_cf_node(block, node) \
2218 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2219 block != nir_cf_node_cf_tree_next(node); \
2220 block = nir_block_cf_tree_next(block))
2222 /* If the following CF node is an if, this function returns that if.
2223 * Otherwise, it returns NULL.
2225 nir_if
*nir_block_get_following_if(nir_block
*block
);
2227 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2229 void nir_index_local_regs(nir_function_impl
*impl
);
2230 void nir_index_global_regs(nir_shader
*shader
);
2231 void nir_index_ssa_defs(nir_function_impl
*impl
);
2232 unsigned nir_index_instrs(nir_function_impl
*impl
);
2234 void nir_index_blocks(nir_function_impl
*impl
);
2236 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2237 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2238 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2240 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2241 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2242 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2243 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2246 void nir_validate_shader(nir_shader
*shader
);
2247 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2248 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2250 #include "util/debug.h"
2252 should_clone_nir(void)
2254 static int should_clone
= -1;
2255 if (should_clone
< 0)
2256 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2258 return should_clone
;
2261 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2262 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2263 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2264 static inline bool should_clone_nir(void) { return false; }
2267 #define _PASS(nir, do_pass) do { \
2269 nir_validate_shader(nir); \
2270 if (should_clone_nir()) { \
2271 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2277 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2278 nir_metadata_set_validation_flag(nir); \
2279 if (pass(nir, ##__VA_ARGS__)) { \
2281 nir_metadata_check_validation_flag(nir); \
2285 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2286 pass(nir, ##__VA_ARGS__); \
2289 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2290 void nir_calc_dominance(nir_shader
*shader
);
2292 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2293 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2295 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2296 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2298 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2299 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2301 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2302 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2304 int nir_gs_count_vertices(const nir_shader
*shader
);
2306 bool nir_split_var_copies(nir_shader
*shader
);
2308 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2309 bool nir_lower_returns(nir_shader
*shader
);
2311 bool nir_inline_functions(nir_shader
*shader
);
2313 bool nir_propagate_invariant(nir_shader
*shader
);
2315 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2316 void nir_lower_var_copies(nir_shader
*shader
);
2318 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2320 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2322 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2324 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2325 nir_function_impl
*entrypoint
,
2326 bool outputs
, bool inputs
);
2328 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2330 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2331 int (*type_size
)(const struct glsl_type
*));
2334 /* If set, this forces all non-flat fragment shader inputs to be
2335 * interpolated as if with the "sample" qualifier. This requires
2336 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2338 nir_lower_io_force_sample_interpolation
= (1 << 1),
2339 } nir_lower_io_options
;
2340 void nir_lower_io(nir_shader
*shader
,
2341 nir_variable_mode modes
,
2342 int (*type_size
)(const struct glsl_type
*),
2343 nir_lower_io_options
);
2344 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2345 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2347 bool nir_is_per_vertex_io(nir_variable
*var
, gl_shader_stage stage
);
2349 void nir_lower_io_types(nir_shader
*shader
);
2350 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2352 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2354 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2355 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2356 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2357 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2359 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2360 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2362 void nir_lower_samplers(nir_shader
*shader
,
2363 const struct gl_shader_program
*shader_program
);
2365 bool nir_lower_system_values(nir_shader
*shader
);
2367 typedef struct nir_lower_tex_options
{
2369 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2370 * sampler types a texture projector is lowered.
2375 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2377 bool lower_txf_offset
;
2380 * If true, lower away nir_tex_src_offset for all rect textures.
2382 bool lower_rect_offset
;
2385 * If true, lower rect textures to 2D, using txs to fetch the
2386 * texture dimensions and dividing the texture coords by the
2387 * texture dims to normalize.
2392 * If true, convert yuv to rgb.
2394 unsigned lower_y_uv_external
;
2395 unsigned lower_y_u_v_external
;
2396 unsigned lower_yx_xuxv_external
;
2399 * To emulate certain texture wrap modes, this can be used
2400 * to saturate the specified tex coord to [0.0, 1.0]. The
2401 * bits are according to sampler #, ie. if, for example:
2403 * (conf->saturate_s & (1 << n))
2405 * is true, then the s coord for sampler n is saturated.
2407 * Note that clamping must happen *after* projector lowering
2408 * so any projected texture sample instruction with a clamped
2409 * coordinate gets automatically lowered, regardless of the
2410 * 'lower_txp' setting.
2412 unsigned saturate_s
;
2413 unsigned saturate_t
;
2414 unsigned saturate_r
;
2416 /* Bitmask of textures that need swizzling.
2418 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2419 * swizzles[texture_index] is applied to the result of the texturing
2422 unsigned swizzle_result
;
2424 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2425 * while 4 and 5 represent 0 and 1 respectively.
2427 uint8_t swizzles
[32][4];
2430 * Bitmap of textures that need srgb to linear conversion. If
2431 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2432 * of the texture are lowered to linear.
2434 unsigned lower_srgb
;
2435 } nir_lower_tex_options
;
2437 bool nir_lower_tex(nir_shader
*shader
,
2438 const nir_lower_tex_options
*options
);
2440 bool nir_lower_idiv(nir_shader
*shader
);
2442 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2443 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2444 void nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2446 void nir_lower_two_sided_color(nir_shader
*shader
);
2448 void nir_lower_clamp_color_outputs(nir_shader
*shader
);
2450 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2452 typedef struct nir_lower_wpos_ytransform_options
{
2453 int state_tokens
[5];
2454 bool fs_coord_origin_upper_left
:1;
2455 bool fs_coord_origin_lower_left
:1;
2456 bool fs_coord_pixel_center_integer
:1;
2457 bool fs_coord_pixel_center_half_integer
:1;
2458 } nir_lower_wpos_ytransform_options
;
2460 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2461 const nir_lower_wpos_ytransform_options
*options
);
2462 bool nir_lower_wpos_center(nir_shader
*shader
);
2464 typedef struct nir_lower_drawpixels_options
{
2465 int texcoord_state_tokens
[5];
2466 int scale_state_tokens
[5];
2467 int bias_state_tokens
[5];
2468 unsigned drawpix_sampler
;
2469 unsigned pixelmap_sampler
;
2471 bool scale_and_bias
:1;
2472 } nir_lower_drawpixels_options
;
2474 void nir_lower_drawpixels(nir_shader
*shader
,
2475 const nir_lower_drawpixels_options
*options
);
2477 typedef struct nir_lower_bitmap_options
{
2480 } nir_lower_bitmap_options
;
2482 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2484 void nir_lower_atomics(nir_shader
*shader
,
2485 const struct gl_shader_program
*shader_program
);
2486 void nir_lower_to_source_mods(nir_shader
*shader
);
2488 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2491 nir_lower_drcp
= (1 << 0),
2492 nir_lower_dsqrt
= (1 << 1),
2493 nir_lower_drsq
= (1 << 2),
2494 nir_lower_dtrunc
= (1 << 3),
2495 nir_lower_dfloor
= (1 << 4),
2496 nir_lower_dceil
= (1 << 5),
2497 nir_lower_dfract
= (1 << 6),
2498 nir_lower_dround_even
= (1 << 7),
2499 nir_lower_dmod
= (1 << 8)
2500 } nir_lower_doubles_options
;
2502 void nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2503 void nir_lower_double_pack(nir_shader
*shader
);
2505 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2507 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2508 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2510 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2511 void nir_convert_to_ssa(nir_shader
*shader
);
2513 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2514 bool nir_repair_ssa(nir_shader
*shader
);
2516 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2517 * registers. If false, convert all values (even those not involved in a phi
2518 * node) to registers.
2520 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2522 bool nir_opt_algebraic(nir_shader
*shader
);
2523 bool nir_opt_algebraic_late(nir_shader
*shader
);
2524 bool nir_opt_constant_folding(nir_shader
*shader
);
2526 bool nir_opt_global_to_local(nir_shader
*shader
);
2528 bool nir_copy_prop(nir_shader
*shader
);
2530 bool nir_opt_cse(nir_shader
*shader
);
2532 bool nir_opt_dce(nir_shader
*shader
);
2534 bool nir_opt_dead_cf(nir_shader
*shader
);
2536 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2538 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2540 bool nir_opt_remove_phis(nir_shader
*shader
);
2542 bool nir_opt_undef(nir_shader
*shader
);
2544 bool nir_opt_conditional_discard(nir_shader
*shader
);
2546 void nir_sweep(nir_shader
*shader
);
2548 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2549 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);