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"
32 #include "GL/gl.h" /* GLenum */
33 #include "util/list.h"
34 #include "util/ralloc.h"
36 #include "util/bitset.h"
37 #include "nir_types.h"
38 #include "glsl/shader_enums.h"
41 #include "nir_opcodes.h"
48 struct gl_shader_program
;
51 #define NIR_TRUE (~0u)
53 /** Defines a cast function
55 * This macro defines a cast function from in_type to out_type where
56 * out_type is some structure type that contains a field of type out_type.
58 * Note that you have to be a bit careful as the generated cast function
61 #define NIR_DEFINE_CAST(name, in_type, out_type, field) \
62 static inline out_type * \
63 name(const in_type *parent) \
65 return exec_node_data(out_type, parent, field); \
68 struct nir_function_overload
;
75 * Description of built-in state associated with a uniform
77 * \sa nir_variable::state_slots
90 nir_var_shader_storage
,
95 * Data stored in an nir_constant
97 union nir_constant_data
{
104 typedef struct nir_constant
{
106 * Value of the constant.
108 * The field used to back the values supplied by the constant is determined
109 * by the type associated with the \c nir_variable. Constants may be
110 * scalars, vectors, or matrices.
112 union nir_constant_data value
;
114 /* Array elements / Structure Fields */
115 struct nir_constant
**elements
;
119 * \brief Layout qualifiers for gl_FragDepth.
121 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
122 * with a layout qualifier.
125 nir_depth_layout_none
, /**< No depth layout is specified. */
126 nir_depth_layout_any
,
127 nir_depth_layout_greater
,
128 nir_depth_layout_less
,
129 nir_depth_layout_unchanged
133 * Either a uniform, global variable, shader input, or shader output. Based on
134 * ir_variable - it should be easy to translate between the two.
138 struct exec_node node
;
141 * Declared type of the variable
143 const struct glsl_type
*type
;
146 * Declared name of the variable
151 * For variables which satisfy the is_interface_instance() predicate, this
152 * points to an array of integers such that if the ith member of the
153 * interface block is an array, max_ifc_array_access[i] is the maximum
154 * array element of that member that has been accessed. If the ith member
155 * of the interface block is not an array, max_ifc_array_access[i] is
158 * For variables whose type is not an interface block, this pointer is
161 unsigned *max_ifc_array_access
;
163 struct nir_variable_data
{
166 * Is the variable read-only?
168 * This is set for variables declared as \c const, shader inputs,
171 unsigned read_only
:1;
174 unsigned invariant
:1;
177 * Storage class of the variable.
179 * \sa nir_variable_mode
181 nir_variable_mode mode
:4;
184 * Interpolation mode for shader inputs / outputs
186 * \sa glsl_interp_qualifier
188 unsigned interpolation
:2;
191 * \name ARB_fragment_coord_conventions
194 unsigned origin_upper_left
:1;
195 unsigned pixel_center_integer
:1;
199 * Was the location explicitly set in the shader?
201 * If the location is explicitly set in the shader, it \b cannot be changed
202 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
205 unsigned explicit_location
:1;
206 unsigned explicit_index
:1;
209 * Was an initial binding explicitly set in the shader?
211 * If so, constant_initializer contains an integer nir_constant
212 * representing the initial binding point.
214 unsigned explicit_binding
:1;
217 * Does this variable have an initializer?
219 * This is used by the linker to cross-validiate initializers of global
222 unsigned has_initializer
:1;
225 * Is this variable a generic output or input that has not yet been matched
226 * up to a variable in another stage of the pipeline?
228 * This is used by the linker as scratch storage while assigning locations
229 * to generic inputs and outputs.
231 unsigned is_unmatched_generic_inout
:1;
234 * If non-zero, then this variable may be packed along with other variables
235 * into a single varying slot, so this offset should be applied when
236 * accessing components. For example, an offset of 1 means that the x
237 * component of this variable is actually stored in component y of the
238 * location specified by \c location.
240 unsigned location_frac
:2;
243 * Non-zero if this variable was created by lowering a named interface
244 * block which was not an array.
246 * Note that this variable and \c from_named_ifc_block_array will never
249 unsigned from_named_ifc_block_nonarray
:1;
252 * Non-zero if this variable was created by lowering a named interface
253 * block which was an array.
255 * Note that this variable and \c from_named_ifc_block_nonarray will never
258 unsigned from_named_ifc_block_array
:1;
261 * \brief Layout qualifier for gl_FragDepth.
263 * This is not equal to \c ir_depth_layout_none if and only if this
264 * variable is \c gl_FragDepth and a layout qualifier is specified.
266 nir_depth_layout depth_layout
;
269 * Storage location of the base of this variable
271 * The precise meaning of this field depends on the nature of the variable.
273 * - Vertex shader input: one of the values from \c gl_vert_attrib.
274 * - Vertex shader output: one of the values from \c gl_varying_slot.
275 * - Geometry shader input: one of the values from \c gl_varying_slot.
276 * - Geometry shader output: one of the values from \c gl_varying_slot.
277 * - Fragment shader input: one of the values from \c gl_varying_slot.
278 * - Fragment shader output: one of the values from \c gl_frag_result.
279 * - Uniforms: Per-stage uniform slot number for default uniform block.
280 * - Uniforms: Index within the uniform block definition for UBO members.
281 * - Non-UBO Uniforms: uniform slot number.
282 * - Other: This field is not currently used.
284 * If the variable is a uniform, shader input, or shader output, and the
285 * slot has not been assigned, the value will be -1.
290 * The actual location of the variable in the IR. Only valid for inputs
293 unsigned int driver_location
;
296 * output index for dual source blending.
301 * Descriptor set binding for sampler or UBO.
306 * Initial binding point for a sampler or UBO.
308 * For array types, this represents the binding point for the first element.
313 * Location an atomic counter is stored at.
316 unsigned buffer_index
;
321 * ARB_shader_image_load_store qualifiers.
324 bool read_only
; /**< "readonly" qualifier. */
325 bool write_only
; /**< "writeonly" qualifier. */
330 /** Image internal format if specified explicitly, otherwise GL_NONE. */
335 * Highest element accessed with a constant expression array index
337 * Not used for non-array variables.
339 unsigned max_array_access
;
344 * Built-in state that backs this uniform
346 * Once set at variable creation, \c state_slots must remain invariant.
347 * This is because, ideally, this array would be shared by all clones of
348 * this variable in the IR tree. In other words, we'd really like for it
349 * to be a fly-weight.
351 * If the variable is not a uniform, \c num_state_slots will be zero and
352 * \c state_slots will be \c NULL.
355 unsigned num_state_slots
; /**< Number of state slots used */
356 nir_state_slot
*state_slots
; /**< State descriptors. */
360 * Constant expression assigned in the initializer of the variable
362 nir_constant
*constant_initializer
;
365 * For variables that are in an interface block or are an instance of an
366 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
368 * \sa ir_variable::location
370 const struct glsl_type
*interface_type
;
374 struct exec_node node
;
376 unsigned num_components
; /** < number of vector components */
377 unsigned num_array_elems
; /** < size of array (0 for no array) */
379 /** generic register index. */
382 /** only for debug purposes, can be NULL */
385 /** whether this register is local (per-function) or global (per-shader) */
389 * If this flag is set to true, then accessing channels >= num_components
390 * is well-defined, and simply spills over to the next array element. This
391 * is useful for backends that can do per-component accessing, in
392 * particular scalar backends. By setting this flag and making
393 * num_components equal to 1, structures can be packed tightly into
394 * registers and then registers can be accessed per-component to get to
395 * each structure member, even if it crosses vec4 boundaries.
399 /** set of nir_instr's where this register is used (read from) */
400 struct list_head uses
;
402 /** set of nir_instr's where this register is defined (written to) */
403 struct list_head defs
;
405 /** set of nir_if's where this register is used as a condition */
406 struct list_head if_uses
;
413 nir_instr_type_intrinsic
,
414 nir_instr_type_load_const
,
416 nir_instr_type_ssa_undef
,
418 nir_instr_type_parallel_copy
,
421 typedef struct nir_instr
{
422 struct exec_node node
;
424 struct nir_block
*block
;
426 /** generic instruction index. */
429 /* A temporary for optimization and analysis passes to use for storing
430 * flags. For instance, DCE uses this to store the "dead/live" info.
435 static inline nir_instr
*
436 nir_instr_next(nir_instr
*instr
)
438 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
439 if (exec_node_is_tail_sentinel(next
))
442 return exec_node_data(nir_instr
, next
, node
);
445 static inline nir_instr
*
446 nir_instr_prev(nir_instr
*instr
)
448 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
449 if (exec_node_is_head_sentinel(prev
))
452 return exec_node_data(nir_instr
, prev
, node
);
456 nir_instr_is_first(nir_instr
*instr
)
458 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
462 nir_instr_is_last(nir_instr
*instr
)
464 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
468 /** for debugging only, can be NULL */
471 /** generic SSA definition index. */
474 /** Index into the live_in and live_out bitfields */
477 nir_instr
*parent_instr
;
479 /** set of nir_instr's where this register is used (read from) */
480 struct list_head uses
;
482 /** set of nir_if's where this register is used as a condition */
483 struct list_head if_uses
;
485 uint8_t num_components
;
492 struct nir_src
*indirect
; /** < NULL for no indirect offset */
493 unsigned base_offset
;
495 /* TODO use-def chain goes here */
499 nir_instr
*parent_instr
;
500 struct list_head def_link
;
503 struct nir_src
*indirect
; /** < NULL for no indirect offset */
504 unsigned base_offset
;
506 /* TODO def-use chain goes here */
511 typedef struct nir_src
{
513 nir_instr
*parent_instr
;
514 struct nir_if
*parent_if
;
517 struct list_head use_link
;
528 # define NIR_SRC_INIT nir_src()
530 # define NIR_SRC_INIT (nir_src) { { NULL } }
533 #define nir_foreach_use(reg_or_ssa_def, src) \
534 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
536 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
537 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
539 #define nir_foreach_if_use(reg_or_ssa_def, src) \
540 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
542 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
543 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
555 # define NIR_DEST_INIT nir_dest()
557 # define NIR_DEST_INIT (nir_dest) { { { NULL } } }
560 #define nir_foreach_def(reg, dest) \
561 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
563 #define nir_foreach_def_safe(reg, dest) \
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 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
601 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
607 * \name input modifiers
611 * For inputs interpreted as floating point, flips the sign bit. For
612 * inputs interpreted as integers, performs the two's complement negation.
617 * Clears the sign bit for floating point values, and computes the integer
618 * absolute value for integers. Note that the negate modifier acts after
619 * the absolute value modifier, therefore if both are set then all inputs
620 * will become negative.
626 * For each input component, says which component of the register it is
627 * chosen from. Note that which elements of the swizzle are used and which
628 * are ignored are based on the write mask for most opcodes - for example,
629 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
630 * a swizzle of {2, x, 1, 0} where x means "don't care."
639 * \name saturate output modifier
641 * Only valid for opcodes that output floating-point numbers. Clamps the
642 * output to between 0.0 and 1.0 inclusive.
647 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
651 nir_type_invalid
= 0, /* Not a valid type */
659 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
660 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
661 } nir_op_algebraic_property
;
669 * The number of components in the output
671 * If non-zero, this is the size of the output and input sizes are
672 * explicitly given; swizzle and writemask are still in effect, but if
673 * the output component is masked out, then the input component may
676 * If zero, the opcode acts in the standard, per-component manner; the
677 * operation is performed on each component (except the ones that are
678 * masked out) with the input being taken from the input swizzle for
681 * The size of some of the inputs may be given (i.e. non-zero) even
682 * though output_size is zero; in that case, the inputs with a zero
683 * size act per-component, while the inputs with non-zero size don't.
685 unsigned output_size
;
688 * The type of vector that the instruction outputs. Note that the
689 * staurate modifier is only allowed on outputs with the float type.
692 nir_alu_type output_type
;
695 * The number of components in each input
697 unsigned input_sizes
[4];
700 * The type of vector that each input takes. Note that negate and
701 * absolute value are only allowed on inputs with int or float type and
702 * behave differently on the two.
704 nir_alu_type input_types
[4];
706 nir_op_algebraic_property algebraic_properties
;
709 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
711 typedef struct nir_alu_instr
{
718 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
719 nir_alu_instr
*instr
);
720 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
721 nir_alu_instr
*instr
);
723 /* is this source channel used? */
725 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
727 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
728 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
730 return (instr
->dest
.write_mask
>> channel
) & 1;
734 * For instructions whose destinations are SSA, get the number of channels
737 static inline unsigned
738 nir_ssa_alu_instr_src_components(nir_alu_instr
*instr
, unsigned src
)
740 assert(instr
->dest
.dest
.is_ssa
);
742 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
743 return nir_op_infos
[instr
->op
].input_sizes
[src
];
745 return instr
->dest
.dest
.ssa
.num_components
;
750 nir_deref_type_array
,
751 nir_deref_type_struct
754 typedef struct nir_deref
{
755 nir_deref_type deref_type
;
756 struct nir_deref
*child
;
757 const struct glsl_type
*type
;
766 /* This enum describes how the array is referenced. If the deref is
767 * direct then the base_offset is used. If the deref is indirect then then
768 * offset is given by base_offset + indirect. If the deref is a wildcard
769 * then the deref refers to all of the elements of the array at the same
770 * time. Wildcard dereferences are only ever allowed in copy_var
771 * intrinsics and the source and destination derefs must have matching
775 nir_deref_array_type_direct
,
776 nir_deref_array_type_indirect
,
777 nir_deref_array_type_wildcard
,
778 } nir_deref_array_type
;
783 nir_deref_array_type deref_array_type
;
784 unsigned base_offset
;
794 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
795 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
796 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
798 /** Returns the tail of a deref chain */
799 static inline nir_deref
*
800 nir_deref_tail(nir_deref
*deref
)
803 deref
= deref
->child
;
811 nir_deref_var
**params
;
812 nir_deref_var
*return_deref
;
814 struct nir_function_overload
*callee
;
817 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
818 num_variables, num_indices, flags) \
819 nir_intrinsic_##name,
821 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
824 #include "nir_intrinsics.h"
825 nir_num_intrinsics
= nir_last_intrinsic
+ 1
829 #undef LAST_INTRINSIC
831 /** Represents an intrinsic
833 * An intrinsic is an instruction type for handling things that are
834 * more-or-less regular operations but don't just consume and produce SSA
835 * values like ALU operations do. Intrinsics are not for things that have
836 * special semantic meaning such as phi nodes and parallel copies.
837 * Examples of intrinsics include variable load/store operations, system
838 * value loads, and the like. Even though texturing more-or-less falls
839 * under this category, texturing is its own instruction type because
840 * trying to represent texturing with intrinsics would lead to a
841 * combinatorial explosion of intrinsic opcodes.
843 * By having a single instruction type for handling a lot of different
844 * cases, optimization passes can look for intrinsics and, for the most
845 * part, completely ignore them. Each intrinsic type also has a few
846 * possible flags that govern whether or not they can be reordered or
847 * eliminated. That way passes like dead code elimination can still work
848 * on intrisics without understanding the meaning of each.
850 * Each intrinsic has some number of constant indices, some number of
851 * variables, and some number of sources. What these sources, variables,
852 * and indices mean depends on the intrinsic and is documented with the
853 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
854 * instructions are the only types of instruction that can operate on
860 nir_intrinsic_op intrinsic
;
864 /** number of components if this is a vectorized intrinsic
866 * Similarly to ALU operations, some intrinsics are vectorized.
867 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
868 * For vectorized intrinsics, the num_components field specifies the
869 * number of destination components and the number of source components
870 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
872 uint8_t num_components
;
876 nir_deref_var
*variables
[2];
879 } nir_intrinsic_instr
;
882 * \name NIR intrinsics semantic flags
884 * information about what the compiler can do with the intrinsics.
886 * \sa nir_intrinsic_info::flags
890 * whether the intrinsic can be safely eliminated if none of its output
891 * value is not being used.
893 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
896 * Whether the intrinsic can be reordered with respect to any other
897 * intrinsic, i.e. whether the only reordering dependencies of the
898 * intrinsic are due to the register reads/writes.
900 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
901 } nir_intrinsic_semantic_flag
;
903 #define NIR_INTRINSIC_MAX_INPUTS 4
908 unsigned num_srcs
; /** < number of register/SSA inputs */
910 /** number of components of each input register
912 * If this value is 0, the number of components is given by the
913 * num_components field of nir_intrinsic_instr.
915 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
919 /** number of components of the output register
921 * If this value is 0, the number of components is given by the
922 * num_components field of nir_intrinsic_instr.
924 unsigned dest_components
;
926 /** the number of inputs/outputs that are variables */
927 unsigned num_variables
;
929 /** the number of constant indices used by the intrinsic */
930 unsigned num_indices
;
932 /** semantic flags for calls to this intrinsic */
933 nir_intrinsic_semantic_flag flags
;
934 } nir_intrinsic_info
;
936 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
939 * \group texture information
941 * This gives semantic information about textures which is useful to the
942 * frontend, the backend, and lowering passes, but not the optimizer.
947 nir_tex_src_projector
,
948 nir_tex_src_comparitor
, /* shadow comparitor */
952 nir_tex_src_ms_index
, /* MSAA sample index */
955 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
956 nir_num_tex_src_types
961 nir_tex_src_type src_type
;
965 nir_texop_tex
, /**< Regular texture look-up */
966 nir_texop_txb
, /**< Texture look-up with LOD bias */
967 nir_texop_txl
, /**< Texture look-up with explicit LOD */
968 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
969 nir_texop_txf
, /**< Texel fetch with explicit LOD */
970 nir_texop_txf_ms
, /**< Multisample texture fetch */
971 nir_texop_txs
, /**< Texture size */
972 nir_texop_lod
, /**< Texture lod query */
973 nir_texop_tg4
, /**< Texture gather */
974 nir_texop_query_levels
, /**< Texture levels query */
975 nir_texop_texture_samples
, /**< Texture samples query */
981 enum glsl_sampler_dim sampler_dim
;
982 nir_alu_type dest_type
;
987 unsigned num_srcs
, coord_components
;
988 bool is_array
, is_shadow
;
991 * If is_shadow is true, whether this is the old-style shadow that outputs 4
992 * components or the new-style shadow that outputs 1 component.
994 bool is_new_style_shadow
;
996 /* constant offset - must be 0 if the offset source is used */
999 /* gather component selector */
1000 unsigned component
: 2;
1002 /* The descriptor set containing this texture */
1003 unsigned sampler_set
;
1005 /** The sampler index
1007 * If this texture instruction has a nir_tex_src_sampler_offset source,
1008 * then the sampler index is given by sampler_index + sampler_offset.
1010 unsigned sampler_index
;
1012 /** The size of the sampler array or 0 if it's not an array */
1013 unsigned sampler_array_size
;
1015 nir_deref_var
*sampler
; /* if this is NULL, use sampler_index instead */
1018 static inline unsigned
1019 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1021 switch (instr
->op
) {
1022 case nir_texop_txs
: {
1024 switch (instr
->sampler_dim
) {
1025 case GLSL_SAMPLER_DIM_1D
:
1026 case GLSL_SAMPLER_DIM_BUF
:
1029 case GLSL_SAMPLER_DIM_2D
:
1030 case GLSL_SAMPLER_DIM_CUBE
:
1031 case GLSL_SAMPLER_DIM_MS
:
1032 case GLSL_SAMPLER_DIM_RECT
:
1033 case GLSL_SAMPLER_DIM_EXTERNAL
:
1036 case GLSL_SAMPLER_DIM_3D
:
1040 unreachable("not reached");
1042 if (instr
->is_array
)
1050 case nir_texop_texture_samples
:
1051 case nir_texop_query_levels
:
1055 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1062 static inline unsigned
1063 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1065 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1066 return instr
->coord_components
;
1069 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1070 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1071 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1072 if (instr
->is_array
)
1073 return instr
->coord_components
- 1;
1075 return instr
->coord_components
;
1082 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1084 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1085 if (instr
->src
[i
].src_type
== type
)
1102 nir_const_value value
;
1105 } nir_load_const_instr
;
1118 /* creates a new SSA variable in an undefined state */
1123 } nir_ssa_undef_instr
;
1126 struct exec_node node
;
1128 /* The predecessor block corresponding to this source */
1129 struct nir_block
*pred
;
1134 #define nir_foreach_phi_src(phi, entry) \
1135 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1136 #define nir_foreach_phi_src_safe(phi, entry) \
1137 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1142 struct exec_list srcs
; /** < list of nir_phi_src */
1148 struct exec_node node
;
1151 } nir_parallel_copy_entry
;
1153 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1154 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1159 /* A list of nir_parallel_copy_entry's. The sources of all of the
1160 * entries are copied to the corresponding destinations "in parallel".
1161 * In other words, if we have two entries: a -> b and b -> a, the values
1164 struct exec_list entries
;
1165 } nir_parallel_copy_instr
;
1167 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1168 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1169 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1170 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1171 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1172 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1173 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1174 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1175 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1176 nir_parallel_copy_instr
, instr
)
1181 * Control flow consists of a tree of control flow nodes, which include
1182 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1183 * instructions that always run start-to-finish. Each basic block also keeps
1184 * track of its successors (blocks which may run immediately after the current
1185 * block) and predecessors (blocks which could have run immediately before the
1186 * current block). Each function also has a start block and an end block which
1187 * all return statements point to (which is always empty). Together, all the
1188 * blocks with their predecessors and successors make up the control flow
1189 * graph (CFG) of the function. There are helpers that modify the tree of
1190 * control flow nodes while modifying the CFG appropriately; these should be
1191 * used instead of modifying the tree directly.
1198 nir_cf_node_function
1201 typedef struct nir_cf_node
{
1202 struct exec_node node
;
1203 nir_cf_node_type type
;
1204 struct nir_cf_node
*parent
;
1207 typedef struct nir_block
{
1208 nir_cf_node cf_node
;
1210 struct exec_list instr_list
; /** < list of nir_instr */
1212 /** generic block index; generated by nir_index_blocks */
1216 * Each block can only have up to 2 successors, so we put them in a simple
1217 * array - no need for anything more complicated.
1219 struct nir_block
*successors
[2];
1221 /* Set of nir_block predecessors in the CFG */
1222 struct set
*predecessors
;
1225 * this node's immediate dominator in the dominance tree - set to NULL for
1228 struct nir_block
*imm_dom
;
1230 /* This node's children in the dominance tree */
1231 unsigned num_dom_children
;
1232 struct nir_block
**dom_children
;
1234 /* Set of nir_block's on the dominance frontier of this block */
1235 struct set
*dom_frontier
;
1238 * These two indices have the property that dom_{pre,post}_index for each
1239 * child of this block in the dominance tree will always be between
1240 * dom_pre_index and dom_post_index for this block, which makes testing if
1241 * a given block is dominated by another block an O(1) operation.
1243 unsigned dom_pre_index
, dom_post_index
;
1245 /* live in and out for this block; used for liveness analysis */
1246 BITSET_WORD
*live_in
;
1247 BITSET_WORD
*live_out
;
1250 static inline nir_instr
*
1251 nir_block_first_instr(nir_block
*block
)
1253 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1254 return exec_node_data(nir_instr
, head
, node
);
1257 static inline nir_instr
*
1258 nir_block_last_instr(nir_block
*block
)
1260 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1261 return exec_node_data(nir_instr
, tail
, node
);
1264 #define nir_foreach_instr(block, instr) \
1265 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1266 #define nir_foreach_instr_reverse(block, instr) \
1267 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1268 #define nir_foreach_instr_safe(block, instr) \
1269 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1270 #define nir_foreach_instr_safe_reverse(block, instr) \
1271 foreach_list_typed_safe_reverse(nir_instr, instr, node, &(block)->instr_list)
1273 typedef struct nir_if
{
1274 nir_cf_node cf_node
;
1277 struct exec_list then_list
; /** < list of nir_cf_node */
1278 struct exec_list else_list
; /** < list of nir_cf_node */
1281 static inline nir_cf_node
*
1282 nir_if_first_then_node(nir_if
*if_stmt
)
1284 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1285 return exec_node_data(nir_cf_node
, head
, node
);
1288 static inline nir_cf_node
*
1289 nir_if_last_then_node(nir_if
*if_stmt
)
1291 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1292 return exec_node_data(nir_cf_node
, tail
, node
);
1295 static inline nir_cf_node
*
1296 nir_if_first_else_node(nir_if
*if_stmt
)
1298 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1299 return exec_node_data(nir_cf_node
, head
, node
);
1302 static inline nir_cf_node
*
1303 nir_if_last_else_node(nir_if
*if_stmt
)
1305 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1306 return exec_node_data(nir_cf_node
, tail
, node
);
1310 nir_cf_node cf_node
;
1312 struct exec_list body
; /** < list of nir_cf_node */
1315 static inline nir_cf_node
*
1316 nir_loop_first_cf_node(nir_loop
*loop
)
1318 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1321 static inline nir_cf_node
*
1322 nir_loop_last_cf_node(nir_loop
*loop
)
1324 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1328 * Various bits of metadata that can may be created or required by
1329 * optimization and analysis passes
1332 nir_metadata_none
= 0x0,
1333 nir_metadata_block_index
= 0x1,
1334 nir_metadata_dominance
= 0x2,
1335 nir_metadata_live_variables
= 0x4,
1339 nir_cf_node cf_node
;
1341 /** pointer to the overload of which this is an implementation */
1342 struct nir_function_overload
*overload
;
1344 struct exec_list body
; /** < list of nir_cf_node */
1346 nir_block
*end_block
;
1348 /** list for all local variables in the function */
1349 struct exec_list locals
;
1351 /** array of variables used as parameters */
1352 unsigned num_params
;
1353 nir_variable
**params
;
1355 /** variable used to hold the result of the function */
1356 nir_variable
*return_var
;
1358 /** list of local registers in the function */
1359 struct exec_list registers
;
1361 /** next available local register index */
1364 /** next available SSA value index */
1367 /* total number of basic blocks, only valid when block_index_dirty = false */
1368 unsigned num_blocks
;
1370 nir_metadata valid_metadata
;
1371 } nir_function_impl
;
1373 static inline nir_block
*
1374 nir_start_block(nir_function_impl
*impl
)
1376 return (nir_block
*) exec_list_get_head(&impl
->body
);
1379 static inline nir_cf_node
*
1380 nir_cf_node_next(nir_cf_node
*node
)
1382 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1383 if (exec_node_is_tail_sentinel(next
))
1386 return exec_node_data(nir_cf_node
, next
, node
);
1389 static inline nir_cf_node
*
1390 nir_cf_node_prev(nir_cf_node
*node
)
1392 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1393 if (exec_node_is_head_sentinel(prev
))
1396 return exec_node_data(nir_cf_node
, prev
, node
);
1400 nir_cf_node_is_first(const nir_cf_node
*node
)
1402 return exec_node_is_head_sentinel(node
->node
.prev
);
1406 nir_cf_node_is_last(const nir_cf_node
*node
)
1408 return exec_node_is_tail_sentinel(node
->node
.next
);
1411 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1412 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1413 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1414 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1419 nir_parameter_inout
,
1420 } nir_parameter_type
;
1423 nir_parameter_type param_type
;
1424 const struct glsl_type
*type
;
1427 typedef struct nir_function_overload
{
1428 struct exec_node node
;
1430 unsigned num_params
;
1431 nir_parameter
*params
;
1432 const struct glsl_type
*return_type
;
1434 nir_function_impl
*impl
; /** < NULL if the overload is only declared yet */
1436 /** pointer to the function of which this is an overload */
1437 struct nir_function
*function
;
1438 } nir_function_overload
;
1440 typedef struct nir_function
{
1441 struct exec_node node
;
1443 struct exec_list overload_list
; /** < list of nir_function_overload */
1445 struct nir_shader
*shader
;
1448 #define nir_function_first_overload(func) \
1449 exec_node_data(nir_function_overload, \
1450 exec_list_get_head(&(func)->overload_list), node)
1452 typedef struct nir_shader_compiler_options
{
1459 /** lowers fneg and ineg to fsub and isub. */
1461 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1464 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1467 /* Does the native fdot instruction replicate its result for four
1468 * components? If so, then opt_algebraic_late will turn all fdotN
1469 * instructions into fdot_replicatedN instructions.
1471 bool fdot_replicates
;
1473 /** lowers ffract to fsub+ffloor: */
1477 * Does the driver support real 32-bit integers? (Otherwise, integers
1478 * are simulated by floats.)
1480 bool native_integers
;
1481 } nir_shader_compiler_options
;
1483 typedef struct nir_shader
{
1484 /** list of uniforms (nir_variable) */
1485 struct exec_list uniforms
;
1487 /** list of inputs (nir_variable) */
1488 struct exec_list inputs
;
1490 /** list of outputs (nir_variable) */
1491 struct exec_list outputs
;
1493 /** Set of driver-specific options for the shader.
1495 * The memory for the options is expected to be kept in a single static
1496 * copy by the driver.
1498 const struct nir_shader_compiler_options
*options
;
1500 /** list of global variables in the shader (nir_variable) */
1501 struct exec_list globals
;
1503 /** list of system value variables in the shader (nir_variable) */
1504 struct exec_list system_values
;
1506 struct exec_list functions
; /** < list of nir_function */
1508 /** list of global register in the shader */
1509 struct exec_list registers
;
1511 /** next available global register index */
1515 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1518 unsigned num_inputs
, num_uniforms
, num_outputs
;
1520 /** The shader stage, such as MESA_SHADER_VERTEX. */
1521 gl_shader_stage stage
;
1524 /** The maximum number of vertices the geometry shader might write. */
1525 unsigned vertices_out
;
1527 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1528 unsigned invocations
;
1532 #define nir_foreach_overload(shader, overload) \
1533 foreach_list_typed(nir_function, func, node, &(shader)->functions) \
1534 foreach_list_typed(nir_function_overload, overload, node, \
1535 &(func)->overload_list)
1537 nir_shader
*nir_shader_create(void *mem_ctx
,
1538 gl_shader_stage stage
,
1539 const nir_shader_compiler_options
*options
);
1541 /** creates a register, including assigning it an index and adding it to the list */
1542 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1544 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1546 void nir_reg_remove(nir_register
*reg
);
1548 /** creates a function and adds it to the shader's list of functions */
1549 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1551 /** creates a null function returning null */
1552 nir_function_overload
*nir_function_overload_create(nir_function
*func
);
1554 nir_function_impl
*nir_function_impl_create(nir_function_overload
*func
);
1556 nir_block
*nir_block_create(void *mem_ctx
);
1557 nir_if
*nir_if_create(void *mem_ctx
);
1558 nir_loop
*nir_loop_create(void *mem_ctx
);
1560 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1562 /** requests that the given pieces of metadata be generated */
1563 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1564 /** dirties all but the preserved metadata */
1565 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1567 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1568 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1570 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1572 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1573 unsigned num_components
);
1575 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1576 nir_intrinsic_op op
);
1578 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1579 nir_function_overload
*callee
);
1581 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1583 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1585 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1587 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1588 unsigned num_components
);
1590 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1591 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1592 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1594 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1596 nir_load_const_instr
*
1597 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1600 * NIR Cursors and Instruction Insertion API
1603 * A tiny struct representing a point to insert/extract instructions or
1604 * control flow nodes. Helps reduce the combinatorial explosion of possible
1605 * points to insert/extract.
1607 * \sa nir_control_flow.h
1610 nir_cursor_before_block
,
1611 nir_cursor_after_block
,
1612 nir_cursor_before_instr
,
1613 nir_cursor_after_instr
,
1614 } nir_cursor_option
;
1617 nir_cursor_option option
;
1624 static inline nir_block
*
1625 nir_cursor_current_block(nir_cursor cursor
)
1627 if (cursor
.option
== nir_cursor_before_instr
||
1628 cursor
.option
== nir_cursor_after_instr
) {
1629 return cursor
.instr
->block
;
1631 return cursor
.block
;
1635 static inline nir_cursor
1636 nir_before_block(nir_block
*block
)
1639 cursor
.option
= nir_cursor_before_block
;
1640 cursor
.block
= block
;
1644 static inline nir_cursor
1645 nir_after_block(nir_block
*block
)
1648 cursor
.option
= nir_cursor_after_block
;
1649 cursor
.block
= block
;
1653 static inline nir_cursor
1654 nir_before_instr(nir_instr
*instr
)
1657 cursor
.option
= nir_cursor_before_instr
;
1658 cursor
.instr
= instr
;
1662 static inline nir_cursor
1663 nir_after_instr(nir_instr
*instr
)
1666 cursor
.option
= nir_cursor_after_instr
;
1667 cursor
.instr
= instr
;
1671 static inline nir_cursor
1672 nir_after_block_before_jump(nir_block
*block
)
1674 nir_instr
*last_instr
= nir_block_last_instr(block
);
1675 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1676 return nir_before_instr(last_instr
);
1678 return nir_after_block(block
);
1682 static inline nir_cursor
1683 nir_before_cf_node(nir_cf_node
*node
)
1685 if (node
->type
== nir_cf_node_block
)
1686 return nir_before_block(nir_cf_node_as_block(node
));
1688 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1691 static inline nir_cursor
1692 nir_after_cf_node(nir_cf_node
*node
)
1694 if (node
->type
== nir_cf_node_block
)
1695 return nir_after_block(nir_cf_node_as_block(node
));
1697 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1700 static inline nir_cursor
1701 nir_before_cf_list(struct exec_list
*cf_list
)
1703 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1704 exec_list_get_head(cf_list
), node
);
1705 return nir_before_cf_node(first_node
);
1708 static inline nir_cursor
1709 nir_after_cf_list(struct exec_list
*cf_list
)
1711 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1712 exec_list_get_tail(cf_list
), node
);
1713 return nir_after_cf_node(last_node
);
1717 * Insert a NIR instruction at the given cursor.
1719 * Note: This does not update the cursor.
1721 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
1724 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1726 nir_instr_insert(nir_before_instr(instr
), before
);
1730 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
1732 nir_instr_insert(nir_after_instr(instr
), after
);
1736 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
1738 nir_instr_insert(nir_before_block(block
), before
);
1742 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
1744 nir_instr_insert(nir_after_block(block
), after
);
1748 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
1750 nir_instr_insert(nir_before_cf_node(node
), before
);
1754 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
1756 nir_instr_insert(nir_after_cf_node(node
), after
);
1760 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
1762 nir_instr_insert(nir_before_cf_list(list
), before
);
1766 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
1768 nir_instr_insert(nir_after_cf_list(list
), after
);
1771 void nir_instr_remove(nir_instr
*instr
);
1775 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
1776 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
1777 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
1778 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
1780 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
1781 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
1783 nir_const_value
*nir_src_as_const_value(nir_src src
);
1784 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
1785 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
1786 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
1787 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
1788 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
1791 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1792 unsigned num_components
, const char *name
);
1793 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1794 unsigned num_components
, const char *name
);
1795 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
1797 /* visits basic blocks in source-code order */
1798 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
1799 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1801 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1803 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1806 /* If the following CF node is an if, this function returns that if.
1807 * Otherwise, it returns NULL.
1809 nir_if
*nir_block_get_following_if(nir_block
*block
);
1811 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
1813 void nir_index_local_regs(nir_function_impl
*impl
);
1814 void nir_index_global_regs(nir_shader
*shader
);
1815 void nir_index_ssa_defs(nir_function_impl
*impl
);
1816 unsigned nir_index_instrs(nir_function_impl
*impl
);
1818 void nir_index_blocks(nir_function_impl
*impl
);
1820 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
1821 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
1824 void nir_validate_shader(nir_shader
*shader
);
1826 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
1829 void nir_calc_dominance_impl(nir_function_impl
*impl
);
1830 void nir_calc_dominance(nir_shader
*shader
);
1832 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
1833 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
1835 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
1836 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
1838 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
1839 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
1841 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
1842 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
1844 int nir_gs_count_vertices(nir_shader
*shader
);
1846 bool nir_split_var_copies(nir_shader
*shader
);
1848 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
1849 void nir_lower_var_copies(nir_shader
*shader
);
1851 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
1853 bool nir_lower_locals_to_regs(nir_shader
*shader
);
1855 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
1857 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
1859 void nir_assign_var_locations(struct exec_list
*var_list
,
1861 int (*type_size
)(const struct glsl_type
*));
1863 void nir_lower_io(nir_shader
*shader
,
1864 int (*type_size
)(const struct glsl_type
*));
1865 void nir_lower_vars_to_ssa(nir_shader
*shader
);
1867 bool nir_remove_dead_variables(nir_shader
*shader
);
1869 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
1870 bool nir_lower_vec_to_movs(nir_shader
*shader
);
1871 void nir_lower_alu_to_scalar(nir_shader
*shader
);
1872 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
1874 void nir_lower_phis_to_scalar(nir_shader
*shader
);
1876 void nir_lower_samplers(nir_shader
*shader
,
1877 const struct gl_shader_program
*shader_program
);
1878 void nir_lower_samplers_for_vk(nir_shader
*shader
);
1880 bool nir_lower_system_values(nir_shader
*shader
);
1882 typedef struct nir_lower_tex_options
{
1884 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
1885 * sampler types a texture projector is lowered.
1890 * If true, lower rect textures to 2D, using txs to fetch the
1891 * texture dimensions and dividing the texture coords by the
1892 * texture dims to normalize.
1897 * To emulate certain texture wrap modes, this can be used
1898 * to saturate the specified tex coord to [0.0, 1.0]. The
1899 * bits are according to sampler #, ie. if, for example:
1901 * (conf->saturate_s & (1 << n))
1903 * is true, then the s coord for sampler n is saturated.
1905 * Note that clamping must happen *after* projector lowering
1906 * so any projected texture sample instruction with a clamped
1907 * coordinate gets automatically lowered, regardless of the
1908 * 'lower_txp' setting.
1910 unsigned saturate_s
;
1911 unsigned saturate_t
;
1912 unsigned saturate_r
;
1913 } nir_lower_tex_options
;
1915 void nir_lower_tex(nir_shader
*shader
,
1916 const nir_lower_tex_options
*options
);
1918 void nir_lower_idiv(nir_shader
*shader
);
1920 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
1921 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
1923 void nir_lower_two_sided_color(nir_shader
*shader
);
1925 void nir_lower_atomics(nir_shader
*shader
);
1926 void nir_lower_to_source_mods(nir_shader
*shader
);
1928 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
1930 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
1932 void nir_live_variables_impl(nir_function_impl
*impl
);
1933 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
1935 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
1936 void nir_convert_to_ssa(nir_shader
*shader
);
1938 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
1939 * registers. If false, convert all values (even those not involved in a phi
1940 * node) to registers.
1942 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
1944 bool nir_opt_algebraic(nir_shader
*shader
);
1945 bool nir_opt_algebraic_late(nir_shader
*shader
);
1946 bool nir_opt_constant_folding(nir_shader
*shader
);
1948 bool nir_opt_global_to_local(nir_shader
*shader
);
1950 bool nir_copy_prop_impl(nir_function_impl
*impl
);
1951 bool nir_copy_prop(nir_shader
*shader
);
1953 bool nir_opt_cse(nir_shader
*shader
);
1955 bool nir_opt_dce_impl(nir_function_impl
*impl
);
1956 bool nir_opt_dce(nir_shader
*shader
);
1958 bool nir_opt_dead_cf(nir_shader
*shader
);
1960 void nir_opt_gcm(nir_shader
*shader
);
1962 bool nir_opt_peephole_select(nir_shader
*shader
);
1963 bool nir_opt_peephole_ffma(nir_shader
*shader
);
1965 bool nir_opt_remove_phis(nir_shader
*shader
);
1967 bool nir_opt_undef(nir_shader
*shader
);
1969 void nir_sweep(nir_shader
*shader
);
1971 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
1972 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);