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 "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
91 nir_var_shader_storage
,
96 * Data stored in an nir_constant
98 union nir_constant_data
{
105 typedef struct nir_constant
{
107 * Value of the constant.
109 * The field used to back the values supplied by the constant is determined
110 * by the type associated with the \c nir_variable. Constants may be
111 * scalars, vectors, or matrices.
113 union nir_constant_data value
;
115 /* we could get this from the var->type but makes clone *much* easier to
116 * not have to care about the type.
118 unsigned num_elements
;
120 /* Array elements / Structure Fields */
121 struct nir_constant
**elements
;
125 * \brief Layout qualifiers for gl_FragDepth.
127 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
128 * with a layout qualifier.
131 nir_depth_layout_none
, /**< No depth layout is specified. */
132 nir_depth_layout_any
,
133 nir_depth_layout_greater
,
134 nir_depth_layout_less
,
135 nir_depth_layout_unchanged
139 * Either a uniform, global variable, shader input, or shader output. Based on
140 * ir_variable - it should be easy to translate between the two.
144 struct exec_node node
;
147 * Declared type of the variable
149 const struct glsl_type
*type
;
152 * Declared name of the variable
156 struct nir_variable_data
{
159 * Is the variable read-only?
161 * This is set for variables declared as \c const, shader inputs,
164 unsigned read_only
:1;
168 unsigned invariant
:1;
171 * Storage class of the variable.
173 * \sa nir_variable_mode
175 nir_variable_mode mode
:4;
178 * Interpolation mode for shader inputs / outputs
180 * \sa glsl_interp_qualifier
182 unsigned interpolation
:2;
185 * \name ARB_fragment_coord_conventions
188 unsigned origin_upper_left
:1;
189 unsigned pixel_center_integer
:1;
193 * Was the location explicitly set in the shader?
195 * If the location is explicitly set in the shader, it \b cannot be changed
196 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
199 unsigned explicit_location
:1;
200 unsigned explicit_index
:1;
203 * Was an initial binding explicitly set in the shader?
205 * If so, constant_initializer contains an integer nir_constant
206 * representing the initial binding point.
208 unsigned explicit_binding
:1;
211 * Does this variable have an initializer?
213 * This is used by the linker to cross-validiate initializers of global
216 unsigned has_initializer
:1;
219 * Is this variable a generic output or input that has not yet been matched
220 * up to a variable in another stage of the pipeline?
222 * This is used by the linker as scratch storage while assigning locations
223 * to generic inputs and outputs.
225 unsigned is_unmatched_generic_inout
:1;
228 * If non-zero, then this variable may be packed along with other variables
229 * into a single varying slot, so this offset should be applied when
230 * accessing components. For example, an offset of 1 means that the x
231 * component of this variable is actually stored in component y of the
232 * location specified by \c location.
234 unsigned location_frac
:2;
237 * Non-zero if this variable was created by lowering a named interface
238 * block which was not an array.
240 * Note that this variable and \c from_named_ifc_block_array will never
243 unsigned from_named_ifc_block_nonarray
:1;
246 * Non-zero if this variable was created by lowering a named interface
247 * block which was an array.
249 * Note that this variable and \c from_named_ifc_block_nonarray will never
252 unsigned from_named_ifc_block_array
:1;
255 * \brief Layout qualifier for gl_FragDepth.
257 * This is not equal to \c ir_depth_layout_none if and only if this
258 * variable is \c gl_FragDepth and a layout qualifier is specified.
260 nir_depth_layout depth_layout
;
263 * Storage location of the base of this variable
265 * The precise meaning of this field depends on the nature of the variable.
267 * - Vertex shader input: one of the values from \c gl_vert_attrib.
268 * - Vertex shader output: one of the values from \c gl_varying_slot.
269 * - Geometry shader input: one of the values from \c gl_varying_slot.
270 * - Geometry shader output: one of the values from \c gl_varying_slot.
271 * - Fragment shader input: one of the values from \c gl_varying_slot.
272 * - Fragment shader output: one of the values from \c gl_frag_result.
273 * - Uniforms: Per-stage uniform slot number for default uniform block.
274 * - Uniforms: Index within the uniform block definition for UBO members.
275 * - Non-UBO Uniforms: uniform slot number.
276 * - Other: This field is not currently used.
278 * If the variable is a uniform, shader input, or shader output, and the
279 * slot has not been assigned, the value will be -1.
284 * The actual location of the variable in the IR. Only valid for inputs
287 unsigned int driver_location
;
290 * output index for dual source blending.
295 * Descriptor set binding for sampler or UBO.
300 * Initial binding point for a sampler or UBO.
302 * For array types, this represents the binding point for the first element.
307 * Location an atomic counter is stored at.
314 * ARB_shader_image_load_store qualifiers.
317 bool read_only
; /**< "readonly" qualifier. */
318 bool write_only
; /**< "writeonly" qualifier. */
323 /** Image internal format if specified explicitly, otherwise GL_NONE. */
328 * Highest element accessed with a constant expression array index
330 * Not used for non-array variables.
332 unsigned max_array_access
;
337 * Built-in state that backs this uniform
339 * Once set at variable creation, \c state_slots must remain invariant.
340 * This is because, ideally, this array would be shared by all clones of
341 * this variable in the IR tree. In other words, we'd really like for it
342 * to be a fly-weight.
344 * If the variable is not a uniform, \c num_state_slots will be zero and
345 * \c state_slots will be \c NULL.
348 unsigned num_state_slots
; /**< Number of state slots used */
349 nir_state_slot
*state_slots
; /**< State descriptors. */
353 * Constant expression assigned in the initializer of the variable
355 nir_constant
*constant_initializer
;
358 * For variables that are in an interface block or are an instance of an
359 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
361 * \sa ir_variable::location
363 const struct glsl_type
*interface_type
;
366 #define nir_foreach_variable(var, var_list) \
367 foreach_list_typed(nir_variable, var, node, var_list)
370 struct exec_node node
;
372 unsigned num_components
; /** < number of vector components */
373 unsigned num_array_elems
; /** < size of array (0 for no array) */
375 /** generic register index. */
378 /** only for debug purposes, can be NULL */
381 /** whether this register is local (per-function) or global (per-shader) */
385 * If this flag is set to true, then accessing channels >= num_components
386 * is well-defined, and simply spills over to the next array element. This
387 * is useful for backends that can do per-component accessing, in
388 * particular scalar backends. By setting this flag and making
389 * num_components equal to 1, structures can be packed tightly into
390 * registers and then registers can be accessed per-component to get to
391 * each structure member, even if it crosses vec4 boundaries.
395 /** set of nir_src's where this register is used (read from) */
396 struct list_head uses
;
398 /** set of nir_dest's where this register is defined (written to) */
399 struct list_head defs
;
401 /** set of nir_if's where this register is used as a condition */
402 struct list_head if_uses
;
409 nir_instr_type_intrinsic
,
410 nir_instr_type_load_const
,
412 nir_instr_type_ssa_undef
,
414 nir_instr_type_parallel_copy
,
417 typedef struct nir_instr
{
418 struct exec_node node
;
420 struct nir_block
*block
;
422 /** generic instruction index. */
425 /* A temporary for optimization and analysis passes to use for storing
426 * flags. For instance, DCE uses this to store the "dead/live" info.
431 static inline nir_instr
*
432 nir_instr_next(nir_instr
*instr
)
434 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
435 if (exec_node_is_tail_sentinel(next
))
438 return exec_node_data(nir_instr
, next
, node
);
441 static inline nir_instr
*
442 nir_instr_prev(nir_instr
*instr
)
444 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
445 if (exec_node_is_head_sentinel(prev
))
448 return exec_node_data(nir_instr
, prev
, node
);
452 nir_instr_is_first(nir_instr
*instr
)
454 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
458 nir_instr_is_last(nir_instr
*instr
)
460 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
464 /** for debugging only, can be NULL */
467 /** generic SSA definition index. */
470 /** Index into the live_in and live_out bitfields */
473 nir_instr
*parent_instr
;
475 /** set of nir_instr's where this register is used (read from) */
476 struct list_head uses
;
478 /** set of nir_if's where this register is used as a condition */
479 struct list_head if_uses
;
481 uint8_t num_components
;
488 struct nir_src
*indirect
; /** < NULL for no indirect offset */
489 unsigned base_offset
;
491 /* TODO use-def chain goes here */
495 nir_instr
*parent_instr
;
496 struct list_head def_link
;
499 struct nir_src
*indirect
; /** < NULL for no indirect offset */
500 unsigned base_offset
;
502 /* TODO def-use chain goes here */
507 typedef struct nir_src
{
509 nir_instr
*parent_instr
;
510 struct nir_if
*parent_if
;
513 struct list_head use_link
;
524 # define NIR_SRC_INIT nir_src()
526 # define NIR_SRC_INIT (nir_src) { { NULL } }
529 #define nir_foreach_use(reg_or_ssa_def, src) \
530 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
532 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
533 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
535 #define nir_foreach_if_use(reg_or_ssa_def, src) \
536 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
538 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
539 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
551 # define NIR_DEST_INIT nir_dest()
553 # define NIR_DEST_INIT (nir_dest) { { { NULL } } }
556 #define nir_foreach_def(reg, dest) \
557 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
559 #define nir_foreach_def_safe(reg, dest) \
560 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
562 static inline nir_src
563 nir_src_for_ssa(nir_ssa_def
*def
)
565 nir_src src
= NIR_SRC_INIT
;
573 static inline nir_src
574 nir_src_for_reg(nir_register
*reg
)
576 nir_src src
= NIR_SRC_INIT
;
580 src
.reg
.indirect
= NULL
;
581 src
.reg
.base_offset
= 0;
586 static inline nir_dest
587 nir_dest_for_reg(nir_register
*reg
)
589 nir_dest dest
= NIR_DEST_INIT
;
596 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
597 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
603 * \name input modifiers
607 * For inputs interpreted as floating point, flips the sign bit. For
608 * inputs interpreted as integers, performs the two's complement negation.
613 * Clears the sign bit for floating point values, and computes the integer
614 * absolute value for integers. Note that the negate modifier acts after
615 * the absolute value modifier, therefore if both are set then all inputs
616 * will become negative.
622 * For each input component, says which component of the register it is
623 * chosen from. Note that which elements of the swizzle are used and which
624 * are ignored are based on the write mask for most opcodes - for example,
625 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
626 * a swizzle of {2, x, 1, 0} where x means "don't care."
635 * \name saturate output modifier
637 * Only valid for opcodes that output floating-point numbers. Clamps the
638 * output to between 0.0 and 1.0 inclusive.
643 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
647 nir_type_invalid
= 0, /* Not a valid type */
655 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
656 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
657 } nir_op_algebraic_property
;
665 * The number of components in the output
667 * If non-zero, this is the size of the output and input sizes are
668 * explicitly given; swizzle and writemask are still in effect, but if
669 * the output component is masked out, then the input component may
672 * If zero, the opcode acts in the standard, per-component manner; the
673 * operation is performed on each component (except the ones that are
674 * masked out) with the input being taken from the input swizzle for
677 * The size of some of the inputs may be given (i.e. non-zero) even
678 * though output_size is zero; in that case, the inputs with a zero
679 * size act per-component, while the inputs with non-zero size don't.
681 unsigned output_size
;
684 * The type of vector that the instruction outputs. Note that the
685 * staurate modifier is only allowed on outputs with the float type.
688 nir_alu_type output_type
;
691 * The number of components in each input
693 unsigned input_sizes
[4];
696 * The type of vector that each input takes. Note that negate and
697 * absolute value are only allowed on inputs with int or float type and
698 * behave differently on the two.
700 nir_alu_type input_types
[4];
702 nir_op_algebraic_property algebraic_properties
;
705 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
707 typedef struct nir_alu_instr
{
714 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
715 nir_alu_instr
*instr
);
716 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
717 nir_alu_instr
*instr
);
719 /* is this source channel used? */
721 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
723 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
724 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
726 return (instr
->dest
.write_mask
>> channel
) & 1;
730 * For instructions whose destinations are SSA, get the number of channels
733 static inline unsigned
734 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
736 assert(instr
->dest
.dest
.is_ssa
);
738 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
739 return nir_op_infos
[instr
->op
].input_sizes
[src
];
741 return instr
->dest
.dest
.ssa
.num_components
;
746 nir_deref_type_array
,
747 nir_deref_type_struct
750 typedef struct nir_deref
{
751 nir_deref_type deref_type
;
752 struct nir_deref
*child
;
753 const struct glsl_type
*type
;
762 /* This enum describes how the array is referenced. If the deref is
763 * direct then the base_offset is used. If the deref is indirect then then
764 * offset is given by base_offset + indirect. If the deref is a wildcard
765 * then the deref refers to all of the elements of the array at the same
766 * time. Wildcard dereferences are only ever allowed in copy_var
767 * intrinsics and the source and destination derefs must have matching
771 nir_deref_array_type_direct
,
772 nir_deref_array_type_indirect
,
773 nir_deref_array_type_wildcard
,
774 } nir_deref_array_type
;
779 nir_deref_array_type deref_array_type
;
780 unsigned base_offset
;
790 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
791 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
792 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
794 /* Returns the last deref in the chain. */
795 static inline nir_deref
*
796 nir_deref_tail(nir_deref
*deref
)
799 deref
= deref
->child
;
807 nir_deref_var
**params
;
808 nir_deref_var
*return_deref
;
810 struct nir_function_overload
*callee
;
813 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
814 num_variables, num_indices, flags) \
815 nir_intrinsic_##name,
817 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
820 #include "nir_intrinsics.h"
821 nir_num_intrinsics
= nir_last_intrinsic
+ 1
825 #undef LAST_INTRINSIC
827 /** Represents an intrinsic
829 * An intrinsic is an instruction type for handling things that are
830 * more-or-less regular operations but don't just consume and produce SSA
831 * values like ALU operations do. Intrinsics are not for things that have
832 * special semantic meaning such as phi nodes and parallel copies.
833 * Examples of intrinsics include variable load/store operations, system
834 * value loads, and the like. Even though texturing more-or-less falls
835 * under this category, texturing is its own instruction type because
836 * trying to represent texturing with intrinsics would lead to a
837 * combinatorial explosion of intrinsic opcodes.
839 * By having a single instruction type for handling a lot of different
840 * cases, optimization passes can look for intrinsics and, for the most
841 * part, completely ignore them. Each intrinsic type also has a few
842 * possible flags that govern whether or not they can be reordered or
843 * eliminated. That way passes like dead code elimination can still work
844 * on intrisics without understanding the meaning of each.
846 * Each intrinsic has some number of constant indices, some number of
847 * variables, and some number of sources. What these sources, variables,
848 * and indices mean depends on the intrinsic and is documented with the
849 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
850 * instructions are the only types of instruction that can operate on
856 nir_intrinsic_op intrinsic
;
860 /** number of components if this is a vectorized intrinsic
862 * Similarly to ALU operations, some intrinsics are vectorized.
863 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
864 * For vectorized intrinsics, the num_components field specifies the
865 * number of destination components and the number of source components
866 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
868 uint8_t num_components
;
872 nir_deref_var
*variables
[2];
875 } nir_intrinsic_instr
;
878 * \name NIR intrinsics semantic flags
880 * information about what the compiler can do with the intrinsics.
882 * \sa nir_intrinsic_info::flags
886 * whether the intrinsic can be safely eliminated if none of its output
887 * value is not being used.
889 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
892 * Whether the intrinsic can be reordered with respect to any other
893 * intrinsic, i.e. whether the only reordering dependencies of the
894 * intrinsic are due to the register reads/writes.
896 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
897 } nir_intrinsic_semantic_flag
;
899 #define NIR_INTRINSIC_MAX_INPUTS 4
904 unsigned num_srcs
; /** < number of register/SSA inputs */
906 /** number of components of each input register
908 * If this value is 0, the number of components is given by the
909 * num_components field of nir_intrinsic_instr.
911 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
915 /** number of components of the output register
917 * If this value is 0, the number of components is given by the
918 * num_components field of nir_intrinsic_instr.
920 unsigned dest_components
;
922 /** the number of inputs/outputs that are variables */
923 unsigned num_variables
;
925 /** the number of constant indices used by the intrinsic */
926 unsigned num_indices
;
928 /** semantic flags for calls to this intrinsic */
929 nir_intrinsic_semantic_flag flags
;
930 } nir_intrinsic_info
;
932 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
935 * \group texture information
937 * This gives semantic information about textures which is useful to the
938 * frontend, the backend, and lowering passes, but not the optimizer.
943 nir_tex_src_projector
,
944 nir_tex_src_comparitor
, /* shadow comparitor */
948 nir_tex_src_ms_index
, /* MSAA sample index */
951 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
952 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
953 nir_num_tex_src_types
958 nir_tex_src_type src_type
;
962 nir_texop_tex
, /**< Regular texture look-up */
963 nir_texop_txb
, /**< Texture look-up with LOD bias */
964 nir_texop_txl
, /**< Texture look-up with explicit LOD */
965 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
966 nir_texop_txf
, /**< Texel fetch with explicit LOD */
967 nir_texop_txf_ms
, /**< Multisample texture fetch */
968 nir_texop_txs
, /**< Texture size */
969 nir_texop_lod
, /**< Texture lod query */
970 nir_texop_tg4
, /**< Texture gather */
971 nir_texop_query_levels
, /**< Texture levels query */
972 nir_texop_texture_samples
, /**< Texture samples query */
973 nir_texop_samples_identical
, /**< Query whether all samples are definitely
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 texture index
1004 * If this texture instruction has a nir_tex_src_texture_offset source,
1005 * then the texture index is given by texture_index + texture_offset.
1007 unsigned texture_index
;
1009 /** The size of the texture array or 0 if it's not an array */
1010 unsigned texture_array_size
;
1012 /** The texture deref
1014 * If both this and `sampler` are both NULL, use texture_index instead.
1015 * If `texture` is NULL, but `sampler` is non-NULL, then the texture is
1016 * implied from the sampler.
1018 nir_deref_var
*texture
;
1020 /** The sampler index
1022 * If this texture instruction has a nir_tex_src_sampler_offset source,
1023 * then the sampler index is given by sampler_index + sampler_offset.
1025 unsigned sampler_index
;
1027 /** The sampler deref
1029 * If this is null, use sampler_index instead.
1031 nir_deref_var
*sampler
;
1034 static inline unsigned
1035 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1037 switch (instr
->op
) {
1038 case nir_texop_txs
: {
1040 switch (instr
->sampler_dim
) {
1041 case GLSL_SAMPLER_DIM_1D
:
1042 case GLSL_SAMPLER_DIM_BUF
:
1045 case GLSL_SAMPLER_DIM_2D
:
1046 case GLSL_SAMPLER_DIM_CUBE
:
1047 case GLSL_SAMPLER_DIM_MS
:
1048 case GLSL_SAMPLER_DIM_RECT
:
1049 case GLSL_SAMPLER_DIM_EXTERNAL
:
1052 case GLSL_SAMPLER_DIM_3D
:
1056 unreachable("not reached");
1058 if (instr
->is_array
)
1066 case nir_texop_texture_samples
:
1067 case nir_texop_query_levels
:
1068 case nir_texop_samples_identical
:
1072 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1079 /* Returns true if this texture operation queries something about the texture
1080 * rather than actually sampling it.
1083 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1085 switch (instr
->op
) {
1088 case nir_texop_texture_samples
:
1089 case nir_texop_query_levels
:
1096 case nir_texop_txf_ms
:
1100 unreachable("Invalid texture opcode");
1104 static inline unsigned
1105 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1107 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1108 return instr
->coord_components
;
1111 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1112 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1113 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1114 if (instr
->is_array
)
1115 return instr
->coord_components
- 1;
1117 return instr
->coord_components
;
1124 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1126 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1127 if (instr
->src
[i
].src_type
== type
)
1144 nir_const_value value
;
1147 } nir_load_const_instr
;
1160 /* creates a new SSA variable in an undefined state */
1165 } nir_ssa_undef_instr
;
1168 struct exec_node node
;
1170 /* The predecessor block corresponding to this source */
1171 struct nir_block
*pred
;
1176 #define nir_foreach_phi_src(phi, entry) \
1177 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1178 #define nir_foreach_phi_src_safe(phi, entry) \
1179 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1184 struct exec_list srcs
; /** < list of nir_phi_src */
1190 struct exec_node node
;
1193 } nir_parallel_copy_entry
;
1195 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1196 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1201 /* A list of nir_parallel_copy_entry's. The sources of all of the
1202 * entries are copied to the corresponding destinations "in parallel".
1203 * In other words, if we have two entries: a -> b and b -> a, the values
1206 struct exec_list entries
;
1207 } nir_parallel_copy_instr
;
1209 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1210 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1211 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1212 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1213 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1214 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1215 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1216 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1217 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1218 nir_parallel_copy_instr
, instr
)
1223 * Control flow consists of a tree of control flow nodes, which include
1224 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1225 * instructions that always run start-to-finish. Each basic block also keeps
1226 * track of its successors (blocks which may run immediately after the current
1227 * block) and predecessors (blocks which could have run immediately before the
1228 * current block). Each function also has a start block and an end block which
1229 * all return statements point to (which is always empty). Together, all the
1230 * blocks with their predecessors and successors make up the control flow
1231 * graph (CFG) of the function. There are helpers that modify the tree of
1232 * control flow nodes while modifying the CFG appropriately; these should be
1233 * used instead of modifying the tree directly.
1240 nir_cf_node_function
1243 typedef struct nir_cf_node
{
1244 struct exec_node node
;
1245 nir_cf_node_type type
;
1246 struct nir_cf_node
*parent
;
1249 typedef struct nir_block
{
1250 nir_cf_node cf_node
;
1252 struct exec_list instr_list
; /** < list of nir_instr */
1254 /** generic block index; generated by nir_index_blocks */
1258 * Each block can only have up to 2 successors, so we put them in a simple
1259 * array - no need for anything more complicated.
1261 struct nir_block
*successors
[2];
1263 /* Set of nir_block predecessors in the CFG */
1264 struct set
*predecessors
;
1267 * this node's immediate dominator in the dominance tree - set to NULL for
1270 struct nir_block
*imm_dom
;
1272 /* This node's children in the dominance tree */
1273 unsigned num_dom_children
;
1274 struct nir_block
**dom_children
;
1276 /* Set of nir_block's on the dominance frontier of this block */
1277 struct set
*dom_frontier
;
1280 * These two indices have the property that dom_{pre,post}_index for each
1281 * child of this block in the dominance tree will always be between
1282 * dom_pre_index and dom_post_index for this block, which makes testing if
1283 * a given block is dominated by another block an O(1) operation.
1285 unsigned dom_pre_index
, dom_post_index
;
1287 /* live in and out for this block; used for liveness analysis */
1288 BITSET_WORD
*live_in
;
1289 BITSET_WORD
*live_out
;
1292 static inline nir_instr
*
1293 nir_block_first_instr(nir_block
*block
)
1295 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1296 return exec_node_data(nir_instr
, head
, node
);
1299 static inline nir_instr
*
1300 nir_block_last_instr(nir_block
*block
)
1302 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1303 return exec_node_data(nir_instr
, tail
, node
);
1306 #define nir_foreach_instr(block, instr) \
1307 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1308 #define nir_foreach_instr_reverse(block, instr) \
1309 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1310 #define nir_foreach_instr_safe(block, instr) \
1311 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1312 #define nir_foreach_instr_safe_reverse(block, instr) \
1313 foreach_list_typed_safe_reverse(nir_instr, instr, node, &(block)->instr_list)
1315 typedef struct nir_if
{
1316 nir_cf_node cf_node
;
1319 struct exec_list then_list
; /** < list of nir_cf_node */
1320 struct exec_list else_list
; /** < list of nir_cf_node */
1323 static inline nir_cf_node
*
1324 nir_if_first_then_node(nir_if
*if_stmt
)
1326 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1327 return exec_node_data(nir_cf_node
, head
, node
);
1330 static inline nir_cf_node
*
1331 nir_if_last_then_node(nir_if
*if_stmt
)
1333 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1334 return exec_node_data(nir_cf_node
, tail
, node
);
1337 static inline nir_cf_node
*
1338 nir_if_first_else_node(nir_if
*if_stmt
)
1340 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1341 return exec_node_data(nir_cf_node
, head
, node
);
1344 static inline nir_cf_node
*
1345 nir_if_last_else_node(nir_if
*if_stmt
)
1347 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1348 return exec_node_data(nir_cf_node
, tail
, node
);
1352 nir_cf_node cf_node
;
1354 struct exec_list body
; /** < list of nir_cf_node */
1357 static inline nir_cf_node
*
1358 nir_loop_first_cf_node(nir_loop
*loop
)
1360 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1363 static inline nir_cf_node
*
1364 nir_loop_last_cf_node(nir_loop
*loop
)
1366 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1370 * Various bits of metadata that can may be created or required by
1371 * optimization and analysis passes
1374 nir_metadata_none
= 0x0,
1375 nir_metadata_block_index
= 0x1,
1376 nir_metadata_dominance
= 0x2,
1377 nir_metadata_live_ssa_defs
= 0x4,
1378 nir_metadata_not_properly_reset
= 0x8,
1382 nir_cf_node cf_node
;
1384 /** pointer to the overload of which this is an implementation */
1385 struct nir_function_overload
*overload
;
1387 struct exec_list body
; /** < list of nir_cf_node */
1389 nir_block
*end_block
;
1391 /** list for all local variables in the function */
1392 struct exec_list locals
;
1394 /** array of variables used as parameters */
1395 unsigned num_params
;
1396 nir_variable
**params
;
1398 /** variable used to hold the result of the function */
1399 nir_variable
*return_var
;
1401 /** list of local registers in the function */
1402 struct exec_list registers
;
1404 /** next available local register index */
1407 /** next available SSA value index */
1410 /* total number of basic blocks, only valid when block_index_dirty = false */
1411 unsigned num_blocks
;
1413 nir_metadata valid_metadata
;
1414 } nir_function_impl
;
1416 static inline nir_block
*
1417 nir_start_block(nir_function_impl
*impl
)
1419 return (nir_block
*) exec_list_get_head(&impl
->body
);
1422 static inline nir_cf_node
*
1423 nir_cf_node_next(nir_cf_node
*node
)
1425 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1426 if (exec_node_is_tail_sentinel(next
))
1429 return exec_node_data(nir_cf_node
, next
, node
);
1432 static inline nir_cf_node
*
1433 nir_cf_node_prev(nir_cf_node
*node
)
1435 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1436 if (exec_node_is_head_sentinel(prev
))
1439 return exec_node_data(nir_cf_node
, prev
, node
);
1443 nir_cf_node_is_first(const nir_cf_node
*node
)
1445 return exec_node_is_head_sentinel(node
->node
.prev
);
1449 nir_cf_node_is_last(const nir_cf_node
*node
)
1451 return exec_node_is_tail_sentinel(node
->node
.next
);
1454 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1455 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1456 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1457 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1462 nir_parameter_inout
,
1463 } nir_parameter_type
;
1466 nir_parameter_type param_type
;
1467 const struct glsl_type
*type
;
1470 typedef struct nir_function_overload
{
1471 struct exec_node node
;
1473 unsigned num_params
;
1474 nir_parameter
*params
;
1475 const struct glsl_type
*return_type
;
1477 nir_function_impl
*impl
; /** < NULL if the overload is only declared yet */
1479 /** pointer to the function of which this is an overload */
1480 struct nir_function
*function
;
1481 } nir_function_overload
;
1483 typedef struct nir_function
{
1484 struct exec_node node
;
1486 struct exec_list overload_list
; /** < list of nir_function_overload */
1488 struct nir_shader
*shader
;
1491 #define nir_function_first_overload(func) \
1492 exec_node_data(nir_function_overload, \
1493 exec_list_get_head(&(func)->overload_list), node)
1495 typedef struct nir_shader_compiler_options
{
1502 /** lowers fneg and ineg to fsub and isub. */
1504 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1507 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1510 /* Does the native fdot instruction replicate its result for four
1511 * components? If so, then opt_algebraic_late will turn all fdotN
1512 * instructions into fdot_replicatedN instructions.
1514 bool fdot_replicates
;
1516 /** lowers ffract to fsub+ffloor: */
1520 * Does the driver support real 32-bit integers? (Otherwise, integers
1521 * are simulated by floats.)
1523 bool native_integers
;
1524 } nir_shader_compiler_options
;
1526 typedef struct nir_shader_info
{
1529 /* Descriptive name provided by the client; may be NULL */
1532 /* Number of textures used by this shader */
1533 unsigned num_textures
;
1534 /* Number of uniform buffers used by this shader */
1536 /* Number of atomic buffers used by this shader */
1538 /* Number of shader storage buffers used by this shader */
1540 /* Number of images used by this shader */
1541 unsigned num_images
;
1543 /* Which inputs are actually read */
1544 uint64_t inputs_read
;
1545 /* Which outputs are actually written */
1546 uint64_t outputs_written
;
1547 /* Which system values are actually read */
1548 uint64_t system_values_read
;
1550 /* Which patch inputs are actually read */
1551 uint32_t patch_inputs_read
;
1552 /* Which patch outputs are actually written */
1553 uint32_t patch_outputs_written
;
1555 /* Whether or not this shader ever uses textureGather() */
1556 bool uses_texture_gather
;
1558 /* Whether or not this shader uses the gl_ClipDistance output */
1559 bool uses_clip_distance_out
;
1561 /* Whether or not separate shader objects were used */
1562 bool separate_shader
;
1564 /** Was this shader linked with any transform feedback varyings? */
1565 bool has_transform_feedback_varyings
;
1569 /** The number of vertices recieves per input primitive */
1570 unsigned vertices_in
;
1572 /** The output primitive type (GL enum value) */
1573 unsigned output_primitive
;
1575 /** The maximum number of vertices the geometry shader might write. */
1576 unsigned vertices_out
;
1578 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1579 unsigned invocations
;
1581 /** Whether or not this shader uses EndPrimitive */
1582 bool uses_end_primitive
;
1584 /** Whether or not this shader uses non-zero streams */
1592 * Whether early fragment tests are enabled as defined by
1593 * ARB_shader_image_load_store.
1595 bool early_fragment_tests
;
1597 /** gl_FragDepth layout for ARB_conservative_depth. */
1598 enum gl_frag_depth_layout depth_layout
;
1602 unsigned local_size
[3];
1606 /** The number of vertices in the TCS output patch. */
1607 unsigned vertices_out
;
1612 typedef struct nir_shader
{
1613 /** list of uniforms (nir_variable) */
1614 struct exec_list uniforms
;
1616 /** list of inputs (nir_variable) */
1617 struct exec_list inputs
;
1619 /** list of outputs (nir_variable) */
1620 struct exec_list outputs
;
1622 /** Set of driver-specific options for the shader.
1624 * The memory for the options is expected to be kept in a single static
1625 * copy by the driver.
1627 const struct nir_shader_compiler_options
*options
;
1629 /** Various bits of compile-time information about a given shader */
1630 struct nir_shader_info info
;
1632 /** list of global variables in the shader (nir_variable) */
1633 struct exec_list globals
;
1635 /** list of system value variables in the shader (nir_variable) */
1636 struct exec_list system_values
;
1638 struct exec_list functions
; /** < list of nir_function */
1640 /** list of global register in the shader */
1641 struct exec_list registers
;
1643 /** next available global register index */
1647 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1650 unsigned num_inputs
, num_uniforms
, num_outputs
;
1652 /** The shader stage, such as MESA_SHADER_VERTEX. */
1653 gl_shader_stage stage
;
1656 #define nir_foreach_overload(shader, overload) \
1657 foreach_list_typed(nir_function, func, node, &(shader)->functions) \
1658 foreach_list_typed(nir_function_overload, overload, node, \
1659 &(func)->overload_list)
1661 nir_shader
*nir_shader_create(void *mem_ctx
,
1662 gl_shader_stage stage
,
1663 const nir_shader_compiler_options
*options
);
1665 /** creates a register, including assigning it an index and adding it to the list */
1666 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1668 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1670 void nir_reg_remove(nir_register
*reg
);
1672 /** Adds a variable to the appropreate list in nir_shader */
1673 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1676 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1678 assert(var
->data
.mode
== nir_var_local
);
1679 exec_list_push_tail(&impl
->locals
, &var
->node
);
1682 /** creates a variable, sets a few defaults, and adds it to the list */
1683 nir_variable
*nir_variable_create(nir_shader
*shader
,
1684 nir_variable_mode mode
,
1685 const struct glsl_type
*type
,
1687 /** creates a local variable and adds it to the list */
1688 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1689 const struct glsl_type
*type
,
1692 /** creates a function and adds it to the shader's list of functions */
1693 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1695 /** creates a null function returning null */
1696 nir_function_overload
*nir_function_overload_create(nir_function
*func
);
1698 nir_function_impl
*nir_function_impl_create(nir_function_overload
*func
);
1700 nir_block
*nir_block_create(nir_shader
*shader
);
1701 nir_if
*nir_if_create(nir_shader
*shader
);
1702 nir_loop
*nir_loop_create(nir_shader
*shader
);
1704 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1706 /** requests that the given pieces of metadata be generated */
1707 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1708 /** dirties all but the preserved metadata */
1709 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1711 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1712 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1714 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1716 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1717 unsigned num_components
);
1719 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1720 nir_intrinsic_op op
);
1722 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1723 nir_function_overload
*callee
);
1725 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1727 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1729 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1731 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1732 unsigned num_components
);
1734 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1735 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1736 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1738 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1740 nir_load_const_instr
*
1741 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1744 * NIR Cursors and Instruction Insertion API
1747 * A tiny struct representing a point to insert/extract instructions or
1748 * control flow nodes. Helps reduce the combinatorial explosion of possible
1749 * points to insert/extract.
1751 * \sa nir_control_flow.h
1754 nir_cursor_before_block
,
1755 nir_cursor_after_block
,
1756 nir_cursor_before_instr
,
1757 nir_cursor_after_instr
,
1758 } nir_cursor_option
;
1761 nir_cursor_option option
;
1768 static inline nir_block
*
1769 nir_cursor_current_block(nir_cursor cursor
)
1771 if (cursor
.option
== nir_cursor_before_instr
||
1772 cursor
.option
== nir_cursor_after_instr
) {
1773 return cursor
.instr
->block
;
1775 return cursor
.block
;
1779 static inline nir_cursor
1780 nir_before_block(nir_block
*block
)
1783 cursor
.option
= nir_cursor_before_block
;
1784 cursor
.block
= block
;
1788 static inline nir_cursor
1789 nir_after_block(nir_block
*block
)
1792 cursor
.option
= nir_cursor_after_block
;
1793 cursor
.block
= block
;
1797 static inline nir_cursor
1798 nir_before_instr(nir_instr
*instr
)
1801 cursor
.option
= nir_cursor_before_instr
;
1802 cursor
.instr
= instr
;
1806 static inline nir_cursor
1807 nir_after_instr(nir_instr
*instr
)
1810 cursor
.option
= nir_cursor_after_instr
;
1811 cursor
.instr
= instr
;
1815 static inline nir_cursor
1816 nir_after_block_before_jump(nir_block
*block
)
1818 nir_instr
*last_instr
= nir_block_last_instr(block
);
1819 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1820 return nir_before_instr(last_instr
);
1822 return nir_after_block(block
);
1826 static inline nir_cursor
1827 nir_before_cf_node(nir_cf_node
*node
)
1829 if (node
->type
== nir_cf_node_block
)
1830 return nir_before_block(nir_cf_node_as_block(node
));
1832 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1835 static inline nir_cursor
1836 nir_after_cf_node(nir_cf_node
*node
)
1838 if (node
->type
== nir_cf_node_block
)
1839 return nir_after_block(nir_cf_node_as_block(node
));
1841 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1844 static inline nir_cursor
1845 nir_before_cf_list(struct exec_list
*cf_list
)
1847 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1848 exec_list_get_head(cf_list
), node
);
1849 return nir_before_cf_node(first_node
);
1852 static inline nir_cursor
1853 nir_after_cf_list(struct exec_list
*cf_list
)
1855 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1856 exec_list_get_tail(cf_list
), node
);
1857 return nir_after_cf_node(last_node
);
1861 * Insert a NIR instruction at the given cursor.
1863 * Note: This does not update the cursor.
1865 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
1868 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1870 nir_instr_insert(nir_before_instr(instr
), before
);
1874 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
1876 nir_instr_insert(nir_after_instr(instr
), after
);
1880 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
1882 nir_instr_insert(nir_before_block(block
), before
);
1886 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
1888 nir_instr_insert(nir_after_block(block
), after
);
1892 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
1894 nir_instr_insert(nir_before_cf_node(node
), before
);
1898 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
1900 nir_instr_insert(nir_after_cf_node(node
), after
);
1904 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
1906 nir_instr_insert(nir_before_cf_list(list
), before
);
1910 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
1912 nir_instr_insert(nir_after_cf_list(list
), after
);
1915 void nir_instr_remove(nir_instr
*instr
);
1919 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
1920 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
1921 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
1922 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
1924 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
1925 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
1927 nir_const_value
*nir_src_as_const_value(nir_src src
);
1928 bool nir_src_is_dynamically_uniform(nir_src src
);
1929 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
1930 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
1931 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
1932 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
1933 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
1936 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1937 unsigned num_components
, const char *name
);
1938 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1939 unsigned num_components
, const char *name
);
1940 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
1941 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
1942 nir_instr
*after_me
);
1944 /* visits basic blocks in source-code order */
1945 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
1946 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1948 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1950 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1953 /* If the following CF node is an if, this function returns that if.
1954 * Otherwise, it returns NULL.
1956 nir_if
*nir_block_get_following_if(nir_block
*block
);
1958 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
1960 void nir_index_local_regs(nir_function_impl
*impl
);
1961 void nir_index_global_regs(nir_shader
*shader
);
1962 void nir_index_ssa_defs(nir_function_impl
*impl
);
1963 unsigned nir_index_instrs(nir_function_impl
*impl
);
1965 void nir_index_blocks(nir_function_impl
*impl
);
1967 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
1968 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
1970 nir_shader
* nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
1973 void nir_validate_shader(nir_shader
*shader
);
1974 void nir_metadata_set_validation_flag(nir_shader
*shader
);
1975 void nir_metadata_check_validation_flag(nir_shader
*shader
);
1977 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
1978 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
1979 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
1982 void nir_calc_dominance_impl(nir_function_impl
*impl
);
1983 void nir_calc_dominance(nir_shader
*shader
);
1985 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
1986 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
1988 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
1989 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
1991 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
1992 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
1994 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
1995 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
1997 int nir_gs_count_vertices(const nir_shader
*shader
);
1999 bool nir_split_var_copies(nir_shader
*shader
);
2001 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2002 void nir_lower_var_copies(nir_shader
*shader
);
2004 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2006 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2008 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
2010 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
2012 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2014 void nir_assign_var_locations(struct exec_list
*var_list
,
2016 int (*type_size
)(const struct glsl_type
*));
2018 void nir_lower_io(nir_shader
*shader
,
2019 nir_variable_mode mode
,
2020 int (*type_size
)(const struct glsl_type
*));
2021 nir_src
*nir_get_io_indirect_src(nir_intrinsic_instr
*instr
);
2022 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2024 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2026 bool nir_remove_dead_variables(nir_shader
*shader
);
2028 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2029 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2030 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2031 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2033 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2035 void nir_lower_samplers(nir_shader
*shader
,
2036 const struct gl_shader_program
*shader_program
);
2038 bool nir_lower_system_values(nir_shader
*shader
);
2040 typedef struct nir_lower_tex_options
{
2042 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2043 * sampler types a texture projector is lowered.
2048 * If true, lower rect textures to 2D, using txs to fetch the
2049 * texture dimensions and dividing the texture coords by the
2050 * texture dims to normalize.
2055 * To emulate certain texture wrap modes, this can be used
2056 * to saturate the specified tex coord to [0.0, 1.0]. The
2057 * bits are according to sampler #, ie. if, for example:
2059 * (conf->saturate_s & (1 << n))
2061 * is true, then the s coord for sampler n is saturated.
2063 * Note that clamping must happen *after* projector lowering
2064 * so any projected texture sample instruction with a clamped
2065 * coordinate gets automatically lowered, regardless of the
2066 * 'lower_txp' setting.
2068 unsigned saturate_s
;
2069 unsigned saturate_t
;
2070 unsigned saturate_r
;
2072 /* Bitmask of samplers that need swizzling.
2074 * If (swizzle_result & (1 << sampler_index)), then the swizzle in
2075 * swizzles[sampler_index] is applied to the result of the texturing
2078 unsigned swizzle_result
;
2080 /* A swizzle for each sampler. Values 0-3 represent x, y, z, or w swizzles
2081 * while 4 and 5 represent 0 and 1 respectively.
2083 uint8_t swizzles
[32][4];
2084 } nir_lower_tex_options
;
2086 bool nir_lower_tex(nir_shader
*shader
,
2087 const nir_lower_tex_options
*options
);
2089 void nir_lower_idiv(nir_shader
*shader
);
2091 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2092 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2094 void nir_lower_two_sided_color(nir_shader
*shader
);
2096 void nir_lower_atomics(nir_shader
*shader
,
2097 const struct gl_shader_program
*shader_program
);
2098 void nir_lower_to_source_mods(nir_shader
*shader
);
2100 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2102 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2104 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2105 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2107 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2108 void nir_convert_to_ssa(nir_shader
*shader
);
2110 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2111 * registers. If false, convert all values (even those not involved in a phi
2112 * node) to registers.
2114 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2116 bool nir_opt_algebraic(nir_shader
*shader
);
2117 bool nir_opt_algebraic_late(nir_shader
*shader
);
2118 bool nir_opt_constant_folding(nir_shader
*shader
);
2120 bool nir_opt_global_to_local(nir_shader
*shader
);
2122 bool nir_copy_prop(nir_shader
*shader
);
2124 bool nir_opt_cse(nir_shader
*shader
);
2126 bool nir_opt_dce(nir_shader
*shader
);
2128 bool nir_opt_dead_cf(nir_shader
*shader
);
2130 void nir_opt_gcm(nir_shader
*shader
);
2132 bool nir_opt_peephole_select(nir_shader
*shader
);
2134 bool nir_opt_remove_phis(nir_shader
*shader
);
2136 bool nir_opt_undef(nir_shader
*shader
);
2138 void nir_sweep(nir_shader
*shader
);
2140 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2141 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);