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 * Returns the bits in the inputs_read, outputs_written, or
371 * system_values_read bitfield corresponding to this variable.
373 static inline uint64_t
374 nir_variable_get_io_mask(nir_variable
*var
, gl_shader_stage stage
)
376 assert(var
->data
.mode
== nir_var_shader_in
||
377 var
->data
.mode
== nir_var_shader_out
||
378 var
->data
.mode
== nir_var_system_value
);
379 assert(var
->data
.location
>= 0);
381 const struct glsl_type
*var_type
= var
->type
;
382 if (stage
== MESA_SHADER_GEOMETRY
&& var
->data
.mode
== nir_var_shader_in
) {
383 /* Most geometry shader inputs are per-vertex arrays */
384 if (var
->data
.location
>= VARYING_SLOT_VAR0
)
385 assert(glsl_type_is_array(var_type
));
387 if (glsl_type_is_array(var_type
))
388 var_type
= glsl_get_array_element(var_type
);
391 unsigned slots
= glsl_count_attribute_slots(var_type
);
392 return ((1ull << slots
) - 1) << var
->data
.location
;
396 struct exec_node node
;
398 unsigned num_components
; /** < number of vector components */
399 unsigned num_array_elems
; /** < size of array (0 for no array) */
401 /** generic register index. */
404 /** only for debug purposes, can be NULL */
407 /** whether this register is local (per-function) or global (per-shader) */
411 * If this flag is set to true, then accessing channels >= num_components
412 * is well-defined, and simply spills over to the next array element. This
413 * is useful for backends that can do per-component accessing, in
414 * particular scalar backends. By setting this flag and making
415 * num_components equal to 1, structures can be packed tightly into
416 * registers and then registers can be accessed per-component to get to
417 * each structure member, even if it crosses vec4 boundaries.
421 /** set of nir_src's where this register is used (read from) */
422 struct list_head uses
;
424 /** set of nir_dest's where this register is defined (written to) */
425 struct list_head defs
;
427 /** set of nir_if's where this register is used as a condition */
428 struct list_head if_uses
;
435 nir_instr_type_intrinsic
,
436 nir_instr_type_load_const
,
438 nir_instr_type_ssa_undef
,
440 nir_instr_type_parallel_copy
,
443 typedef struct nir_instr
{
444 struct exec_node node
;
446 struct nir_block
*block
;
448 /** generic instruction index. */
451 /* A temporary for optimization and analysis passes to use for storing
452 * flags. For instance, DCE uses this to store the "dead/live" info.
457 static inline nir_instr
*
458 nir_instr_next(nir_instr
*instr
)
460 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
461 if (exec_node_is_tail_sentinel(next
))
464 return exec_node_data(nir_instr
, next
, node
);
467 static inline nir_instr
*
468 nir_instr_prev(nir_instr
*instr
)
470 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
471 if (exec_node_is_head_sentinel(prev
))
474 return exec_node_data(nir_instr
, prev
, node
);
478 nir_instr_is_first(nir_instr
*instr
)
480 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
484 nir_instr_is_last(nir_instr
*instr
)
486 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
490 /** for debugging only, can be NULL */
493 /** generic SSA definition index. */
496 /** Index into the live_in and live_out bitfields */
499 nir_instr
*parent_instr
;
501 /** set of nir_instr's where this register is used (read from) */
502 struct list_head uses
;
504 /** set of nir_if's where this register is used as a condition */
505 struct list_head if_uses
;
507 uint8_t num_components
;
514 struct nir_src
*indirect
; /** < NULL for no indirect offset */
515 unsigned base_offset
;
517 /* TODO use-def chain goes here */
521 nir_instr
*parent_instr
;
522 struct list_head def_link
;
525 struct nir_src
*indirect
; /** < NULL for no indirect offset */
526 unsigned base_offset
;
528 /* TODO def-use chain goes here */
533 typedef struct nir_src
{
535 nir_instr
*parent_instr
;
536 struct nir_if
*parent_if
;
539 struct list_head use_link
;
550 # define NIR_SRC_INIT nir_src()
552 # define NIR_SRC_INIT (nir_src) { { NULL } }
555 #define nir_foreach_use(reg_or_ssa_def, src) \
556 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
558 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
559 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
561 #define nir_foreach_if_use(reg_or_ssa_def, src) \
562 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
564 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
565 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
577 # define NIR_DEST_INIT nir_dest()
579 # define NIR_DEST_INIT (nir_dest) { { { NULL } } }
582 #define nir_foreach_def(reg, dest) \
583 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
585 #define nir_foreach_def_safe(reg, dest) \
586 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
588 static inline nir_src
589 nir_src_for_ssa(nir_ssa_def
*def
)
591 nir_src src
= NIR_SRC_INIT
;
599 static inline nir_src
600 nir_src_for_reg(nir_register
*reg
)
602 nir_src src
= NIR_SRC_INIT
;
606 src
.reg
.indirect
= NULL
;
607 src
.reg
.base_offset
= 0;
612 static inline nir_dest
613 nir_dest_for_reg(nir_register
*reg
)
615 nir_dest dest
= NIR_DEST_INIT
;
622 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
623 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
629 * \name input modifiers
633 * For inputs interpreted as floating point, flips the sign bit. For
634 * inputs interpreted as integers, performs the two's complement negation.
639 * Clears the sign bit for floating point values, and computes the integer
640 * absolute value for integers. Note that the negate modifier acts after
641 * the absolute value modifier, therefore if both are set then all inputs
642 * will become negative.
648 * For each input component, says which component of the register it is
649 * chosen from. Note that which elements of the swizzle are used and which
650 * are ignored are based on the write mask for most opcodes - for example,
651 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
652 * a swizzle of {2, x, 1, 0} where x means "don't care."
661 * \name saturate output modifier
663 * Only valid for opcodes that output floating-point numbers. Clamps the
664 * output to between 0.0 and 1.0 inclusive.
669 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
673 nir_type_invalid
= 0, /* Not a valid type */
681 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
682 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
683 } nir_op_algebraic_property
;
691 * The number of components in the output
693 * If non-zero, this is the size of the output and input sizes are
694 * explicitly given; swizzle and writemask are still in effect, but if
695 * the output component is masked out, then the input component may
698 * If zero, the opcode acts in the standard, per-component manner; the
699 * operation is performed on each component (except the ones that are
700 * masked out) with the input being taken from the input swizzle for
703 * The size of some of the inputs may be given (i.e. non-zero) even
704 * though output_size is zero; in that case, the inputs with a zero
705 * size act per-component, while the inputs with non-zero size don't.
707 unsigned output_size
;
710 * The type of vector that the instruction outputs. Note that the
711 * staurate modifier is only allowed on outputs with the float type.
714 nir_alu_type output_type
;
717 * The number of components in each input
719 unsigned input_sizes
[4];
722 * The type of vector that each input takes. Note that negate and
723 * absolute value are only allowed on inputs with int or float type and
724 * behave differently on the two.
726 nir_alu_type input_types
[4];
728 nir_op_algebraic_property algebraic_properties
;
731 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
733 typedef struct nir_alu_instr
{
740 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
741 nir_alu_instr
*instr
);
742 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
743 nir_alu_instr
*instr
);
745 /* is this source channel used? */
747 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
749 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
750 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
752 return (instr
->dest
.write_mask
>> channel
) & 1;
756 * For instructions whose destinations are SSA, get the number of channels
759 static inline unsigned
760 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
762 assert(instr
->dest
.dest
.is_ssa
);
764 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
765 return nir_op_infos
[instr
->op
].input_sizes
[src
];
767 return instr
->dest
.dest
.ssa
.num_components
;
772 nir_deref_type_array
,
773 nir_deref_type_struct
776 typedef struct nir_deref
{
777 nir_deref_type deref_type
;
778 struct nir_deref
*child
;
779 const struct glsl_type
*type
;
788 /* This enum describes how the array is referenced. If the deref is
789 * direct then the base_offset is used. If the deref is indirect then then
790 * offset is given by base_offset + indirect. If the deref is a wildcard
791 * then the deref refers to all of the elements of the array at the same
792 * time. Wildcard dereferences are only ever allowed in copy_var
793 * intrinsics and the source and destination derefs must have matching
797 nir_deref_array_type_direct
,
798 nir_deref_array_type_indirect
,
799 nir_deref_array_type_wildcard
,
800 } nir_deref_array_type
;
805 nir_deref_array_type deref_array_type
;
806 unsigned base_offset
;
816 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
817 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
818 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
820 /* Returns the last deref in the chain. */
821 static inline nir_deref
*
822 nir_deref_tail(nir_deref
*deref
)
825 deref
= deref
->child
;
833 nir_deref_var
**params
;
834 nir_deref_var
*return_deref
;
836 struct nir_function_overload
*callee
;
839 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
840 num_variables, num_indices, flags) \
841 nir_intrinsic_##name,
843 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
846 #include "nir_intrinsics.h"
847 nir_num_intrinsics
= nir_last_intrinsic
+ 1
851 #undef LAST_INTRINSIC
853 /** Represents an intrinsic
855 * An intrinsic is an instruction type for handling things that are
856 * more-or-less regular operations but don't just consume and produce SSA
857 * values like ALU operations do. Intrinsics are not for things that have
858 * special semantic meaning such as phi nodes and parallel copies.
859 * Examples of intrinsics include variable load/store operations, system
860 * value loads, and the like. Even though texturing more-or-less falls
861 * under this category, texturing is its own instruction type because
862 * trying to represent texturing with intrinsics would lead to a
863 * combinatorial explosion of intrinsic opcodes.
865 * By having a single instruction type for handling a lot of different
866 * cases, optimization passes can look for intrinsics and, for the most
867 * part, completely ignore them. Each intrinsic type also has a few
868 * possible flags that govern whether or not they can be reordered or
869 * eliminated. That way passes like dead code elimination can still work
870 * on intrisics without understanding the meaning of each.
872 * Each intrinsic has some number of constant indices, some number of
873 * variables, and some number of sources. What these sources, variables,
874 * and indices mean depends on the intrinsic and is documented with the
875 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
876 * instructions are the only types of instruction that can operate on
882 nir_intrinsic_op intrinsic
;
886 /** number of components if this is a vectorized intrinsic
888 * Similarly to ALU operations, some intrinsics are vectorized.
889 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
890 * For vectorized intrinsics, the num_components field specifies the
891 * number of destination components and the number of source components
892 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
894 uint8_t num_components
;
898 nir_deref_var
*variables
[2];
901 } nir_intrinsic_instr
;
904 * \name NIR intrinsics semantic flags
906 * information about what the compiler can do with the intrinsics.
908 * \sa nir_intrinsic_info::flags
912 * whether the intrinsic can be safely eliminated if none of its output
913 * value is not being used.
915 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
918 * Whether the intrinsic can be reordered with respect to any other
919 * intrinsic, i.e. whether the only reordering dependencies of the
920 * intrinsic are due to the register reads/writes.
922 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
923 } nir_intrinsic_semantic_flag
;
925 #define NIR_INTRINSIC_MAX_INPUTS 4
930 unsigned num_srcs
; /** < number of register/SSA inputs */
932 /** number of components of each input register
934 * If this value is 0, the number of components is given by the
935 * num_components field of nir_intrinsic_instr.
937 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
941 /** number of components of the output register
943 * If this value is 0, the number of components is given by the
944 * num_components field of nir_intrinsic_instr.
946 unsigned dest_components
;
948 /** the number of inputs/outputs that are variables */
949 unsigned num_variables
;
951 /** the number of constant indices used by the intrinsic */
952 unsigned num_indices
;
954 /** semantic flags for calls to this intrinsic */
955 nir_intrinsic_semantic_flag flags
;
956 } nir_intrinsic_info
;
958 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
961 * \group texture information
963 * This gives semantic information about textures which is useful to the
964 * frontend, the backend, and lowering passes, but not the optimizer.
969 nir_tex_src_projector
,
970 nir_tex_src_comparitor
, /* shadow comparitor */
974 nir_tex_src_ms_index
, /* MSAA sample index */
977 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
978 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
979 nir_num_tex_src_types
984 nir_tex_src_type src_type
;
988 nir_texop_tex
, /**< Regular texture look-up */
989 nir_texop_txb
, /**< Texture look-up with LOD bias */
990 nir_texop_txl
, /**< Texture look-up with explicit LOD */
991 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
992 nir_texop_txf
, /**< Texel fetch with explicit LOD */
993 nir_texop_txf_ms
, /**< Multisample texture fetch */
994 nir_texop_txs
, /**< Texture size */
995 nir_texop_lod
, /**< Texture lod query */
996 nir_texop_tg4
, /**< Texture gather */
997 nir_texop_query_levels
, /**< Texture levels query */
998 nir_texop_texture_samples
, /**< Texture samples query */
999 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1007 enum glsl_sampler_dim sampler_dim
;
1008 nir_alu_type dest_type
;
1013 unsigned num_srcs
, coord_components
;
1014 bool is_array
, is_shadow
;
1017 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1018 * components or the new-style shadow that outputs 1 component.
1020 bool is_new_style_shadow
;
1022 /* constant offset - must be 0 if the offset source is used */
1023 int const_offset
[4];
1025 /* gather component selector */
1026 unsigned component
: 2;
1028 /** The texture index
1030 * If this texture instruction has a nir_tex_src_texture_offset source,
1031 * then the texture index is given by texture_index + texture_offset.
1033 unsigned texture_index
;
1035 /** The size of the texture array or 0 if it's not an array */
1036 unsigned texture_array_size
;
1038 /** The texture deref
1040 * If both this and `sampler` are both NULL, use texture_index instead.
1041 * If `texture` is NULL, but `sampler` is non-NULL, then the texture is
1042 * implied from the sampler.
1044 nir_deref_var
*texture
;
1046 /** The sampler index
1048 * If this texture instruction has a nir_tex_src_sampler_offset source,
1049 * then the sampler index is given by sampler_index + sampler_offset.
1051 unsigned sampler_index
;
1053 /** The sampler deref
1055 * If this is null, use sampler_index instead.
1057 nir_deref_var
*sampler
;
1060 static inline unsigned
1061 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1063 switch (instr
->op
) {
1064 case nir_texop_txs
: {
1066 switch (instr
->sampler_dim
) {
1067 case GLSL_SAMPLER_DIM_1D
:
1068 case GLSL_SAMPLER_DIM_BUF
:
1071 case GLSL_SAMPLER_DIM_2D
:
1072 case GLSL_SAMPLER_DIM_CUBE
:
1073 case GLSL_SAMPLER_DIM_MS
:
1074 case GLSL_SAMPLER_DIM_RECT
:
1075 case GLSL_SAMPLER_DIM_EXTERNAL
:
1078 case GLSL_SAMPLER_DIM_3D
:
1082 unreachable("not reached");
1084 if (instr
->is_array
)
1092 case nir_texop_texture_samples
:
1093 case nir_texop_query_levels
:
1094 case nir_texop_samples_identical
:
1098 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1105 /* Returns true if this texture operation queries something about the texture
1106 * rather than actually sampling it.
1109 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1111 switch (instr
->op
) {
1114 case nir_texop_texture_samples
:
1115 case nir_texop_query_levels
:
1122 case nir_texop_txf_ms
:
1126 unreachable("Invalid texture opcode");
1130 static inline unsigned
1131 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1133 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1134 return instr
->coord_components
;
1137 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1138 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1139 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1140 if (instr
->is_array
)
1141 return instr
->coord_components
- 1;
1143 return instr
->coord_components
;
1150 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1152 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1153 if (instr
->src
[i
].src_type
== type
)
1170 nir_const_value value
;
1173 } nir_load_const_instr
;
1186 /* creates a new SSA variable in an undefined state */
1191 } nir_ssa_undef_instr
;
1194 struct exec_node node
;
1196 /* The predecessor block corresponding to this source */
1197 struct nir_block
*pred
;
1202 #define nir_foreach_phi_src(phi, entry) \
1203 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1204 #define nir_foreach_phi_src_safe(phi, entry) \
1205 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1210 struct exec_list srcs
; /** < list of nir_phi_src */
1216 struct exec_node node
;
1219 } nir_parallel_copy_entry
;
1221 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1222 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1227 /* A list of nir_parallel_copy_entry's. The sources of all of the
1228 * entries are copied to the corresponding destinations "in parallel".
1229 * In other words, if we have two entries: a -> b and b -> a, the values
1232 struct exec_list entries
;
1233 } nir_parallel_copy_instr
;
1235 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1236 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1237 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1238 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1239 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1240 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1241 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1242 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1243 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1244 nir_parallel_copy_instr
, instr
)
1249 * Control flow consists of a tree of control flow nodes, which include
1250 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1251 * instructions that always run start-to-finish. Each basic block also keeps
1252 * track of its successors (blocks which may run immediately after the current
1253 * block) and predecessors (blocks which could have run immediately before the
1254 * current block). Each function also has a start block and an end block which
1255 * all return statements point to (which is always empty). Together, all the
1256 * blocks with their predecessors and successors make up the control flow
1257 * graph (CFG) of the function. There are helpers that modify the tree of
1258 * control flow nodes while modifying the CFG appropriately; these should be
1259 * used instead of modifying the tree directly.
1266 nir_cf_node_function
1269 typedef struct nir_cf_node
{
1270 struct exec_node node
;
1271 nir_cf_node_type type
;
1272 struct nir_cf_node
*parent
;
1275 typedef struct nir_block
{
1276 nir_cf_node cf_node
;
1278 struct exec_list instr_list
; /** < list of nir_instr */
1280 /** generic block index; generated by nir_index_blocks */
1284 * Each block can only have up to 2 successors, so we put them in a simple
1285 * array - no need for anything more complicated.
1287 struct nir_block
*successors
[2];
1289 /* Set of nir_block predecessors in the CFG */
1290 struct set
*predecessors
;
1293 * this node's immediate dominator in the dominance tree - set to NULL for
1296 struct nir_block
*imm_dom
;
1298 /* This node's children in the dominance tree */
1299 unsigned num_dom_children
;
1300 struct nir_block
**dom_children
;
1302 /* Set of nir_block's on the dominance frontier of this block */
1303 struct set
*dom_frontier
;
1306 * These two indices have the property that dom_{pre,post}_index for each
1307 * child of this block in the dominance tree will always be between
1308 * dom_pre_index and dom_post_index for this block, which makes testing if
1309 * a given block is dominated by another block an O(1) operation.
1311 unsigned dom_pre_index
, dom_post_index
;
1313 /* live in and out for this block; used for liveness analysis */
1314 BITSET_WORD
*live_in
;
1315 BITSET_WORD
*live_out
;
1318 static inline nir_instr
*
1319 nir_block_first_instr(nir_block
*block
)
1321 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1322 return exec_node_data(nir_instr
, head
, node
);
1325 static inline nir_instr
*
1326 nir_block_last_instr(nir_block
*block
)
1328 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1329 return exec_node_data(nir_instr
, tail
, node
);
1332 #define nir_foreach_instr(block, instr) \
1333 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1334 #define nir_foreach_instr_reverse(block, instr) \
1335 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1336 #define nir_foreach_instr_safe(block, instr) \
1337 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1338 #define nir_foreach_instr_reverse_safe(block, instr) \
1339 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1341 typedef struct nir_if
{
1342 nir_cf_node cf_node
;
1345 struct exec_list then_list
; /** < list of nir_cf_node */
1346 struct exec_list else_list
; /** < list of nir_cf_node */
1349 static inline nir_cf_node
*
1350 nir_if_first_then_node(nir_if
*if_stmt
)
1352 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1353 return exec_node_data(nir_cf_node
, head
, node
);
1356 static inline nir_cf_node
*
1357 nir_if_last_then_node(nir_if
*if_stmt
)
1359 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1360 return exec_node_data(nir_cf_node
, tail
, node
);
1363 static inline nir_cf_node
*
1364 nir_if_first_else_node(nir_if
*if_stmt
)
1366 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1367 return exec_node_data(nir_cf_node
, head
, node
);
1370 static inline nir_cf_node
*
1371 nir_if_last_else_node(nir_if
*if_stmt
)
1373 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1374 return exec_node_data(nir_cf_node
, tail
, node
);
1378 nir_cf_node cf_node
;
1380 struct exec_list body
; /** < list of nir_cf_node */
1383 static inline nir_cf_node
*
1384 nir_loop_first_cf_node(nir_loop
*loop
)
1386 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1389 static inline nir_cf_node
*
1390 nir_loop_last_cf_node(nir_loop
*loop
)
1392 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1396 * Various bits of metadata that can may be created or required by
1397 * optimization and analysis passes
1400 nir_metadata_none
= 0x0,
1401 nir_metadata_block_index
= 0x1,
1402 nir_metadata_dominance
= 0x2,
1403 nir_metadata_live_ssa_defs
= 0x4,
1404 nir_metadata_not_properly_reset
= 0x8,
1408 nir_cf_node cf_node
;
1410 /** pointer to the overload of which this is an implementation */
1411 struct nir_function_overload
*overload
;
1413 struct exec_list body
; /** < list of nir_cf_node */
1415 nir_block
*end_block
;
1417 /** list for all local variables in the function */
1418 struct exec_list locals
;
1420 /** array of variables used as parameters */
1421 unsigned num_params
;
1422 nir_variable
**params
;
1424 /** variable used to hold the result of the function */
1425 nir_variable
*return_var
;
1427 /** list of local registers in the function */
1428 struct exec_list registers
;
1430 /** next available local register index */
1433 /** next available SSA value index */
1436 /* total number of basic blocks, only valid when block_index_dirty = false */
1437 unsigned num_blocks
;
1439 nir_metadata valid_metadata
;
1440 } nir_function_impl
;
1442 static inline nir_block
*
1443 nir_start_block(nir_function_impl
*impl
)
1445 return (nir_block
*) exec_list_get_head(&impl
->body
);
1448 static inline nir_cf_node
*
1449 nir_cf_node_next(nir_cf_node
*node
)
1451 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1452 if (exec_node_is_tail_sentinel(next
))
1455 return exec_node_data(nir_cf_node
, next
, node
);
1458 static inline nir_cf_node
*
1459 nir_cf_node_prev(nir_cf_node
*node
)
1461 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1462 if (exec_node_is_head_sentinel(prev
))
1465 return exec_node_data(nir_cf_node
, prev
, node
);
1469 nir_cf_node_is_first(const nir_cf_node
*node
)
1471 return exec_node_is_head_sentinel(node
->node
.prev
);
1475 nir_cf_node_is_last(const nir_cf_node
*node
)
1477 return exec_node_is_tail_sentinel(node
->node
.next
);
1480 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1481 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1482 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1483 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1488 nir_parameter_inout
,
1489 } nir_parameter_type
;
1492 nir_parameter_type param_type
;
1493 const struct glsl_type
*type
;
1496 typedef struct nir_function_overload
{
1497 struct exec_node node
;
1499 unsigned num_params
;
1500 nir_parameter
*params
;
1501 const struct glsl_type
*return_type
;
1503 nir_function_impl
*impl
; /** < NULL if the overload is only declared yet */
1505 /** pointer to the function of which this is an overload */
1506 struct nir_function
*function
;
1507 } nir_function_overload
;
1509 typedef struct nir_function
{
1510 struct exec_node node
;
1512 struct exec_list overload_list
; /** < list of nir_function_overload */
1514 struct nir_shader
*shader
;
1517 #define nir_function_first_overload(func) \
1518 exec_node_data(nir_function_overload, \
1519 exec_list_get_head(&(func)->overload_list), node)
1521 typedef struct nir_shader_compiler_options
{
1528 /** lowers fneg and ineg to fsub and isub. */
1530 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1533 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1536 /* Does the native fdot instruction replicate its result for four
1537 * components? If so, then opt_algebraic_late will turn all fdotN
1538 * instructions into fdot_replicatedN instructions.
1540 bool fdot_replicates
;
1542 /** lowers ffract to fsub+ffloor: */
1546 * Does the driver support real 32-bit integers? (Otherwise, integers
1547 * are simulated by floats.)
1549 bool native_integers
;
1550 } nir_shader_compiler_options
;
1552 typedef struct nir_shader_info
{
1555 /* Descriptive name provided by the client; may be NULL */
1558 /* Number of textures used by this shader */
1559 unsigned num_textures
;
1560 /* Number of uniform buffers used by this shader */
1562 /* Number of atomic buffers used by this shader */
1564 /* Number of shader storage buffers used by this shader */
1566 /* Number of images used by this shader */
1567 unsigned num_images
;
1569 /* Which inputs are actually read */
1570 uint64_t inputs_read
;
1571 /* Which outputs are actually written */
1572 uint64_t outputs_written
;
1573 /* Which system values are actually read */
1574 uint64_t system_values_read
;
1576 /* Which patch inputs are actually read */
1577 uint32_t patch_inputs_read
;
1578 /* Which patch outputs are actually written */
1579 uint32_t patch_outputs_written
;
1581 /* Whether or not this shader ever uses textureGather() */
1582 bool uses_texture_gather
;
1584 /* Whether or not this shader uses the gl_ClipDistance output */
1585 bool uses_clip_distance_out
;
1587 /* Whether or not separate shader objects were used */
1588 bool separate_shader
;
1590 /** Was this shader linked with any transform feedback varyings? */
1591 bool has_transform_feedback_varyings
;
1595 /** The number of vertices recieves per input primitive */
1596 unsigned vertices_in
;
1598 /** The output primitive type (GL enum value) */
1599 unsigned output_primitive
;
1601 /** The maximum number of vertices the geometry shader might write. */
1602 unsigned vertices_out
;
1604 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1605 unsigned invocations
;
1607 /** Whether or not this shader uses EndPrimitive */
1608 bool uses_end_primitive
;
1610 /** Whether or not this shader uses non-zero streams */
1618 * Whether early fragment tests are enabled as defined by
1619 * ARB_shader_image_load_store.
1621 bool early_fragment_tests
;
1623 /** gl_FragDepth layout for ARB_conservative_depth. */
1624 enum gl_frag_depth_layout depth_layout
;
1628 unsigned local_size
[3];
1632 /** The number of vertices in the TCS output patch. */
1633 unsigned vertices_out
;
1638 typedef struct nir_shader
{
1639 /** list of uniforms (nir_variable) */
1640 struct exec_list uniforms
;
1642 /** list of inputs (nir_variable) */
1643 struct exec_list inputs
;
1645 /** list of outputs (nir_variable) */
1646 struct exec_list outputs
;
1648 /** Set of driver-specific options for the shader.
1650 * The memory for the options is expected to be kept in a single static
1651 * copy by the driver.
1653 const struct nir_shader_compiler_options
*options
;
1655 /** Various bits of compile-time information about a given shader */
1656 struct nir_shader_info info
;
1658 /** list of global variables in the shader (nir_variable) */
1659 struct exec_list globals
;
1661 /** list of system value variables in the shader (nir_variable) */
1662 struct exec_list system_values
;
1664 struct exec_list functions
; /** < list of nir_function */
1666 /** list of global register in the shader */
1667 struct exec_list registers
;
1669 /** next available global register index */
1673 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1676 unsigned num_inputs
, num_uniforms
, num_outputs
;
1678 /** The shader stage, such as MESA_SHADER_VERTEX. */
1679 gl_shader_stage stage
;
1682 #define nir_foreach_overload(shader, overload) \
1683 foreach_list_typed(nir_function, func, node, &(shader)->functions) \
1684 foreach_list_typed(nir_function_overload, overload, node, \
1685 &(func)->overload_list)
1687 nir_shader
*nir_shader_create(void *mem_ctx
,
1688 gl_shader_stage stage
,
1689 const nir_shader_compiler_options
*options
);
1691 /** creates a register, including assigning it an index and adding it to the list */
1692 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1694 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1696 void nir_reg_remove(nir_register
*reg
);
1698 /** Adds a variable to the appropreate list in nir_shader */
1699 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1702 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1704 assert(var
->data
.mode
== nir_var_local
);
1705 exec_list_push_tail(&impl
->locals
, &var
->node
);
1708 /** creates a variable, sets a few defaults, and adds it to the list */
1709 nir_variable
*nir_variable_create(nir_shader
*shader
,
1710 nir_variable_mode mode
,
1711 const struct glsl_type
*type
,
1713 /** creates a local variable and adds it to the list */
1714 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1715 const struct glsl_type
*type
,
1718 /** creates a function and adds it to the shader's list of functions */
1719 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1721 /** creates a null function returning null */
1722 nir_function_overload
*nir_function_overload_create(nir_function
*func
);
1724 nir_function_impl
*nir_function_impl_create(nir_function_overload
*func
);
1725 /** creates a function_impl that isn't tied to any particular overload */
1726 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1728 nir_block
*nir_block_create(nir_shader
*shader
);
1729 nir_if
*nir_if_create(nir_shader
*shader
);
1730 nir_loop
*nir_loop_create(nir_shader
*shader
);
1732 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1734 /** requests that the given pieces of metadata be generated */
1735 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1736 /** dirties all but the preserved metadata */
1737 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1739 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1740 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1742 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1744 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1745 unsigned num_components
);
1747 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1748 nir_intrinsic_op op
);
1750 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1751 nir_function_overload
*callee
);
1753 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1755 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1757 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1759 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1760 unsigned num_components
);
1762 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1763 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1764 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1766 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1768 nir_load_const_instr
*
1769 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1772 * NIR Cursors and Instruction Insertion API
1775 * A tiny struct representing a point to insert/extract instructions or
1776 * control flow nodes. Helps reduce the combinatorial explosion of possible
1777 * points to insert/extract.
1779 * \sa nir_control_flow.h
1782 nir_cursor_before_block
,
1783 nir_cursor_after_block
,
1784 nir_cursor_before_instr
,
1785 nir_cursor_after_instr
,
1786 } nir_cursor_option
;
1789 nir_cursor_option option
;
1796 static inline nir_block
*
1797 nir_cursor_current_block(nir_cursor cursor
)
1799 if (cursor
.option
== nir_cursor_before_instr
||
1800 cursor
.option
== nir_cursor_after_instr
) {
1801 return cursor
.instr
->block
;
1803 return cursor
.block
;
1807 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
1809 static inline nir_cursor
1810 nir_before_block(nir_block
*block
)
1813 cursor
.option
= nir_cursor_before_block
;
1814 cursor
.block
= block
;
1818 static inline nir_cursor
1819 nir_after_block(nir_block
*block
)
1822 cursor
.option
= nir_cursor_after_block
;
1823 cursor
.block
= block
;
1827 static inline nir_cursor
1828 nir_before_instr(nir_instr
*instr
)
1831 cursor
.option
= nir_cursor_before_instr
;
1832 cursor
.instr
= instr
;
1836 static inline nir_cursor
1837 nir_after_instr(nir_instr
*instr
)
1840 cursor
.option
= nir_cursor_after_instr
;
1841 cursor
.instr
= instr
;
1845 static inline nir_cursor
1846 nir_after_block_before_jump(nir_block
*block
)
1848 nir_instr
*last_instr
= nir_block_last_instr(block
);
1849 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1850 return nir_before_instr(last_instr
);
1852 return nir_after_block(block
);
1856 static inline nir_cursor
1857 nir_before_cf_node(nir_cf_node
*node
)
1859 if (node
->type
== nir_cf_node_block
)
1860 return nir_before_block(nir_cf_node_as_block(node
));
1862 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1865 static inline nir_cursor
1866 nir_after_cf_node(nir_cf_node
*node
)
1868 if (node
->type
== nir_cf_node_block
)
1869 return nir_after_block(nir_cf_node_as_block(node
));
1871 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1874 static inline nir_cursor
1875 nir_before_cf_list(struct exec_list
*cf_list
)
1877 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1878 exec_list_get_head(cf_list
), node
);
1879 return nir_before_cf_node(first_node
);
1882 static inline nir_cursor
1883 nir_after_cf_list(struct exec_list
*cf_list
)
1885 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1886 exec_list_get_tail(cf_list
), node
);
1887 return nir_after_cf_node(last_node
);
1891 * Insert a NIR instruction at the given cursor.
1893 * Note: This does not update the cursor.
1895 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
1898 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1900 nir_instr_insert(nir_before_instr(instr
), before
);
1904 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
1906 nir_instr_insert(nir_after_instr(instr
), after
);
1910 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
1912 nir_instr_insert(nir_before_block(block
), before
);
1916 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
1918 nir_instr_insert(nir_after_block(block
), after
);
1922 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
1924 nir_instr_insert(nir_before_cf_node(node
), before
);
1928 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
1930 nir_instr_insert(nir_after_cf_node(node
), after
);
1934 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
1936 nir_instr_insert(nir_before_cf_list(list
), before
);
1940 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
1942 nir_instr_insert(nir_after_cf_list(list
), after
);
1945 void nir_instr_remove(nir_instr
*instr
);
1949 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
1950 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
1951 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
1952 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
1954 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
1955 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
1957 nir_const_value
*nir_src_as_const_value(nir_src src
);
1958 bool nir_src_is_dynamically_uniform(nir_src src
);
1959 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
1960 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
1961 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
1962 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
1963 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
1966 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1967 unsigned num_components
, const char *name
);
1968 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1969 unsigned num_components
, const char *name
);
1970 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
1971 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
1972 nir_instr
*after_me
);
1974 /* visits basic blocks in source-code order */
1975 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
1976 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1978 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1980 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1983 /* If the following CF node is an if, this function returns that if.
1984 * Otherwise, it returns NULL.
1986 nir_if
*nir_block_get_following_if(nir_block
*block
);
1988 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
1990 void nir_index_local_regs(nir_function_impl
*impl
);
1991 void nir_index_global_regs(nir_shader
*shader
);
1992 void nir_index_ssa_defs(nir_function_impl
*impl
);
1993 unsigned nir_index_instrs(nir_function_impl
*impl
);
1995 void nir_index_blocks(nir_function_impl
*impl
);
1997 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
1998 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2000 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2001 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*impl
);
2004 void nir_validate_shader(nir_shader
*shader
);
2005 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2006 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2008 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2009 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2010 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2013 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2014 void nir_calc_dominance(nir_shader
*shader
);
2016 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2017 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2019 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2020 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2022 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2023 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2025 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2026 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2028 int nir_gs_count_vertices(const nir_shader
*shader
);
2030 bool nir_split_var_copies(nir_shader
*shader
);
2032 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2033 bool nir_lower_returns(nir_shader
*shader
);
2035 bool nir_inline_functions_impl(nir_function_impl
*impl
);
2036 bool nir_inline_functions(nir_shader
*shader
);
2038 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2039 void nir_lower_var_copies(nir_shader
*shader
);
2041 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2043 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2045 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
2047 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
2049 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2051 void nir_assign_var_locations(struct exec_list
*var_list
,
2053 int (*type_size
)(const struct glsl_type
*));
2055 void nir_lower_io(nir_shader
*shader
,
2056 nir_variable_mode mode
,
2057 int (*type_size
)(const struct glsl_type
*));
2058 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2059 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2061 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2063 bool nir_remove_dead_variables(nir_shader
*shader
);
2065 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2066 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2067 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2068 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2070 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2072 void nir_lower_samplers(nir_shader
*shader
,
2073 const struct gl_shader_program
*shader_program
);
2075 bool nir_lower_system_values(nir_shader
*shader
);
2077 typedef struct nir_lower_tex_options
{
2079 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2080 * sampler types a texture projector is lowered.
2085 * If true, lower rect textures to 2D, using txs to fetch the
2086 * texture dimensions and dividing the texture coords by the
2087 * texture dims to normalize.
2092 * To emulate certain texture wrap modes, this can be used
2093 * to saturate the specified tex coord to [0.0, 1.0]. The
2094 * bits are according to sampler #, ie. if, for example:
2096 * (conf->saturate_s & (1 << n))
2098 * is true, then the s coord for sampler n is saturated.
2100 * Note that clamping must happen *after* projector lowering
2101 * so any projected texture sample instruction with a clamped
2102 * coordinate gets automatically lowered, regardless of the
2103 * 'lower_txp' setting.
2105 unsigned saturate_s
;
2106 unsigned saturate_t
;
2107 unsigned saturate_r
;
2109 /* Bitmask of samplers that need swizzling.
2111 * If (swizzle_result & (1 << sampler_index)), then the swizzle in
2112 * swizzles[sampler_index] is applied to the result of the texturing
2115 unsigned swizzle_result
;
2117 /* A swizzle for each sampler. Values 0-3 represent x, y, z, or w swizzles
2118 * while 4 and 5 represent 0 and 1 respectively.
2120 uint8_t swizzles
[32][4];
2121 } nir_lower_tex_options
;
2123 bool nir_lower_tex(nir_shader
*shader
,
2124 const nir_lower_tex_options
*options
);
2126 void nir_lower_idiv(nir_shader
*shader
);
2128 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2129 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2131 void nir_lower_two_sided_color(nir_shader
*shader
);
2133 void nir_lower_atomics(nir_shader
*shader
,
2134 const struct gl_shader_program
*shader_program
);
2135 void nir_lower_to_source_mods(nir_shader
*shader
);
2137 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2139 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2141 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2142 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2144 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2145 void nir_convert_to_ssa(nir_shader
*shader
);
2147 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2148 * registers. If false, convert all values (even those not involved in a phi
2149 * node) to registers.
2151 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2153 bool nir_opt_algebraic(nir_shader
*shader
);
2154 bool nir_opt_algebraic_late(nir_shader
*shader
);
2155 bool nir_opt_constant_folding(nir_shader
*shader
);
2157 bool nir_opt_global_to_local(nir_shader
*shader
);
2159 bool nir_copy_prop(nir_shader
*shader
);
2161 bool nir_opt_cse(nir_shader
*shader
);
2163 bool nir_opt_dce(nir_shader
*shader
);
2165 bool nir_opt_dead_cf(nir_shader
*shader
);
2167 void nir_opt_gcm(nir_shader
*shader
);
2169 bool nir_opt_peephole_select(nir_shader
*shader
);
2171 bool nir_opt_remove_phis(nir_shader
*shader
);
2173 bool nir_opt_undef(nir_shader
*shader
);
2175 void nir_sweep(nir_shader
*shader
);
2177 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2178 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);