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); \
74 * Description of built-in state associated with a uniform
76 * \sa nir_variable::state_slots
90 nir_var_shader_storage
,
95 * Data stored in an nir_constant
97 union nir_constant_data
{
104 typedef struct nir_constant
{
106 * Value of the constant.
108 * The field used to back the values supplied by the constant is determined
109 * by the type associated with the \c nir_variable. Constants may be
110 * scalars, vectors, or matrices.
112 union nir_constant_data value
;
114 /* we could get this from the var->type but makes clone *much* easier to
115 * not have to care about the type.
117 unsigned num_elements
;
119 /* Array elements / Structure Fields */
120 struct nir_constant
**elements
;
124 * \brief Layout qualifiers for gl_FragDepth.
126 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
127 * with a layout qualifier.
130 nir_depth_layout_none
, /**< No depth layout is specified. */
131 nir_depth_layout_any
,
132 nir_depth_layout_greater
,
133 nir_depth_layout_less
,
134 nir_depth_layout_unchanged
138 * Either a uniform, global variable, shader input, or shader output. Based on
139 * ir_variable - it should be easy to translate between the two.
143 struct exec_node node
;
146 * Declared type of the variable
148 const struct glsl_type
*type
;
151 * Declared name of the variable
155 struct nir_variable_data
{
158 * Is the variable read-only?
160 * This is set for variables declared as \c const, shader inputs,
163 unsigned read_only
:1;
167 unsigned invariant
:1;
170 * Storage class of the variable.
172 * \sa nir_variable_mode
174 nir_variable_mode mode
:4;
177 * Interpolation mode for shader inputs / outputs
179 * \sa glsl_interp_qualifier
181 unsigned interpolation
:2;
184 * \name ARB_fragment_coord_conventions
187 unsigned origin_upper_left
:1;
188 unsigned pixel_center_integer
:1;
192 * Was the location explicitly set in the shader?
194 * If the location is explicitly set in the shader, it \b cannot be changed
195 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
198 unsigned explicit_location
:1;
199 unsigned explicit_index
:1;
202 * Was an initial binding explicitly set in the shader?
204 * If so, constant_initializer contains an integer nir_constant
205 * representing the initial binding point.
207 unsigned explicit_binding
:1;
210 * Does this variable have an initializer?
212 * This is used by the linker to cross-validiate initializers of global
215 unsigned has_initializer
:1;
218 * If non-zero, then this variable may be packed along with other variables
219 * into a single varying slot, so this offset should be applied when
220 * accessing components. For example, an offset of 1 means that the x
221 * component of this variable is actually stored in component y of the
222 * location specified by \c location.
224 unsigned location_frac
:2;
227 * Non-zero if this variable was created by lowering a named interface
228 * block which was not an array.
230 * Note that this variable and \c from_named_ifc_block_array will never
233 unsigned from_named_ifc_block_nonarray
:1;
236 * Non-zero if this variable was created by lowering a named interface
237 * block which was an array.
239 * Note that this variable and \c from_named_ifc_block_nonarray will never
242 unsigned from_named_ifc_block_array
:1;
245 * \brief Layout qualifier for gl_FragDepth.
247 * This is not equal to \c ir_depth_layout_none if and only if this
248 * variable is \c gl_FragDepth and a layout qualifier is specified.
250 nir_depth_layout depth_layout
;
253 * Storage location of the base of this variable
255 * The precise meaning of this field depends on the nature of the variable.
257 * - Vertex shader input: one of the values from \c gl_vert_attrib.
258 * - Vertex shader output: one of the values from \c gl_varying_slot.
259 * - Geometry shader input: one of the values from \c gl_varying_slot.
260 * - Geometry shader output: one of the values from \c gl_varying_slot.
261 * - Fragment shader input: one of the values from \c gl_varying_slot.
262 * - Fragment shader output: one of the values from \c gl_frag_result.
263 * - Uniforms: Per-stage uniform slot number for default uniform block.
264 * - Uniforms: Index within the uniform block definition for UBO members.
265 * - Non-UBO Uniforms: uniform slot number.
266 * - Other: This field is not currently used.
268 * If the variable is a uniform, shader input, or shader output, and the
269 * slot has not been assigned, the value will be -1.
274 * The actual location of the variable in the IR. Only valid for inputs
277 unsigned int driver_location
;
280 * output index for dual source blending.
285 * Descriptor set binding for sampler or UBO.
290 * Initial binding point for a sampler or UBO.
292 * For array types, this represents the binding point for the first element.
297 * Location an atomic counter is stored at.
304 * ARB_shader_image_load_store qualifiers.
307 bool read_only
; /**< "readonly" qualifier. */
308 bool write_only
; /**< "writeonly" qualifier. */
313 /** Image internal format if specified explicitly, otherwise GL_NONE. */
318 * Highest element accessed with a constant expression array index
320 * Not used for non-array variables.
322 unsigned max_array_access
;
327 * Built-in state that backs this uniform
329 * Once set at variable creation, \c state_slots must remain invariant.
330 * This is because, ideally, this array would be shared by all clones of
331 * this variable in the IR tree. In other words, we'd really like for it
332 * to be a fly-weight.
334 * If the variable is not a uniform, \c num_state_slots will be zero and
335 * \c state_slots will be \c NULL.
338 unsigned num_state_slots
; /**< Number of state slots used */
339 nir_state_slot
*state_slots
; /**< State descriptors. */
343 * Constant expression assigned in the initializer of the variable
345 nir_constant
*constant_initializer
;
348 * For variables that are in an interface block or are an instance of an
349 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
351 * \sa ir_variable::location
353 const struct glsl_type
*interface_type
;
356 #define nir_foreach_variable(var, var_list) \
357 foreach_list_typed(nir_variable, var, node, var_list)
360 * Returns the bits in the inputs_read, outputs_written, or
361 * system_values_read bitfield corresponding to this variable.
363 static inline uint64_t
364 nir_variable_get_io_mask(nir_variable
*var
, gl_shader_stage stage
)
366 assert(var
->data
.mode
== nir_var_shader_in
||
367 var
->data
.mode
== nir_var_shader_out
||
368 var
->data
.mode
== nir_var_system_value
);
369 assert(var
->data
.location
>= 0);
371 const struct glsl_type
*var_type
= var
->type
;
372 if (stage
== MESA_SHADER_GEOMETRY
&& var
->data
.mode
== nir_var_shader_in
) {
373 /* Most geometry shader inputs are per-vertex arrays */
374 if (var
->data
.location
>= VARYING_SLOT_VAR0
)
375 assert(glsl_type_is_array(var_type
));
377 if (glsl_type_is_array(var_type
))
378 var_type
= glsl_get_array_element(var_type
);
381 bool is_vertex_input
= (var
->data
.mode
== nir_var_shader_in
&&
382 stage
== MESA_SHADER_VERTEX
);
383 unsigned slots
= glsl_count_attribute_slots(var_type
, is_vertex_input
);
384 return ((1ull << slots
) - 1) << var
->data
.location
;
388 struct exec_node node
;
390 unsigned num_components
; /** < number of vector components */
391 unsigned num_array_elems
; /** < size of array (0 for no array) */
393 /** generic register index. */
396 /** only for debug purposes, can be NULL */
399 /** whether this register is local (per-function) or global (per-shader) */
403 * If this flag is set to true, then accessing channels >= num_components
404 * is well-defined, and simply spills over to the next array element. This
405 * is useful for backends that can do per-component accessing, in
406 * particular scalar backends. By setting this flag and making
407 * num_components equal to 1, structures can be packed tightly into
408 * registers and then registers can be accessed per-component to get to
409 * each structure member, even if it crosses vec4 boundaries.
413 /** set of nir_src's where this register is used (read from) */
414 struct list_head uses
;
416 /** set of nir_dest's where this register is defined (written to) */
417 struct list_head defs
;
419 /** set of nir_if's where this register is used as a condition */
420 struct list_head if_uses
;
427 nir_instr_type_intrinsic
,
428 nir_instr_type_load_const
,
430 nir_instr_type_ssa_undef
,
432 nir_instr_type_parallel_copy
,
435 typedef struct nir_instr
{
436 struct exec_node node
;
438 struct nir_block
*block
;
440 /** generic instruction index. */
443 /* A temporary for optimization and analysis passes to use for storing
444 * flags. For instance, DCE uses this to store the "dead/live" info.
449 static inline nir_instr
*
450 nir_instr_next(nir_instr
*instr
)
452 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
453 if (exec_node_is_tail_sentinel(next
))
456 return exec_node_data(nir_instr
, next
, node
);
459 static inline nir_instr
*
460 nir_instr_prev(nir_instr
*instr
)
462 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
463 if (exec_node_is_head_sentinel(prev
))
466 return exec_node_data(nir_instr
, prev
, node
);
470 nir_instr_is_first(nir_instr
*instr
)
472 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
476 nir_instr_is_last(nir_instr
*instr
)
478 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
482 /** for debugging only, can be NULL */
485 /** generic SSA definition index. */
488 /** Index into the live_in and live_out bitfields */
491 nir_instr
*parent_instr
;
493 /** set of nir_instr's where this register is used (read from) */
494 struct list_head uses
;
496 /** set of nir_if's where this register is used as a condition */
497 struct list_head if_uses
;
499 uint8_t num_components
;
506 struct nir_src
*indirect
; /** < NULL for no indirect offset */
507 unsigned base_offset
;
509 /* TODO use-def chain goes here */
513 nir_instr
*parent_instr
;
514 struct list_head def_link
;
517 struct nir_src
*indirect
; /** < NULL for no indirect offset */
518 unsigned base_offset
;
520 /* TODO def-use chain goes here */
525 typedef struct nir_src
{
527 nir_instr
*parent_instr
;
528 struct nir_if
*parent_if
;
531 struct list_head use_link
;
542 # define NIR_SRC_INIT nir_src()
544 # define NIR_SRC_INIT (nir_src) { { NULL } }
547 #define nir_foreach_use(reg_or_ssa_def, src) \
548 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
550 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
551 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
553 #define nir_foreach_if_use(reg_or_ssa_def, src) \
554 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
556 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
557 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
569 # define NIR_DEST_INIT nir_dest()
571 # define NIR_DEST_INIT (nir_dest) { { { NULL } } }
574 #define nir_foreach_def(reg, dest) \
575 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
577 #define nir_foreach_def_safe(reg, dest) \
578 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
580 static inline nir_src
581 nir_src_for_ssa(nir_ssa_def
*def
)
583 nir_src src
= NIR_SRC_INIT
;
591 static inline nir_src
592 nir_src_for_reg(nir_register
*reg
)
594 nir_src src
= NIR_SRC_INIT
;
598 src
.reg
.indirect
= NULL
;
599 src
.reg
.base_offset
= 0;
604 static inline nir_dest
605 nir_dest_for_reg(nir_register
*reg
)
607 nir_dest dest
= NIR_DEST_INIT
;
614 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
615 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
621 * \name input modifiers
625 * For inputs interpreted as floating point, flips the sign bit. For
626 * inputs interpreted as integers, performs the two's complement negation.
631 * Clears the sign bit for floating point values, and computes the integer
632 * absolute value for integers. Note that the negate modifier acts after
633 * the absolute value modifier, therefore if both are set then all inputs
634 * will become negative.
640 * For each input component, says which component of the register it is
641 * chosen from. Note that which elements of the swizzle are used and which
642 * are ignored are based on the write mask for most opcodes - for example,
643 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
644 * a swizzle of {2, x, 1, 0} where x means "don't care."
653 * \name saturate output modifier
655 * Only valid for opcodes that output floating-point numbers. Clamps the
656 * output to between 0.0 and 1.0 inclusive.
661 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
665 nir_type_invalid
= 0, /* Not a valid type */
673 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
674 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
675 } nir_op_algebraic_property
;
683 * The number of components in the output
685 * If non-zero, this is the size of the output and input sizes are
686 * explicitly given; swizzle and writemask are still in effect, but if
687 * the output component is masked out, then the input component may
690 * If zero, the opcode acts in the standard, per-component manner; the
691 * operation is performed on each component (except the ones that are
692 * masked out) with the input being taken from the input swizzle for
695 * The size of some of the inputs may be given (i.e. non-zero) even
696 * though output_size is zero; in that case, the inputs with a zero
697 * size act per-component, while the inputs with non-zero size don't.
699 unsigned output_size
;
702 * The type of vector that the instruction outputs. Note that the
703 * staurate modifier is only allowed on outputs with the float type.
706 nir_alu_type output_type
;
709 * The number of components in each input
711 unsigned input_sizes
[4];
714 * The type of vector that each input takes. Note that negate and
715 * absolute value are only allowed on inputs with int or float type and
716 * behave differently on the two.
718 nir_alu_type input_types
[4];
720 nir_op_algebraic_property algebraic_properties
;
723 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
725 typedef struct nir_alu_instr
{
732 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
733 nir_alu_instr
*instr
);
734 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
735 nir_alu_instr
*instr
);
737 /* is this source channel used? */
739 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
741 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
742 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
744 return (instr
->dest
.write_mask
>> channel
) & 1;
748 * For instructions whose destinations are SSA, get the number of channels
751 static inline unsigned
752 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
754 assert(instr
->dest
.dest
.is_ssa
);
756 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
757 return nir_op_infos
[instr
->op
].input_sizes
[src
];
759 return instr
->dest
.dest
.ssa
.num_components
;
764 nir_deref_type_array
,
765 nir_deref_type_struct
768 typedef struct nir_deref
{
769 nir_deref_type deref_type
;
770 struct nir_deref
*child
;
771 const struct glsl_type
*type
;
780 /* This enum describes how the array is referenced. If the deref is
781 * direct then the base_offset is used. If the deref is indirect then then
782 * offset is given by base_offset + indirect. If the deref is a wildcard
783 * then the deref refers to all of the elements of the array at the same
784 * time. Wildcard dereferences are only ever allowed in copy_var
785 * intrinsics and the source and destination derefs must have matching
789 nir_deref_array_type_direct
,
790 nir_deref_array_type_indirect
,
791 nir_deref_array_type_wildcard
,
792 } nir_deref_array_type
;
797 nir_deref_array_type deref_array_type
;
798 unsigned base_offset
;
808 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
809 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
810 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
812 /* Returns the last deref in the chain. */
813 static inline nir_deref
*
814 nir_deref_tail(nir_deref
*deref
)
817 deref
= deref
->child
;
825 nir_deref_var
**params
;
826 nir_deref_var
*return_deref
;
828 struct nir_function
*callee
;
831 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
832 num_variables, num_indices, flags) \
833 nir_intrinsic_##name,
835 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
838 #include "nir_intrinsics.h"
839 nir_num_intrinsics
= nir_last_intrinsic
+ 1
843 #undef LAST_INTRINSIC
845 /** Represents an intrinsic
847 * An intrinsic is an instruction type for handling things that are
848 * more-or-less regular operations but don't just consume and produce SSA
849 * values like ALU operations do. Intrinsics are not for things that have
850 * special semantic meaning such as phi nodes and parallel copies.
851 * Examples of intrinsics include variable load/store operations, system
852 * value loads, and the like. Even though texturing more-or-less falls
853 * under this category, texturing is its own instruction type because
854 * trying to represent texturing with intrinsics would lead to a
855 * combinatorial explosion of intrinsic opcodes.
857 * By having a single instruction type for handling a lot of different
858 * cases, optimization passes can look for intrinsics and, for the most
859 * part, completely ignore them. Each intrinsic type also has a few
860 * possible flags that govern whether or not they can be reordered or
861 * eliminated. That way passes like dead code elimination can still work
862 * on intrisics without understanding the meaning of each.
864 * Each intrinsic has some number of constant indices, some number of
865 * variables, and some number of sources. What these sources, variables,
866 * and indices mean depends on the intrinsic and is documented with the
867 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
868 * instructions are the only types of instruction that can operate on
874 nir_intrinsic_op intrinsic
;
878 /** number of components if this is a vectorized intrinsic
880 * Similarly to ALU operations, some intrinsics are vectorized.
881 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
882 * For vectorized intrinsics, the num_components field specifies the
883 * number of destination components and the number of source components
884 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
886 uint8_t num_components
;
890 nir_deref_var
*variables
[2];
893 } nir_intrinsic_instr
;
896 * \name NIR intrinsics semantic flags
898 * information about what the compiler can do with the intrinsics.
900 * \sa nir_intrinsic_info::flags
904 * whether the intrinsic can be safely eliminated if none of its output
905 * value is not being used.
907 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
910 * Whether the intrinsic can be reordered with respect to any other
911 * intrinsic, i.e. whether the only reordering dependencies of the
912 * intrinsic are due to the register reads/writes.
914 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
915 } nir_intrinsic_semantic_flag
;
917 #define NIR_INTRINSIC_MAX_INPUTS 4
922 unsigned num_srcs
; /** < number of register/SSA inputs */
924 /** number of components of each input register
926 * If this value is 0, the number of components is given by the
927 * num_components field of nir_intrinsic_instr.
929 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
933 /** number of components of the output register
935 * If this value is 0, the number of components is given by the
936 * num_components field of nir_intrinsic_instr.
938 unsigned dest_components
;
940 /** the number of inputs/outputs that are variables */
941 unsigned num_variables
;
943 /** the number of constant indices used by the intrinsic */
944 unsigned num_indices
;
946 /** semantic flags for calls to this intrinsic */
947 nir_intrinsic_semantic_flag flags
;
948 } nir_intrinsic_info
;
950 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
953 * \group texture information
955 * This gives semantic information about textures which is useful to the
956 * frontend, the backend, and lowering passes, but not the optimizer.
961 nir_tex_src_projector
,
962 nir_tex_src_comparitor
, /* shadow comparitor */
966 nir_tex_src_ms_index
, /* MSAA sample index */
969 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
970 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
971 nir_num_tex_src_types
976 nir_tex_src_type src_type
;
980 nir_texop_tex
, /**< Regular texture look-up */
981 nir_texop_txb
, /**< Texture look-up with LOD bias */
982 nir_texop_txl
, /**< Texture look-up with explicit LOD */
983 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
984 nir_texop_txf
, /**< Texel fetch with explicit LOD */
985 nir_texop_txf_ms
, /**< Multisample texture fetch */
986 nir_texop_txs
, /**< Texture size */
987 nir_texop_lod
, /**< Texture lod query */
988 nir_texop_tg4
, /**< Texture gather */
989 nir_texop_query_levels
, /**< Texture levels query */
990 nir_texop_texture_samples
, /**< Texture samples query */
991 nir_texop_samples_identical
, /**< Query whether all samples are definitely
999 enum glsl_sampler_dim sampler_dim
;
1000 nir_alu_type dest_type
;
1005 unsigned num_srcs
, coord_components
;
1006 bool is_array
, is_shadow
;
1009 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1010 * components or the new-style shadow that outputs 1 component.
1012 bool is_new_style_shadow
;
1014 /* constant offset - must be 0 if the offset source is used */
1015 int const_offset
[4];
1017 /* gather component selector */
1018 unsigned component
: 2;
1020 /** The texture index
1022 * If this texture instruction has a nir_tex_src_texture_offset source,
1023 * then the texture index is given by texture_index + texture_offset.
1025 unsigned texture_index
;
1027 /** The size of the texture array or 0 if it's not an array */
1028 unsigned texture_array_size
;
1030 /** The texture deref
1032 * If both this and `sampler` are both NULL, use texture_index instead.
1033 * If `texture` is NULL, but `sampler` is non-NULL, then the texture is
1034 * implied from the sampler.
1036 nir_deref_var
*texture
;
1038 /** The sampler index
1040 * If this texture instruction has a nir_tex_src_sampler_offset source,
1041 * then the sampler index is given by sampler_index + sampler_offset.
1043 unsigned sampler_index
;
1045 /** The sampler deref
1047 * If this is null, use sampler_index instead.
1049 nir_deref_var
*sampler
;
1052 static inline unsigned
1053 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1055 switch (instr
->op
) {
1056 case nir_texop_txs
: {
1058 switch (instr
->sampler_dim
) {
1059 case GLSL_SAMPLER_DIM_1D
:
1060 case GLSL_SAMPLER_DIM_BUF
:
1063 case GLSL_SAMPLER_DIM_2D
:
1064 case GLSL_SAMPLER_DIM_CUBE
:
1065 case GLSL_SAMPLER_DIM_MS
:
1066 case GLSL_SAMPLER_DIM_RECT
:
1067 case GLSL_SAMPLER_DIM_EXTERNAL
:
1070 case GLSL_SAMPLER_DIM_3D
:
1074 unreachable("not reached");
1076 if (instr
->is_array
)
1084 case nir_texop_texture_samples
:
1085 case nir_texop_query_levels
:
1086 case nir_texop_samples_identical
:
1090 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1097 /* Returns true if this texture operation queries something about the texture
1098 * rather than actually sampling it.
1101 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1103 switch (instr
->op
) {
1106 case nir_texop_texture_samples
:
1107 case nir_texop_query_levels
:
1114 case nir_texop_txf_ms
:
1118 unreachable("Invalid texture opcode");
1122 static inline unsigned
1123 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1125 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1126 return instr
->coord_components
;
1129 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1130 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1131 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1132 if (instr
->is_array
)
1133 return instr
->coord_components
- 1;
1135 return instr
->coord_components
;
1142 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1144 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1145 if (instr
->src
[i
].src_type
== type
)
1162 nir_const_value value
;
1165 } nir_load_const_instr
;
1178 /* creates a new SSA variable in an undefined state */
1183 } nir_ssa_undef_instr
;
1186 struct exec_node node
;
1188 /* The predecessor block corresponding to this source */
1189 struct nir_block
*pred
;
1194 #define nir_foreach_phi_src(phi, entry) \
1195 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1196 #define nir_foreach_phi_src_safe(phi, entry) \
1197 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1202 struct exec_list srcs
; /** < list of nir_phi_src */
1208 struct exec_node node
;
1211 } nir_parallel_copy_entry
;
1213 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1214 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1219 /* A list of nir_parallel_copy_entry's. The sources of all of the
1220 * entries are copied to the corresponding destinations "in parallel".
1221 * In other words, if we have two entries: a -> b and b -> a, the values
1224 struct exec_list entries
;
1225 } nir_parallel_copy_instr
;
1227 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1228 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1229 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1230 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1231 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1232 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1233 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1234 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1235 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1236 nir_parallel_copy_instr
, instr
)
1241 * Control flow consists of a tree of control flow nodes, which include
1242 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1243 * instructions that always run start-to-finish. Each basic block also keeps
1244 * track of its successors (blocks which may run immediately after the current
1245 * block) and predecessors (blocks which could have run immediately before the
1246 * current block). Each function also has a start block and an end block which
1247 * all return statements point to (which is always empty). Together, all the
1248 * blocks with their predecessors and successors make up the control flow
1249 * graph (CFG) of the function. There are helpers that modify the tree of
1250 * control flow nodes while modifying the CFG appropriately; these should be
1251 * used instead of modifying the tree directly.
1258 nir_cf_node_function
1261 typedef struct nir_cf_node
{
1262 struct exec_node node
;
1263 nir_cf_node_type type
;
1264 struct nir_cf_node
*parent
;
1267 typedef struct nir_block
{
1268 nir_cf_node cf_node
;
1270 struct exec_list instr_list
; /** < list of nir_instr */
1272 /** generic block index; generated by nir_index_blocks */
1276 * Each block can only have up to 2 successors, so we put them in a simple
1277 * array - no need for anything more complicated.
1279 struct nir_block
*successors
[2];
1281 /* Set of nir_block predecessors in the CFG */
1282 struct set
*predecessors
;
1285 * this node's immediate dominator in the dominance tree - set to NULL for
1288 struct nir_block
*imm_dom
;
1290 /* This node's children in the dominance tree */
1291 unsigned num_dom_children
;
1292 struct nir_block
**dom_children
;
1294 /* Set of nir_block's on the dominance frontier of this block */
1295 struct set
*dom_frontier
;
1298 * These two indices have the property that dom_{pre,post}_index for each
1299 * child of this block in the dominance tree will always be between
1300 * dom_pre_index and dom_post_index for this block, which makes testing if
1301 * a given block is dominated by another block an O(1) operation.
1303 unsigned dom_pre_index
, dom_post_index
;
1305 /* live in and out for this block; used for liveness analysis */
1306 BITSET_WORD
*live_in
;
1307 BITSET_WORD
*live_out
;
1310 static inline nir_instr
*
1311 nir_block_first_instr(nir_block
*block
)
1313 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1314 return exec_node_data(nir_instr
, head
, node
);
1317 static inline nir_instr
*
1318 nir_block_last_instr(nir_block
*block
)
1320 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1321 return exec_node_data(nir_instr
, tail
, node
);
1324 #define nir_foreach_instr(block, instr) \
1325 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1326 #define nir_foreach_instr_reverse(block, instr) \
1327 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1328 #define nir_foreach_instr_safe(block, instr) \
1329 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1330 #define nir_foreach_instr_reverse_safe(block, instr) \
1331 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1333 typedef struct nir_if
{
1334 nir_cf_node cf_node
;
1337 struct exec_list then_list
; /** < list of nir_cf_node */
1338 struct exec_list else_list
; /** < list of nir_cf_node */
1341 static inline nir_cf_node
*
1342 nir_if_first_then_node(nir_if
*if_stmt
)
1344 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1345 return exec_node_data(nir_cf_node
, head
, node
);
1348 static inline nir_cf_node
*
1349 nir_if_last_then_node(nir_if
*if_stmt
)
1351 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1352 return exec_node_data(nir_cf_node
, tail
, node
);
1355 static inline nir_cf_node
*
1356 nir_if_first_else_node(nir_if
*if_stmt
)
1358 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1359 return exec_node_data(nir_cf_node
, head
, node
);
1362 static inline nir_cf_node
*
1363 nir_if_last_else_node(nir_if
*if_stmt
)
1365 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1366 return exec_node_data(nir_cf_node
, tail
, node
);
1370 nir_cf_node cf_node
;
1372 struct exec_list body
; /** < list of nir_cf_node */
1375 static inline nir_cf_node
*
1376 nir_loop_first_cf_node(nir_loop
*loop
)
1378 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1381 static inline nir_cf_node
*
1382 nir_loop_last_cf_node(nir_loop
*loop
)
1384 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1388 * Various bits of metadata that can may be created or required by
1389 * optimization and analysis passes
1392 nir_metadata_none
= 0x0,
1393 nir_metadata_block_index
= 0x1,
1394 nir_metadata_dominance
= 0x2,
1395 nir_metadata_live_ssa_defs
= 0x4,
1396 nir_metadata_not_properly_reset
= 0x8,
1400 nir_cf_node cf_node
;
1402 /** pointer to the function of which this is an implementation */
1403 struct nir_function
*function
;
1405 struct exec_list body
; /** < list of nir_cf_node */
1407 nir_block
*end_block
;
1409 /** list for all local variables in the function */
1410 struct exec_list locals
;
1412 /** array of variables used as parameters */
1413 unsigned num_params
;
1414 nir_variable
**params
;
1416 /** variable used to hold the result of the function */
1417 nir_variable
*return_var
;
1419 /** list of local registers in the function */
1420 struct exec_list registers
;
1422 /** next available local register index */
1425 /** next available SSA value index */
1428 /* total number of basic blocks, only valid when block_index_dirty = false */
1429 unsigned num_blocks
;
1431 nir_metadata valid_metadata
;
1432 } nir_function_impl
;
1434 static inline nir_block
*
1435 nir_start_block(nir_function_impl
*impl
)
1437 return (nir_block
*) exec_list_get_head(&impl
->body
);
1440 static inline nir_cf_node
*
1441 nir_cf_node_next(nir_cf_node
*node
)
1443 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1444 if (exec_node_is_tail_sentinel(next
))
1447 return exec_node_data(nir_cf_node
, next
, node
);
1450 static inline nir_cf_node
*
1451 nir_cf_node_prev(nir_cf_node
*node
)
1453 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1454 if (exec_node_is_head_sentinel(prev
))
1457 return exec_node_data(nir_cf_node
, prev
, node
);
1461 nir_cf_node_is_first(const nir_cf_node
*node
)
1463 return exec_node_is_head_sentinel(node
->node
.prev
);
1467 nir_cf_node_is_last(const nir_cf_node
*node
)
1469 return exec_node_is_tail_sentinel(node
->node
.next
);
1472 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1473 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1474 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1475 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1480 nir_parameter_inout
,
1481 } nir_parameter_type
;
1484 nir_parameter_type param_type
;
1485 const struct glsl_type
*type
;
1488 typedef struct nir_function
{
1489 struct exec_node node
;
1492 struct nir_shader
*shader
;
1494 unsigned num_params
;
1495 nir_parameter
*params
;
1496 const struct glsl_type
*return_type
;
1498 /** The implementation of this function.
1500 * If the function is only declared and not implemented, this is NULL.
1502 nir_function_impl
*impl
;
1505 typedef struct nir_shader_compiler_options
{
1512 /** lowers fneg and ineg to fsub and isub. */
1514 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1517 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1520 /* Does the native fdot instruction replicate its result for four
1521 * components? If so, then opt_algebraic_late will turn all fdotN
1522 * instructions into fdot_replicatedN instructions.
1524 bool fdot_replicates
;
1526 /** lowers ffract to fsub+ffloor: */
1530 * Does the driver support real 32-bit integers? (Otherwise, integers
1531 * are simulated by floats.)
1533 bool native_integers
;
1534 } nir_shader_compiler_options
;
1536 typedef struct nir_shader_info
{
1539 /* Descriptive name provided by the client; may be NULL */
1542 /* Number of textures used by this shader */
1543 unsigned num_textures
;
1544 /* Number of uniform buffers used by this shader */
1546 /* Number of atomic buffers used by this shader */
1548 /* Number of shader storage buffers used by this shader */
1550 /* Number of images used by this shader */
1551 unsigned num_images
;
1553 /* Which inputs are actually read */
1554 uint64_t inputs_read
;
1555 /* Which outputs are actually written */
1556 uint64_t outputs_written
;
1557 /* Which system values are actually read */
1558 uint64_t system_values_read
;
1560 /* Which patch inputs are actually read */
1561 uint32_t patch_inputs_read
;
1562 /* Which patch outputs are actually written */
1563 uint32_t patch_outputs_written
;
1565 /* Whether or not this shader ever uses textureGather() */
1566 bool uses_texture_gather
;
1568 /* Whether or not this shader uses the gl_ClipDistance output */
1569 bool uses_clip_distance_out
;
1571 /* Whether or not separate shader objects were used */
1572 bool separate_shader
;
1574 /** Was this shader linked with any transform feedback varyings? */
1575 bool has_transform_feedback_varyings
;
1579 /** The number of vertices recieves per input primitive */
1580 unsigned vertices_in
;
1582 /** The output primitive type (GL enum value) */
1583 unsigned output_primitive
;
1585 /** The maximum number of vertices the geometry shader might write. */
1586 unsigned vertices_out
;
1588 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1589 unsigned invocations
;
1591 /** Whether or not this shader uses EndPrimitive */
1592 bool uses_end_primitive
;
1594 /** Whether or not this shader uses non-zero streams */
1602 * Whether early fragment tests are enabled as defined by
1603 * ARB_shader_image_load_store.
1605 bool early_fragment_tests
;
1607 /** gl_FragDepth layout for ARB_conservative_depth. */
1608 enum gl_frag_depth_layout depth_layout
;
1612 unsigned local_size
[3];
1616 /** The number of vertices in the TCS output patch. */
1617 unsigned vertices_out
;
1622 typedef struct nir_shader
{
1623 /** list of uniforms (nir_variable) */
1624 struct exec_list uniforms
;
1626 /** list of inputs (nir_variable) */
1627 struct exec_list inputs
;
1629 /** list of outputs (nir_variable) */
1630 struct exec_list outputs
;
1632 /** Set of driver-specific options for the shader.
1634 * The memory for the options is expected to be kept in a single static
1635 * copy by the driver.
1637 const struct nir_shader_compiler_options
*options
;
1639 /** Various bits of compile-time information about a given shader */
1640 struct nir_shader_info info
;
1642 /** list of global variables in the shader (nir_variable) */
1643 struct exec_list globals
;
1645 /** list of system value variables in the shader (nir_variable) */
1646 struct exec_list system_values
;
1648 struct exec_list functions
; /** < list of nir_function */
1650 /** list of global register in the shader */
1651 struct exec_list registers
;
1653 /** next available global register index */
1657 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1660 unsigned num_inputs
, num_uniforms
, num_outputs
;
1662 /** The shader stage, such as MESA_SHADER_VERTEX. */
1663 gl_shader_stage stage
;
1666 #define nir_foreach_function(shader, func) \
1667 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1669 nir_shader
*nir_shader_create(void *mem_ctx
,
1670 gl_shader_stage stage
,
1671 const nir_shader_compiler_options
*options
);
1673 /** creates a register, including assigning it an index and adding it to the list */
1674 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1676 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1678 void nir_reg_remove(nir_register
*reg
);
1680 /** Adds a variable to the appropreate list in nir_shader */
1681 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1684 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1686 assert(var
->data
.mode
== nir_var_local
);
1687 exec_list_push_tail(&impl
->locals
, &var
->node
);
1690 /** creates a variable, sets a few defaults, and adds it to the list */
1691 nir_variable
*nir_variable_create(nir_shader
*shader
,
1692 nir_variable_mode mode
,
1693 const struct glsl_type
*type
,
1695 /** creates a local variable and adds it to the list */
1696 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1697 const struct glsl_type
*type
,
1700 /** creates a function and adds it to the shader's list of functions */
1701 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1703 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1704 /** creates a function_impl that isn't tied to any particular function */
1705 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1707 nir_block
*nir_block_create(nir_shader
*shader
);
1708 nir_if
*nir_if_create(nir_shader
*shader
);
1709 nir_loop
*nir_loop_create(nir_shader
*shader
);
1711 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1713 /** requests that the given pieces of metadata be generated */
1714 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1715 /** dirties all but the preserved metadata */
1716 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1718 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1719 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1721 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1723 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1724 unsigned num_components
);
1726 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1727 nir_intrinsic_op op
);
1729 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1730 nir_function
*callee
);
1732 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1734 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1736 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1738 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1739 unsigned num_components
);
1741 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1742 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1743 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1745 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1747 nir_load_const_instr
*
1748 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1751 * NIR Cursors and Instruction Insertion API
1754 * A tiny struct representing a point to insert/extract instructions or
1755 * control flow nodes. Helps reduce the combinatorial explosion of possible
1756 * points to insert/extract.
1758 * \sa nir_control_flow.h
1761 nir_cursor_before_block
,
1762 nir_cursor_after_block
,
1763 nir_cursor_before_instr
,
1764 nir_cursor_after_instr
,
1765 } nir_cursor_option
;
1768 nir_cursor_option option
;
1775 static inline nir_block
*
1776 nir_cursor_current_block(nir_cursor cursor
)
1778 if (cursor
.option
== nir_cursor_before_instr
||
1779 cursor
.option
== nir_cursor_after_instr
) {
1780 return cursor
.instr
->block
;
1782 return cursor
.block
;
1786 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
1788 static inline nir_cursor
1789 nir_before_block(nir_block
*block
)
1792 cursor
.option
= nir_cursor_before_block
;
1793 cursor
.block
= block
;
1797 static inline nir_cursor
1798 nir_after_block(nir_block
*block
)
1801 cursor
.option
= nir_cursor_after_block
;
1802 cursor
.block
= block
;
1806 static inline nir_cursor
1807 nir_before_instr(nir_instr
*instr
)
1810 cursor
.option
= nir_cursor_before_instr
;
1811 cursor
.instr
= instr
;
1815 static inline nir_cursor
1816 nir_after_instr(nir_instr
*instr
)
1819 cursor
.option
= nir_cursor_after_instr
;
1820 cursor
.instr
= instr
;
1824 static inline nir_cursor
1825 nir_after_block_before_jump(nir_block
*block
)
1827 nir_instr
*last_instr
= nir_block_last_instr(block
);
1828 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1829 return nir_before_instr(last_instr
);
1831 return nir_after_block(block
);
1835 static inline nir_cursor
1836 nir_before_cf_node(nir_cf_node
*node
)
1838 if (node
->type
== nir_cf_node_block
)
1839 return nir_before_block(nir_cf_node_as_block(node
));
1841 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1844 static inline nir_cursor
1845 nir_after_cf_node(nir_cf_node
*node
)
1847 if (node
->type
== nir_cf_node_block
)
1848 return nir_after_block(nir_cf_node_as_block(node
));
1850 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1853 static inline nir_cursor
1854 nir_after_cf_node_and_phis(nir_cf_node
*node
)
1856 if (node
->type
== nir_cf_node_block
)
1857 return nir_after_block(nir_cf_node_as_block(node
));
1859 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
1860 assert(block
->cf_node
.type
== nir_cf_node_block
);
1862 nir_foreach_instr(block
, instr
) {
1863 if (instr
->type
!= nir_instr_type_phi
)
1864 return nir_before_instr(instr
);
1866 return nir_after_block(block
);
1869 static inline nir_cursor
1870 nir_before_cf_list(struct exec_list
*cf_list
)
1872 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1873 exec_list_get_head(cf_list
), node
);
1874 return nir_before_cf_node(first_node
);
1877 static inline nir_cursor
1878 nir_after_cf_list(struct exec_list
*cf_list
)
1880 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1881 exec_list_get_tail(cf_list
), node
);
1882 return nir_after_cf_node(last_node
);
1886 * Insert a NIR instruction at the given cursor.
1888 * Note: This does not update the cursor.
1890 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
1893 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1895 nir_instr_insert(nir_before_instr(instr
), before
);
1899 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
1901 nir_instr_insert(nir_after_instr(instr
), after
);
1905 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
1907 nir_instr_insert(nir_before_block(block
), before
);
1911 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
1913 nir_instr_insert(nir_after_block(block
), after
);
1917 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
1919 nir_instr_insert(nir_before_cf_node(node
), before
);
1923 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
1925 nir_instr_insert(nir_after_cf_node(node
), after
);
1929 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
1931 nir_instr_insert(nir_before_cf_list(list
), before
);
1935 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
1937 nir_instr_insert(nir_after_cf_list(list
), after
);
1940 void nir_instr_remove(nir_instr
*instr
);
1944 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
1945 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
1946 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
1947 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
1949 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
1950 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
1952 nir_const_value
*nir_src_as_const_value(nir_src src
);
1953 bool nir_src_is_dynamically_uniform(nir_src src
);
1954 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
1955 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
1956 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
1957 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
1958 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
1961 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1962 unsigned num_components
, const char *name
);
1963 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1964 unsigned num_components
, const char *name
);
1965 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
1966 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
1967 nir_instr
*after_me
);
1969 /* visits basic blocks in source-code order */
1970 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
1971 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1973 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1975 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1978 /* If the following CF node is an if, this function returns that if.
1979 * Otherwise, it returns NULL.
1981 nir_if
*nir_block_get_following_if(nir_block
*block
);
1983 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
1985 void nir_index_local_regs(nir_function_impl
*impl
);
1986 void nir_index_global_regs(nir_shader
*shader
);
1987 void nir_index_ssa_defs(nir_function_impl
*impl
);
1988 unsigned nir_index_instrs(nir_function_impl
*impl
);
1990 void nir_index_blocks(nir_function_impl
*impl
);
1992 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
1993 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
1995 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
1996 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*impl
);
1999 void nir_validate_shader(nir_shader
*shader
);
2000 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2001 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2003 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2004 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2005 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2008 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2009 void nir_calc_dominance(nir_shader
*shader
);
2011 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2012 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2014 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2015 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2017 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2018 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2020 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2021 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2023 int nir_gs_count_vertices(const nir_shader
*shader
);
2025 bool nir_split_var_copies(nir_shader
*shader
);
2027 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2028 bool nir_lower_returns(nir_shader
*shader
);
2030 bool nir_inline_functions_impl(nir_function_impl
*impl
);
2031 bool nir_inline_functions(nir_shader
*shader
);
2033 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2034 void nir_lower_var_copies(nir_shader
*shader
);
2036 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2038 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2040 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
2042 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
2044 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2046 void nir_assign_var_locations(struct exec_list
*var_list
,
2048 int (*type_size
)(const struct glsl_type
*));
2050 void nir_lower_io(nir_shader
*shader
,
2051 nir_variable_mode mode
,
2052 int (*type_size
)(const struct glsl_type
*));
2053 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2054 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2056 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2058 bool nir_remove_dead_variables(nir_shader
*shader
);
2060 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2061 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2062 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2063 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2065 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2067 void nir_lower_samplers(nir_shader
*shader
,
2068 const struct gl_shader_program
*shader_program
);
2070 bool nir_lower_system_values(nir_shader
*shader
);
2072 typedef struct nir_lower_tex_options
{
2074 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2075 * sampler types a texture projector is lowered.
2080 * If true, lower rect textures to 2D, using txs to fetch the
2081 * texture dimensions and dividing the texture coords by the
2082 * texture dims to normalize.
2087 * To emulate certain texture wrap modes, this can be used
2088 * to saturate the specified tex coord to [0.0, 1.0]. The
2089 * bits are according to sampler #, ie. if, for example:
2091 * (conf->saturate_s & (1 << n))
2093 * is true, then the s coord for sampler n is saturated.
2095 * Note that clamping must happen *after* projector lowering
2096 * so any projected texture sample instruction with a clamped
2097 * coordinate gets automatically lowered, regardless of the
2098 * 'lower_txp' setting.
2100 unsigned saturate_s
;
2101 unsigned saturate_t
;
2102 unsigned saturate_r
;
2104 /* Bitmask of samplers that need swizzling.
2106 * If (swizzle_result & (1 << sampler_index)), then the swizzle in
2107 * swizzles[sampler_index] is applied to the result of the texturing
2110 unsigned swizzle_result
;
2112 /* A swizzle for each sampler. Values 0-3 represent x, y, z, or w swizzles
2113 * while 4 and 5 represent 0 and 1 respectively.
2115 uint8_t swizzles
[32][4];
2116 } nir_lower_tex_options
;
2118 bool nir_lower_tex(nir_shader
*shader
,
2119 const nir_lower_tex_options
*options
);
2121 void nir_lower_idiv(nir_shader
*shader
);
2123 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2124 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2126 void nir_lower_two_sided_color(nir_shader
*shader
);
2128 void nir_lower_atomics(nir_shader
*shader
,
2129 const struct gl_shader_program
*shader_program
);
2130 void nir_lower_to_source_mods(nir_shader
*shader
);
2132 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2134 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2136 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2137 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2139 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2140 void nir_convert_to_ssa(nir_shader
*shader
);
2142 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2143 * registers. If false, convert all values (even those not involved in a phi
2144 * node) to registers.
2146 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2148 bool nir_opt_algebraic(nir_shader
*shader
);
2149 bool nir_opt_algebraic_late(nir_shader
*shader
);
2150 bool nir_opt_constant_folding(nir_shader
*shader
);
2152 bool nir_opt_global_to_local(nir_shader
*shader
);
2154 bool nir_copy_prop(nir_shader
*shader
);
2156 bool nir_opt_cse(nir_shader
*shader
);
2158 bool nir_opt_dce(nir_shader
*shader
);
2160 bool nir_opt_dead_cf(nir_shader
*shader
);
2162 void nir_opt_gcm(nir_shader
*shader
);
2164 bool nir_opt_peephole_select(nir_shader
*shader
);
2166 bool nir_opt_remove_phis(nir_shader
*shader
);
2168 bool nir_opt_undef(nir_shader
*shader
);
2170 void nir_sweep(nir_shader
*shader
);
2172 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2173 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);