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.
142 typedef struct nir_variable
{
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 * Initial binding point for a sampler or UBO.
287 * For array types, this represents the binding point for the first element.
292 * Location an atomic counter is stored at.
297 * ARB_shader_image_load_store qualifiers.
300 bool read_only
; /**< "readonly" qualifier. */
301 bool write_only
; /**< "writeonly" qualifier. */
306 /** Image internal format if specified explicitly, otherwise GL_NONE. */
311 * Highest element accessed with a constant expression array index
313 * Not used for non-array variables.
315 unsigned max_array_access
;
320 * Built-in state that backs this uniform
322 * Once set at variable creation, \c state_slots must remain invariant.
323 * This is because, ideally, this array would be shared by all clones of
324 * this variable in the IR tree. In other words, we'd really like for it
325 * to be a fly-weight.
327 * If the variable is not a uniform, \c num_state_slots will be zero and
328 * \c state_slots will be \c NULL.
331 unsigned num_state_slots
; /**< Number of state slots used */
332 nir_state_slot
*state_slots
; /**< State descriptors. */
336 * Constant expression assigned in the initializer of the variable
338 nir_constant
*constant_initializer
;
341 * For variables that are in an interface block or are an instance of an
342 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
344 * \sa ir_variable::location
346 const struct glsl_type
*interface_type
;
349 #define nir_foreach_variable(var, var_list) \
350 foreach_list_typed(nir_variable, var, node, var_list)
352 typedef struct nir_register
{
353 struct exec_node node
;
355 unsigned num_components
; /** < number of vector components */
356 unsigned num_array_elems
; /** < size of array (0 for no array) */
358 /** generic register index. */
361 /** only for debug purposes, can be NULL */
364 /** whether this register is local (per-function) or global (per-shader) */
368 * If this flag is set to true, then accessing channels >= num_components
369 * is well-defined, and simply spills over to the next array element. This
370 * is useful for backends that can do per-component accessing, in
371 * particular scalar backends. By setting this flag and making
372 * num_components equal to 1, structures can be packed tightly into
373 * registers and then registers can be accessed per-component to get to
374 * each structure member, even if it crosses vec4 boundaries.
378 /** set of nir_src's where this register is used (read from) */
379 struct list_head uses
;
381 /** set of nir_dest's where this register is defined (written to) */
382 struct list_head defs
;
384 /** set of nir_if's where this register is used as a condition */
385 struct list_head if_uses
;
392 nir_instr_type_intrinsic
,
393 nir_instr_type_load_const
,
395 nir_instr_type_ssa_undef
,
397 nir_instr_type_parallel_copy
,
400 typedef struct nir_instr
{
401 struct exec_node node
;
403 struct nir_block
*block
;
405 /** generic instruction index. */
408 /* A temporary for optimization and analysis passes to use for storing
409 * flags. For instance, DCE uses this to store the "dead/live" info.
414 static inline nir_instr
*
415 nir_instr_next(nir_instr
*instr
)
417 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
418 if (exec_node_is_tail_sentinel(next
))
421 return exec_node_data(nir_instr
, next
, node
);
424 static inline nir_instr
*
425 nir_instr_prev(nir_instr
*instr
)
427 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
428 if (exec_node_is_head_sentinel(prev
))
431 return exec_node_data(nir_instr
, prev
, node
);
435 nir_instr_is_first(nir_instr
*instr
)
437 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
441 nir_instr_is_last(nir_instr
*instr
)
443 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
446 typedef struct nir_ssa_def
{
447 /** for debugging only, can be NULL */
450 /** generic SSA definition index. */
453 /** Index into the live_in and live_out bitfields */
456 nir_instr
*parent_instr
;
458 /** set of nir_instr's where this register is used (read from) */
459 struct list_head uses
;
461 /** set of nir_if's where this register is used as a condition */
462 struct list_head if_uses
;
464 uint8_t num_components
;
471 struct nir_src
*indirect
; /** < NULL for no indirect offset */
472 unsigned base_offset
;
474 /* TODO use-def chain goes here */
478 nir_instr
*parent_instr
;
479 struct list_head def_link
;
482 struct nir_src
*indirect
; /** < NULL for no indirect offset */
483 unsigned base_offset
;
485 /* TODO def-use chain goes here */
490 typedef struct nir_src
{
492 nir_instr
*parent_instr
;
493 struct nir_if
*parent_if
;
496 struct list_head use_link
;
506 #define NIR_SRC_INIT (nir_src) { { NULL } }
508 #define nir_foreach_use(reg_or_ssa_def, src) \
509 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
511 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
512 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
514 #define nir_foreach_if_use(reg_or_ssa_def, src) \
515 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
517 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
518 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
529 #define NIR_DEST_INIT (nir_dest) { { { NULL } } }
531 #define nir_foreach_def(reg, dest) \
532 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
534 #define nir_foreach_def_safe(reg, dest) \
535 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
537 static inline nir_src
538 nir_src_for_ssa(nir_ssa_def
*def
)
540 nir_src src
= NIR_SRC_INIT
;
548 static inline nir_src
549 nir_src_for_reg(nir_register
*reg
)
551 nir_src src
= NIR_SRC_INIT
;
555 src
.reg
.indirect
= NULL
;
556 src
.reg
.base_offset
= 0;
561 static inline nir_dest
562 nir_dest_for_reg(nir_register
*reg
)
564 nir_dest dest
= NIR_DEST_INIT
;
571 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
572 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
578 * \name input modifiers
582 * For inputs interpreted as floating point, flips the sign bit. For
583 * inputs interpreted as integers, performs the two's complement negation.
588 * Clears the sign bit for floating point values, and computes the integer
589 * absolute value for integers. Note that the negate modifier acts after
590 * the absolute value modifier, therefore if both are set then all inputs
591 * will become negative.
597 * For each input component, says which component of the register it is
598 * chosen from. Note that which elements of the swizzle are used and which
599 * are ignored are based on the write mask for most opcodes - for example,
600 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
601 * a swizzle of {2, x, 1, 0} where x means "don't care."
610 * \name saturate output modifier
612 * Only valid for opcodes that output floating-point numbers. Clamps the
613 * output to between 0.0 and 1.0 inclusive.
618 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
622 nir_type_invalid
= 0, /* Not a valid type */
630 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
631 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
632 } nir_op_algebraic_property
;
640 * The number of components in the output
642 * If non-zero, this is the size of the output and input sizes are
643 * explicitly given; swizzle and writemask are still in effect, but if
644 * the output component is masked out, then the input component may
647 * If zero, the opcode acts in the standard, per-component manner; the
648 * operation is performed on each component (except the ones that are
649 * masked out) with the input being taken from the input swizzle for
652 * The size of some of the inputs may be given (i.e. non-zero) even
653 * though output_size is zero; in that case, the inputs with a zero
654 * size act per-component, while the inputs with non-zero size don't.
656 unsigned output_size
;
659 * The type of vector that the instruction outputs. Note that the
660 * staurate modifier is only allowed on outputs with the float type.
663 nir_alu_type output_type
;
666 * The number of components in each input
668 unsigned input_sizes
[4];
671 * The type of vector that each input takes. Note that negate and
672 * absolute value are only allowed on inputs with int or float type and
673 * behave differently on the two.
675 nir_alu_type input_types
[4];
677 nir_op_algebraic_property algebraic_properties
;
680 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
682 typedef struct nir_alu_instr
{
689 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
690 nir_alu_instr
*instr
);
691 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
692 nir_alu_instr
*instr
);
694 /* is this source channel used? */
696 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
698 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
699 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
701 return (instr
->dest
.write_mask
>> channel
) & 1;
705 * For instructions whose destinations are SSA, get the number of channels
708 static inline unsigned
709 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
711 assert(instr
->dest
.dest
.is_ssa
);
713 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
714 return nir_op_infos
[instr
->op
].input_sizes
[src
];
716 return instr
->dest
.dest
.ssa
.num_components
;
721 nir_deref_type_array
,
722 nir_deref_type_struct
725 typedef struct nir_deref
{
726 nir_deref_type deref_type
;
727 struct nir_deref
*child
;
728 const struct glsl_type
*type
;
737 /* This enum describes how the array is referenced. If the deref is
738 * direct then the base_offset is used. If the deref is indirect then then
739 * offset is given by base_offset + indirect. If the deref is a wildcard
740 * then the deref refers to all of the elements of the array at the same
741 * time. Wildcard dereferences are only ever allowed in copy_var
742 * intrinsics and the source and destination derefs must have matching
746 nir_deref_array_type_direct
,
747 nir_deref_array_type_indirect
,
748 nir_deref_array_type_wildcard
,
749 } nir_deref_array_type
;
754 nir_deref_array_type deref_array_type
;
755 unsigned base_offset
;
765 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
766 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
767 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
769 /* Returns the last deref in the chain. */
770 static inline nir_deref
*
771 nir_deref_tail(nir_deref
*deref
)
774 deref
= deref
->child
;
782 nir_deref_var
**params
;
783 nir_deref_var
*return_deref
;
785 struct nir_function
*callee
;
788 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
789 num_variables, num_indices, flags) \
790 nir_intrinsic_##name,
792 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
795 #include "nir_intrinsics.h"
796 nir_num_intrinsics
= nir_last_intrinsic
+ 1
800 #undef LAST_INTRINSIC
802 /** Represents an intrinsic
804 * An intrinsic is an instruction type for handling things that are
805 * more-or-less regular operations but don't just consume and produce SSA
806 * values like ALU operations do. Intrinsics are not for things that have
807 * special semantic meaning such as phi nodes and parallel copies.
808 * Examples of intrinsics include variable load/store operations, system
809 * value loads, and the like. Even though texturing more-or-less falls
810 * under this category, texturing is its own instruction type because
811 * trying to represent texturing with intrinsics would lead to a
812 * combinatorial explosion of intrinsic opcodes.
814 * By having a single instruction type for handling a lot of different
815 * cases, optimization passes can look for intrinsics and, for the most
816 * part, completely ignore them. Each intrinsic type also has a few
817 * possible flags that govern whether or not they can be reordered or
818 * eliminated. That way passes like dead code elimination can still work
819 * on intrisics without understanding the meaning of each.
821 * Each intrinsic has some number of constant indices, some number of
822 * variables, and some number of sources. What these sources, variables,
823 * and indices mean depends on the intrinsic and is documented with the
824 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
825 * instructions are the only types of instruction that can operate on
831 nir_intrinsic_op intrinsic
;
835 /** number of components if this is a vectorized intrinsic
837 * Similarly to ALU operations, some intrinsics are vectorized.
838 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
839 * For vectorized intrinsics, the num_components field specifies the
840 * number of destination components and the number of source components
841 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
843 uint8_t num_components
;
847 nir_deref_var
*variables
[2];
850 } nir_intrinsic_instr
;
853 * \name NIR intrinsics semantic flags
855 * information about what the compiler can do with the intrinsics.
857 * \sa nir_intrinsic_info::flags
861 * whether the intrinsic can be safely eliminated if none of its output
862 * value is not being used.
864 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
867 * Whether the intrinsic can be reordered with respect to any other
868 * intrinsic, i.e. whether the only reordering dependencies of the
869 * intrinsic are due to the register reads/writes.
871 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
872 } nir_intrinsic_semantic_flag
;
874 #define NIR_INTRINSIC_MAX_INPUTS 4
879 unsigned num_srcs
; /** < number of register/SSA inputs */
881 /** number of components of each input register
883 * If this value is 0, the number of components is given by the
884 * num_components field of nir_intrinsic_instr.
886 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
890 /** number of components of the output register
892 * If this value is 0, the number of components is given by the
893 * num_components field of nir_intrinsic_instr.
895 unsigned dest_components
;
897 /** the number of inputs/outputs that are variables */
898 unsigned num_variables
;
900 /** the number of constant indices used by the intrinsic */
901 unsigned num_indices
;
903 /** semantic flags for calls to this intrinsic */
904 nir_intrinsic_semantic_flag flags
;
905 } nir_intrinsic_info
;
907 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
910 * \group texture information
912 * This gives semantic information about textures which is useful to the
913 * frontend, the backend, and lowering passes, but not the optimizer.
918 nir_tex_src_projector
,
919 nir_tex_src_comparitor
, /* shadow comparitor */
923 nir_tex_src_ms_index
, /* MSAA sample index */
926 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
927 nir_num_tex_src_types
932 nir_tex_src_type src_type
;
936 nir_texop_tex
, /**< Regular texture look-up */
937 nir_texop_txb
, /**< Texture look-up with LOD bias */
938 nir_texop_txl
, /**< Texture look-up with explicit LOD */
939 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
940 nir_texop_txf
, /**< Texel fetch with explicit LOD */
941 nir_texop_txf_ms
, /**< Multisample texture fetch */
942 nir_texop_txs
, /**< Texture size */
943 nir_texop_lod
, /**< Texture lod query */
944 nir_texop_tg4
, /**< Texture gather */
945 nir_texop_query_levels
, /**< Texture levels query */
946 nir_texop_texture_samples
, /**< Texture samples query */
947 nir_texop_samples_identical
, /**< Query whether all samples are definitely
955 enum glsl_sampler_dim sampler_dim
;
956 nir_alu_type dest_type
;
961 unsigned num_srcs
, coord_components
;
962 bool is_array
, is_shadow
;
965 * If is_shadow is true, whether this is the old-style shadow that outputs 4
966 * components or the new-style shadow that outputs 1 component.
968 bool is_new_style_shadow
;
970 /* constant offset - must be 0 if the offset source is used */
973 /* gather component selector */
974 unsigned component
: 2;
976 /** The sampler index
978 * If this texture instruction has a nir_tex_src_sampler_offset source,
979 * then the sampler index is given by sampler_index + sampler_offset.
981 unsigned sampler_index
;
983 /** The size of the sampler array or 0 if it's not an array */
984 unsigned sampler_array_size
;
986 nir_deref_var
*sampler
; /* if this is NULL, use sampler_index instead */
989 static inline unsigned
990 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
993 case nir_texop_txs
: {
995 switch (instr
->sampler_dim
) {
996 case GLSL_SAMPLER_DIM_1D
:
997 case GLSL_SAMPLER_DIM_BUF
:
1000 case GLSL_SAMPLER_DIM_2D
:
1001 case GLSL_SAMPLER_DIM_CUBE
:
1002 case GLSL_SAMPLER_DIM_MS
:
1003 case GLSL_SAMPLER_DIM_RECT
:
1004 case GLSL_SAMPLER_DIM_EXTERNAL
:
1007 case GLSL_SAMPLER_DIM_3D
:
1011 unreachable("not reached");
1013 if (instr
->is_array
)
1021 case nir_texop_texture_samples
:
1022 case nir_texop_query_levels
:
1023 case nir_texop_samples_identical
:
1027 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1034 /* Returns true if this texture operation queries something about the texture
1035 * rather than actually sampling it.
1038 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1040 switch (instr
->op
) {
1043 case nir_texop_texture_samples
:
1044 case nir_texop_query_levels
:
1051 case nir_texop_txf_ms
:
1055 unreachable("Invalid texture opcode");
1059 static inline unsigned
1060 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1062 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1063 return instr
->coord_components
;
1066 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1067 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1068 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1069 if (instr
->is_array
)
1070 return instr
->coord_components
- 1;
1072 return instr
->coord_components
;
1079 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1081 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1082 if (instr
->src
[i
].src_type
== type
)
1099 nir_const_value value
;
1102 } nir_load_const_instr
;
1115 /* creates a new SSA variable in an undefined state */
1120 } nir_ssa_undef_instr
;
1123 struct exec_node node
;
1125 /* The predecessor block corresponding to this source */
1126 struct nir_block
*pred
;
1131 #define nir_foreach_phi_src(phi, entry) \
1132 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1133 #define nir_foreach_phi_src_safe(phi, entry) \
1134 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1139 struct exec_list srcs
; /** < list of nir_phi_src */
1145 struct exec_node node
;
1148 } nir_parallel_copy_entry
;
1150 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1151 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1156 /* A list of nir_parallel_copy_entry's. The sources of all of the
1157 * entries are copied to the corresponding destinations "in parallel".
1158 * In other words, if we have two entries: a -> b and b -> a, the values
1161 struct exec_list entries
;
1162 } nir_parallel_copy_instr
;
1164 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1165 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1166 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1167 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1168 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1169 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1170 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1171 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1172 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1173 nir_parallel_copy_instr
, instr
)
1178 * Control flow consists of a tree of control flow nodes, which include
1179 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1180 * instructions that always run start-to-finish. Each basic block also keeps
1181 * track of its successors (blocks which may run immediately after the current
1182 * block) and predecessors (blocks which could have run immediately before the
1183 * current block). Each function also has a start block and an end block which
1184 * all return statements point to (which is always empty). Together, all the
1185 * blocks with their predecessors and successors make up the control flow
1186 * graph (CFG) of the function. There are helpers that modify the tree of
1187 * control flow nodes while modifying the CFG appropriately; these should be
1188 * used instead of modifying the tree directly.
1195 nir_cf_node_function
1198 typedef struct nir_cf_node
{
1199 struct exec_node node
;
1200 nir_cf_node_type type
;
1201 struct nir_cf_node
*parent
;
1204 typedef struct nir_block
{
1205 nir_cf_node cf_node
;
1207 struct exec_list instr_list
; /** < list of nir_instr */
1209 /** generic block index; generated by nir_index_blocks */
1213 * Each block can only have up to 2 successors, so we put them in a simple
1214 * array - no need for anything more complicated.
1216 struct nir_block
*successors
[2];
1218 /* Set of nir_block predecessors in the CFG */
1219 struct set
*predecessors
;
1222 * this node's immediate dominator in the dominance tree - set to NULL for
1225 struct nir_block
*imm_dom
;
1227 /* This node's children in the dominance tree */
1228 unsigned num_dom_children
;
1229 struct nir_block
**dom_children
;
1231 /* Set of nir_block's on the dominance frontier of this block */
1232 struct set
*dom_frontier
;
1235 * These two indices have the property that dom_{pre,post}_index for each
1236 * child of this block in the dominance tree will always be between
1237 * dom_pre_index and dom_post_index for this block, which makes testing if
1238 * a given block is dominated by another block an O(1) operation.
1240 unsigned dom_pre_index
, dom_post_index
;
1242 /* live in and out for this block; used for liveness analysis */
1243 BITSET_WORD
*live_in
;
1244 BITSET_WORD
*live_out
;
1247 static inline nir_instr
*
1248 nir_block_first_instr(nir_block
*block
)
1250 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1251 return exec_node_data(nir_instr
, head
, node
);
1254 static inline nir_instr
*
1255 nir_block_last_instr(nir_block
*block
)
1257 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1258 return exec_node_data(nir_instr
, tail
, node
);
1261 #define nir_foreach_instr(block, instr) \
1262 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1263 #define nir_foreach_instr_reverse(block, instr) \
1264 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1265 #define nir_foreach_instr_safe(block, instr) \
1266 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1267 #define nir_foreach_instr_reverse_safe(block, instr) \
1268 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1270 typedef struct nir_if
{
1271 nir_cf_node cf_node
;
1274 struct exec_list then_list
; /** < list of nir_cf_node */
1275 struct exec_list else_list
; /** < list of nir_cf_node */
1278 static inline nir_cf_node
*
1279 nir_if_first_then_node(nir_if
*if_stmt
)
1281 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1282 return exec_node_data(nir_cf_node
, head
, node
);
1285 static inline nir_cf_node
*
1286 nir_if_last_then_node(nir_if
*if_stmt
)
1288 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1289 return exec_node_data(nir_cf_node
, tail
, node
);
1292 static inline nir_cf_node
*
1293 nir_if_first_else_node(nir_if
*if_stmt
)
1295 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1296 return exec_node_data(nir_cf_node
, head
, node
);
1299 static inline nir_cf_node
*
1300 nir_if_last_else_node(nir_if
*if_stmt
)
1302 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1303 return exec_node_data(nir_cf_node
, tail
, node
);
1307 nir_cf_node cf_node
;
1309 struct exec_list body
; /** < list of nir_cf_node */
1312 static inline nir_cf_node
*
1313 nir_loop_first_cf_node(nir_loop
*loop
)
1315 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1318 static inline nir_cf_node
*
1319 nir_loop_last_cf_node(nir_loop
*loop
)
1321 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1325 * Various bits of metadata that can may be created or required by
1326 * optimization and analysis passes
1329 nir_metadata_none
= 0x0,
1330 nir_metadata_block_index
= 0x1,
1331 nir_metadata_dominance
= 0x2,
1332 nir_metadata_live_ssa_defs
= 0x4,
1333 nir_metadata_not_properly_reset
= 0x8,
1337 nir_cf_node cf_node
;
1339 /** pointer to the function of which this is an implementation */
1340 struct nir_function
*function
;
1342 struct exec_list body
; /** < list of nir_cf_node */
1344 nir_block
*end_block
;
1346 /** list for all local variables in the function */
1347 struct exec_list locals
;
1349 /** array of variables used as parameters */
1350 unsigned num_params
;
1351 nir_variable
**params
;
1353 /** variable used to hold the result of the function */
1354 nir_variable
*return_var
;
1356 /** list of local registers in the function */
1357 struct exec_list registers
;
1359 /** next available local register index */
1362 /** next available SSA value index */
1365 /* total number of basic blocks, only valid when block_index_dirty = false */
1366 unsigned num_blocks
;
1368 nir_metadata valid_metadata
;
1369 } nir_function_impl
;
1371 static inline nir_block
*
1372 nir_start_block(nir_function_impl
*impl
)
1374 return (nir_block
*) exec_list_get_head(&impl
->body
);
1377 static inline nir_cf_node
*
1378 nir_cf_node_next(nir_cf_node
*node
)
1380 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1381 if (exec_node_is_tail_sentinel(next
))
1384 return exec_node_data(nir_cf_node
, next
, node
);
1387 static inline nir_cf_node
*
1388 nir_cf_node_prev(nir_cf_node
*node
)
1390 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1391 if (exec_node_is_head_sentinel(prev
))
1394 return exec_node_data(nir_cf_node
, prev
, node
);
1398 nir_cf_node_is_first(const nir_cf_node
*node
)
1400 return exec_node_is_head_sentinel(node
->node
.prev
);
1404 nir_cf_node_is_last(const nir_cf_node
*node
)
1406 return exec_node_is_tail_sentinel(node
->node
.next
);
1409 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1410 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1411 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1412 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1417 nir_parameter_inout
,
1418 } nir_parameter_type
;
1421 nir_parameter_type param_type
;
1422 const struct glsl_type
*type
;
1425 typedef struct nir_function
{
1426 struct exec_node node
;
1429 struct nir_shader
*shader
;
1431 unsigned num_params
;
1432 nir_parameter
*params
;
1433 const struct glsl_type
*return_type
;
1435 /** The implementation of this function.
1437 * If the function is only declared and not implemented, this is NULL.
1439 nir_function_impl
*impl
;
1442 typedef struct nir_shader_compiler_options
{
1450 bool lower_bitfield_extract
;
1451 bool lower_bitfield_insert
;
1452 bool lower_uadd_carry
;
1453 bool lower_usub_borrow
;
1454 /** lowers fneg and ineg to fsub and isub. */
1456 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1459 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1462 /* Does the native fdot instruction replicate its result for four
1463 * components? If so, then opt_algebraic_late will turn all fdotN
1464 * instructions into fdot_replicatedN instructions.
1466 bool fdot_replicates
;
1468 /** lowers ffract to fsub+ffloor: */
1472 * Does the driver support real 32-bit integers? (Otherwise, integers
1473 * are simulated by floats.)
1475 bool native_integers
;
1476 } nir_shader_compiler_options
;
1478 typedef struct nir_shader_info
{
1481 /* Descriptive name provided by the client; may be NULL */
1484 /* Number of textures used by this shader */
1485 unsigned num_textures
;
1486 /* Number of uniform buffers used by this shader */
1488 /* Number of atomic buffers used by this shader */
1490 /* Number of shader storage buffers used by this shader */
1492 /* Number of images used by this shader */
1493 unsigned num_images
;
1495 /* Which inputs are actually read */
1496 uint64_t inputs_read
;
1497 /* Which outputs are actually written */
1498 uint64_t outputs_written
;
1499 /* Which system values are actually read */
1500 uint64_t system_values_read
;
1502 /* Which patch inputs are actually read */
1503 uint32_t patch_inputs_read
;
1504 /* Which patch outputs are actually written */
1505 uint32_t patch_outputs_written
;
1507 /* Whether or not this shader ever uses textureGather() */
1508 bool uses_texture_gather
;
1510 /* Whether or not this shader uses the gl_ClipDistance output */
1511 bool uses_clip_distance_out
;
1513 /* Whether or not separate shader objects were used */
1514 bool separate_shader
;
1516 /** Was this shader linked with any transform feedback varyings? */
1517 bool has_transform_feedback_varyings
;
1521 /** The number of vertices recieves per input primitive */
1522 unsigned vertices_in
;
1524 /** The output primitive type (GL enum value) */
1525 unsigned output_primitive
;
1527 /** The maximum number of vertices the geometry shader might write. */
1528 unsigned vertices_out
;
1530 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1531 unsigned invocations
;
1533 /** Whether or not this shader uses EndPrimitive */
1534 bool uses_end_primitive
;
1536 /** Whether or not this shader uses non-zero streams */
1544 * Whether early fragment tests are enabled as defined by
1545 * ARB_shader_image_load_store.
1547 bool early_fragment_tests
;
1549 /** gl_FragDepth layout for ARB_conservative_depth. */
1550 enum gl_frag_depth_layout depth_layout
;
1554 unsigned local_size
[3];
1558 /** The number of vertices in the TCS output patch. */
1559 unsigned vertices_out
;
1564 typedef struct nir_shader
{
1565 /** list of uniforms (nir_variable) */
1566 struct exec_list uniforms
;
1568 /** list of inputs (nir_variable) */
1569 struct exec_list inputs
;
1571 /** list of outputs (nir_variable) */
1572 struct exec_list outputs
;
1574 /** Set of driver-specific options for the shader.
1576 * The memory for the options is expected to be kept in a single static
1577 * copy by the driver.
1579 const struct nir_shader_compiler_options
*options
;
1581 /** Various bits of compile-time information about a given shader */
1582 struct nir_shader_info info
;
1584 /** list of global variables in the shader (nir_variable) */
1585 struct exec_list globals
;
1587 /** list of system value variables in the shader (nir_variable) */
1588 struct exec_list system_values
;
1590 struct exec_list functions
; /** < list of nir_function */
1592 /** list of global register in the shader */
1593 struct exec_list registers
;
1595 /** next available global register index */
1599 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1602 unsigned num_inputs
, num_uniforms
, num_outputs
;
1604 /** The shader stage, such as MESA_SHADER_VERTEX. */
1605 gl_shader_stage stage
;
1608 #define nir_foreach_function(shader, func) \
1609 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1611 nir_shader
*nir_shader_create(void *mem_ctx
,
1612 gl_shader_stage stage
,
1613 const nir_shader_compiler_options
*options
);
1615 /** creates a register, including assigning it an index and adding it to the list */
1616 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1618 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1620 void nir_reg_remove(nir_register
*reg
);
1622 /** Adds a variable to the appropreate list in nir_shader */
1623 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1626 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1628 assert(var
->data
.mode
== nir_var_local
);
1629 exec_list_push_tail(&impl
->locals
, &var
->node
);
1632 /** creates a variable, sets a few defaults, and adds it to the list */
1633 nir_variable
*nir_variable_create(nir_shader
*shader
,
1634 nir_variable_mode mode
,
1635 const struct glsl_type
*type
,
1637 /** creates a local variable and adds it to the list */
1638 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1639 const struct glsl_type
*type
,
1642 /** creates a function and adds it to the shader's list of functions */
1643 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1645 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1647 nir_block
*nir_block_create(nir_shader
*shader
);
1648 nir_if
*nir_if_create(nir_shader
*shader
);
1649 nir_loop
*nir_loop_create(nir_shader
*shader
);
1651 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1653 /** requests that the given pieces of metadata be generated */
1654 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1655 /** dirties all but the preserved metadata */
1656 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1658 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1659 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1661 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1663 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1664 unsigned num_components
);
1666 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1667 nir_intrinsic_op op
);
1669 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1670 nir_function
*callee
);
1672 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1674 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1676 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1678 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1679 unsigned num_components
);
1681 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1682 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1683 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1685 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1687 nir_load_const_instr
*
1688 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1691 * NIR Cursors and Instruction Insertion API
1694 * A tiny struct representing a point to insert/extract instructions or
1695 * control flow nodes. Helps reduce the combinatorial explosion of possible
1696 * points to insert/extract.
1698 * \sa nir_control_flow.h
1701 nir_cursor_before_block
,
1702 nir_cursor_after_block
,
1703 nir_cursor_before_instr
,
1704 nir_cursor_after_instr
,
1705 } nir_cursor_option
;
1708 nir_cursor_option option
;
1715 static inline nir_cursor
1716 nir_before_block(nir_block
*block
)
1719 cursor
.option
= nir_cursor_before_block
;
1720 cursor
.block
= block
;
1724 static inline nir_cursor
1725 nir_after_block(nir_block
*block
)
1728 cursor
.option
= nir_cursor_after_block
;
1729 cursor
.block
= block
;
1733 static inline nir_cursor
1734 nir_before_instr(nir_instr
*instr
)
1737 cursor
.option
= nir_cursor_before_instr
;
1738 cursor
.instr
= instr
;
1742 static inline nir_cursor
1743 nir_after_instr(nir_instr
*instr
)
1746 cursor
.option
= nir_cursor_after_instr
;
1747 cursor
.instr
= instr
;
1751 static inline nir_cursor
1752 nir_after_block_before_jump(nir_block
*block
)
1754 nir_instr
*last_instr
= nir_block_last_instr(block
);
1755 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1756 return nir_before_instr(last_instr
);
1758 return nir_after_block(block
);
1762 static inline nir_cursor
1763 nir_before_cf_node(nir_cf_node
*node
)
1765 if (node
->type
== nir_cf_node_block
)
1766 return nir_before_block(nir_cf_node_as_block(node
));
1768 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1771 static inline nir_cursor
1772 nir_after_cf_node(nir_cf_node
*node
)
1774 if (node
->type
== nir_cf_node_block
)
1775 return nir_after_block(nir_cf_node_as_block(node
));
1777 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1780 static inline nir_cursor
1781 nir_before_cf_list(struct exec_list
*cf_list
)
1783 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1784 exec_list_get_head(cf_list
), node
);
1785 return nir_before_cf_node(first_node
);
1788 static inline nir_cursor
1789 nir_after_cf_list(struct exec_list
*cf_list
)
1791 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1792 exec_list_get_tail(cf_list
), node
);
1793 return nir_after_cf_node(last_node
);
1797 * Insert a NIR instruction at the given cursor.
1799 * Note: This does not update the cursor.
1801 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
1804 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1806 nir_instr_insert(nir_before_instr(instr
), before
);
1810 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
1812 nir_instr_insert(nir_after_instr(instr
), after
);
1816 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
1818 nir_instr_insert(nir_before_block(block
), before
);
1822 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
1824 nir_instr_insert(nir_after_block(block
), after
);
1828 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
1830 nir_instr_insert(nir_before_cf_node(node
), before
);
1834 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
1836 nir_instr_insert(nir_after_cf_node(node
), after
);
1840 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
1842 nir_instr_insert(nir_before_cf_list(list
), before
);
1846 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
1848 nir_instr_insert(nir_after_cf_list(list
), after
);
1851 void nir_instr_remove(nir_instr
*instr
);
1855 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
1856 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
1857 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
1858 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
1860 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
1861 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
1863 nir_const_value
*nir_src_as_const_value(nir_src src
);
1864 bool nir_src_is_dynamically_uniform(nir_src src
);
1865 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
1866 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
1867 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
1868 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
1869 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
1872 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1873 unsigned num_components
, const char *name
);
1874 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1875 unsigned num_components
, const char *name
);
1876 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
1877 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
1878 nir_instr
*after_me
);
1880 /* visits basic blocks in source-code order */
1881 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
1882 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1884 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1886 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1889 /* If the following CF node is an if, this function returns that if.
1890 * Otherwise, it returns NULL.
1892 nir_if
*nir_block_get_following_if(nir_block
*block
);
1894 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
1896 void nir_index_local_regs(nir_function_impl
*impl
);
1897 void nir_index_global_regs(nir_shader
*shader
);
1898 void nir_index_ssa_defs(nir_function_impl
*impl
);
1899 unsigned nir_index_instrs(nir_function_impl
*impl
);
1901 void nir_index_blocks(nir_function_impl
*impl
);
1903 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
1904 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
1906 nir_shader
* nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
1909 void nir_validate_shader(nir_shader
*shader
);
1910 void nir_metadata_set_validation_flag(nir_shader
*shader
);
1911 void nir_metadata_check_validation_flag(nir_shader
*shader
);
1913 #include "util/debug.h"
1915 should_clone_nir(void)
1917 static int should_clone
= -1;
1918 if (should_clone
< 0)
1919 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
1921 return should_clone
;
1924 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
1925 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
1926 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
1927 static inline bool should_clone_nir(void) { return false; }
1930 #define _PASS(nir, do_pass) do { \
1932 nir_validate_shader(nir); \
1933 if (should_clone_nir()) { \
1934 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
1940 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
1941 nir_metadata_set_validation_flag(nir); \
1942 if (pass(nir, ##__VA_ARGS__)) { \
1944 nir_metadata_check_validation_flag(nir); \
1948 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
1949 pass(nir, ##__VA_ARGS__); \
1952 void nir_calc_dominance_impl(nir_function_impl
*impl
);
1953 void nir_calc_dominance(nir_shader
*shader
);
1955 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
1956 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
1958 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
1959 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
1961 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
1962 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
1964 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
1965 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
1967 int nir_gs_count_vertices(const nir_shader
*shader
);
1969 bool nir_split_var_copies(nir_shader
*shader
);
1971 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
1972 void nir_lower_var_copies(nir_shader
*shader
);
1974 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
1976 bool nir_lower_locals_to_regs(nir_shader
*shader
);
1978 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
1980 void nir_assign_var_locations(struct exec_list
*var_list
,
1982 int (*type_size
)(const struct glsl_type
*));
1984 void nir_lower_io(nir_shader
*shader
,
1985 nir_variable_mode mode
,
1986 int (*type_size
)(const struct glsl_type
*));
1987 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
1988 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
1990 void nir_lower_vars_to_ssa(nir_shader
*shader
);
1992 bool nir_remove_dead_variables(nir_shader
*shader
);
1994 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
1995 bool nir_lower_vec_to_movs(nir_shader
*shader
);
1996 void nir_lower_alu_to_scalar(nir_shader
*shader
);
1997 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
1999 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2001 void nir_lower_samplers(nir_shader
*shader
,
2002 const struct gl_shader_program
*shader_program
);
2004 bool nir_lower_system_values(nir_shader
*shader
);
2006 typedef struct nir_lower_tex_options
{
2008 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2009 * sampler types a texture projector is lowered.
2014 * If true, lower rect textures to 2D, using txs to fetch the
2015 * texture dimensions and dividing the texture coords by the
2016 * texture dims to normalize.
2021 * To emulate certain texture wrap modes, this can be used
2022 * to saturate the specified tex coord to [0.0, 1.0]. The
2023 * bits are according to sampler #, ie. if, for example:
2025 * (conf->saturate_s & (1 << n))
2027 * is true, then the s coord for sampler n is saturated.
2029 * Note that clamping must happen *after* projector lowering
2030 * so any projected texture sample instruction with a clamped
2031 * coordinate gets automatically lowered, regardless of the
2032 * 'lower_txp' setting.
2034 unsigned saturate_s
;
2035 unsigned saturate_t
;
2036 unsigned saturate_r
;
2038 /* Bitmask of samplers that need swizzling.
2040 * If (swizzle_result & (1 << sampler_index)), then the swizzle in
2041 * swizzles[sampler_index] is applied to the result of the texturing
2044 unsigned swizzle_result
;
2046 /* A swizzle for each sampler. Values 0-3 represent x, y, z, or w swizzles
2047 * while 4 and 5 represent 0 and 1 respectively.
2049 uint8_t swizzles
[32][4];
2050 } nir_lower_tex_options
;
2052 bool nir_lower_tex(nir_shader
*shader
,
2053 const nir_lower_tex_options
*options
);
2055 void nir_lower_idiv(nir_shader
*shader
);
2057 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2058 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2060 void nir_lower_two_sided_color(nir_shader
*shader
);
2062 void nir_lower_atomics(nir_shader
*shader
,
2063 const struct gl_shader_program
*shader_program
);
2064 void nir_lower_to_source_mods(nir_shader
*shader
);
2066 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2068 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2070 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2071 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2073 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2074 void nir_convert_to_ssa(nir_shader
*shader
);
2076 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2077 * registers. If false, convert all values (even those not involved in a phi
2078 * node) to registers.
2080 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2082 bool nir_opt_algebraic(nir_shader
*shader
);
2083 bool nir_opt_algebraic_late(nir_shader
*shader
);
2084 bool nir_opt_constant_folding(nir_shader
*shader
);
2086 bool nir_opt_global_to_local(nir_shader
*shader
);
2088 bool nir_copy_prop(nir_shader
*shader
);
2090 bool nir_opt_cse(nir_shader
*shader
);
2092 bool nir_opt_dce(nir_shader
*shader
);
2094 bool nir_opt_dead_cf(nir_shader
*shader
);
2096 void nir_opt_gcm(nir_shader
*shader
);
2098 bool nir_opt_peephole_select(nir_shader
*shader
);
2100 bool nir_opt_remove_phis(nir_shader
*shader
);
2102 bool nir_opt_undef(nir_shader
*shader
);
2104 void nir_sweep(nir_shader
*shader
);
2106 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2107 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);