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"
31 #include "compiler/glsl/list.h"
32 #include "GL/gl.h" /* GLenum */
33 #include "util/list.h"
34 #include "util/ralloc.h"
36 #include "util/bitset.h"
37 #include "compiler/nir_types.h"
38 #include "compiler/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
,
97 * Data stored in an nir_constant
99 union nir_constant_data
{
106 typedef struct nir_constant
{
108 * Value of the constant.
110 * The field used to back the values supplied by the constant is determined
111 * by the type associated with the \c nir_variable. Constants may be
112 * scalars, vectors, or matrices.
114 union nir_constant_data value
;
116 /* we could get this from the var->type but makes clone *much* easier to
117 * not have to care about the type.
119 unsigned num_elements
;
121 /* Array elements / Structure Fields */
122 struct nir_constant
**elements
;
126 * \brief Layout qualifiers for gl_FragDepth.
128 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
129 * with a layout qualifier.
132 nir_depth_layout_none
, /**< No depth layout is specified. */
133 nir_depth_layout_any
,
134 nir_depth_layout_greater
,
135 nir_depth_layout_less
,
136 nir_depth_layout_unchanged
140 * Either a uniform, global variable, shader input, or shader output. Based on
141 * ir_variable - it should be easy to translate between the two.
144 typedef struct nir_variable
{
145 struct exec_node node
;
148 * Declared type of the variable
150 const struct glsl_type
*type
;
153 * Declared name of the variable
157 struct nir_variable_data
{
160 * Is the variable read-only?
162 * This is set for variables declared as \c const, shader inputs,
165 unsigned read_only
:1;
169 unsigned invariant
:1;
172 * Storage class of the variable.
174 * \sa nir_variable_mode
176 nir_variable_mode mode
:5;
179 * Interpolation mode for shader inputs / outputs
181 * \sa glsl_interp_qualifier
183 unsigned interpolation
:2;
186 * \name ARB_fragment_coord_conventions
189 unsigned origin_upper_left
:1;
190 unsigned pixel_center_integer
:1;
194 * Was the location explicitly set in the shader?
196 * If the location is explicitly set in the shader, it \b cannot be changed
197 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
200 unsigned explicit_location
:1;
201 unsigned explicit_index
:1;
204 * Was an initial binding explicitly set in the shader?
206 * If so, constant_initializer contains an integer nir_constant
207 * representing the initial binding point.
209 unsigned explicit_binding
:1;
212 * Does this variable have an initializer?
214 * This is used by the linker to cross-validiate initializers of global
217 unsigned has_initializer
:1;
220 * If non-zero, then this variable may be packed along with other variables
221 * into a single varying slot, so this offset should be applied when
222 * accessing components. For example, an offset of 1 means that the x
223 * component of this variable is actually stored in component y of the
224 * location specified by \c location.
226 unsigned location_frac
:2;
229 * \brief Layout qualifier for gl_FragDepth.
231 * This is not equal to \c ir_depth_layout_none if and only if this
232 * variable is \c gl_FragDepth and a layout qualifier is specified.
234 nir_depth_layout depth_layout
;
237 * Storage location of the base of this variable
239 * The precise meaning of this field depends on the nature of the variable.
241 * - Vertex shader input: one of the values from \c gl_vert_attrib.
242 * - Vertex shader output: one of the values from \c gl_varying_slot.
243 * - Geometry shader input: one of the values from \c gl_varying_slot.
244 * - Geometry shader output: one of the values from \c gl_varying_slot.
245 * - Fragment shader input: one of the values from \c gl_varying_slot.
246 * - Fragment shader output: one of the values from \c gl_frag_result.
247 * - Uniforms: Per-stage uniform slot number for default uniform block.
248 * - Uniforms: Index within the uniform block definition for UBO members.
249 * - Non-UBO Uniforms: uniform slot number.
250 * - Other: This field is not currently used.
252 * If the variable is a uniform, shader input, or shader output, and the
253 * slot has not been assigned, the value will be -1.
258 * The actual location of the variable in the IR. Only valid for inputs
261 unsigned int driver_location
;
264 * output index for dual source blending.
269 * Initial binding point for a sampler or UBO.
271 * For array types, this represents the binding point for the first element.
276 * Location an atomic counter is stored at.
281 * ARB_shader_image_load_store qualifiers.
284 bool read_only
; /**< "readonly" qualifier. */
285 bool write_only
; /**< "writeonly" qualifier. */
290 /** Image internal format if specified explicitly, otherwise GL_NONE. */
295 * Highest element accessed with a constant expression array index
297 * Not used for non-array variables.
299 unsigned max_array_access
;
304 * Built-in state that backs this uniform
306 * Once set at variable creation, \c state_slots must remain invariant.
307 * This is because, ideally, this array would be shared by all clones of
308 * this variable in the IR tree. In other words, we'd really like for it
309 * to be a fly-weight.
311 * If the variable is not a uniform, \c num_state_slots will be zero and
312 * \c state_slots will be \c NULL.
315 unsigned num_state_slots
; /**< Number of state slots used */
316 nir_state_slot
*state_slots
; /**< State descriptors. */
320 * Constant expression assigned in the initializer of the variable
322 nir_constant
*constant_initializer
;
325 * For variables that are in an interface block or are an instance of an
326 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
328 * \sa ir_variable::location
330 const struct glsl_type
*interface_type
;
333 #define nir_foreach_variable(var, var_list) \
334 foreach_list_typed(nir_variable, var, node, var_list)
337 nir_variable_is_global(const nir_variable
*var
)
339 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
342 typedef struct nir_register
{
343 struct exec_node node
;
345 unsigned num_components
; /** < number of vector components */
346 unsigned num_array_elems
; /** < size of array (0 for no array) */
348 /** generic register index. */
351 /** only for debug purposes, can be NULL */
354 /** whether this register is local (per-function) or global (per-shader) */
358 * If this flag is set to true, then accessing channels >= num_components
359 * is well-defined, and simply spills over to the next array element. This
360 * is useful for backends that can do per-component accessing, in
361 * particular scalar backends. By setting this flag and making
362 * num_components equal to 1, structures can be packed tightly into
363 * registers and then registers can be accessed per-component to get to
364 * each structure member, even if it crosses vec4 boundaries.
368 /** set of nir_src's where this register is used (read from) */
369 struct list_head uses
;
371 /** set of nir_dest's where this register is defined (written to) */
372 struct list_head defs
;
374 /** set of nir_if's where this register is used as a condition */
375 struct list_head if_uses
;
382 nir_instr_type_intrinsic
,
383 nir_instr_type_load_const
,
385 nir_instr_type_ssa_undef
,
387 nir_instr_type_parallel_copy
,
390 typedef struct nir_instr
{
391 struct exec_node node
;
393 struct nir_block
*block
;
395 /** generic instruction index. */
398 /* A temporary for optimization and analysis passes to use for storing
399 * flags. For instance, DCE uses this to store the "dead/live" info.
404 static inline nir_instr
*
405 nir_instr_next(nir_instr
*instr
)
407 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
408 if (exec_node_is_tail_sentinel(next
))
411 return exec_node_data(nir_instr
, next
, node
);
414 static inline nir_instr
*
415 nir_instr_prev(nir_instr
*instr
)
417 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
418 if (exec_node_is_head_sentinel(prev
))
421 return exec_node_data(nir_instr
, prev
, node
);
425 nir_instr_is_first(nir_instr
*instr
)
427 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
431 nir_instr_is_last(nir_instr
*instr
)
433 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
436 typedef struct nir_ssa_def
{
437 /** for debugging only, can be NULL */
440 /** generic SSA definition index. */
443 /** Index into the live_in and live_out bitfields */
446 nir_instr
*parent_instr
;
448 /** set of nir_instr's where this register is used (read from) */
449 struct list_head uses
;
451 /** set of nir_if's where this register is used as a condition */
452 struct list_head if_uses
;
454 uint8_t num_components
;
461 struct nir_src
*indirect
; /** < NULL for no indirect offset */
462 unsigned base_offset
;
464 /* TODO use-def chain goes here */
468 nir_instr
*parent_instr
;
469 struct list_head def_link
;
472 struct nir_src
*indirect
; /** < NULL for no indirect offset */
473 unsigned base_offset
;
475 /* TODO def-use chain goes here */
480 typedef struct nir_src
{
482 nir_instr
*parent_instr
;
483 struct nir_if
*parent_if
;
486 struct list_head use_link
;
496 #define NIR_SRC_INIT (nir_src) { { NULL } }
498 #define nir_foreach_use(reg_or_ssa_def, src) \
499 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
501 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
502 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
504 #define nir_foreach_if_use(reg_or_ssa_def, src) \
505 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
507 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
508 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
519 #define NIR_DEST_INIT (nir_dest) { { { NULL } } }
521 #define nir_foreach_def(reg, dest) \
522 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
524 #define nir_foreach_def_safe(reg, dest) \
525 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
527 static inline nir_src
528 nir_src_for_ssa(nir_ssa_def
*def
)
530 nir_src src
= NIR_SRC_INIT
;
538 static inline nir_src
539 nir_src_for_reg(nir_register
*reg
)
541 nir_src src
= NIR_SRC_INIT
;
545 src
.reg
.indirect
= NULL
;
546 src
.reg
.base_offset
= 0;
551 static inline nir_dest
552 nir_dest_for_reg(nir_register
*reg
)
554 nir_dest dest
= NIR_DEST_INIT
;
561 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
562 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
568 * \name input modifiers
572 * For inputs interpreted as floating point, flips the sign bit. For
573 * inputs interpreted as integers, performs the two's complement negation.
578 * Clears the sign bit for floating point values, and computes the integer
579 * absolute value for integers. Note that the negate modifier acts after
580 * the absolute value modifier, therefore if both are set then all inputs
581 * will become negative.
587 * For each input component, says which component of the register it is
588 * chosen from. Note that which elements of the swizzle are used and which
589 * are ignored are based on the write mask for most opcodes - for example,
590 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
591 * a swizzle of {2, x, 1, 0} where x means "don't care."
600 * \name saturate output modifier
602 * Only valid for opcodes that output floating-point numbers. Clamps the
603 * output to between 0.0 and 1.0 inclusive.
608 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
612 nir_type_invalid
= 0, /* Not a valid type */
620 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
621 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
622 } nir_op_algebraic_property
;
630 * The number of components in the output
632 * If non-zero, this is the size of the output and input sizes are
633 * explicitly given; swizzle and writemask are still in effect, but if
634 * the output component is masked out, then the input component may
637 * If zero, the opcode acts in the standard, per-component manner; the
638 * operation is performed on each component (except the ones that are
639 * masked out) with the input being taken from the input swizzle for
642 * The size of some of the inputs may be given (i.e. non-zero) even
643 * though output_size is zero; in that case, the inputs with a zero
644 * size act per-component, while the inputs with non-zero size don't.
646 unsigned output_size
;
649 * The type of vector that the instruction outputs. Note that the
650 * staurate modifier is only allowed on outputs with the float type.
653 nir_alu_type output_type
;
656 * The number of components in each input
658 unsigned input_sizes
[4];
661 * The type of vector that each input takes. Note that negate and
662 * absolute value are only allowed on inputs with int or float type and
663 * behave differently on the two.
665 nir_alu_type input_types
[4];
667 nir_op_algebraic_property algebraic_properties
;
670 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
672 typedef struct nir_alu_instr
{
679 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
680 nir_alu_instr
*instr
);
681 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
682 nir_alu_instr
*instr
);
684 /* is this source channel used? */
686 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
688 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
689 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
691 return (instr
->dest
.write_mask
>> channel
) & 1;
695 * For instructions whose destinations are SSA, get the number of channels
698 static inline unsigned
699 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
701 assert(instr
->dest
.dest
.is_ssa
);
703 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
704 return nir_op_infos
[instr
->op
].input_sizes
[src
];
706 return instr
->dest
.dest
.ssa
.num_components
;
711 nir_deref_type_array
,
712 nir_deref_type_struct
715 typedef struct nir_deref
{
716 nir_deref_type deref_type
;
717 struct nir_deref
*child
;
718 const struct glsl_type
*type
;
727 /* This enum describes how the array is referenced. If the deref is
728 * direct then the base_offset is used. If the deref is indirect then then
729 * offset is given by base_offset + indirect. If the deref is a wildcard
730 * then the deref refers to all of the elements of the array at the same
731 * time. Wildcard dereferences are only ever allowed in copy_var
732 * intrinsics and the source and destination derefs must have matching
736 nir_deref_array_type_direct
,
737 nir_deref_array_type_indirect
,
738 nir_deref_array_type_wildcard
,
739 } nir_deref_array_type
;
744 nir_deref_array_type deref_array_type
;
745 unsigned base_offset
;
755 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
756 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
757 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
759 /* Returns the last deref in the chain. */
760 static inline nir_deref
*
761 nir_deref_tail(nir_deref
*deref
)
764 deref
= deref
->child
;
772 nir_deref_var
**params
;
773 nir_deref_var
*return_deref
;
775 struct nir_function
*callee
;
778 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
779 num_variables, num_indices, idx0, idx1, idx2, flags) \
780 nir_intrinsic_##name,
782 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
785 #include "nir_intrinsics.h"
786 nir_num_intrinsics
= nir_last_intrinsic
+ 1
790 #undef LAST_INTRINSIC
792 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
794 /** Represents an intrinsic
796 * An intrinsic is an instruction type for handling things that are
797 * more-or-less regular operations but don't just consume and produce SSA
798 * values like ALU operations do. Intrinsics are not for things that have
799 * special semantic meaning such as phi nodes and parallel copies.
800 * Examples of intrinsics include variable load/store operations, system
801 * value loads, and the like. Even though texturing more-or-less falls
802 * under this category, texturing is its own instruction type because
803 * trying to represent texturing with intrinsics would lead to a
804 * combinatorial explosion of intrinsic opcodes.
806 * By having a single instruction type for handling a lot of different
807 * cases, optimization passes can look for intrinsics and, for the most
808 * part, completely ignore them. Each intrinsic type also has a few
809 * possible flags that govern whether or not they can be reordered or
810 * eliminated. That way passes like dead code elimination can still work
811 * on intrisics without understanding the meaning of each.
813 * Each intrinsic has some number of constant indices, some number of
814 * variables, and some number of sources. What these sources, variables,
815 * and indices mean depends on the intrinsic and is documented with the
816 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
817 * instructions are the only types of instruction that can operate on
823 nir_intrinsic_op intrinsic
;
827 /** number of components if this is a vectorized intrinsic
829 * Similarly to ALU operations, some intrinsics are vectorized.
830 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
831 * For vectorized intrinsics, the num_components field specifies the
832 * number of destination components and the number of source components
833 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
835 uint8_t num_components
;
837 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
839 nir_deref_var
*variables
[2];
842 } nir_intrinsic_instr
;
845 * \name NIR intrinsics semantic flags
847 * information about what the compiler can do with the intrinsics.
849 * \sa nir_intrinsic_info::flags
853 * whether the intrinsic can be safely eliminated if none of its output
854 * value is not being used.
856 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
859 * Whether the intrinsic can be reordered with respect to any other
860 * intrinsic, i.e. whether the only reordering dependencies of the
861 * intrinsic are due to the register reads/writes.
863 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
864 } nir_intrinsic_semantic_flag
;
867 * \name NIR intrinsics const-index flag
869 * Indicates the usage of a const_index slot.
871 * \sa nir_intrinsic_info::index_map
875 * Generally instructions that take a offset src argument, can encode
876 * a constant 'base' value which is added to the offset.
878 NIR_INTRINSIC_BASE
= 1,
881 * For store instructions, a writemask for the store.
883 NIR_INTRINSIC_WRMASK
= 2,
886 * The stream-id for GS emit_vertex/end_primitive intrinsics.
888 NIR_INTRINSIC_STREAM_ID
= 3,
891 * The clip-plane id for load_user_clip_plane intrinsic.
893 NIR_INTRINSIC_UCP_ID
= 4,
895 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
897 } nir_intrinsic_index_flag
;
899 #define NIR_INTRINSIC_MAX_INPUTS 4
904 unsigned num_srcs
; /** < number of register/SSA inputs */
906 /** number of components of each input register
908 * If this value is 0, the number of components is given by the
909 * num_components field of nir_intrinsic_instr.
911 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
915 /** number of components of the output register
917 * If this value is 0, the number of components is given by the
918 * num_components field of nir_intrinsic_instr.
920 unsigned dest_components
;
922 /** the number of inputs/outputs that are variables */
923 unsigned num_variables
;
925 /** the number of constant indices used by the intrinsic */
926 unsigned num_indices
;
928 /** indicates the usage of intr->const_index[n] */
929 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
931 /** semantic flags for calls to this intrinsic */
932 nir_intrinsic_semantic_flag flags
;
933 } nir_intrinsic_info
;
935 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
938 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
940 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
942 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
943 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
944 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
947 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
949 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
950 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
951 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
954 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
955 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
956 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
957 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
960 * \group texture information
962 * This gives semantic information about textures which is useful to the
963 * frontend, the backend, and lowering passes, but not the optimizer.
968 nir_tex_src_projector
,
969 nir_tex_src_comparitor
, /* shadow comparitor */
973 nir_tex_src_ms_index
, /* MSAA sample index */
976 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
977 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
978 nir_num_tex_src_types
983 nir_tex_src_type src_type
;
987 nir_texop_tex
, /**< Regular texture look-up */
988 nir_texop_txb
, /**< Texture look-up with LOD bias */
989 nir_texop_txl
, /**< Texture look-up with explicit LOD */
990 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
991 nir_texop_txf
, /**< Texel fetch with explicit LOD */
992 nir_texop_txf_ms
, /**< Multisample texture fetch */
993 nir_texop_txs
, /**< Texture size */
994 nir_texop_lod
, /**< Texture lod query */
995 nir_texop_tg4
, /**< Texture gather */
996 nir_texop_query_levels
, /**< Texture levels query */
997 nir_texop_texture_samples
, /**< Texture samples query */
998 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1006 enum glsl_sampler_dim sampler_dim
;
1007 nir_alu_type dest_type
;
1012 unsigned num_srcs
, coord_components
;
1013 bool is_array
, is_shadow
;
1016 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1017 * components or the new-style shadow that outputs 1 component.
1019 bool is_new_style_shadow
;
1021 /* gather component selector */
1022 unsigned component
: 2;
1024 /** The texture index
1026 * If this texture instruction has a nir_tex_src_texture_offset source,
1027 * then the texture index is given by texture_index + texture_offset.
1029 unsigned texture_index
;
1031 /** The size of the texture array or 0 if it's not an array */
1032 unsigned texture_array_size
;
1034 /** The texture deref
1036 * If this is null, use texture_index instead.
1038 nir_deref_var
*texture
;
1040 /** The sampler index
1042 * The following operations do not require a sampler and, as such, this
1043 * field should be ignored:
1045 * - nir_texop_txf_ms
1049 * - nir_texop_query_levels
1050 * - nir_texop_texture_samples
1051 * - nir_texop_samples_identical
1053 * If this texture instruction has a nir_tex_src_sampler_offset source,
1054 * then the sampler index is given by sampler_index + sampler_offset.
1056 unsigned sampler_index
;
1058 /** The sampler deref
1060 * If this is null, use sampler_index instead.
1062 nir_deref_var
*sampler
;
1065 static inline unsigned
1066 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1068 switch (instr
->op
) {
1069 case nir_texop_txs
: {
1071 switch (instr
->sampler_dim
) {
1072 case GLSL_SAMPLER_DIM_1D
:
1073 case GLSL_SAMPLER_DIM_BUF
:
1076 case GLSL_SAMPLER_DIM_2D
:
1077 case GLSL_SAMPLER_DIM_CUBE
:
1078 case GLSL_SAMPLER_DIM_MS
:
1079 case GLSL_SAMPLER_DIM_RECT
:
1080 case GLSL_SAMPLER_DIM_EXTERNAL
:
1083 case GLSL_SAMPLER_DIM_3D
:
1087 unreachable("not reached");
1089 if (instr
->is_array
)
1097 case nir_texop_texture_samples
:
1098 case nir_texop_query_levels
:
1099 case nir_texop_samples_identical
:
1103 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1110 /* Returns true if this texture operation queries something about the texture
1111 * rather than actually sampling it.
1114 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1116 switch (instr
->op
) {
1119 case nir_texop_texture_samples
:
1120 case nir_texop_query_levels
:
1127 case nir_texop_txf_ms
:
1131 unreachable("Invalid texture opcode");
1135 static inline unsigned
1136 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1138 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1139 return instr
->coord_components
;
1142 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1143 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1144 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1145 if (instr
->is_array
)
1146 return instr
->coord_components
- 1;
1148 return instr
->coord_components
;
1155 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1157 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1158 if (instr
->src
[i
].src_type
== type
)
1175 nir_const_value value
;
1178 } nir_load_const_instr
;
1191 /* creates a new SSA variable in an undefined state */
1196 } nir_ssa_undef_instr
;
1199 struct exec_node node
;
1201 /* The predecessor block corresponding to this source */
1202 struct nir_block
*pred
;
1207 #define nir_foreach_phi_src(phi, entry) \
1208 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1209 #define nir_foreach_phi_src_safe(phi, entry) \
1210 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1215 struct exec_list srcs
; /** < list of nir_phi_src */
1221 struct exec_node node
;
1224 } nir_parallel_copy_entry
;
1226 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1227 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1232 /* A list of nir_parallel_copy_entry's. The sources of all of the
1233 * entries are copied to the corresponding destinations "in parallel".
1234 * In other words, if we have two entries: a -> b and b -> a, the values
1237 struct exec_list entries
;
1238 } nir_parallel_copy_instr
;
1240 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1241 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1242 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1243 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1244 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1245 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1246 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1247 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1248 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1249 nir_parallel_copy_instr
, instr
)
1254 * Control flow consists of a tree of control flow nodes, which include
1255 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1256 * instructions that always run start-to-finish. Each basic block also keeps
1257 * track of its successors (blocks which may run immediately after the current
1258 * block) and predecessors (blocks which could have run immediately before the
1259 * current block). Each function also has a start block and an end block which
1260 * all return statements point to (which is always empty). Together, all the
1261 * blocks with their predecessors and successors make up the control flow
1262 * graph (CFG) of the function. There are helpers that modify the tree of
1263 * control flow nodes while modifying the CFG appropriately; these should be
1264 * used instead of modifying the tree directly.
1271 nir_cf_node_function
1274 typedef struct nir_cf_node
{
1275 struct exec_node node
;
1276 nir_cf_node_type type
;
1277 struct nir_cf_node
*parent
;
1280 typedef struct nir_block
{
1281 nir_cf_node cf_node
;
1283 struct exec_list instr_list
; /** < list of nir_instr */
1285 /** generic block index; generated by nir_index_blocks */
1289 * Each block can only have up to 2 successors, so we put them in a simple
1290 * array - no need for anything more complicated.
1292 struct nir_block
*successors
[2];
1294 /* Set of nir_block predecessors in the CFG */
1295 struct set
*predecessors
;
1298 * this node's immediate dominator in the dominance tree - set to NULL for
1301 struct nir_block
*imm_dom
;
1303 /* This node's children in the dominance tree */
1304 unsigned num_dom_children
;
1305 struct nir_block
**dom_children
;
1307 /* Set of nir_block's on the dominance frontier of this block */
1308 struct set
*dom_frontier
;
1311 * These two indices have the property that dom_{pre,post}_index for each
1312 * child of this block in the dominance tree will always be between
1313 * dom_pre_index and dom_post_index for this block, which makes testing if
1314 * a given block is dominated by another block an O(1) operation.
1316 unsigned dom_pre_index
, dom_post_index
;
1318 /* live in and out for this block; used for liveness analysis */
1319 BITSET_WORD
*live_in
;
1320 BITSET_WORD
*live_out
;
1323 static inline nir_instr
*
1324 nir_block_first_instr(nir_block
*block
)
1326 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1327 return exec_node_data(nir_instr
, head
, node
);
1330 static inline nir_instr
*
1331 nir_block_last_instr(nir_block
*block
)
1333 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1334 return exec_node_data(nir_instr
, tail
, node
);
1337 #define nir_foreach_instr(block, instr) \
1338 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1339 #define nir_foreach_instr_reverse(block, instr) \
1340 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1341 #define nir_foreach_instr_safe(block, instr) \
1342 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1343 #define nir_foreach_instr_reverse_safe(block, instr) \
1344 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1346 typedef struct nir_if
{
1347 nir_cf_node cf_node
;
1350 struct exec_list then_list
; /** < list of nir_cf_node */
1351 struct exec_list else_list
; /** < list of nir_cf_node */
1354 static inline nir_cf_node
*
1355 nir_if_first_then_node(nir_if
*if_stmt
)
1357 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1358 return exec_node_data(nir_cf_node
, head
, node
);
1361 static inline nir_cf_node
*
1362 nir_if_last_then_node(nir_if
*if_stmt
)
1364 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1365 return exec_node_data(nir_cf_node
, tail
, node
);
1368 static inline nir_cf_node
*
1369 nir_if_first_else_node(nir_if
*if_stmt
)
1371 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1372 return exec_node_data(nir_cf_node
, head
, node
);
1375 static inline nir_cf_node
*
1376 nir_if_last_else_node(nir_if
*if_stmt
)
1378 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1379 return exec_node_data(nir_cf_node
, tail
, node
);
1383 nir_cf_node cf_node
;
1385 struct exec_list body
; /** < list of nir_cf_node */
1388 static inline nir_cf_node
*
1389 nir_loop_first_cf_node(nir_loop
*loop
)
1391 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1394 static inline nir_cf_node
*
1395 nir_loop_last_cf_node(nir_loop
*loop
)
1397 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1401 * Various bits of metadata that can may be created or required by
1402 * optimization and analysis passes
1405 nir_metadata_none
= 0x0,
1406 nir_metadata_block_index
= 0x1,
1407 nir_metadata_dominance
= 0x2,
1408 nir_metadata_live_ssa_defs
= 0x4,
1409 nir_metadata_not_properly_reset
= 0x8,
1413 nir_cf_node cf_node
;
1415 /** pointer to the function of which this is an implementation */
1416 struct nir_function
*function
;
1418 struct exec_list body
; /** < list of nir_cf_node */
1420 nir_block
*end_block
;
1422 /** list for all local variables in the function */
1423 struct exec_list locals
;
1425 /** array of variables used as parameters */
1426 unsigned num_params
;
1427 nir_variable
**params
;
1429 /** variable used to hold the result of the function */
1430 nir_variable
*return_var
;
1432 /** list of local registers in the function */
1433 struct exec_list registers
;
1435 /** next available local register index */
1438 /** next available SSA value index */
1441 /* total number of basic blocks, only valid when block_index_dirty = false */
1442 unsigned num_blocks
;
1444 nir_metadata valid_metadata
;
1445 } nir_function_impl
;
1447 static inline nir_block
*
1448 nir_start_block(nir_function_impl
*impl
)
1450 return (nir_block
*) exec_list_get_head(&impl
->body
);
1453 static inline nir_cf_node
*
1454 nir_cf_node_next(nir_cf_node
*node
)
1456 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1457 if (exec_node_is_tail_sentinel(next
))
1460 return exec_node_data(nir_cf_node
, next
, node
);
1463 static inline nir_cf_node
*
1464 nir_cf_node_prev(nir_cf_node
*node
)
1466 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1467 if (exec_node_is_head_sentinel(prev
))
1470 return exec_node_data(nir_cf_node
, prev
, node
);
1474 nir_cf_node_is_first(const nir_cf_node
*node
)
1476 return exec_node_is_head_sentinel(node
->node
.prev
);
1480 nir_cf_node_is_last(const nir_cf_node
*node
)
1482 return exec_node_is_tail_sentinel(node
->node
.next
);
1485 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1486 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1487 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1488 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1493 nir_parameter_inout
,
1494 } nir_parameter_type
;
1497 nir_parameter_type param_type
;
1498 const struct glsl_type
*type
;
1501 typedef struct nir_function
{
1502 struct exec_node node
;
1505 struct nir_shader
*shader
;
1507 unsigned num_params
;
1508 nir_parameter
*params
;
1509 const struct glsl_type
*return_type
;
1511 /** The implementation of this function.
1513 * If the function is only declared and not implemented, this is NULL.
1515 nir_function_impl
*impl
;
1518 typedef struct nir_shader_compiler_options
{
1526 bool lower_bitfield_extract
;
1527 bool lower_bitfield_insert
;
1528 bool lower_uadd_carry
;
1529 bool lower_usub_borrow
;
1530 /** lowers fneg and ineg to fsub and isub. */
1532 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1535 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1538 /* Does the native fdot instruction replicate its result for four
1539 * components? If so, then opt_algebraic_late will turn all fdotN
1540 * instructions into fdot_replicatedN instructions.
1542 bool fdot_replicates
;
1544 /** lowers ffract to fsub+ffloor: */
1547 bool lower_pack_half_2x16
;
1548 bool lower_pack_unorm_2x16
;
1549 bool lower_pack_snorm_2x16
;
1550 bool lower_pack_unorm_4x8
;
1551 bool lower_pack_snorm_4x8
;
1552 bool lower_unpack_half_2x16
;
1553 bool lower_unpack_unorm_2x16
;
1554 bool lower_unpack_snorm_2x16
;
1555 bool lower_unpack_unorm_4x8
;
1556 bool lower_unpack_snorm_4x8
;
1558 bool lower_extract_byte
;
1559 bool lower_extract_word
;
1562 * Does the driver support real 32-bit integers? (Otherwise, integers
1563 * are simulated by floats.)
1565 bool native_integers
;
1566 } nir_shader_compiler_options
;
1568 typedef struct nir_shader_info
{
1571 /* Descriptive name provided by the client; may be NULL */
1574 /* Number of textures used by this shader */
1575 unsigned num_textures
;
1576 /* Number of uniform buffers used by this shader */
1578 /* Number of atomic buffers used by this shader */
1580 /* Number of shader storage buffers used by this shader */
1582 /* Number of images used by this shader */
1583 unsigned num_images
;
1585 /* Which inputs are actually read */
1586 uint64_t inputs_read
;
1587 /* Which outputs are actually written */
1588 uint64_t outputs_written
;
1589 /* Which system values are actually read */
1590 uint64_t system_values_read
;
1592 /* Which patch inputs are actually read */
1593 uint32_t patch_inputs_read
;
1594 /* Which patch outputs are actually written */
1595 uint32_t patch_outputs_written
;
1597 /* Whether or not this shader ever uses textureGather() */
1598 bool uses_texture_gather
;
1600 /* Whether or not this shader uses the gl_ClipDistance output */
1601 bool uses_clip_distance_out
;
1603 /* Whether or not separate shader objects were used */
1604 bool separate_shader
;
1606 /** Was this shader linked with any transform feedback varyings? */
1607 bool has_transform_feedback_varyings
;
1611 /** The number of vertices recieves per input primitive */
1612 unsigned vertices_in
;
1614 /** The output primitive type (GL enum value) */
1615 unsigned output_primitive
;
1617 /** The maximum number of vertices the geometry shader might write. */
1618 unsigned vertices_out
;
1620 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1621 unsigned invocations
;
1623 /** Whether or not this shader uses EndPrimitive */
1624 bool uses_end_primitive
;
1626 /** Whether or not this shader uses non-zero streams */
1634 * Whether early fragment tests are enabled as defined by
1635 * ARB_shader_image_load_store.
1637 bool early_fragment_tests
;
1639 /** gl_FragDepth layout for ARB_conservative_depth. */
1640 enum gl_frag_depth_layout depth_layout
;
1644 unsigned local_size
[3];
1648 /** The number of vertices in the TCS output patch. */
1649 unsigned vertices_out
;
1654 typedef struct nir_shader
{
1655 /** list of uniforms (nir_variable) */
1656 struct exec_list uniforms
;
1658 /** list of inputs (nir_variable) */
1659 struct exec_list inputs
;
1661 /** list of outputs (nir_variable) */
1662 struct exec_list outputs
;
1664 /** list of shared compute variables (nir_variable) */
1665 struct exec_list shared
;
1667 /** Set of driver-specific options for the shader.
1669 * The memory for the options is expected to be kept in a single static
1670 * copy by the driver.
1672 const struct nir_shader_compiler_options
*options
;
1674 /** Various bits of compile-time information about a given shader */
1675 struct nir_shader_info info
;
1677 /** list of global variables in the shader (nir_variable) */
1678 struct exec_list globals
;
1680 /** list of system value variables in the shader (nir_variable) */
1681 struct exec_list system_values
;
1683 struct exec_list functions
; /** < list of nir_function */
1685 /** list of global register in the shader */
1686 struct exec_list registers
;
1688 /** next available global register index */
1692 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1695 unsigned num_inputs
, num_uniforms
, num_outputs
;
1697 /** The shader stage, such as MESA_SHADER_VERTEX. */
1698 gl_shader_stage stage
;
1701 #define nir_foreach_function(shader, func) \
1702 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1704 nir_shader
*nir_shader_create(void *mem_ctx
,
1705 gl_shader_stage stage
,
1706 const nir_shader_compiler_options
*options
);
1708 /** creates a register, including assigning it an index and adding it to the list */
1709 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1711 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1713 void nir_reg_remove(nir_register
*reg
);
1715 /** Adds a variable to the appropreate list in nir_shader */
1716 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1719 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1721 assert(var
->data
.mode
== nir_var_local
);
1722 exec_list_push_tail(&impl
->locals
, &var
->node
);
1725 /** creates a variable, sets a few defaults, and adds it to the list */
1726 nir_variable
*nir_variable_create(nir_shader
*shader
,
1727 nir_variable_mode mode
,
1728 const struct glsl_type
*type
,
1730 /** creates a local variable and adds it to the list */
1731 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1732 const struct glsl_type
*type
,
1735 /** creates a function and adds it to the shader's list of functions */
1736 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1738 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1739 /** creates a function_impl that isn't tied to any particular function */
1740 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1742 nir_block
*nir_block_create(nir_shader
*shader
);
1743 nir_if
*nir_if_create(nir_shader
*shader
);
1744 nir_loop
*nir_loop_create(nir_shader
*shader
);
1746 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1748 /** requests that the given pieces of metadata be generated */
1749 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1750 /** dirties all but the preserved metadata */
1751 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1753 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1754 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1756 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1758 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1759 unsigned num_components
);
1761 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1762 nir_intrinsic_op op
);
1764 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1765 nir_function
*callee
);
1767 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1769 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1771 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1773 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1774 unsigned num_components
);
1776 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1777 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1778 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1780 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1782 nir_load_const_instr
*
1783 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1786 * NIR Cursors and Instruction Insertion API
1789 * A tiny struct representing a point to insert/extract instructions or
1790 * control flow nodes. Helps reduce the combinatorial explosion of possible
1791 * points to insert/extract.
1793 * \sa nir_control_flow.h
1796 nir_cursor_before_block
,
1797 nir_cursor_after_block
,
1798 nir_cursor_before_instr
,
1799 nir_cursor_after_instr
,
1800 } nir_cursor_option
;
1803 nir_cursor_option option
;
1810 static inline nir_cursor
1811 nir_before_block(nir_block
*block
)
1814 cursor
.option
= nir_cursor_before_block
;
1815 cursor
.block
= block
;
1819 static inline nir_cursor
1820 nir_after_block(nir_block
*block
)
1823 cursor
.option
= nir_cursor_after_block
;
1824 cursor
.block
= block
;
1828 static inline nir_cursor
1829 nir_before_instr(nir_instr
*instr
)
1832 cursor
.option
= nir_cursor_before_instr
;
1833 cursor
.instr
= instr
;
1837 static inline nir_cursor
1838 nir_after_instr(nir_instr
*instr
)
1841 cursor
.option
= nir_cursor_after_instr
;
1842 cursor
.instr
= instr
;
1846 static inline nir_cursor
1847 nir_after_block_before_jump(nir_block
*block
)
1849 nir_instr
*last_instr
= nir_block_last_instr(block
);
1850 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1851 return nir_before_instr(last_instr
);
1853 return nir_after_block(block
);
1857 static inline nir_cursor
1858 nir_before_cf_node(nir_cf_node
*node
)
1860 if (node
->type
== nir_cf_node_block
)
1861 return nir_before_block(nir_cf_node_as_block(node
));
1863 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1866 static inline nir_cursor
1867 nir_after_cf_node(nir_cf_node
*node
)
1869 if (node
->type
== nir_cf_node_block
)
1870 return nir_after_block(nir_cf_node_as_block(node
));
1872 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1875 static inline nir_cursor
1876 nir_before_cf_list(struct exec_list
*cf_list
)
1878 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1879 exec_list_get_head(cf_list
), node
);
1880 return nir_before_cf_node(first_node
);
1883 static inline nir_cursor
1884 nir_after_cf_list(struct exec_list
*cf_list
)
1886 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1887 exec_list_get_tail(cf_list
), node
);
1888 return nir_after_cf_node(last_node
);
1892 * Insert a NIR instruction at the given cursor.
1894 * Note: This does not update the cursor.
1896 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
1899 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1901 nir_instr_insert(nir_before_instr(instr
), before
);
1905 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
1907 nir_instr_insert(nir_after_instr(instr
), after
);
1911 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
1913 nir_instr_insert(nir_before_block(block
), before
);
1917 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
1919 nir_instr_insert(nir_after_block(block
), after
);
1923 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
1925 nir_instr_insert(nir_before_cf_node(node
), before
);
1929 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
1931 nir_instr_insert(nir_after_cf_node(node
), after
);
1935 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
1937 nir_instr_insert(nir_before_cf_list(list
), before
);
1941 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
1943 nir_instr_insert(nir_after_cf_list(list
), after
);
1946 void nir_instr_remove(nir_instr
*instr
);
1950 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
1951 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
1952 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
1953 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
1955 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
1956 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
1958 nir_const_value
*nir_src_as_const_value(nir_src src
);
1959 bool nir_src_is_dynamically_uniform(nir_src src
);
1960 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
1961 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
1962 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
1963 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
1964 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
1967 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1968 unsigned num_components
, const char *name
);
1969 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1970 unsigned num_components
, const char *name
);
1971 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
1972 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
1973 nir_instr
*after_me
);
1975 /* visits basic blocks in source-code order */
1976 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
1977 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1979 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1981 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1984 /* If the following CF node is an if, this function returns that if.
1985 * Otherwise, it returns NULL.
1987 nir_if
*nir_block_get_following_if(nir_block
*block
);
1989 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
1991 void nir_index_local_regs(nir_function_impl
*impl
);
1992 void nir_index_global_regs(nir_shader
*shader
);
1993 void nir_index_ssa_defs(nir_function_impl
*impl
);
1994 unsigned nir_index_instrs(nir_function_impl
*impl
);
1996 void nir_index_blocks(nir_function_impl
*impl
);
1998 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
1999 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2001 nir_shader
* nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2002 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2005 void nir_validate_shader(nir_shader
*shader
);
2006 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2007 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2009 #include "util/debug.h"
2011 should_clone_nir(void)
2013 static int should_clone
= -1;
2014 if (should_clone
< 0)
2015 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2017 return should_clone
;
2020 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2021 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2022 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2023 static inline bool should_clone_nir(void) { return false; }
2026 #define _PASS(nir, do_pass) do { \
2028 nir_validate_shader(nir); \
2029 if (should_clone_nir()) { \
2030 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2036 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2037 nir_metadata_set_validation_flag(nir); \
2038 if (pass(nir, ##__VA_ARGS__)) { \
2040 nir_metadata_check_validation_flag(nir); \
2044 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2045 pass(nir, ##__VA_ARGS__); \
2048 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2049 void nir_calc_dominance(nir_shader
*shader
);
2051 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2052 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2054 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2055 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2057 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2058 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2060 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2061 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2063 int nir_gs_count_vertices(const nir_shader
*shader
);
2065 bool nir_split_var_copies(nir_shader
*shader
);
2067 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2068 void nir_lower_var_copies(nir_shader
*shader
);
2070 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2072 bool nir_lower_indirect_derefs(nir_shader
*shader
, uint32_t mode_mask
);
2074 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2076 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
2078 void nir_assign_var_locations(struct exec_list
*var_list
,
2080 int (*type_size
)(const struct glsl_type
*));
2082 void nir_lower_io(nir_shader
*shader
,
2083 nir_variable_mode mode
,
2084 int (*type_size
)(const struct glsl_type
*));
2085 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2086 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2088 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2090 bool nir_remove_dead_variables(nir_shader
*shader
);
2092 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2093 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2094 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2095 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2097 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2099 void nir_lower_samplers(nir_shader
*shader
,
2100 const struct gl_shader_program
*shader_program
);
2102 bool nir_lower_system_values(nir_shader
*shader
);
2104 typedef struct nir_lower_tex_options
{
2106 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2107 * sampler types a texture projector is lowered.
2112 * If true, lower rect textures to 2D, using txs to fetch the
2113 * texture dimensions and dividing the texture coords by the
2114 * texture dims to normalize.
2119 * To emulate certain texture wrap modes, this can be used
2120 * to saturate the specified tex coord to [0.0, 1.0]. The
2121 * bits are according to sampler #, ie. if, for example:
2123 * (conf->saturate_s & (1 << n))
2125 * is true, then the s coord for sampler n is saturated.
2127 * Note that clamping must happen *after* projector lowering
2128 * so any projected texture sample instruction with a clamped
2129 * coordinate gets automatically lowered, regardless of the
2130 * 'lower_txp' setting.
2132 unsigned saturate_s
;
2133 unsigned saturate_t
;
2134 unsigned saturate_r
;
2136 /* Bitmask of textures that need swizzling.
2138 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2139 * swizzles[texture_index] is applied to the result of the texturing
2142 unsigned swizzle_result
;
2144 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2145 * while 4 and 5 represent 0 and 1 respectively.
2147 uint8_t swizzles
[32][4];
2148 } nir_lower_tex_options
;
2150 bool nir_lower_tex(nir_shader
*shader
,
2151 const nir_lower_tex_options
*options
);
2153 void nir_lower_idiv(nir_shader
*shader
);
2155 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2156 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2158 void nir_lower_two_sided_color(nir_shader
*shader
);
2160 void nir_lower_atomics(nir_shader
*shader
,
2161 const struct gl_shader_program
*shader_program
);
2162 void nir_lower_to_source_mods(nir_shader
*shader
);
2164 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2166 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2168 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2169 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2171 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2172 void nir_convert_to_ssa(nir_shader
*shader
);
2174 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2175 * registers. If false, convert all values (even those not involved in a phi
2176 * node) to registers.
2178 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2180 bool nir_opt_algebraic(nir_shader
*shader
);
2181 bool nir_opt_algebraic_late(nir_shader
*shader
);
2182 bool nir_opt_constant_folding(nir_shader
*shader
);
2184 bool nir_opt_global_to_local(nir_shader
*shader
);
2186 bool nir_copy_prop(nir_shader
*shader
);
2188 bool nir_opt_cse(nir_shader
*shader
);
2190 bool nir_opt_dce(nir_shader
*shader
);
2192 bool nir_opt_dead_cf(nir_shader
*shader
);
2194 void nir_opt_gcm(nir_shader
*shader
);
2196 bool nir_opt_peephole_select(nir_shader
*shader
);
2198 bool nir_opt_remove_phis(nir_shader
*shader
);
2200 bool nir_opt_undef(nir_shader
*shader
);
2202 void nir_sweep(nir_shader
*shader
);
2204 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2205 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);