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
{
107 typedef struct nir_constant
{
109 * Value of the constant.
111 * The field used to back the values supplied by the constant is determined
112 * by the type associated with the \c nir_variable. Constants may be
113 * scalars, vectors, or matrices.
115 union nir_constant_data value
;
117 /* we could get this from the var->type but makes clone *much* easier to
118 * not have to care about the type.
120 unsigned num_elements
;
122 /* Array elements / Structure Fields */
123 struct nir_constant
**elements
;
127 * \brief Layout qualifiers for gl_FragDepth.
129 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
130 * with a layout qualifier.
133 nir_depth_layout_none
, /**< No depth layout is specified. */
134 nir_depth_layout_any
,
135 nir_depth_layout_greater
,
136 nir_depth_layout_less
,
137 nir_depth_layout_unchanged
141 * Either a uniform, global variable, shader input, or shader output. Based on
142 * ir_variable - it should be easy to translate between the two.
145 typedef struct nir_variable
{
146 struct exec_node node
;
149 * Declared type of the variable
151 const struct glsl_type
*type
;
154 * Declared name of the variable
158 struct nir_variable_data
{
161 * Is the variable read-only?
163 * This is set for variables declared as \c const, shader inputs,
166 unsigned read_only
:1;
170 unsigned invariant
:1;
173 * Storage class of the variable.
175 * \sa nir_variable_mode
177 nir_variable_mode mode
:5;
180 * Interpolation mode for shader inputs / outputs
182 * \sa glsl_interp_qualifier
184 unsigned interpolation
:2;
187 * \name ARB_fragment_coord_conventions
190 unsigned origin_upper_left
:1;
191 unsigned pixel_center_integer
:1;
195 * Was the location explicitly set in the shader?
197 * If the location is explicitly set in the shader, it \b cannot be changed
198 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
201 unsigned explicit_location
:1;
202 unsigned explicit_index
:1;
205 * Was an initial binding explicitly set in the shader?
207 * If so, constant_initializer contains an integer nir_constant
208 * representing the initial binding point.
210 unsigned explicit_binding
:1;
213 * Does this variable have an initializer?
215 * This is used by the linker to cross-validiate initializers of global
218 unsigned has_initializer
:1;
221 * If non-zero, then this variable may be packed along with other variables
222 * into a single varying slot, so this offset should be applied when
223 * accessing components. For example, an offset of 1 means that the x
224 * component of this variable is actually stored in component y of the
225 * location specified by \c location.
227 unsigned location_frac
:2;
230 * \brief Layout qualifier for gl_FragDepth.
232 * This is not equal to \c ir_depth_layout_none if and only if this
233 * variable is \c gl_FragDepth and a layout qualifier is specified.
235 nir_depth_layout depth_layout
;
238 * Storage location of the base of this variable
240 * The precise meaning of this field depends on the nature of the variable.
242 * - Vertex shader input: one of the values from \c gl_vert_attrib.
243 * - Vertex shader output: one of the values from \c gl_varying_slot.
244 * - Geometry shader input: one of the values from \c gl_varying_slot.
245 * - Geometry shader output: one of the values from \c gl_varying_slot.
246 * - Fragment shader input: one of the values from \c gl_varying_slot.
247 * - Fragment shader output: one of the values from \c gl_frag_result.
248 * - Uniforms: Per-stage uniform slot number for default uniform block.
249 * - Uniforms: Index within the uniform block definition for UBO members.
250 * - Non-UBO Uniforms: uniform slot number.
251 * - Other: This field is not currently used.
253 * If the variable is a uniform, shader input, or shader output, and the
254 * slot has not been assigned, the value will be -1.
259 * The actual location of the variable in the IR. Only valid for inputs
262 unsigned int driver_location
;
265 * output index for dual source blending.
270 * Initial binding point for a sampler or UBO.
272 * For array types, this represents the binding point for the first element.
277 * Location an atomic counter is stored at.
282 * ARB_shader_image_load_store qualifiers.
285 bool read_only
; /**< "readonly" qualifier. */
286 bool write_only
; /**< "writeonly" qualifier. */
291 /** Image internal format if specified explicitly, otherwise GL_NONE. */
296 * Highest element accessed with a constant expression array index
298 * Not used for non-array variables.
300 unsigned max_array_access
;
305 * Built-in state that backs this uniform
307 * Once set at variable creation, \c state_slots must remain invariant.
308 * This is because, ideally, this array would be shared by all clones of
309 * this variable in the IR tree. In other words, we'd really like for it
310 * to be a fly-weight.
312 * If the variable is not a uniform, \c num_state_slots will be zero and
313 * \c state_slots will be \c NULL.
316 unsigned num_state_slots
; /**< Number of state slots used */
317 nir_state_slot
*state_slots
; /**< State descriptors. */
321 * Constant expression assigned in the initializer of the variable
323 nir_constant
*constant_initializer
;
326 * For variables that are in an interface block or are an instance of an
327 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
329 * \sa ir_variable::location
331 const struct glsl_type
*interface_type
;
334 #define nir_foreach_variable(var, var_list) \
335 foreach_list_typed(nir_variable, var, node, var_list)
338 nir_variable_is_global(const nir_variable
*var
)
340 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
343 typedef struct nir_register
{
344 struct exec_node node
;
346 unsigned num_components
; /** < number of vector components */
347 unsigned num_array_elems
; /** < size of array (0 for no array) */
349 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
352 /** generic register index. */
355 /** only for debug purposes, can be NULL */
358 /** whether this register is local (per-function) or global (per-shader) */
362 * If this flag is set to true, then accessing channels >= num_components
363 * is well-defined, and simply spills over to the next array element. This
364 * is useful for backends that can do per-component accessing, in
365 * particular scalar backends. By setting this flag and making
366 * num_components equal to 1, structures can be packed tightly into
367 * registers and then registers can be accessed per-component to get to
368 * each structure member, even if it crosses vec4 boundaries.
372 /** set of nir_src's where this register is used (read from) */
373 struct list_head uses
;
375 /** set of nir_dest's where this register is defined (written to) */
376 struct list_head defs
;
378 /** set of nir_if's where this register is used as a condition */
379 struct list_head if_uses
;
386 nir_instr_type_intrinsic
,
387 nir_instr_type_load_const
,
389 nir_instr_type_ssa_undef
,
391 nir_instr_type_parallel_copy
,
394 typedef struct nir_instr
{
395 struct exec_node node
;
397 struct nir_block
*block
;
399 /** generic instruction index. */
402 /* A temporary for optimization and analysis passes to use for storing
403 * flags. For instance, DCE uses this to store the "dead/live" info.
408 static inline nir_instr
*
409 nir_instr_next(nir_instr
*instr
)
411 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
412 if (exec_node_is_tail_sentinel(next
))
415 return exec_node_data(nir_instr
, next
, node
);
418 static inline nir_instr
*
419 nir_instr_prev(nir_instr
*instr
)
421 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
422 if (exec_node_is_head_sentinel(prev
))
425 return exec_node_data(nir_instr
, prev
, node
);
429 nir_instr_is_first(nir_instr
*instr
)
431 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
435 nir_instr_is_last(nir_instr
*instr
)
437 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
440 typedef struct nir_ssa_def
{
441 /** for debugging only, can be NULL */
444 /** generic SSA definition index. */
447 /** Index into the live_in and live_out bitfields */
450 nir_instr
*parent_instr
;
452 /** set of nir_instr's where this register is used (read from) */
453 struct list_head uses
;
455 /** set of nir_if's where this register is used as a condition */
456 struct list_head if_uses
;
458 uint8_t num_components
;
460 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
468 struct nir_src
*indirect
; /** < NULL for no indirect offset */
469 unsigned base_offset
;
471 /* TODO use-def chain goes here */
475 nir_instr
*parent_instr
;
476 struct list_head def_link
;
479 struct nir_src
*indirect
; /** < NULL for no indirect offset */
480 unsigned base_offset
;
482 /* TODO def-use chain goes here */
487 typedef struct nir_src
{
489 nir_instr
*parent_instr
;
490 struct nir_if
*parent_if
;
493 struct list_head use_link
;
503 #define NIR_SRC_INIT (nir_src) { { NULL } }
505 #define nir_foreach_use(reg_or_ssa_def, src) \
506 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
508 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
509 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
511 #define nir_foreach_if_use(reg_or_ssa_def, src) \
512 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
514 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
515 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
526 #define NIR_DEST_INIT (nir_dest) { { { NULL } } }
528 #define nir_foreach_def(reg, dest) \
529 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
531 #define nir_foreach_def_safe(reg, dest) \
532 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
534 static inline nir_src
535 nir_src_for_ssa(nir_ssa_def
*def
)
537 nir_src src
= NIR_SRC_INIT
;
545 static inline nir_src
546 nir_src_for_reg(nir_register
*reg
)
548 nir_src src
= NIR_SRC_INIT
;
552 src
.reg
.indirect
= NULL
;
553 src
.reg
.base_offset
= 0;
558 static inline nir_dest
559 nir_dest_for_reg(nir_register
*reg
)
561 nir_dest dest
= NIR_DEST_INIT
;
568 static inline unsigned
569 nir_src_bit_size(nir_src src
)
571 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
574 static inline unsigned
575 nir_dest_bit_size(nir_dest dest
)
577 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
580 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
581 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
587 * \name input modifiers
591 * For inputs interpreted as floating point, flips the sign bit. For
592 * inputs interpreted as integers, performs the two's complement negation.
597 * Clears the sign bit for floating point values, and computes the integer
598 * absolute value for integers. Note that the negate modifier acts after
599 * the absolute value modifier, therefore if both are set then all inputs
600 * will become negative.
606 * For each input component, says which component of the register it is
607 * chosen from. Note that which elements of the swizzle are used and which
608 * are ignored are based on the write mask for most opcodes - for example,
609 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
610 * a swizzle of {2, x, 1, 0} where x means "don't care."
619 * \name saturate output modifier
621 * Only valid for opcodes that output floating-point numbers. Clamps the
622 * output to between 0.0 and 1.0 inclusive.
627 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
631 nir_type_invalid
= 0, /* Not a valid type */
636 nir_type_bool32
= 32 | nir_type_bool
,
637 nir_type_int8
= 8 | nir_type_int
,
638 nir_type_int16
= 16 | nir_type_int
,
639 nir_type_int32
= 32 | nir_type_int
,
640 nir_type_int64
= 64 | nir_type_int
,
641 nir_type_uint8
= 8 | nir_type_uint
,
642 nir_type_uint16
= 16 | nir_type_uint
,
643 nir_type_uint32
= 32 | nir_type_uint
,
644 nir_type_uint64
= 64 | nir_type_uint
,
645 nir_type_float16
= 16 | nir_type_float
,
646 nir_type_float32
= 32 | nir_type_float
,
647 nir_type_float64
= 64 | nir_type_float
,
650 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
651 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
653 static inline unsigned
654 nir_alu_type_get_type_size(nir_alu_type type
)
656 return type
& NIR_ALU_TYPE_SIZE_MASK
;
659 static inline unsigned
660 nir_alu_type_get_base_type(nir_alu_type type
)
662 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
666 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
667 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
668 } nir_op_algebraic_property
;
676 * The number of components in the output
678 * If non-zero, this is the size of the output and input sizes are
679 * explicitly given; swizzle and writemask are still in effect, but if
680 * the output component is masked out, then the input component may
683 * If zero, the opcode acts in the standard, per-component manner; the
684 * operation is performed on each component (except the ones that are
685 * masked out) with the input being taken from the input swizzle for
688 * The size of some of the inputs may be given (i.e. non-zero) even
689 * though output_size is zero; in that case, the inputs with a zero
690 * size act per-component, while the inputs with non-zero size don't.
692 unsigned output_size
;
695 * The type of vector that the instruction outputs. Note that the
696 * staurate modifier is only allowed on outputs with the float type.
699 nir_alu_type output_type
;
702 * The number of components in each input
704 unsigned input_sizes
[4];
707 * The type of vector that each input takes. Note that negate and
708 * absolute value are only allowed on inputs with int or float type and
709 * behave differently on the two.
711 nir_alu_type input_types
[4];
713 nir_op_algebraic_property algebraic_properties
;
716 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
718 typedef struct nir_alu_instr
{
725 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
726 nir_alu_instr
*instr
);
727 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
728 nir_alu_instr
*instr
);
730 /* is this source channel used? */
732 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
734 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
735 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
737 return (instr
->dest
.write_mask
>> channel
) & 1;
741 * For instructions whose destinations are SSA, get the number of channels
744 static inline unsigned
745 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
747 assert(instr
->dest
.dest
.is_ssa
);
749 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
750 return nir_op_infos
[instr
->op
].input_sizes
[src
];
752 return instr
->dest
.dest
.ssa
.num_components
;
757 nir_deref_type_array
,
758 nir_deref_type_struct
761 typedef struct nir_deref
{
762 nir_deref_type deref_type
;
763 struct nir_deref
*child
;
764 const struct glsl_type
*type
;
773 /* This enum describes how the array is referenced. If the deref is
774 * direct then the base_offset is used. If the deref is indirect then then
775 * offset is given by base_offset + indirect. If the deref is a wildcard
776 * then the deref refers to all of the elements of the array at the same
777 * time. Wildcard dereferences are only ever allowed in copy_var
778 * intrinsics and the source and destination derefs must have matching
782 nir_deref_array_type_direct
,
783 nir_deref_array_type_indirect
,
784 nir_deref_array_type_wildcard
,
785 } nir_deref_array_type
;
790 nir_deref_array_type deref_array_type
;
791 unsigned base_offset
;
801 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
802 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
803 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
805 /* Returns the last deref in the chain. */
806 static inline nir_deref
*
807 nir_deref_tail(nir_deref
*deref
)
810 deref
= deref
->child
;
818 nir_deref_var
**params
;
819 nir_deref_var
*return_deref
;
821 struct nir_function
*callee
;
824 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
825 num_variables, num_indices, idx0, idx1, idx2, flags) \
826 nir_intrinsic_##name,
828 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
831 #include "nir_intrinsics.h"
832 nir_num_intrinsics
= nir_last_intrinsic
+ 1
836 #undef LAST_INTRINSIC
838 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
840 /** Represents an intrinsic
842 * An intrinsic is an instruction type for handling things that are
843 * more-or-less regular operations but don't just consume and produce SSA
844 * values like ALU operations do. Intrinsics are not for things that have
845 * special semantic meaning such as phi nodes and parallel copies.
846 * Examples of intrinsics include variable load/store operations, system
847 * value loads, and the like. Even though texturing more-or-less falls
848 * under this category, texturing is its own instruction type because
849 * trying to represent texturing with intrinsics would lead to a
850 * combinatorial explosion of intrinsic opcodes.
852 * By having a single instruction type for handling a lot of different
853 * cases, optimization passes can look for intrinsics and, for the most
854 * part, completely ignore them. Each intrinsic type also has a few
855 * possible flags that govern whether or not they can be reordered or
856 * eliminated. That way passes like dead code elimination can still work
857 * on intrisics without understanding the meaning of each.
859 * Each intrinsic has some number of constant indices, some number of
860 * variables, and some number of sources. What these sources, variables,
861 * and indices mean depends on the intrinsic and is documented with the
862 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
863 * instructions are the only types of instruction that can operate on
869 nir_intrinsic_op intrinsic
;
873 /** number of components if this is a vectorized intrinsic
875 * Similarly to ALU operations, some intrinsics are vectorized.
876 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
877 * For vectorized intrinsics, the num_components field specifies the
878 * number of destination components and the number of source components
879 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
881 uint8_t num_components
;
883 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
885 nir_deref_var
*variables
[2];
888 } nir_intrinsic_instr
;
891 * \name NIR intrinsics semantic flags
893 * information about what the compiler can do with the intrinsics.
895 * \sa nir_intrinsic_info::flags
899 * whether the intrinsic can be safely eliminated if none of its output
900 * value is not being used.
902 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
905 * Whether the intrinsic can be reordered with respect to any other
906 * intrinsic, i.e. whether the only reordering dependencies of the
907 * intrinsic are due to the register reads/writes.
909 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
910 } nir_intrinsic_semantic_flag
;
913 * \name NIR intrinsics const-index flag
915 * Indicates the usage of a const_index slot.
917 * \sa nir_intrinsic_info::index_map
921 * Generally instructions that take a offset src argument, can encode
922 * a constant 'base' value which is added to the offset.
924 NIR_INTRINSIC_BASE
= 1,
927 * For store instructions, a writemask for the store.
929 NIR_INTRINSIC_WRMASK
= 2,
932 * The stream-id for GS emit_vertex/end_primitive intrinsics.
934 NIR_INTRINSIC_STREAM_ID
= 3,
937 * The clip-plane id for load_user_clip_plane intrinsic.
939 NIR_INTRINSIC_UCP_ID
= 4,
941 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
943 } nir_intrinsic_index_flag
;
945 #define NIR_INTRINSIC_MAX_INPUTS 4
950 unsigned num_srcs
; /** < number of register/SSA inputs */
952 /** number of components of each input register
954 * If this value is 0, the number of components is given by the
955 * num_components field of nir_intrinsic_instr.
957 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
961 /** number of components of the output register
963 * If this value is 0, the number of components is given by the
964 * num_components field of nir_intrinsic_instr.
966 unsigned dest_components
;
968 /** the number of inputs/outputs that are variables */
969 unsigned num_variables
;
971 /** the number of constant indices used by the intrinsic */
972 unsigned num_indices
;
974 /** indicates the usage of intr->const_index[n] */
975 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
977 /** semantic flags for calls to this intrinsic */
978 nir_intrinsic_semantic_flag flags
;
979 } nir_intrinsic_info
;
981 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
984 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
986 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
988 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
989 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
990 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
993 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
995 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
996 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
997 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1000 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1001 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1002 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1003 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1006 * \group texture information
1008 * This gives semantic information about textures which is useful to the
1009 * frontend, the backend, and lowering passes, but not the optimizer.
1014 nir_tex_src_projector
,
1015 nir_tex_src_comparitor
, /* shadow comparitor */
1019 nir_tex_src_ms_index
, /* MSAA sample index */
1022 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1023 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1024 nir_num_tex_src_types
1029 nir_tex_src_type src_type
;
1033 nir_texop_tex
, /**< Regular texture look-up */
1034 nir_texop_txb
, /**< Texture look-up with LOD bias */
1035 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1036 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1037 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1038 nir_texop_txf_ms
, /**< Multisample texture fetch */
1039 nir_texop_txs
, /**< Texture size */
1040 nir_texop_lod
, /**< Texture lod query */
1041 nir_texop_tg4
, /**< Texture gather */
1042 nir_texop_query_levels
, /**< Texture levels query */
1043 nir_texop_texture_samples
, /**< Texture samples query */
1044 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1052 enum glsl_sampler_dim sampler_dim
;
1053 nir_alu_type dest_type
;
1058 unsigned num_srcs
, coord_components
;
1059 bool is_array
, is_shadow
;
1062 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1063 * components or the new-style shadow that outputs 1 component.
1065 bool is_new_style_shadow
;
1067 /* gather component selector */
1068 unsigned component
: 2;
1070 /** The texture index
1072 * If this texture instruction has a nir_tex_src_texture_offset source,
1073 * then the texture index is given by texture_index + texture_offset.
1075 unsigned texture_index
;
1077 /** The size of the texture array or 0 if it's not an array */
1078 unsigned texture_array_size
;
1080 /** The texture deref
1082 * If this is null, use texture_index instead.
1084 nir_deref_var
*texture
;
1086 /** The sampler index
1088 * The following operations do not require a sampler and, as such, this
1089 * field should be ignored:
1091 * - nir_texop_txf_ms
1095 * - nir_texop_query_levels
1096 * - nir_texop_texture_samples
1097 * - nir_texop_samples_identical
1099 * If this texture instruction has a nir_tex_src_sampler_offset source,
1100 * then the sampler index is given by sampler_index + sampler_offset.
1102 unsigned sampler_index
;
1104 /** The sampler deref
1106 * If this is null, use sampler_index instead.
1108 nir_deref_var
*sampler
;
1111 static inline unsigned
1112 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1114 switch (instr
->op
) {
1115 case nir_texop_txs
: {
1117 switch (instr
->sampler_dim
) {
1118 case GLSL_SAMPLER_DIM_1D
:
1119 case GLSL_SAMPLER_DIM_BUF
:
1122 case GLSL_SAMPLER_DIM_2D
:
1123 case GLSL_SAMPLER_DIM_CUBE
:
1124 case GLSL_SAMPLER_DIM_MS
:
1125 case GLSL_SAMPLER_DIM_RECT
:
1126 case GLSL_SAMPLER_DIM_EXTERNAL
:
1129 case GLSL_SAMPLER_DIM_3D
:
1133 unreachable("not reached");
1135 if (instr
->is_array
)
1143 case nir_texop_texture_samples
:
1144 case nir_texop_query_levels
:
1145 case nir_texop_samples_identical
:
1149 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1156 /* Returns true if this texture operation queries something about the texture
1157 * rather than actually sampling it.
1160 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1162 switch (instr
->op
) {
1165 case nir_texop_texture_samples
:
1166 case nir_texop_query_levels
:
1173 case nir_texop_txf_ms
:
1177 unreachable("Invalid texture opcode");
1181 static inline unsigned
1182 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1184 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1185 return instr
->coord_components
;
1188 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1189 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1190 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1191 if (instr
->is_array
)
1192 return instr
->coord_components
- 1;
1194 return instr
->coord_components
;
1201 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1203 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1204 if (instr
->src
[i
].src_type
== type
)
1224 nir_const_value value
;
1227 } nir_load_const_instr
;
1240 /* creates a new SSA variable in an undefined state */
1245 } nir_ssa_undef_instr
;
1248 struct exec_node node
;
1250 /* The predecessor block corresponding to this source */
1251 struct nir_block
*pred
;
1256 #define nir_foreach_phi_src(phi, entry) \
1257 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1258 #define nir_foreach_phi_src_safe(phi, entry) \
1259 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1264 struct exec_list srcs
; /** < list of nir_phi_src */
1270 struct exec_node node
;
1273 } nir_parallel_copy_entry
;
1275 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1276 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1281 /* A list of nir_parallel_copy_entry's. The sources of all of the
1282 * entries are copied to the corresponding destinations "in parallel".
1283 * In other words, if we have two entries: a -> b and b -> a, the values
1286 struct exec_list entries
;
1287 } nir_parallel_copy_instr
;
1289 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1290 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1291 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1292 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1293 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1294 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1295 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1296 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1297 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1298 nir_parallel_copy_instr
, instr
)
1303 * Control flow consists of a tree of control flow nodes, which include
1304 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1305 * instructions that always run start-to-finish. Each basic block also keeps
1306 * track of its successors (blocks which may run immediately after the current
1307 * block) and predecessors (blocks which could have run immediately before the
1308 * current block). Each function also has a start block and an end block which
1309 * all return statements point to (which is always empty). Together, all the
1310 * blocks with their predecessors and successors make up the control flow
1311 * graph (CFG) of the function. There are helpers that modify the tree of
1312 * control flow nodes while modifying the CFG appropriately; these should be
1313 * used instead of modifying the tree directly.
1320 nir_cf_node_function
1323 typedef struct nir_cf_node
{
1324 struct exec_node node
;
1325 nir_cf_node_type type
;
1326 struct nir_cf_node
*parent
;
1329 typedef struct nir_block
{
1330 nir_cf_node cf_node
;
1332 struct exec_list instr_list
; /** < list of nir_instr */
1334 /** generic block index; generated by nir_index_blocks */
1338 * Each block can only have up to 2 successors, so we put them in a simple
1339 * array - no need for anything more complicated.
1341 struct nir_block
*successors
[2];
1343 /* Set of nir_block predecessors in the CFG */
1344 struct set
*predecessors
;
1347 * this node's immediate dominator in the dominance tree - set to NULL for
1350 struct nir_block
*imm_dom
;
1352 /* This node's children in the dominance tree */
1353 unsigned num_dom_children
;
1354 struct nir_block
**dom_children
;
1356 /* Set of nir_block's on the dominance frontier of this block */
1357 struct set
*dom_frontier
;
1360 * These two indices have the property that dom_{pre,post}_index for each
1361 * child of this block in the dominance tree will always be between
1362 * dom_pre_index and dom_post_index for this block, which makes testing if
1363 * a given block is dominated by another block an O(1) operation.
1365 unsigned dom_pre_index
, dom_post_index
;
1367 /* live in and out for this block; used for liveness analysis */
1368 BITSET_WORD
*live_in
;
1369 BITSET_WORD
*live_out
;
1372 static inline nir_instr
*
1373 nir_block_first_instr(nir_block
*block
)
1375 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1376 return exec_node_data(nir_instr
, head
, node
);
1379 static inline nir_instr
*
1380 nir_block_last_instr(nir_block
*block
)
1382 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1383 return exec_node_data(nir_instr
, tail
, node
);
1386 #define nir_foreach_instr(block, instr) \
1387 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1388 #define nir_foreach_instr_reverse(block, instr) \
1389 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1390 #define nir_foreach_instr_safe(block, instr) \
1391 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1392 #define nir_foreach_instr_reverse_safe(block, instr) \
1393 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1395 typedef struct nir_if
{
1396 nir_cf_node cf_node
;
1399 struct exec_list then_list
; /** < list of nir_cf_node */
1400 struct exec_list else_list
; /** < list of nir_cf_node */
1403 static inline nir_cf_node
*
1404 nir_if_first_then_node(nir_if
*if_stmt
)
1406 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1407 return exec_node_data(nir_cf_node
, head
, node
);
1410 static inline nir_cf_node
*
1411 nir_if_last_then_node(nir_if
*if_stmt
)
1413 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1414 return exec_node_data(nir_cf_node
, tail
, node
);
1417 static inline nir_cf_node
*
1418 nir_if_first_else_node(nir_if
*if_stmt
)
1420 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1421 return exec_node_data(nir_cf_node
, head
, node
);
1424 static inline nir_cf_node
*
1425 nir_if_last_else_node(nir_if
*if_stmt
)
1427 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1428 return exec_node_data(nir_cf_node
, tail
, node
);
1432 nir_cf_node cf_node
;
1434 struct exec_list body
; /** < list of nir_cf_node */
1437 static inline nir_cf_node
*
1438 nir_loop_first_cf_node(nir_loop
*loop
)
1440 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1443 static inline nir_cf_node
*
1444 nir_loop_last_cf_node(nir_loop
*loop
)
1446 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1450 * Various bits of metadata that can may be created or required by
1451 * optimization and analysis passes
1454 nir_metadata_none
= 0x0,
1455 nir_metadata_block_index
= 0x1,
1456 nir_metadata_dominance
= 0x2,
1457 nir_metadata_live_ssa_defs
= 0x4,
1458 nir_metadata_not_properly_reset
= 0x8,
1462 nir_cf_node cf_node
;
1464 /** pointer to the function of which this is an implementation */
1465 struct nir_function
*function
;
1467 struct exec_list body
; /** < list of nir_cf_node */
1469 nir_block
*end_block
;
1471 /** list for all local variables in the function */
1472 struct exec_list locals
;
1474 /** array of variables used as parameters */
1475 unsigned num_params
;
1476 nir_variable
**params
;
1478 /** variable used to hold the result of the function */
1479 nir_variable
*return_var
;
1481 /** list of local registers in the function */
1482 struct exec_list registers
;
1484 /** next available local register index */
1487 /** next available SSA value index */
1490 /* total number of basic blocks, only valid when block_index_dirty = false */
1491 unsigned num_blocks
;
1493 nir_metadata valid_metadata
;
1494 } nir_function_impl
;
1496 static inline nir_block
*
1497 nir_start_block(nir_function_impl
*impl
)
1499 return (nir_block
*) exec_list_get_head(&impl
->body
);
1502 static inline nir_cf_node
*
1503 nir_cf_node_next(nir_cf_node
*node
)
1505 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1506 if (exec_node_is_tail_sentinel(next
))
1509 return exec_node_data(nir_cf_node
, next
, node
);
1512 static inline nir_cf_node
*
1513 nir_cf_node_prev(nir_cf_node
*node
)
1515 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1516 if (exec_node_is_head_sentinel(prev
))
1519 return exec_node_data(nir_cf_node
, prev
, node
);
1523 nir_cf_node_is_first(const nir_cf_node
*node
)
1525 return exec_node_is_head_sentinel(node
->node
.prev
);
1529 nir_cf_node_is_last(const nir_cf_node
*node
)
1531 return exec_node_is_tail_sentinel(node
->node
.next
);
1534 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1535 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1536 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1537 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1542 nir_parameter_inout
,
1543 } nir_parameter_type
;
1546 nir_parameter_type param_type
;
1547 const struct glsl_type
*type
;
1550 typedef struct nir_function
{
1551 struct exec_node node
;
1554 struct nir_shader
*shader
;
1556 unsigned num_params
;
1557 nir_parameter
*params
;
1558 const struct glsl_type
*return_type
;
1560 /** The implementation of this function.
1562 * If the function is only declared and not implemented, this is NULL.
1564 nir_function_impl
*impl
;
1567 typedef struct nir_shader_compiler_options
{
1575 bool lower_bitfield_extract
;
1576 bool lower_bitfield_insert
;
1577 bool lower_uadd_carry
;
1578 bool lower_usub_borrow
;
1579 /** lowers fneg and ineg to fsub and isub. */
1581 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1584 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1587 /* Does the native fdot instruction replicate its result for four
1588 * components? If so, then opt_algebraic_late will turn all fdotN
1589 * instructions into fdot_replicatedN instructions.
1591 bool fdot_replicates
;
1593 /** lowers ffract to fsub+ffloor: */
1596 bool lower_pack_half_2x16
;
1597 bool lower_pack_unorm_2x16
;
1598 bool lower_pack_snorm_2x16
;
1599 bool lower_pack_unorm_4x8
;
1600 bool lower_pack_snorm_4x8
;
1601 bool lower_unpack_half_2x16
;
1602 bool lower_unpack_unorm_2x16
;
1603 bool lower_unpack_snorm_2x16
;
1604 bool lower_unpack_unorm_4x8
;
1605 bool lower_unpack_snorm_4x8
;
1607 bool lower_extract_byte
;
1608 bool lower_extract_word
;
1611 * Does the driver support real 32-bit integers? (Otherwise, integers
1612 * are simulated by floats.)
1614 bool native_integers
;
1615 } nir_shader_compiler_options
;
1617 typedef struct nir_shader_info
{
1620 /* Descriptive name provided by the client; may be NULL */
1623 /* Number of textures used by this shader */
1624 unsigned num_textures
;
1625 /* Number of uniform buffers used by this shader */
1627 /* Number of atomic buffers used by this shader */
1629 /* Number of shader storage buffers used by this shader */
1631 /* Number of images used by this shader */
1632 unsigned num_images
;
1634 /* Which inputs are actually read */
1635 uint64_t inputs_read
;
1636 /* Which outputs are actually written */
1637 uint64_t outputs_written
;
1638 /* Which system values are actually read */
1639 uint64_t system_values_read
;
1641 /* Which patch inputs are actually read */
1642 uint32_t patch_inputs_read
;
1643 /* Which patch outputs are actually written */
1644 uint32_t patch_outputs_written
;
1646 /* Whether or not this shader ever uses textureGather() */
1647 bool uses_texture_gather
;
1649 /* Whether or not this shader uses the gl_ClipDistance output */
1650 bool uses_clip_distance_out
;
1652 /* Whether or not separate shader objects were used */
1653 bool separate_shader
;
1655 /** Was this shader linked with any transform feedback varyings? */
1656 bool has_transform_feedback_varyings
;
1660 /** The number of vertices recieves per input primitive */
1661 unsigned vertices_in
;
1663 /** The output primitive type (GL enum value) */
1664 unsigned output_primitive
;
1666 /** The maximum number of vertices the geometry shader might write. */
1667 unsigned vertices_out
;
1669 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1670 unsigned invocations
;
1672 /** Whether or not this shader uses EndPrimitive */
1673 bool uses_end_primitive
;
1675 /** Whether or not this shader uses non-zero streams */
1683 * Whether early fragment tests are enabled as defined by
1684 * ARB_shader_image_load_store.
1686 bool early_fragment_tests
;
1688 /** gl_FragDepth layout for ARB_conservative_depth. */
1689 enum gl_frag_depth_layout depth_layout
;
1693 unsigned local_size
[3];
1697 /** The number of vertices in the TCS output patch. */
1698 unsigned vertices_out
;
1703 typedef struct nir_shader
{
1704 /** list of uniforms (nir_variable) */
1705 struct exec_list uniforms
;
1707 /** list of inputs (nir_variable) */
1708 struct exec_list inputs
;
1710 /** list of outputs (nir_variable) */
1711 struct exec_list outputs
;
1713 /** list of shared compute variables (nir_variable) */
1714 struct exec_list shared
;
1716 /** Set of driver-specific options for the shader.
1718 * The memory for the options is expected to be kept in a single static
1719 * copy by the driver.
1721 const struct nir_shader_compiler_options
*options
;
1723 /** Various bits of compile-time information about a given shader */
1724 struct nir_shader_info info
;
1726 /** list of global variables in the shader (nir_variable) */
1727 struct exec_list globals
;
1729 /** list of system value variables in the shader (nir_variable) */
1730 struct exec_list system_values
;
1732 struct exec_list functions
; /** < list of nir_function */
1734 /** list of global register in the shader */
1735 struct exec_list registers
;
1737 /** next available global register index */
1741 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1744 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1746 /** The shader stage, such as MESA_SHADER_VERTEX. */
1747 gl_shader_stage stage
;
1750 #define nir_foreach_function(shader, func) \
1751 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1753 nir_shader
*nir_shader_create(void *mem_ctx
,
1754 gl_shader_stage stage
,
1755 const nir_shader_compiler_options
*options
);
1757 /** creates a register, including assigning it an index and adding it to the list */
1758 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1760 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1762 void nir_reg_remove(nir_register
*reg
);
1764 /** Adds a variable to the appropreate list in nir_shader */
1765 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1768 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1770 assert(var
->data
.mode
== nir_var_local
);
1771 exec_list_push_tail(&impl
->locals
, &var
->node
);
1774 /** creates a variable, sets a few defaults, and adds it to the list */
1775 nir_variable
*nir_variable_create(nir_shader
*shader
,
1776 nir_variable_mode mode
,
1777 const struct glsl_type
*type
,
1779 /** creates a local variable and adds it to the list */
1780 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1781 const struct glsl_type
*type
,
1784 /** creates a function and adds it to the shader's list of functions */
1785 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1787 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1788 /** creates a function_impl that isn't tied to any particular function */
1789 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1791 nir_block
*nir_block_create(nir_shader
*shader
);
1792 nir_if
*nir_if_create(nir_shader
*shader
);
1793 nir_loop
*nir_loop_create(nir_shader
*shader
);
1795 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1797 /** requests that the given pieces of metadata be generated */
1798 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1799 /** dirties all but the preserved metadata */
1800 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1802 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1803 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1805 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1807 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1808 unsigned num_components
);
1810 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1811 nir_intrinsic_op op
);
1813 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1814 nir_function
*callee
);
1816 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1818 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1820 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1822 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1823 unsigned num_components
);
1825 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1826 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1827 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1829 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1831 nir_load_const_instr
*
1832 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1835 * NIR Cursors and Instruction Insertion API
1838 * A tiny struct representing a point to insert/extract instructions or
1839 * control flow nodes. Helps reduce the combinatorial explosion of possible
1840 * points to insert/extract.
1842 * \sa nir_control_flow.h
1845 nir_cursor_before_block
,
1846 nir_cursor_after_block
,
1847 nir_cursor_before_instr
,
1848 nir_cursor_after_instr
,
1849 } nir_cursor_option
;
1852 nir_cursor_option option
;
1859 static inline nir_cursor
1860 nir_before_block(nir_block
*block
)
1863 cursor
.option
= nir_cursor_before_block
;
1864 cursor
.block
= block
;
1868 static inline nir_cursor
1869 nir_after_block(nir_block
*block
)
1872 cursor
.option
= nir_cursor_after_block
;
1873 cursor
.block
= block
;
1877 static inline nir_cursor
1878 nir_before_instr(nir_instr
*instr
)
1881 cursor
.option
= nir_cursor_before_instr
;
1882 cursor
.instr
= instr
;
1886 static inline nir_cursor
1887 nir_after_instr(nir_instr
*instr
)
1890 cursor
.option
= nir_cursor_after_instr
;
1891 cursor
.instr
= instr
;
1895 static inline nir_cursor
1896 nir_after_block_before_jump(nir_block
*block
)
1898 nir_instr
*last_instr
= nir_block_last_instr(block
);
1899 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1900 return nir_before_instr(last_instr
);
1902 return nir_after_block(block
);
1906 static inline nir_cursor
1907 nir_before_cf_node(nir_cf_node
*node
)
1909 if (node
->type
== nir_cf_node_block
)
1910 return nir_before_block(nir_cf_node_as_block(node
));
1912 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1915 static inline nir_cursor
1916 nir_after_cf_node(nir_cf_node
*node
)
1918 if (node
->type
== nir_cf_node_block
)
1919 return nir_after_block(nir_cf_node_as_block(node
));
1921 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1924 static inline nir_cursor
1925 nir_before_cf_list(struct exec_list
*cf_list
)
1927 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1928 exec_list_get_head(cf_list
), node
);
1929 return nir_before_cf_node(first_node
);
1932 static inline nir_cursor
1933 nir_after_cf_list(struct exec_list
*cf_list
)
1935 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1936 exec_list_get_tail(cf_list
), node
);
1937 return nir_after_cf_node(last_node
);
1941 * Insert a NIR instruction at the given cursor.
1943 * Note: This does not update the cursor.
1945 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
1948 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1950 nir_instr_insert(nir_before_instr(instr
), before
);
1954 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
1956 nir_instr_insert(nir_after_instr(instr
), after
);
1960 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
1962 nir_instr_insert(nir_before_block(block
), before
);
1966 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
1968 nir_instr_insert(nir_after_block(block
), after
);
1972 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
1974 nir_instr_insert(nir_before_cf_node(node
), before
);
1978 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
1980 nir_instr_insert(nir_after_cf_node(node
), after
);
1984 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
1986 nir_instr_insert(nir_before_cf_list(list
), before
);
1990 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
1992 nir_instr_insert(nir_after_cf_list(list
), after
);
1995 void nir_instr_remove(nir_instr
*instr
);
1999 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2000 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2001 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2002 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2004 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2005 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2007 nir_const_value
*nir_src_as_const_value(nir_src src
);
2008 bool nir_src_is_dynamically_uniform(nir_src src
);
2009 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2010 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2011 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2012 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2013 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2016 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2017 unsigned num_components
, unsigned bit_size
,
2019 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2020 unsigned num_components
, unsigned bit_size
,
2022 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2023 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2024 nir_instr
*after_me
);
2026 /* visits basic blocks in source-code order */
2027 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
2028 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
2030 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
2032 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
2035 /* If the following CF node is an if, this function returns that if.
2036 * Otherwise, it returns NULL.
2038 nir_if
*nir_block_get_following_if(nir_block
*block
);
2040 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2042 void nir_index_local_regs(nir_function_impl
*impl
);
2043 void nir_index_global_regs(nir_shader
*shader
);
2044 void nir_index_ssa_defs(nir_function_impl
*impl
);
2045 unsigned nir_index_instrs(nir_function_impl
*impl
);
2047 void nir_index_blocks(nir_function_impl
*impl
);
2049 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2050 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2052 nir_shader
* nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2053 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2056 void nir_validate_shader(nir_shader
*shader
);
2057 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2058 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2060 #include "util/debug.h"
2062 should_clone_nir(void)
2064 static int should_clone
= -1;
2065 if (should_clone
< 0)
2066 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2068 return should_clone
;
2071 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2072 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2073 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2074 static inline bool should_clone_nir(void) { return false; }
2077 #define _PASS(nir, do_pass) do { \
2079 nir_validate_shader(nir); \
2080 if (should_clone_nir()) { \
2081 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2087 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2088 nir_metadata_set_validation_flag(nir); \
2089 if (pass(nir, ##__VA_ARGS__)) { \
2091 nir_metadata_check_validation_flag(nir); \
2095 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2096 pass(nir, ##__VA_ARGS__); \
2099 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2100 void nir_calc_dominance(nir_shader
*shader
);
2102 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2103 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2105 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2106 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2108 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2109 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2111 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2112 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2114 int nir_gs_count_vertices(const nir_shader
*shader
);
2116 bool nir_split_var_copies(nir_shader
*shader
);
2118 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2119 void nir_lower_var_copies(nir_shader
*shader
);
2121 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2123 bool nir_lower_indirect_derefs(nir_shader
*shader
, uint32_t mode_mask
);
2125 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2127 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
2129 void nir_assign_var_locations(struct exec_list
*var_list
,
2131 int (*type_size
)(const struct glsl_type
*));
2133 void nir_lower_io(nir_shader
*shader
,
2134 nir_variable_mode mode
,
2135 int (*type_size
)(const struct glsl_type
*));
2136 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2137 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2139 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2141 bool nir_remove_dead_variables(nir_shader
*shader
);
2143 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2144 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2145 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2146 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2148 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2150 void nir_lower_samplers(nir_shader
*shader
,
2151 const struct gl_shader_program
*shader_program
);
2153 bool nir_lower_system_values(nir_shader
*shader
);
2155 typedef struct nir_lower_tex_options
{
2157 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2158 * sampler types a texture projector is lowered.
2163 * If true, lower rect textures to 2D, using txs to fetch the
2164 * texture dimensions and dividing the texture coords by the
2165 * texture dims to normalize.
2170 * To emulate certain texture wrap modes, this can be used
2171 * to saturate the specified tex coord to [0.0, 1.0]. The
2172 * bits are according to sampler #, ie. if, for example:
2174 * (conf->saturate_s & (1 << n))
2176 * is true, then the s coord for sampler n is saturated.
2178 * Note that clamping must happen *after* projector lowering
2179 * so any projected texture sample instruction with a clamped
2180 * coordinate gets automatically lowered, regardless of the
2181 * 'lower_txp' setting.
2183 unsigned saturate_s
;
2184 unsigned saturate_t
;
2185 unsigned saturate_r
;
2187 /* Bitmask of textures that need swizzling.
2189 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2190 * swizzles[texture_index] is applied to the result of the texturing
2193 unsigned swizzle_result
;
2195 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2196 * while 4 and 5 represent 0 and 1 respectively.
2198 uint8_t swizzles
[32][4];
2199 } nir_lower_tex_options
;
2201 bool nir_lower_tex(nir_shader
*shader
,
2202 const nir_lower_tex_options
*options
);
2204 void nir_lower_idiv(nir_shader
*shader
);
2206 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2207 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2209 void nir_lower_two_sided_color(nir_shader
*shader
);
2211 void nir_lower_atomics(nir_shader
*shader
,
2212 const struct gl_shader_program
*shader_program
);
2213 void nir_lower_to_source_mods(nir_shader
*shader
);
2215 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2217 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2219 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2220 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2222 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2223 void nir_convert_to_ssa(nir_shader
*shader
);
2225 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2226 * registers. If false, convert all values (even those not involved in a phi
2227 * node) to registers.
2229 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2231 bool nir_opt_algebraic(nir_shader
*shader
);
2232 bool nir_opt_algebraic_late(nir_shader
*shader
);
2233 bool nir_opt_constant_folding(nir_shader
*shader
);
2235 bool nir_opt_global_to_local(nir_shader
*shader
);
2237 bool nir_copy_prop(nir_shader
*shader
);
2239 bool nir_opt_cse(nir_shader
*shader
);
2241 bool nir_opt_dce(nir_shader
*shader
);
2243 bool nir_opt_dead_cf(nir_shader
*shader
);
2245 void nir_opt_gcm(nir_shader
*shader
);
2247 bool nir_opt_peephole_select(nir_shader
*shader
);
2249 bool nir_opt_remove_phis(nir_shader
*shader
);
2251 bool nir_opt_undef(nir_shader
*shader
);
2253 void nir_sweep(nir_shader
*shader
);
2255 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2256 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);