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 "util/macros.h"
38 #include "compiler/nir_types.h"
39 #include "compiler/shader_enums.h"
40 #include "compiler/shader_info.h"
43 #include "nir_opcodes.h"
50 struct gl_shader_program
;
53 #define NIR_TRUE (~0u)
55 /** Defines a cast function
57 * This macro defines a cast function from in_type to out_type where
58 * out_type is some structure type that contains a field of type out_type.
60 * Note that you have to be a bit careful as the generated cast function
63 #define NIR_DEFINE_CAST(name, in_type, out_type, field, \
64 type_field, type_value) \
65 static inline out_type * \
66 name(const in_type *parent) \
68 assert(parent && parent->type_field == type_value); \
69 return exec_node_data(out_type, parent, field); \
78 * Description of built-in state associated with a uniform
80 * \sa nir_variable::state_slots
88 nir_var_shader_in
= (1 << 0),
89 nir_var_shader_out
= (1 << 1),
90 nir_var_global
= (1 << 2),
91 nir_var_local
= (1 << 3),
92 nir_var_uniform
= (1 << 4),
93 nir_var_shader_storage
= (1 << 5),
94 nir_var_system_value
= (1 << 6),
95 nir_var_param
= (1 << 7),
96 nir_var_shared
= (1 << 8),
110 typedef struct nir_constant
{
112 * Value of the constant.
114 * The field used to back the values supplied by the constant is determined
115 * by the type associated with the \c nir_variable. Constants may be
116 * scalars, vectors, or matrices.
118 nir_const_value values
[4];
120 /* we could get this from the var->type but makes clone *much* easier to
121 * not have to care about the type.
123 unsigned num_elements
;
125 /* Array elements / Structure Fields */
126 struct nir_constant
**elements
;
130 * \brief Layout qualifiers for gl_FragDepth.
132 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
133 * with a layout qualifier.
136 nir_depth_layout_none
, /**< No depth layout is specified. */
137 nir_depth_layout_any
,
138 nir_depth_layout_greater
,
139 nir_depth_layout_less
,
140 nir_depth_layout_unchanged
144 * Either a uniform, global variable, shader input, or shader output. Based on
145 * ir_variable - it should be easy to translate between the two.
148 typedef struct nir_variable
{
149 struct exec_node node
;
152 * Declared type of the variable
154 const struct glsl_type
*type
;
157 * Declared name of the variable
161 struct nir_variable_data
{
163 * Storage class of the variable.
165 * \sa nir_variable_mode
167 nir_variable_mode mode
;
170 * Is the variable read-only?
172 * This is set for variables declared as \c const, shader inputs,
175 unsigned read_only
:1;
179 unsigned invariant
:1;
182 * Interpolation mode for shader inputs / outputs
184 * \sa glsl_interp_mode
186 unsigned interpolation
:2;
189 * \name ARB_fragment_coord_conventions
192 unsigned origin_upper_left
:1;
193 unsigned pixel_center_integer
:1;
197 * Was the location explicitly set in the shader?
199 * If the location is explicitly set in the shader, it \b cannot be changed
200 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
203 unsigned explicit_location
:1;
204 unsigned explicit_index
:1;
207 * Was an initial binding explicitly set in the shader?
209 * If so, constant_initializer contains an integer nir_constant
210 * representing the initial binding point.
212 unsigned explicit_binding
:1;
215 * Does this variable have an initializer?
217 * This is used by the linker to cross-validiate initializers of global
220 unsigned has_initializer
:1;
223 * If non-zero, then this variable may be packed along with other variables
224 * into a single varying slot, so this offset should be applied when
225 * accessing components. For example, an offset of 1 means that the x
226 * component of this variable is actually stored in component y of the
227 * location specified by \c location.
229 unsigned location_frac
:2;
232 * If true, this variable represents an array of scalars that should
233 * be tightly packed. In other words, consecutive array elements
234 * should be stored one component apart, rather than one slot apart.
239 * Whether this is a fragment shader output implicitly initialized with
240 * the previous contents of the specified render target at the
241 * framebuffer location corresponding to this shader invocation.
243 unsigned fb_fetch_output
:1;
246 * \brief Layout qualifier for gl_FragDepth.
248 * This is not equal to \c ir_depth_layout_none if and only if this
249 * variable is \c gl_FragDepth and a layout qualifier is specified.
251 nir_depth_layout depth_layout
;
254 * Storage location of the base of this variable
256 * The precise meaning of this field depends on the nature of the variable.
258 * - Vertex shader input: one of the values from \c gl_vert_attrib.
259 * - Vertex shader output: one of the values from \c gl_varying_slot.
260 * - Geometry shader input: one of the values from \c gl_varying_slot.
261 * - Geometry shader output: one of the values from \c gl_varying_slot.
262 * - Fragment shader input: one of the values from \c gl_varying_slot.
263 * - Fragment shader output: one of the values from \c gl_frag_result.
264 * - Uniforms: Per-stage uniform slot number for default uniform block.
265 * - Uniforms: Index within the uniform block definition for UBO members.
266 * - Non-UBO Uniforms: uniform slot number.
267 * - Other: This field is not currently used.
269 * If the variable is a uniform, shader input, or shader output, and the
270 * slot has not been assigned, the value will be -1.
275 * The actual location of the variable in the IR. Only valid for inputs
278 unsigned int driver_location
;
281 * output index for dual source blending.
286 * Descriptor set binding for sampler or UBO.
291 * Initial binding point for a sampler or UBO.
293 * For array types, this represents the binding point for the first element.
298 * Location an atomic counter is stored at.
303 * ARB_shader_image_load_store qualifiers.
306 bool read_only
; /**< "readonly" qualifier. */
307 bool write_only
; /**< "writeonly" qualifier. */
312 /** Image internal format if specified explicitly, otherwise GL_NONE. */
317 * Highest element accessed with a constant expression array index
319 * Not used for non-array variables.
321 unsigned max_array_access
;
326 * Built-in state that backs this uniform
328 * Once set at variable creation, \c state_slots must remain invariant.
329 * This is because, ideally, this array would be shared by all clones of
330 * this variable in the IR tree. In other words, we'd really like for it
331 * to be a fly-weight.
333 * If the variable is not a uniform, \c num_state_slots will be zero and
334 * \c state_slots will be \c NULL.
337 unsigned num_state_slots
; /**< Number of state slots used */
338 nir_state_slot
*state_slots
; /**< State descriptors. */
342 * Constant expression assigned in the initializer of the variable
344 nir_constant
*constant_initializer
;
347 * For variables that are in an interface block or are an instance of an
348 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
350 * \sa ir_variable::location
352 const struct glsl_type
*interface_type
;
355 #define nir_foreach_variable(var, var_list) \
356 foreach_list_typed(nir_variable, var, node, var_list)
358 #define nir_foreach_variable_safe(var, var_list) \
359 foreach_list_typed_safe(nir_variable, var, node, var_list)
362 nir_variable_is_global(const nir_variable
*var
)
364 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
367 typedef struct nir_register
{
368 struct exec_node node
;
370 unsigned num_components
; /** < number of vector components */
371 unsigned num_array_elems
; /** < size of array (0 for no array) */
373 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
376 /** generic register index. */
379 /** only for debug purposes, can be NULL */
382 /** whether this register is local (per-function) or global (per-shader) */
386 * If this flag is set to true, then accessing channels >= num_components
387 * is well-defined, and simply spills over to the next array element. This
388 * is useful for backends that can do per-component accessing, in
389 * particular scalar backends. By setting this flag and making
390 * num_components equal to 1, structures can be packed tightly into
391 * registers and then registers can be accessed per-component to get to
392 * each structure member, even if it crosses vec4 boundaries.
396 /** set of nir_src's where this register is used (read from) */
397 struct list_head uses
;
399 /** set of nir_dest's where this register is defined (written to) */
400 struct list_head defs
;
402 /** set of nir_if's where this register is used as a condition */
403 struct list_head if_uses
;
410 nir_instr_type_intrinsic
,
411 nir_instr_type_load_const
,
413 nir_instr_type_ssa_undef
,
415 nir_instr_type_parallel_copy
,
418 typedef struct nir_instr
{
419 struct exec_node node
;
421 struct nir_block
*block
;
423 /** generic instruction index. */
426 /* A temporary for optimization and analysis passes to use for storing
427 * flags. For instance, DCE uses this to store the "dead/live" info.
432 static inline nir_instr
*
433 nir_instr_next(nir_instr
*instr
)
435 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
436 if (exec_node_is_tail_sentinel(next
))
439 return exec_node_data(nir_instr
, next
, node
);
442 static inline nir_instr
*
443 nir_instr_prev(nir_instr
*instr
)
445 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
446 if (exec_node_is_head_sentinel(prev
))
449 return exec_node_data(nir_instr
, prev
, node
);
453 nir_instr_is_first(nir_instr
*instr
)
455 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
459 nir_instr_is_last(nir_instr
*instr
)
461 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
464 typedef struct nir_ssa_def
{
465 /** for debugging only, can be NULL */
468 /** generic SSA definition index. */
471 /** Index into the live_in and live_out bitfields */
474 nir_instr
*parent_instr
;
476 /** set of nir_instr's where this register is used (read from) */
477 struct list_head uses
;
479 /** set of nir_if's where this register is used as a condition */
480 struct list_head if_uses
;
482 uint8_t num_components
;
484 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
492 struct nir_src
*indirect
; /** < NULL for no indirect offset */
493 unsigned base_offset
;
495 /* TODO use-def chain goes here */
499 nir_instr
*parent_instr
;
500 struct list_head def_link
;
503 struct nir_src
*indirect
; /** < NULL for no indirect offset */
504 unsigned base_offset
;
506 /* TODO def-use chain goes here */
511 typedef struct nir_src
{
513 nir_instr
*parent_instr
;
514 struct nir_if
*parent_if
;
517 struct list_head use_link
;
527 static inline nir_src
530 nir_src src
= { { NULL
} };
534 #define NIR_SRC_INIT nir_src_init()
536 #define nir_foreach_use(src, reg_or_ssa_def) \
537 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
539 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
540 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
542 #define nir_foreach_if_use(src, reg_or_ssa_def) \
543 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
545 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
546 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
557 static inline nir_dest
560 nir_dest dest
= { { { NULL
} } };
564 #define NIR_DEST_INIT nir_dest_init()
566 #define nir_foreach_def(dest, reg) \
567 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
569 #define nir_foreach_def_safe(dest, reg) \
570 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
572 static inline nir_src
573 nir_src_for_ssa(nir_ssa_def
*def
)
575 nir_src src
= NIR_SRC_INIT
;
583 static inline nir_src
584 nir_src_for_reg(nir_register
*reg
)
586 nir_src src
= NIR_SRC_INIT
;
590 src
.reg
.indirect
= NULL
;
591 src
.reg
.base_offset
= 0;
596 static inline nir_dest
597 nir_dest_for_reg(nir_register
*reg
)
599 nir_dest dest
= NIR_DEST_INIT
;
606 static inline unsigned
607 nir_src_bit_size(nir_src src
)
609 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
612 static inline unsigned
613 nir_dest_bit_size(nir_dest dest
)
615 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
618 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
619 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
625 * \name input modifiers
629 * For inputs interpreted as floating point, flips the sign bit. For
630 * inputs interpreted as integers, performs the two's complement negation.
635 * Clears the sign bit for floating point values, and computes the integer
636 * absolute value for integers. Note that the negate modifier acts after
637 * the absolute value modifier, therefore if both are set then all inputs
638 * will become negative.
644 * For each input component, says which component of the register it is
645 * chosen from. Note that which elements of the swizzle are used and which
646 * are ignored are based on the write mask for most opcodes - for example,
647 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
648 * a swizzle of {2, x, 1, 0} where x means "don't care."
657 * \name saturate output modifier
659 * Only valid for opcodes that output floating-point numbers. Clamps the
660 * output to between 0.0 and 1.0 inclusive.
665 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
669 nir_type_invalid
= 0, /* Not a valid type */
674 nir_type_bool32
= 32 | nir_type_bool
,
675 nir_type_int8
= 8 | nir_type_int
,
676 nir_type_int16
= 16 | nir_type_int
,
677 nir_type_int32
= 32 | nir_type_int
,
678 nir_type_int64
= 64 | nir_type_int
,
679 nir_type_uint8
= 8 | nir_type_uint
,
680 nir_type_uint16
= 16 | nir_type_uint
,
681 nir_type_uint32
= 32 | nir_type_uint
,
682 nir_type_uint64
= 64 | nir_type_uint
,
683 nir_type_float16
= 16 | nir_type_float
,
684 nir_type_float32
= 32 | nir_type_float
,
685 nir_type_float64
= 64 | nir_type_float
,
688 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
689 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
691 static inline unsigned
692 nir_alu_type_get_type_size(nir_alu_type type
)
694 return type
& NIR_ALU_TYPE_SIZE_MASK
;
697 static inline unsigned
698 nir_alu_type_get_base_type(nir_alu_type type
)
700 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
704 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
705 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
706 } nir_op_algebraic_property
;
714 * The number of components in the output
716 * If non-zero, this is the size of the output and input sizes are
717 * explicitly given; swizzle and writemask are still in effect, but if
718 * the output component is masked out, then the input component may
721 * If zero, the opcode acts in the standard, per-component manner; the
722 * operation is performed on each component (except the ones that are
723 * masked out) with the input being taken from the input swizzle for
726 * The size of some of the inputs may be given (i.e. non-zero) even
727 * though output_size is zero; in that case, the inputs with a zero
728 * size act per-component, while the inputs with non-zero size don't.
730 unsigned output_size
;
733 * The type of vector that the instruction outputs. Note that the
734 * staurate modifier is only allowed on outputs with the float type.
737 nir_alu_type output_type
;
740 * The number of components in each input
742 unsigned input_sizes
[4];
745 * The type of vector that each input takes. Note that negate and
746 * absolute value are only allowed on inputs with int or float type and
747 * behave differently on the two.
749 nir_alu_type input_types
[4];
751 nir_op_algebraic_property algebraic_properties
;
754 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
756 typedef struct nir_alu_instr
{
760 /** Indicates that this ALU instruction generates an exact value
762 * This is kind of a mixture of GLSL "precise" and "invariant" and not
763 * really equivalent to either. This indicates that the value generated by
764 * this operation is high-precision and any code transformations that touch
765 * it must ensure that the resulting value is bit-for-bit identical to the
774 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
775 nir_alu_instr
*instr
);
776 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
777 nir_alu_instr
*instr
);
779 /* is this source channel used? */
781 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
783 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
784 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
786 return (instr
->dest
.write_mask
>> channel
) & 1;
790 * For instructions whose destinations are SSA, get the number of channels
793 static inline unsigned
794 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
796 assert(instr
->dest
.dest
.is_ssa
);
798 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
799 return nir_op_infos
[instr
->op
].input_sizes
[src
];
801 return instr
->dest
.dest
.ssa
.num_components
;
804 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
805 unsigned src1
, unsigned src2
);
809 nir_deref_type_array
,
810 nir_deref_type_struct
813 typedef struct nir_deref
{
814 nir_deref_type deref_type
;
815 struct nir_deref
*child
;
816 const struct glsl_type
*type
;
825 /* This enum describes how the array is referenced. If the deref is
826 * direct then the base_offset is used. If the deref is indirect then
827 * offset is given by base_offset + indirect. If the deref is a wildcard
828 * then the deref refers to all of the elements of the array at the same
829 * time. Wildcard dereferences are only ever allowed in copy_var
830 * intrinsics and the source and destination derefs must have matching
834 nir_deref_array_type_direct
,
835 nir_deref_array_type_indirect
,
836 nir_deref_array_type_wildcard
,
837 } nir_deref_array_type
;
842 nir_deref_array_type deref_array_type
;
843 unsigned base_offset
;
853 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
,
854 deref_type
, nir_deref_type_var
)
855 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
,
856 deref_type
, nir_deref_type_array
)
857 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
,
858 deref_type
, nir_deref_type_struct
)
860 /* Returns the last deref in the chain. */
861 static inline nir_deref
*
862 nir_deref_tail(nir_deref
*deref
)
865 deref
= deref
->child
;
873 nir_deref_var
**params
;
874 nir_deref_var
*return_deref
;
876 struct nir_function
*callee
;
879 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
880 num_variables, num_indices, idx0, idx1, idx2, flags) \
881 nir_intrinsic_##name,
883 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
886 #include "nir_intrinsics.h"
887 nir_num_intrinsics
= nir_last_intrinsic
+ 1
890 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
892 /** Represents an intrinsic
894 * An intrinsic is an instruction type for handling things that are
895 * more-or-less regular operations but don't just consume and produce SSA
896 * values like ALU operations do. Intrinsics are not for things that have
897 * special semantic meaning such as phi nodes and parallel copies.
898 * Examples of intrinsics include variable load/store operations, system
899 * value loads, and the like. Even though texturing more-or-less falls
900 * under this category, texturing is its own instruction type because
901 * trying to represent texturing with intrinsics would lead to a
902 * combinatorial explosion of intrinsic opcodes.
904 * By having a single instruction type for handling a lot of different
905 * cases, optimization passes can look for intrinsics and, for the most
906 * part, completely ignore them. Each intrinsic type also has a few
907 * possible flags that govern whether or not they can be reordered or
908 * eliminated. That way passes like dead code elimination can still work
909 * on intrisics without understanding the meaning of each.
911 * Each intrinsic has some number of constant indices, some number of
912 * variables, and some number of sources. What these sources, variables,
913 * and indices mean depends on the intrinsic and is documented with the
914 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
915 * instructions are the only types of instruction that can operate on
921 nir_intrinsic_op intrinsic
;
925 /** number of components if this is a vectorized intrinsic
927 * Similarly to ALU operations, some intrinsics are vectorized.
928 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
929 * For vectorized intrinsics, the num_components field specifies the
930 * number of destination components and the number of source components
931 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
933 uint8_t num_components
;
935 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
937 nir_deref_var
*variables
[2];
940 } nir_intrinsic_instr
;
943 * \name NIR intrinsics semantic flags
945 * information about what the compiler can do with the intrinsics.
947 * \sa nir_intrinsic_info::flags
951 * whether the intrinsic can be safely eliminated if none of its output
952 * value is not being used.
954 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
957 * Whether the intrinsic can be reordered with respect to any other
958 * intrinsic, i.e. whether the only reordering dependencies of the
959 * intrinsic are due to the register reads/writes.
961 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
962 } nir_intrinsic_semantic_flag
;
965 * \name NIR intrinsics const-index flag
967 * Indicates the usage of a const_index slot.
969 * \sa nir_intrinsic_info::index_map
973 * Generally instructions that take a offset src argument, can encode
974 * a constant 'base' value which is added to the offset.
976 NIR_INTRINSIC_BASE
= 1,
979 * For store instructions, a writemask for the store.
981 NIR_INTRINSIC_WRMASK
= 2,
984 * The stream-id for GS emit_vertex/end_primitive intrinsics.
986 NIR_INTRINSIC_STREAM_ID
= 3,
989 * The clip-plane id for load_user_clip_plane intrinsic.
991 NIR_INTRINSIC_UCP_ID
= 4,
994 * The amount of data, starting from BASE, that this instruction may
995 * access. This is used to provide bounds if the offset is not constant.
997 NIR_INTRINSIC_RANGE
= 5,
1000 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1002 NIR_INTRINSIC_DESC_SET
= 6,
1005 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1007 NIR_INTRINSIC_BINDING
= 7,
1012 NIR_INTRINSIC_COMPONENT
= 8,
1015 * Interpolation mode (only meaningful for FS inputs).
1017 NIR_INTRINSIC_INTERP_MODE
= 9,
1019 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1021 } nir_intrinsic_index_flag
;
1023 #define NIR_INTRINSIC_MAX_INPUTS 4
1028 unsigned num_srcs
; /** < number of register/SSA inputs */
1030 /** number of components of each input register
1032 * If this value is 0, the number of components is given by the
1033 * num_components field of nir_intrinsic_instr.
1035 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1039 /** number of components of the output register
1041 * If this value is 0, the number of components is given by the
1042 * num_components field of nir_intrinsic_instr.
1044 unsigned dest_components
;
1046 /** the number of inputs/outputs that are variables */
1047 unsigned num_variables
;
1049 /** the number of constant indices used by the intrinsic */
1050 unsigned num_indices
;
1052 /** indicates the usage of intr->const_index[n] */
1053 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1055 /** semantic flags for calls to this intrinsic */
1056 nir_intrinsic_semantic_flag flags
;
1057 } nir_intrinsic_info
;
1059 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1062 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1063 static inline type \
1064 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
1066 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1067 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1068 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1070 static inline void \
1071 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1073 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1074 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1075 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1078 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1079 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1080 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1081 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1082 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1083 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1084 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1085 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1086 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1089 * \group texture information
1091 * This gives semantic information about textures which is useful to the
1092 * frontend, the backend, and lowering passes, but not the optimizer.
1097 nir_tex_src_projector
,
1098 nir_tex_src_comparitor
, /* shadow comparitor */
1102 nir_tex_src_ms_index
, /* MSAA sample index */
1103 nir_tex_src_ms_mcs
, /* MSAA compression value */
1106 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1107 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1108 nir_tex_src_plane
, /* < selects plane for planar textures */
1109 nir_num_tex_src_types
1114 nir_tex_src_type src_type
;
1118 nir_texop_tex
, /**< Regular texture look-up */
1119 nir_texop_txb
, /**< Texture look-up with LOD bias */
1120 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1121 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1122 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1123 nir_texop_txf_ms
, /**< Multisample texture fetch */
1124 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1125 nir_texop_txs
, /**< Texture size */
1126 nir_texop_lod
, /**< Texture lod query */
1127 nir_texop_tg4
, /**< Texture gather */
1128 nir_texop_query_levels
, /**< Texture levels query */
1129 nir_texop_texture_samples
, /**< Texture samples query */
1130 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1138 enum glsl_sampler_dim sampler_dim
;
1139 nir_alu_type dest_type
;
1144 unsigned num_srcs
, coord_components
;
1145 bool is_array
, is_shadow
;
1148 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1149 * components or the new-style shadow that outputs 1 component.
1151 bool is_new_style_shadow
;
1153 /* gather component selector */
1154 unsigned component
: 2;
1156 /** The texture index
1158 * If this texture instruction has a nir_tex_src_texture_offset source,
1159 * then the texture index is given by texture_index + texture_offset.
1161 unsigned texture_index
;
1163 /** The size of the texture array or 0 if it's not an array */
1164 unsigned texture_array_size
;
1166 /** The texture deref
1168 * If this is null, use texture_index instead.
1170 nir_deref_var
*texture
;
1172 /** The sampler index
1174 * The following operations do not require a sampler and, as such, this
1175 * field should be ignored:
1177 * - nir_texop_txf_ms
1181 * - nir_texop_query_levels
1182 * - nir_texop_texture_samples
1183 * - nir_texop_samples_identical
1185 * If this texture instruction has a nir_tex_src_sampler_offset source,
1186 * then the sampler index is given by sampler_index + sampler_offset.
1188 unsigned sampler_index
;
1190 /** The sampler deref
1192 * If this is null, use sampler_index instead.
1194 nir_deref_var
*sampler
;
1197 static inline unsigned
1198 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1200 switch (instr
->op
) {
1201 case nir_texop_txs
: {
1203 switch (instr
->sampler_dim
) {
1204 case GLSL_SAMPLER_DIM_1D
:
1205 case GLSL_SAMPLER_DIM_BUF
:
1208 case GLSL_SAMPLER_DIM_2D
:
1209 case GLSL_SAMPLER_DIM_CUBE
:
1210 case GLSL_SAMPLER_DIM_MS
:
1211 case GLSL_SAMPLER_DIM_RECT
:
1212 case GLSL_SAMPLER_DIM_EXTERNAL
:
1213 case GLSL_SAMPLER_DIM_SUBPASS
:
1216 case GLSL_SAMPLER_DIM_3D
:
1220 unreachable("not reached");
1222 if (instr
->is_array
)
1230 case nir_texop_texture_samples
:
1231 case nir_texop_query_levels
:
1232 case nir_texop_samples_identical
:
1236 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1243 /* Returns true if this texture operation queries something about the texture
1244 * rather than actually sampling it.
1247 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1249 switch (instr
->op
) {
1252 case nir_texop_texture_samples
:
1253 case nir_texop_query_levels
:
1254 case nir_texop_txf_ms_mcs
:
1261 case nir_texop_txf_ms
:
1265 unreachable("Invalid texture opcode");
1269 static inline nir_alu_type
1270 nir_tex_instr_src_type(nir_tex_instr
*instr
, unsigned src
)
1272 switch (instr
->src
[src
].src_type
) {
1273 case nir_tex_src_coord
:
1274 switch (instr
->op
) {
1276 case nir_texop_txf_ms
:
1277 case nir_texop_txf_ms_mcs
:
1278 case nir_texop_samples_identical
:
1279 return nir_type_int
;
1282 return nir_type_float
;
1285 case nir_tex_src_lod
:
1286 switch (instr
->op
) {
1289 return nir_type_int
;
1292 return nir_type_float
;
1295 case nir_tex_src_projector
:
1296 case nir_tex_src_comparitor
:
1297 case nir_tex_src_bias
:
1298 case nir_tex_src_ddx
:
1299 case nir_tex_src_ddy
:
1300 return nir_type_float
;
1302 case nir_tex_src_offset
:
1303 case nir_tex_src_ms_index
:
1304 case nir_tex_src_texture_offset
:
1305 case nir_tex_src_sampler_offset
:
1306 return nir_type_int
;
1309 unreachable("Invalid texture source type");
1313 static inline unsigned
1314 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1316 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1317 return instr
->coord_components
;
1319 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1320 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1323 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1324 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1325 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1326 if (instr
->is_array
)
1327 return instr
->coord_components
- 1;
1329 return instr
->coord_components
;
1336 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1338 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1339 if (instr
->src
[i
].src_type
== type
)
1345 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1350 nir_const_value value
;
1353 } nir_load_const_instr
;
1366 /* creates a new SSA variable in an undefined state */
1371 } nir_ssa_undef_instr
;
1374 struct exec_node node
;
1376 /* The predecessor block corresponding to this source */
1377 struct nir_block
*pred
;
1382 #define nir_foreach_phi_src(phi_src, phi) \
1383 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1384 #define nir_foreach_phi_src_safe(phi_src, phi) \
1385 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1390 struct exec_list srcs
; /** < list of nir_phi_src */
1396 struct exec_node node
;
1399 } nir_parallel_copy_entry
;
1401 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1402 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1407 /* A list of nir_parallel_copy_entry's. The sources of all of the
1408 * entries are copied to the corresponding destinations "in parallel".
1409 * In other words, if we have two entries: a -> b and b -> a, the values
1412 struct exec_list entries
;
1413 } nir_parallel_copy_instr
;
1415 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1416 type
, nir_instr_type_alu
)
1417 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1418 type
, nir_instr_type_call
)
1419 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1420 type
, nir_instr_type_jump
)
1421 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1422 type
, nir_instr_type_tex
)
1423 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1424 type
, nir_instr_type_intrinsic
)
1425 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1426 type
, nir_instr_type_load_const
)
1427 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1428 type
, nir_instr_type_ssa_undef
)
1429 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1430 type
, nir_instr_type_phi
)
1431 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1432 nir_parallel_copy_instr
, instr
,
1433 type
, nir_instr_type_parallel_copy
)
1438 * Control flow consists of a tree of control flow nodes, which include
1439 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1440 * instructions that always run start-to-finish. Each basic block also keeps
1441 * track of its successors (blocks which may run immediately after the current
1442 * block) and predecessors (blocks which could have run immediately before the
1443 * current block). Each function also has a start block and an end block which
1444 * all return statements point to (which is always empty). Together, all the
1445 * blocks with their predecessors and successors make up the control flow
1446 * graph (CFG) of the function. There are helpers that modify the tree of
1447 * control flow nodes while modifying the CFG appropriately; these should be
1448 * used instead of modifying the tree directly.
1455 nir_cf_node_function
1458 typedef struct nir_cf_node
{
1459 struct exec_node node
;
1460 nir_cf_node_type type
;
1461 struct nir_cf_node
*parent
;
1464 typedef struct nir_block
{
1465 nir_cf_node cf_node
;
1467 struct exec_list instr_list
; /** < list of nir_instr */
1469 /** generic block index; generated by nir_index_blocks */
1473 * Each block can only have up to 2 successors, so we put them in a simple
1474 * array - no need for anything more complicated.
1476 struct nir_block
*successors
[2];
1478 /* Set of nir_block predecessors in the CFG */
1479 struct set
*predecessors
;
1482 * this node's immediate dominator in the dominance tree - set to NULL for
1485 struct nir_block
*imm_dom
;
1487 /* This node's children in the dominance tree */
1488 unsigned num_dom_children
;
1489 struct nir_block
**dom_children
;
1491 /* Set of nir_block's on the dominance frontier of this block */
1492 struct set
*dom_frontier
;
1495 * These two indices have the property that dom_{pre,post}_index for each
1496 * child of this block in the dominance tree will always be between
1497 * dom_pre_index and dom_post_index for this block, which makes testing if
1498 * a given block is dominated by another block an O(1) operation.
1500 unsigned dom_pre_index
, dom_post_index
;
1502 /* live in and out for this block; used for liveness analysis */
1503 BITSET_WORD
*live_in
;
1504 BITSET_WORD
*live_out
;
1507 static inline nir_instr
*
1508 nir_block_first_instr(nir_block
*block
)
1510 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1511 return exec_node_data(nir_instr
, head
, node
);
1514 static inline nir_instr
*
1515 nir_block_last_instr(nir_block
*block
)
1517 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1518 return exec_node_data(nir_instr
, tail
, node
);
1521 #define nir_foreach_instr(instr, block) \
1522 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1523 #define nir_foreach_instr_reverse(instr, block) \
1524 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1525 #define nir_foreach_instr_safe(instr, block) \
1526 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1527 #define nir_foreach_instr_reverse_safe(instr, block) \
1528 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1530 typedef struct nir_if
{
1531 nir_cf_node cf_node
;
1534 struct exec_list then_list
; /** < list of nir_cf_node */
1535 struct exec_list else_list
; /** < list of nir_cf_node */
1539 nir_cf_node cf_node
;
1541 struct exec_list body
; /** < list of nir_cf_node */
1545 * Various bits of metadata that can may be created or required by
1546 * optimization and analysis passes
1549 nir_metadata_none
= 0x0,
1550 nir_metadata_block_index
= 0x1,
1551 nir_metadata_dominance
= 0x2,
1552 nir_metadata_live_ssa_defs
= 0x4,
1553 nir_metadata_not_properly_reset
= 0x8,
1557 nir_cf_node cf_node
;
1559 /** pointer to the function of which this is an implementation */
1560 struct nir_function
*function
;
1562 struct exec_list body
; /** < list of nir_cf_node */
1564 nir_block
*end_block
;
1566 /** list for all local variables in the function */
1567 struct exec_list locals
;
1569 /** array of variables used as parameters */
1570 unsigned num_params
;
1571 nir_variable
**params
;
1573 /** variable used to hold the result of the function */
1574 nir_variable
*return_var
;
1576 /** list of local registers in the function */
1577 struct exec_list registers
;
1579 /** next available local register index */
1582 /** next available SSA value index */
1585 /* total number of basic blocks, only valid when block_index_dirty = false */
1586 unsigned num_blocks
;
1588 nir_metadata valid_metadata
;
1589 } nir_function_impl
;
1591 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1592 nir_start_block(nir_function_impl
*impl
)
1594 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1597 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1598 nir_impl_last_block(nir_function_impl
*impl
)
1600 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1603 static inline nir_cf_node
*
1604 nir_cf_node_next(nir_cf_node
*node
)
1606 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1607 if (exec_node_is_tail_sentinel(next
))
1610 return exec_node_data(nir_cf_node
, next
, node
);
1613 static inline nir_cf_node
*
1614 nir_cf_node_prev(nir_cf_node
*node
)
1616 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1617 if (exec_node_is_head_sentinel(prev
))
1620 return exec_node_data(nir_cf_node
, prev
, node
);
1624 nir_cf_node_is_first(const nir_cf_node
*node
)
1626 return exec_node_is_head_sentinel(node
->node
.prev
);
1630 nir_cf_node_is_last(const nir_cf_node
*node
)
1632 return exec_node_is_tail_sentinel(node
->node
.next
);
1635 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1636 type
, nir_cf_node_block
)
1637 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1638 type
, nir_cf_node_if
)
1639 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1640 type
, nir_cf_node_loop
)
1641 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1642 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1644 static inline nir_block
*
1645 nir_if_first_then_block(nir_if
*if_stmt
)
1647 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1648 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1651 static inline nir_block
*
1652 nir_if_last_then_block(nir_if
*if_stmt
)
1654 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1655 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1658 static inline nir_block
*
1659 nir_if_first_else_block(nir_if
*if_stmt
)
1661 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1662 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1665 static inline nir_block
*
1666 nir_if_last_else_block(nir_if
*if_stmt
)
1668 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1669 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1672 static inline nir_block
*
1673 nir_loop_first_block(nir_loop
*loop
)
1675 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1676 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1679 static inline nir_block
*
1680 nir_loop_last_block(nir_loop
*loop
)
1682 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1683 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1689 nir_parameter_inout
,
1690 } nir_parameter_type
;
1693 nir_parameter_type param_type
;
1694 const struct glsl_type
*type
;
1697 typedef struct nir_function
{
1698 struct exec_node node
;
1701 struct nir_shader
*shader
;
1703 unsigned num_params
;
1704 nir_parameter
*params
;
1705 const struct glsl_type
*return_type
;
1707 /** The implementation of this function.
1709 * If the function is only declared and not implemented, this is NULL.
1711 nir_function_impl
*impl
;
1714 typedef struct nir_shader_compiler_options
{
1719 /** Lowers flrp when it does not support doubles */
1726 bool lower_bitfield_extract
;
1727 bool lower_bitfield_insert
;
1728 bool lower_uadd_carry
;
1729 bool lower_usub_borrow
;
1730 /** lowers fneg and ineg to fsub and isub. */
1732 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1735 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1738 /** enables rules to lower idiv by power-of-two: */
1741 /* Does the native fdot instruction replicate its result for four
1742 * components? If so, then opt_algebraic_late will turn all fdotN
1743 * instructions into fdot_replicatedN instructions.
1745 bool fdot_replicates
;
1747 /** lowers ffract to fsub+ffloor: */
1750 bool lower_pack_half_2x16
;
1751 bool lower_pack_unorm_2x16
;
1752 bool lower_pack_snorm_2x16
;
1753 bool lower_pack_unorm_4x8
;
1754 bool lower_pack_snorm_4x8
;
1755 bool lower_unpack_half_2x16
;
1756 bool lower_unpack_unorm_2x16
;
1757 bool lower_unpack_snorm_2x16
;
1758 bool lower_unpack_unorm_4x8
;
1759 bool lower_unpack_snorm_4x8
;
1761 bool lower_extract_byte
;
1762 bool lower_extract_word
;
1765 * Does the driver support real 32-bit integers? (Otherwise, integers
1766 * are simulated by floats.)
1768 bool native_integers
;
1770 /* Indicates that the driver only has zero-based vertex id */
1771 bool vertex_id_zero_based
;
1773 bool lower_cs_local_index_from_id
;
1776 * Should nir_lower_io() create load_interpolated_input intrinsics?
1778 * If not, it generates regular load_input intrinsics and interpolation
1779 * information must be inferred from the list of input nir_variables.
1781 bool use_interpolated_input_intrinsics
;
1782 } nir_shader_compiler_options
;
1784 typedef struct nir_shader
{
1785 /** list of uniforms (nir_variable) */
1786 struct exec_list uniforms
;
1788 /** list of inputs (nir_variable) */
1789 struct exec_list inputs
;
1791 /** list of outputs (nir_variable) */
1792 struct exec_list outputs
;
1794 /** list of shared compute variables (nir_variable) */
1795 struct exec_list shared
;
1797 /** Set of driver-specific options for the shader.
1799 * The memory for the options is expected to be kept in a single static
1800 * copy by the driver.
1802 const struct nir_shader_compiler_options
*options
;
1804 /** Various bits of compile-time information about a given shader */
1805 struct shader_info
*info
;
1807 /** list of global variables in the shader (nir_variable) */
1808 struct exec_list globals
;
1810 /** list of system value variables in the shader (nir_variable) */
1811 struct exec_list system_values
;
1813 struct exec_list functions
; /** < list of nir_function */
1815 /** list of global register in the shader */
1816 struct exec_list registers
;
1818 /** next available global register index */
1822 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1825 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1827 /** The shader stage, such as MESA_SHADER_VERTEX. */
1828 gl_shader_stage stage
;
1831 static inline nir_function_impl
*
1832 nir_shader_get_entrypoint(nir_shader
*shader
)
1834 assert(exec_list_length(&shader
->functions
) == 1);
1835 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1836 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1837 assert(func
->return_type
== glsl_void_type());
1838 assert(func
->num_params
== 0);
1843 #define nir_foreach_function(func, shader) \
1844 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1846 nir_shader
*nir_shader_create(void *mem_ctx
,
1847 gl_shader_stage stage
,
1848 const nir_shader_compiler_options
*options
,
1851 /** creates a register, including assigning it an index and adding it to the list */
1852 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1854 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1856 void nir_reg_remove(nir_register
*reg
);
1858 /** Adds a variable to the appropreate list in nir_shader */
1859 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1862 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1864 assert(var
->data
.mode
== nir_var_local
);
1865 exec_list_push_tail(&impl
->locals
, &var
->node
);
1868 /** creates a variable, sets a few defaults, and adds it to the list */
1869 nir_variable
*nir_variable_create(nir_shader
*shader
,
1870 nir_variable_mode mode
,
1871 const struct glsl_type
*type
,
1873 /** creates a local variable and adds it to the list */
1874 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1875 const struct glsl_type
*type
,
1878 /** creates a function and adds it to the shader's list of functions */
1879 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1881 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1882 /** creates a function_impl that isn't tied to any particular function */
1883 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1885 nir_block
*nir_block_create(nir_shader
*shader
);
1886 nir_if
*nir_if_create(nir_shader
*shader
);
1887 nir_loop
*nir_loop_create(nir_shader
*shader
);
1889 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1891 /** requests that the given pieces of metadata be generated */
1892 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1893 /** dirties all but the preserved metadata */
1894 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1896 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1897 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1899 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1901 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1902 unsigned num_components
,
1905 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1906 nir_intrinsic_op op
);
1908 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1909 nir_function
*callee
);
1911 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1913 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1915 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1917 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1918 unsigned num_components
,
1921 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1922 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1923 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1925 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1927 typedef bool (*nir_deref_foreach_leaf_cb
)(nir_deref_var
*deref
, void *state
);
1928 bool nir_deref_foreach_leaf(nir_deref_var
*deref
,
1929 nir_deref_foreach_leaf_cb cb
, void *state
);
1931 nir_load_const_instr
*
1932 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1935 * NIR Cursors and Instruction Insertion API
1938 * A tiny struct representing a point to insert/extract instructions or
1939 * control flow nodes. Helps reduce the combinatorial explosion of possible
1940 * points to insert/extract.
1942 * \sa nir_control_flow.h
1945 nir_cursor_before_block
,
1946 nir_cursor_after_block
,
1947 nir_cursor_before_instr
,
1948 nir_cursor_after_instr
,
1949 } nir_cursor_option
;
1952 nir_cursor_option option
;
1959 static inline nir_block
*
1960 nir_cursor_current_block(nir_cursor cursor
)
1962 if (cursor
.option
== nir_cursor_before_instr
||
1963 cursor
.option
== nir_cursor_after_instr
) {
1964 return cursor
.instr
->block
;
1966 return cursor
.block
;
1970 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
1972 static inline nir_cursor
1973 nir_before_block(nir_block
*block
)
1976 cursor
.option
= nir_cursor_before_block
;
1977 cursor
.block
= block
;
1981 static inline nir_cursor
1982 nir_after_block(nir_block
*block
)
1985 cursor
.option
= nir_cursor_after_block
;
1986 cursor
.block
= block
;
1990 static inline nir_cursor
1991 nir_before_instr(nir_instr
*instr
)
1994 cursor
.option
= nir_cursor_before_instr
;
1995 cursor
.instr
= instr
;
1999 static inline nir_cursor
2000 nir_after_instr(nir_instr
*instr
)
2003 cursor
.option
= nir_cursor_after_instr
;
2004 cursor
.instr
= instr
;
2008 static inline nir_cursor
2009 nir_after_block_before_jump(nir_block
*block
)
2011 nir_instr
*last_instr
= nir_block_last_instr(block
);
2012 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2013 return nir_before_instr(last_instr
);
2015 return nir_after_block(block
);
2019 static inline nir_cursor
2020 nir_before_cf_node(nir_cf_node
*node
)
2022 if (node
->type
== nir_cf_node_block
)
2023 return nir_before_block(nir_cf_node_as_block(node
));
2025 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2028 static inline nir_cursor
2029 nir_after_cf_node(nir_cf_node
*node
)
2031 if (node
->type
== nir_cf_node_block
)
2032 return nir_after_block(nir_cf_node_as_block(node
));
2034 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2037 static inline nir_cursor
2038 nir_after_phis(nir_block
*block
)
2040 nir_foreach_instr(instr
, block
) {
2041 if (instr
->type
!= nir_instr_type_phi
)
2042 return nir_before_instr(instr
);
2044 return nir_after_block(block
);
2047 static inline nir_cursor
2048 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2050 if (node
->type
== nir_cf_node_block
)
2051 return nir_after_block(nir_cf_node_as_block(node
));
2053 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2055 return nir_after_phis(block
);
2058 static inline nir_cursor
2059 nir_before_cf_list(struct exec_list
*cf_list
)
2061 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2062 exec_list_get_head(cf_list
), node
);
2063 return nir_before_cf_node(first_node
);
2066 static inline nir_cursor
2067 nir_after_cf_list(struct exec_list
*cf_list
)
2069 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2070 exec_list_get_tail(cf_list
), node
);
2071 return nir_after_cf_node(last_node
);
2075 * Insert a NIR instruction at the given cursor.
2077 * Note: This does not update the cursor.
2079 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2082 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2084 nir_instr_insert(nir_before_instr(instr
), before
);
2088 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2090 nir_instr_insert(nir_after_instr(instr
), after
);
2094 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2096 nir_instr_insert(nir_before_block(block
), before
);
2100 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2102 nir_instr_insert(nir_after_block(block
), after
);
2106 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2108 nir_instr_insert(nir_before_cf_node(node
), before
);
2112 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2114 nir_instr_insert(nir_after_cf_node(node
), after
);
2118 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2120 nir_instr_insert(nir_before_cf_list(list
), before
);
2124 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2126 nir_instr_insert(nir_after_cf_list(list
), after
);
2129 void nir_instr_remove(nir_instr
*instr
);
2133 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2134 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2135 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2136 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2138 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2139 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2141 nir_const_value
*nir_src_as_const_value(nir_src src
);
2142 bool nir_src_is_dynamically_uniform(nir_src src
);
2143 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2144 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2145 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2146 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2147 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2150 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2151 unsigned num_components
, unsigned bit_size
,
2153 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2154 unsigned num_components
, unsigned bit_size
,
2156 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2157 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2158 nir_instr
*after_me
);
2160 uint8_t nir_ssa_def_components_read(nir_ssa_def
*def
);
2163 * finds the next basic block in source-code order, returns NULL if there is
2167 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2169 /* Performs the opposite of nir_block_cf_tree_next() */
2171 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2173 /* Gets the first block in a CF node in source-code order */
2175 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2177 /* Gets the last block in a CF node in source-code order */
2179 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2181 /* Gets the next block after a CF node in source-code order */
2183 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2185 /* Macros for loops that visit blocks in source-code order */
2187 #define nir_foreach_block(block, impl) \
2188 for (nir_block *block = nir_start_block(impl); block != NULL; \
2189 block = nir_block_cf_tree_next(block))
2191 #define nir_foreach_block_safe(block, impl) \
2192 for (nir_block *block = nir_start_block(impl), \
2193 *next = nir_block_cf_tree_next(block); \
2195 block = next, next = nir_block_cf_tree_next(block))
2197 #define nir_foreach_block_reverse(block, impl) \
2198 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2199 block = nir_block_cf_tree_prev(block))
2201 #define nir_foreach_block_reverse_safe(block, impl) \
2202 for (nir_block *block = nir_impl_last_block(impl), \
2203 *prev = nir_block_cf_tree_prev(block); \
2205 block = prev, prev = nir_block_cf_tree_prev(block))
2207 #define nir_foreach_block_in_cf_node(block, node) \
2208 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2209 block != nir_cf_node_cf_tree_next(node); \
2210 block = nir_block_cf_tree_next(block))
2212 /* If the following CF node is an if, this function returns that if.
2213 * Otherwise, it returns NULL.
2215 nir_if
*nir_block_get_following_if(nir_block
*block
);
2217 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2219 void nir_index_local_regs(nir_function_impl
*impl
);
2220 void nir_index_global_regs(nir_shader
*shader
);
2221 void nir_index_ssa_defs(nir_function_impl
*impl
);
2222 unsigned nir_index_instrs(nir_function_impl
*impl
);
2224 void nir_index_blocks(nir_function_impl
*impl
);
2226 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2227 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2228 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2230 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2231 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2232 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2233 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2236 void nir_validate_shader(nir_shader
*shader
);
2237 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2238 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2240 #include "util/debug.h"
2242 should_clone_nir(void)
2244 static int should_clone
= -1;
2245 if (should_clone
< 0)
2246 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2248 return should_clone
;
2251 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2252 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2253 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2254 static inline bool should_clone_nir(void) { return false; }
2257 #define _PASS(nir, do_pass) do { \
2259 nir_validate_shader(nir); \
2260 if (should_clone_nir()) { \
2261 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2267 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2268 nir_metadata_set_validation_flag(nir); \
2269 if (pass(nir, ##__VA_ARGS__)) { \
2271 nir_metadata_check_validation_flag(nir); \
2275 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2276 pass(nir, ##__VA_ARGS__); \
2279 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2280 void nir_calc_dominance(nir_shader
*shader
);
2282 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2283 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2285 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2286 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2288 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2289 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2291 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2292 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2294 int nir_gs_count_vertices(const nir_shader
*shader
);
2296 bool nir_split_var_copies(nir_shader
*shader
);
2298 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2299 bool nir_lower_returns(nir_shader
*shader
);
2301 bool nir_inline_functions(nir_shader
*shader
);
2303 bool nir_propagate_invariant(nir_shader
*shader
);
2305 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2306 void nir_lower_var_copies(nir_shader
*shader
);
2308 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2310 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2312 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2314 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2315 nir_function_impl
*entrypoint
,
2316 bool outputs
, bool inputs
);
2318 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2320 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2321 int (*type_size
)(const struct glsl_type
*));
2324 /* If set, this forces all non-flat fragment shader inputs to be
2325 * interpolated as if with the "sample" qualifier. This requires
2326 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2328 nir_lower_io_force_sample_interpolation
= (1 << 1),
2329 } nir_lower_io_options
;
2330 void nir_lower_io(nir_shader
*shader
,
2331 nir_variable_mode modes
,
2332 int (*type_size
)(const struct glsl_type
*),
2333 nir_lower_io_options
);
2334 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2335 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2337 bool nir_is_per_vertex_io(nir_variable
*var
, gl_shader_stage stage
);
2339 void nir_lower_io_types(nir_shader
*shader
);
2340 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2342 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2343 bool nir_lower_constant_initializers(nir_shader
*shader
,
2344 nir_variable_mode modes
);
2346 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2347 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2348 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2349 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2351 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2352 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2354 void nir_lower_samplers(nir_shader
*shader
,
2355 const struct gl_shader_program
*shader_program
);
2357 bool nir_lower_system_values(nir_shader
*shader
);
2359 typedef struct nir_lower_tex_options
{
2361 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2362 * sampler types a texture projector is lowered.
2367 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2369 bool lower_txf_offset
;
2372 * If true, lower away nir_tex_src_offset for all rect textures.
2374 bool lower_rect_offset
;
2377 * If true, lower rect textures to 2D, using txs to fetch the
2378 * texture dimensions and dividing the texture coords by the
2379 * texture dims to normalize.
2384 * If true, convert yuv to rgb.
2386 unsigned lower_y_uv_external
;
2387 unsigned lower_y_u_v_external
;
2388 unsigned lower_yx_xuxv_external
;
2391 * To emulate certain texture wrap modes, this can be used
2392 * to saturate the specified tex coord to [0.0, 1.0]. The
2393 * bits are according to sampler #, ie. if, for example:
2395 * (conf->saturate_s & (1 << n))
2397 * is true, then the s coord for sampler n is saturated.
2399 * Note that clamping must happen *after* projector lowering
2400 * so any projected texture sample instruction with a clamped
2401 * coordinate gets automatically lowered, regardless of the
2402 * 'lower_txp' setting.
2404 unsigned saturate_s
;
2405 unsigned saturate_t
;
2406 unsigned saturate_r
;
2408 /* Bitmask of textures that need swizzling.
2410 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2411 * swizzles[texture_index] is applied to the result of the texturing
2414 unsigned swizzle_result
;
2416 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2417 * while 4 and 5 represent 0 and 1 respectively.
2419 uint8_t swizzles
[32][4];
2422 * Bitmap of textures that need srgb to linear conversion. If
2423 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2424 * of the texture are lowered to linear.
2426 unsigned lower_srgb
;
2427 } nir_lower_tex_options
;
2429 bool nir_lower_tex(nir_shader
*shader
,
2430 const nir_lower_tex_options
*options
);
2432 bool nir_lower_idiv(nir_shader
*shader
);
2434 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2435 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2436 void nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2438 void nir_lower_two_sided_color(nir_shader
*shader
);
2440 void nir_lower_clamp_color_outputs(nir_shader
*shader
);
2442 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2444 typedef struct nir_lower_wpos_ytransform_options
{
2445 int state_tokens
[5];
2446 bool fs_coord_origin_upper_left
:1;
2447 bool fs_coord_origin_lower_left
:1;
2448 bool fs_coord_pixel_center_integer
:1;
2449 bool fs_coord_pixel_center_half_integer
:1;
2450 } nir_lower_wpos_ytransform_options
;
2452 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2453 const nir_lower_wpos_ytransform_options
*options
);
2454 bool nir_lower_wpos_center(nir_shader
*shader
);
2456 typedef struct nir_lower_drawpixels_options
{
2457 int texcoord_state_tokens
[5];
2458 int scale_state_tokens
[5];
2459 int bias_state_tokens
[5];
2460 unsigned drawpix_sampler
;
2461 unsigned pixelmap_sampler
;
2463 bool scale_and_bias
:1;
2464 } nir_lower_drawpixels_options
;
2466 void nir_lower_drawpixels(nir_shader
*shader
,
2467 const nir_lower_drawpixels_options
*options
);
2469 typedef struct nir_lower_bitmap_options
{
2472 } nir_lower_bitmap_options
;
2474 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2476 void nir_lower_atomics(nir_shader
*shader
,
2477 const struct gl_shader_program
*shader_program
);
2478 void nir_lower_to_source_mods(nir_shader
*shader
);
2480 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2483 nir_lower_drcp
= (1 << 0),
2484 nir_lower_dsqrt
= (1 << 1),
2485 nir_lower_drsq
= (1 << 2),
2486 nir_lower_dtrunc
= (1 << 3),
2487 nir_lower_dfloor
= (1 << 4),
2488 nir_lower_dceil
= (1 << 5),
2489 nir_lower_dfract
= (1 << 6),
2490 nir_lower_dround_even
= (1 << 7),
2491 nir_lower_dmod
= (1 << 8)
2492 } nir_lower_doubles_options
;
2494 void nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2495 void nir_lower_double_pack(nir_shader
*shader
);
2497 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2499 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2500 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2502 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2503 void nir_convert_to_ssa(nir_shader
*shader
);
2505 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2506 bool nir_repair_ssa(nir_shader
*shader
);
2508 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2509 * registers. If false, convert all values (even those not involved in a phi
2510 * node) to registers.
2512 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2514 bool nir_opt_algebraic(nir_shader
*shader
);
2515 bool nir_opt_algebraic_late(nir_shader
*shader
);
2516 bool nir_opt_constant_folding(nir_shader
*shader
);
2518 bool nir_opt_global_to_local(nir_shader
*shader
);
2520 bool nir_copy_prop(nir_shader
*shader
);
2522 bool nir_opt_cse(nir_shader
*shader
);
2524 bool nir_opt_dce(nir_shader
*shader
);
2526 bool nir_opt_dead_cf(nir_shader
*shader
);
2528 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2530 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2532 bool nir_opt_remove_phis(nir_shader
*shader
);
2534 bool nir_opt_undef(nir_shader
*shader
);
2536 bool nir_opt_conditional_discard(nir_shader
*shader
);
2538 void nir_sweep(nir_shader
*shader
);
2540 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2541 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);