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"
42 #include "nir_opcodes.h"
49 struct gl_shader_program
;
52 #define NIR_TRUE (~0u)
54 /** Defines a cast function
56 * This macro defines a cast function from in_type to out_type where
57 * out_type is some structure type that contains a field of type out_type.
59 * Note that you have to be a bit careful as the generated cast function
62 #define NIR_DEFINE_CAST(name, in_type, out_type, field) \
63 static inline out_type * \
64 name(const in_type *parent) \
66 return exec_node_data(out_type, parent, field); \
75 * Description of built-in state associated with a uniform
77 * \sa nir_variable::state_slots
85 nir_var_shader_in
= (1 << 0),
86 nir_var_shader_out
= (1 << 1),
87 nir_var_global
= (1 << 2),
88 nir_var_local
= (1 << 3),
89 nir_var_uniform
= (1 << 4),
90 nir_var_shader_storage
= (1 << 5),
91 nir_var_system_value
= (1 << 6),
92 nir_var_param
= (1 << 7),
93 nir_var_shared
= (1 << 8),
98 * Data stored in an nir_constant
100 union nir_constant_data
{
108 typedef struct nir_constant
{
110 * Value of the constant.
112 * The field used to back the values supplied by the constant is determined
113 * by the type associated with the \c nir_variable. Constants may be
114 * scalars, vectors, or matrices.
116 union nir_constant_data value
;
118 /* we could get this from the var->type but makes clone *much* easier to
119 * not have to care about the type.
121 unsigned num_elements
;
123 /* Array elements / Structure Fields */
124 struct nir_constant
**elements
;
128 * \brief Layout qualifiers for gl_FragDepth.
130 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
131 * with a layout qualifier.
134 nir_depth_layout_none
, /**< No depth layout is specified. */
135 nir_depth_layout_any
,
136 nir_depth_layout_greater
,
137 nir_depth_layout_less
,
138 nir_depth_layout_unchanged
142 * Either a uniform, global variable, shader input, or shader output. Based on
143 * ir_variable - it should be easy to translate between the two.
146 typedef struct nir_variable
{
147 struct exec_node node
;
150 * Declared type of the variable
152 const struct glsl_type
*type
;
155 * Declared name of the variable
159 struct nir_variable_data
{
161 * Storage class of the variable.
163 * \sa nir_variable_mode
165 nir_variable_mode mode
;
168 * Is the variable read-only?
170 * This is set for variables declared as \c const, shader inputs,
173 unsigned read_only
:1;
177 unsigned invariant
:1;
180 * Interpolation mode for shader inputs / outputs
182 * \sa glsl_interp_mode
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 * Whether this is a fragment shader output implicitly initialized with
231 * the previous contents of the specified render target at the
232 * framebuffer location corresponding to this shader invocation.
234 unsigned fb_fetch_output
:1;
237 * \brief Layout qualifier for gl_FragDepth.
239 * This is not equal to \c ir_depth_layout_none if and only if this
240 * variable is \c gl_FragDepth and a layout qualifier is specified.
242 nir_depth_layout depth_layout
;
245 * Storage location of the base of this variable
247 * The precise meaning of this field depends on the nature of the variable.
249 * - Vertex shader input: one of the values from \c gl_vert_attrib.
250 * - Vertex shader output: one of the values from \c gl_varying_slot.
251 * - Geometry shader input: one of the values from \c gl_varying_slot.
252 * - Geometry shader output: one of the values from \c gl_varying_slot.
253 * - Fragment shader input: one of the values from \c gl_varying_slot.
254 * - Fragment shader output: one of the values from \c gl_frag_result.
255 * - Uniforms: Per-stage uniform slot number for default uniform block.
256 * - Uniforms: Index within the uniform block definition for UBO members.
257 * - Non-UBO Uniforms: uniform slot number.
258 * - Other: This field is not currently used.
260 * If the variable is a uniform, shader input, or shader output, and the
261 * slot has not been assigned, the value will be -1.
266 * The actual location of the variable in the IR. Only valid for inputs
269 unsigned int driver_location
;
272 * output index for dual source blending.
277 * Descriptor set binding for sampler or UBO.
282 * Initial binding point for a sampler or UBO.
284 * For array types, this represents the binding point for the first element.
289 * Location an atomic counter is stored at.
294 * ARB_shader_image_load_store qualifiers.
297 bool read_only
; /**< "readonly" qualifier. */
298 bool write_only
; /**< "writeonly" qualifier. */
303 /** Image internal format if specified explicitly, otherwise GL_NONE. */
308 * Highest element accessed with a constant expression array index
310 * Not used for non-array variables.
312 unsigned max_array_access
;
317 * Built-in state that backs this uniform
319 * Once set at variable creation, \c state_slots must remain invariant.
320 * This is because, ideally, this array would be shared by all clones of
321 * this variable in the IR tree. In other words, we'd really like for it
322 * to be a fly-weight.
324 * If the variable is not a uniform, \c num_state_slots will be zero and
325 * \c state_slots will be \c NULL.
328 unsigned num_state_slots
; /**< Number of state slots used */
329 nir_state_slot
*state_slots
; /**< State descriptors. */
333 * Constant expression assigned in the initializer of the variable
335 nir_constant
*constant_initializer
;
338 * For variables that are in an interface block or are an instance of an
339 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
341 * \sa ir_variable::location
343 const struct glsl_type
*interface_type
;
346 #define nir_foreach_variable(var, var_list) \
347 foreach_list_typed(nir_variable, var, node, var_list)
349 #define nir_foreach_variable_safe(var, var_list) \
350 foreach_list_typed_safe(nir_variable, var, node, var_list)
353 nir_variable_is_global(const nir_variable
*var
)
355 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
358 typedef struct nir_register
{
359 struct exec_node node
;
361 unsigned num_components
; /** < number of vector components */
362 unsigned num_array_elems
; /** < size of array (0 for no array) */
364 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
367 /** generic register index. */
370 /** only for debug purposes, can be NULL */
373 /** whether this register is local (per-function) or global (per-shader) */
377 * If this flag is set to true, then accessing channels >= num_components
378 * is well-defined, and simply spills over to the next array element. This
379 * is useful for backends that can do per-component accessing, in
380 * particular scalar backends. By setting this flag and making
381 * num_components equal to 1, structures can be packed tightly into
382 * registers and then registers can be accessed per-component to get to
383 * each structure member, even if it crosses vec4 boundaries.
387 /** set of nir_src's where this register is used (read from) */
388 struct list_head uses
;
390 /** set of nir_dest's where this register is defined (written to) */
391 struct list_head defs
;
393 /** set of nir_if's where this register is used as a condition */
394 struct list_head if_uses
;
401 nir_instr_type_intrinsic
,
402 nir_instr_type_load_const
,
404 nir_instr_type_ssa_undef
,
406 nir_instr_type_parallel_copy
,
409 typedef struct nir_instr
{
410 struct exec_node node
;
412 struct nir_block
*block
;
414 /** generic instruction index. */
417 /* A temporary for optimization and analysis passes to use for storing
418 * flags. For instance, DCE uses this to store the "dead/live" info.
423 static inline nir_instr
*
424 nir_instr_next(nir_instr
*instr
)
426 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
427 if (exec_node_is_tail_sentinel(next
))
430 return exec_node_data(nir_instr
, next
, node
);
433 static inline nir_instr
*
434 nir_instr_prev(nir_instr
*instr
)
436 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
437 if (exec_node_is_head_sentinel(prev
))
440 return exec_node_data(nir_instr
, prev
, node
);
444 nir_instr_is_first(nir_instr
*instr
)
446 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
450 nir_instr_is_last(nir_instr
*instr
)
452 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
455 typedef struct nir_ssa_def
{
456 /** for debugging only, can be NULL */
459 /** generic SSA definition index. */
462 /** Index into the live_in and live_out bitfields */
465 nir_instr
*parent_instr
;
467 /** set of nir_instr's where this register is used (read from) */
468 struct list_head uses
;
470 /** set of nir_if's where this register is used as a condition */
471 struct list_head if_uses
;
473 uint8_t num_components
;
475 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
483 struct nir_src
*indirect
; /** < NULL for no indirect offset */
484 unsigned base_offset
;
486 /* TODO use-def chain goes here */
490 nir_instr
*parent_instr
;
491 struct list_head def_link
;
494 struct nir_src
*indirect
; /** < NULL for no indirect offset */
495 unsigned base_offset
;
497 /* TODO def-use chain goes here */
502 typedef struct nir_src
{
504 nir_instr
*parent_instr
;
505 struct nir_if
*parent_if
;
508 struct list_head use_link
;
518 static inline nir_src
521 nir_src src
= { { NULL
} };
525 #define NIR_SRC_INIT nir_src_init()
527 #define nir_foreach_use(src, reg_or_ssa_def) \
528 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
530 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
531 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
533 #define nir_foreach_if_use(src, reg_or_ssa_def) \
534 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
536 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
537 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
548 static inline nir_dest
551 nir_dest dest
= { { { NULL
} } };
555 #define NIR_DEST_INIT nir_dest_init()
557 #define nir_foreach_def(dest, reg) \
558 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
560 #define nir_foreach_def_safe(dest, reg) \
561 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
563 static inline nir_src
564 nir_src_for_ssa(nir_ssa_def
*def
)
566 nir_src src
= NIR_SRC_INIT
;
574 static inline nir_src
575 nir_src_for_reg(nir_register
*reg
)
577 nir_src src
= NIR_SRC_INIT
;
581 src
.reg
.indirect
= NULL
;
582 src
.reg
.base_offset
= 0;
587 static inline nir_dest
588 nir_dest_for_reg(nir_register
*reg
)
590 nir_dest dest
= NIR_DEST_INIT
;
597 static inline unsigned
598 nir_src_bit_size(nir_src src
)
600 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
603 static inline unsigned
604 nir_dest_bit_size(nir_dest dest
)
606 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
609 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
610 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
616 * \name input modifiers
620 * For inputs interpreted as floating point, flips the sign bit. For
621 * inputs interpreted as integers, performs the two's complement negation.
626 * Clears the sign bit for floating point values, and computes the integer
627 * absolute value for integers. Note that the negate modifier acts after
628 * the absolute value modifier, therefore if both are set then all inputs
629 * will become negative.
635 * For each input component, says which component of the register it is
636 * chosen from. Note that which elements of the swizzle are used and which
637 * are ignored are based on the write mask for most opcodes - for example,
638 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
639 * a swizzle of {2, x, 1, 0} where x means "don't care."
648 * \name saturate output modifier
650 * Only valid for opcodes that output floating-point numbers. Clamps the
651 * output to between 0.0 and 1.0 inclusive.
656 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
660 nir_type_invalid
= 0, /* Not a valid type */
665 nir_type_bool32
= 32 | nir_type_bool
,
666 nir_type_int8
= 8 | nir_type_int
,
667 nir_type_int16
= 16 | nir_type_int
,
668 nir_type_int32
= 32 | nir_type_int
,
669 nir_type_int64
= 64 | nir_type_int
,
670 nir_type_uint8
= 8 | nir_type_uint
,
671 nir_type_uint16
= 16 | nir_type_uint
,
672 nir_type_uint32
= 32 | nir_type_uint
,
673 nir_type_uint64
= 64 | nir_type_uint
,
674 nir_type_float16
= 16 | nir_type_float
,
675 nir_type_float32
= 32 | nir_type_float
,
676 nir_type_float64
= 64 | nir_type_float
,
679 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
680 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
682 static inline unsigned
683 nir_alu_type_get_type_size(nir_alu_type type
)
685 return type
& NIR_ALU_TYPE_SIZE_MASK
;
688 static inline unsigned
689 nir_alu_type_get_base_type(nir_alu_type type
)
691 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
695 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
696 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
697 } nir_op_algebraic_property
;
705 * The number of components in the output
707 * If non-zero, this is the size of the output and input sizes are
708 * explicitly given; swizzle and writemask are still in effect, but if
709 * the output component is masked out, then the input component may
712 * If zero, the opcode acts in the standard, per-component manner; the
713 * operation is performed on each component (except the ones that are
714 * masked out) with the input being taken from the input swizzle for
717 * The size of some of the inputs may be given (i.e. non-zero) even
718 * though output_size is zero; in that case, the inputs with a zero
719 * size act per-component, while the inputs with non-zero size don't.
721 unsigned output_size
;
724 * The type of vector that the instruction outputs. Note that the
725 * staurate modifier is only allowed on outputs with the float type.
728 nir_alu_type output_type
;
731 * The number of components in each input
733 unsigned input_sizes
[4];
736 * The type of vector that each input takes. Note that negate and
737 * absolute value are only allowed on inputs with int or float type and
738 * behave differently on the two.
740 nir_alu_type input_types
[4];
742 nir_op_algebraic_property algebraic_properties
;
745 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
747 typedef struct nir_alu_instr
{
751 /** Indicates that this ALU instruction generates an exact value
753 * This is kind of a mixture of GLSL "precise" and "invariant" and not
754 * really equivalent to either. This indicates that the value generated by
755 * this operation is high-precision and any code transformations that touch
756 * it must ensure that the resulting value is bit-for-bit identical to the
765 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
766 nir_alu_instr
*instr
);
767 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
768 nir_alu_instr
*instr
);
770 /* is this source channel used? */
772 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
774 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
775 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
777 return (instr
->dest
.write_mask
>> channel
) & 1;
781 * For instructions whose destinations are SSA, get the number of channels
784 static inline unsigned
785 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
787 assert(instr
->dest
.dest
.is_ssa
);
789 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
790 return nir_op_infos
[instr
->op
].input_sizes
[src
];
792 return instr
->dest
.dest
.ssa
.num_components
;
795 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
796 unsigned src1
, unsigned src2
);
800 nir_deref_type_array
,
801 nir_deref_type_struct
804 typedef struct nir_deref
{
805 nir_deref_type deref_type
;
806 struct nir_deref
*child
;
807 const struct glsl_type
*type
;
816 /* This enum describes how the array is referenced. If the deref is
817 * direct then the base_offset is used. If the deref is indirect then
818 * offset is given by base_offset + indirect. If the deref is a wildcard
819 * then the deref refers to all of the elements of the array at the same
820 * time. Wildcard dereferences are only ever allowed in copy_var
821 * intrinsics and the source and destination derefs must have matching
825 nir_deref_array_type_direct
,
826 nir_deref_array_type_indirect
,
827 nir_deref_array_type_wildcard
,
828 } nir_deref_array_type
;
833 nir_deref_array_type deref_array_type
;
834 unsigned base_offset
;
844 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
845 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
846 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
848 /* Returns the last deref in the chain. */
849 static inline nir_deref
*
850 nir_deref_tail(nir_deref
*deref
)
853 deref
= deref
->child
;
861 nir_deref_var
**params
;
862 nir_deref_var
*return_deref
;
864 struct nir_function
*callee
;
867 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
868 num_variables, num_indices, idx0, idx1, idx2, flags) \
869 nir_intrinsic_##name,
871 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
874 #include "nir_intrinsics.h"
875 nir_num_intrinsics
= nir_last_intrinsic
+ 1
878 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
880 /** Represents an intrinsic
882 * An intrinsic is an instruction type for handling things that are
883 * more-or-less regular operations but don't just consume and produce SSA
884 * values like ALU operations do. Intrinsics are not for things that have
885 * special semantic meaning such as phi nodes and parallel copies.
886 * Examples of intrinsics include variable load/store operations, system
887 * value loads, and the like. Even though texturing more-or-less falls
888 * under this category, texturing is its own instruction type because
889 * trying to represent texturing with intrinsics would lead to a
890 * combinatorial explosion of intrinsic opcodes.
892 * By having a single instruction type for handling a lot of different
893 * cases, optimization passes can look for intrinsics and, for the most
894 * part, completely ignore them. Each intrinsic type also has a few
895 * possible flags that govern whether or not they can be reordered or
896 * eliminated. That way passes like dead code elimination can still work
897 * on intrisics without understanding the meaning of each.
899 * Each intrinsic has some number of constant indices, some number of
900 * variables, and some number of sources. What these sources, variables,
901 * and indices mean depends on the intrinsic and is documented with the
902 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
903 * instructions are the only types of instruction that can operate on
909 nir_intrinsic_op intrinsic
;
913 /** number of components if this is a vectorized intrinsic
915 * Similarly to ALU operations, some intrinsics are vectorized.
916 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
917 * For vectorized intrinsics, the num_components field specifies the
918 * number of destination components and the number of source components
919 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
921 uint8_t num_components
;
923 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
925 nir_deref_var
*variables
[2];
928 } nir_intrinsic_instr
;
931 * \name NIR intrinsics semantic flags
933 * information about what the compiler can do with the intrinsics.
935 * \sa nir_intrinsic_info::flags
939 * whether the intrinsic can be safely eliminated if none of its output
940 * value is not being used.
942 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
945 * Whether the intrinsic can be reordered with respect to any other
946 * intrinsic, i.e. whether the only reordering dependencies of the
947 * intrinsic are due to the register reads/writes.
949 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
950 } nir_intrinsic_semantic_flag
;
953 * \name NIR intrinsics const-index flag
955 * Indicates the usage of a const_index slot.
957 * \sa nir_intrinsic_info::index_map
961 * Generally instructions that take a offset src argument, can encode
962 * a constant 'base' value which is added to the offset.
964 NIR_INTRINSIC_BASE
= 1,
967 * For store instructions, a writemask for the store.
969 NIR_INTRINSIC_WRMASK
= 2,
972 * The stream-id for GS emit_vertex/end_primitive intrinsics.
974 NIR_INTRINSIC_STREAM_ID
= 3,
977 * The clip-plane id for load_user_clip_plane intrinsic.
979 NIR_INTRINSIC_UCP_ID
= 4,
982 * The amount of data, starting from BASE, that this instruction may
983 * access. This is used to provide bounds if the offset is not constant.
985 NIR_INTRINSIC_RANGE
= 5,
988 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
990 NIR_INTRINSIC_DESC_SET
= 6,
993 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
995 NIR_INTRINSIC_BINDING
= 7,
1000 NIR_INTRINSIC_COMPONENT
= 8,
1003 * Interpolation mode (only meaningful for FS inputs).
1005 NIR_INTRINSIC_INTERP_MODE
= 9,
1007 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1009 } nir_intrinsic_index_flag
;
1011 #define NIR_INTRINSIC_MAX_INPUTS 4
1016 unsigned num_srcs
; /** < number of register/SSA inputs */
1018 /** number of components of each input register
1020 * If this value is 0, the number of components is given by the
1021 * num_components field of nir_intrinsic_instr.
1023 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1027 /** number of components of the output register
1029 * If this value is 0, the number of components is given by the
1030 * num_components field of nir_intrinsic_instr.
1032 unsigned dest_components
;
1034 /** the number of inputs/outputs that are variables */
1035 unsigned num_variables
;
1037 /** the number of constant indices used by the intrinsic */
1038 unsigned num_indices
;
1040 /** indicates the usage of intr->const_index[n] */
1041 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1043 /** semantic flags for calls to this intrinsic */
1044 nir_intrinsic_semantic_flag flags
;
1045 } nir_intrinsic_info
;
1047 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1050 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1051 static inline type \
1052 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
1054 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1055 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1056 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1058 static inline void \
1059 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1061 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1062 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1063 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1066 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1067 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1068 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1069 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1070 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1071 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1072 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1073 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1074 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1077 * \group texture information
1079 * This gives semantic information about textures which is useful to the
1080 * frontend, the backend, and lowering passes, but not the optimizer.
1085 nir_tex_src_projector
,
1086 nir_tex_src_comparitor
, /* shadow comparitor */
1090 nir_tex_src_ms_index
, /* MSAA sample index */
1091 nir_tex_src_ms_mcs
, /* MSAA compression value */
1094 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1095 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1096 nir_tex_src_plane
, /* < selects plane for planar textures */
1097 nir_num_tex_src_types
1102 nir_tex_src_type src_type
;
1106 nir_texop_tex
, /**< Regular texture look-up */
1107 nir_texop_txb
, /**< Texture look-up with LOD bias */
1108 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1109 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1110 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1111 nir_texop_txf_ms
, /**< Multisample texture fetch */
1112 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1113 nir_texop_txs
, /**< Texture size */
1114 nir_texop_lod
, /**< Texture lod query */
1115 nir_texop_tg4
, /**< Texture gather */
1116 nir_texop_query_levels
, /**< Texture levels query */
1117 nir_texop_texture_samples
, /**< Texture samples query */
1118 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1126 enum glsl_sampler_dim sampler_dim
;
1127 nir_alu_type dest_type
;
1132 unsigned num_srcs
, coord_components
;
1133 bool is_array
, is_shadow
;
1136 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1137 * components or the new-style shadow that outputs 1 component.
1139 bool is_new_style_shadow
;
1141 /* gather component selector */
1142 unsigned component
: 2;
1144 /** The texture index
1146 * If this texture instruction has a nir_tex_src_texture_offset source,
1147 * then the texture index is given by texture_index + texture_offset.
1149 unsigned texture_index
;
1151 /** The size of the texture array or 0 if it's not an array */
1152 unsigned texture_array_size
;
1154 /** The texture deref
1156 * If this is null, use texture_index instead.
1158 nir_deref_var
*texture
;
1160 /** The sampler index
1162 * The following operations do not require a sampler and, as such, this
1163 * field should be ignored:
1165 * - nir_texop_txf_ms
1169 * - nir_texop_query_levels
1170 * - nir_texop_texture_samples
1171 * - nir_texop_samples_identical
1173 * If this texture instruction has a nir_tex_src_sampler_offset source,
1174 * then the sampler index is given by sampler_index + sampler_offset.
1176 unsigned sampler_index
;
1178 /** The sampler deref
1180 * If this is null, use sampler_index instead.
1182 nir_deref_var
*sampler
;
1185 static inline unsigned
1186 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1188 switch (instr
->op
) {
1189 case nir_texop_txs
: {
1191 switch (instr
->sampler_dim
) {
1192 case GLSL_SAMPLER_DIM_1D
:
1193 case GLSL_SAMPLER_DIM_BUF
:
1196 case GLSL_SAMPLER_DIM_2D
:
1197 case GLSL_SAMPLER_DIM_CUBE
:
1198 case GLSL_SAMPLER_DIM_MS
:
1199 case GLSL_SAMPLER_DIM_RECT
:
1200 case GLSL_SAMPLER_DIM_EXTERNAL
:
1203 case GLSL_SAMPLER_DIM_3D
:
1207 unreachable("not reached");
1209 if (instr
->is_array
)
1217 case nir_texop_texture_samples
:
1218 case nir_texop_query_levels
:
1219 case nir_texop_samples_identical
:
1223 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1230 /* Returns true if this texture operation queries something about the texture
1231 * rather than actually sampling it.
1234 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1236 switch (instr
->op
) {
1239 case nir_texop_texture_samples
:
1240 case nir_texop_query_levels
:
1241 case nir_texop_txf_ms_mcs
:
1248 case nir_texop_txf_ms
:
1252 unreachable("Invalid texture opcode");
1256 static inline nir_alu_type
1257 nir_tex_instr_src_type(nir_tex_instr
*instr
, unsigned src
)
1259 switch (instr
->src
[src
].src_type
) {
1260 case nir_tex_src_coord
:
1261 switch (instr
->op
) {
1263 case nir_texop_txf_ms
:
1264 case nir_texop_txf_ms_mcs
:
1265 case nir_texop_samples_identical
:
1266 return nir_type_int
;
1269 return nir_type_float
;
1272 case nir_tex_src_lod
:
1273 switch (instr
->op
) {
1276 return nir_type_int
;
1279 return nir_type_float
;
1282 case nir_tex_src_projector
:
1283 case nir_tex_src_comparitor
:
1284 case nir_tex_src_bias
:
1285 case nir_tex_src_ddx
:
1286 case nir_tex_src_ddy
:
1287 return nir_type_float
;
1289 case nir_tex_src_offset
:
1290 case nir_tex_src_ms_index
:
1291 case nir_tex_src_texture_offset
:
1292 case nir_tex_src_sampler_offset
:
1293 return nir_type_int
;
1296 unreachable("Invalid texture source type");
1300 static inline unsigned
1301 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1303 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1304 return instr
->coord_components
;
1306 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1307 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1310 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1311 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1312 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1313 if (instr
->is_array
)
1314 return instr
->coord_components
- 1;
1316 return instr
->coord_components
;
1323 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1325 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1326 if (instr
->src
[i
].src_type
== type
)
1344 nir_const_value value
;
1347 } nir_load_const_instr
;
1360 /* creates a new SSA variable in an undefined state */
1365 } nir_ssa_undef_instr
;
1368 struct exec_node node
;
1370 /* The predecessor block corresponding to this source */
1371 struct nir_block
*pred
;
1376 #define nir_foreach_phi_src(phi_src, phi) \
1377 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1378 #define nir_foreach_phi_src_safe(phi_src, phi) \
1379 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1384 struct exec_list srcs
; /** < list of nir_phi_src */
1390 struct exec_node node
;
1393 } nir_parallel_copy_entry
;
1395 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1396 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1401 /* A list of nir_parallel_copy_entry's. The sources of all of the
1402 * entries are copied to the corresponding destinations "in parallel".
1403 * In other words, if we have two entries: a -> b and b -> a, the values
1406 struct exec_list entries
;
1407 } nir_parallel_copy_instr
;
1409 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1410 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1411 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1412 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1413 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1414 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1415 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1416 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1417 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1418 nir_parallel_copy_instr
, instr
)
1423 * Control flow consists of a tree of control flow nodes, which include
1424 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1425 * instructions that always run start-to-finish. Each basic block also keeps
1426 * track of its successors (blocks which may run immediately after the current
1427 * block) and predecessors (blocks which could have run immediately before the
1428 * current block). Each function also has a start block and an end block which
1429 * all return statements point to (which is always empty). Together, all the
1430 * blocks with their predecessors and successors make up the control flow
1431 * graph (CFG) of the function. There are helpers that modify the tree of
1432 * control flow nodes while modifying the CFG appropriately; these should be
1433 * used instead of modifying the tree directly.
1440 nir_cf_node_function
1443 typedef struct nir_cf_node
{
1444 struct exec_node node
;
1445 nir_cf_node_type type
;
1446 struct nir_cf_node
*parent
;
1449 typedef struct nir_block
{
1450 nir_cf_node cf_node
;
1452 struct exec_list instr_list
; /** < list of nir_instr */
1454 /** generic block index; generated by nir_index_blocks */
1458 * Each block can only have up to 2 successors, so we put them in a simple
1459 * array - no need for anything more complicated.
1461 struct nir_block
*successors
[2];
1463 /* Set of nir_block predecessors in the CFG */
1464 struct set
*predecessors
;
1467 * this node's immediate dominator in the dominance tree - set to NULL for
1470 struct nir_block
*imm_dom
;
1472 /* This node's children in the dominance tree */
1473 unsigned num_dom_children
;
1474 struct nir_block
**dom_children
;
1476 /* Set of nir_block's on the dominance frontier of this block */
1477 struct set
*dom_frontier
;
1480 * These two indices have the property that dom_{pre,post}_index for each
1481 * child of this block in the dominance tree will always be between
1482 * dom_pre_index and dom_post_index for this block, which makes testing if
1483 * a given block is dominated by another block an O(1) operation.
1485 unsigned dom_pre_index
, dom_post_index
;
1487 /* live in and out for this block; used for liveness analysis */
1488 BITSET_WORD
*live_in
;
1489 BITSET_WORD
*live_out
;
1492 static inline nir_instr
*
1493 nir_block_first_instr(nir_block
*block
)
1495 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1496 return exec_node_data(nir_instr
, head
, node
);
1499 static inline nir_instr
*
1500 nir_block_last_instr(nir_block
*block
)
1502 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1503 return exec_node_data(nir_instr
, tail
, node
);
1506 #define nir_foreach_instr(instr, block) \
1507 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1508 #define nir_foreach_instr_reverse(instr, block) \
1509 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1510 #define nir_foreach_instr_safe(instr, block) \
1511 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1512 #define nir_foreach_instr_reverse_safe(instr, block) \
1513 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1515 typedef struct nir_if
{
1516 nir_cf_node cf_node
;
1519 struct exec_list then_list
; /** < list of nir_cf_node */
1520 struct exec_list else_list
; /** < list of nir_cf_node */
1523 static inline nir_cf_node
*
1524 nir_if_first_then_node(nir_if
*if_stmt
)
1526 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1527 return exec_node_data(nir_cf_node
, head
, node
);
1530 static inline nir_cf_node
*
1531 nir_if_last_then_node(nir_if
*if_stmt
)
1533 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1534 return exec_node_data(nir_cf_node
, tail
, node
);
1537 static inline nir_cf_node
*
1538 nir_if_first_else_node(nir_if
*if_stmt
)
1540 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1541 return exec_node_data(nir_cf_node
, head
, node
);
1544 static inline nir_cf_node
*
1545 nir_if_last_else_node(nir_if
*if_stmt
)
1547 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1548 return exec_node_data(nir_cf_node
, tail
, node
);
1552 nir_cf_node cf_node
;
1554 struct exec_list body
; /** < list of nir_cf_node */
1557 static inline nir_cf_node
*
1558 nir_loop_first_cf_node(nir_loop
*loop
)
1560 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1563 static inline nir_cf_node
*
1564 nir_loop_last_cf_node(nir_loop
*loop
)
1566 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1570 * Various bits of metadata that can may be created or required by
1571 * optimization and analysis passes
1574 nir_metadata_none
= 0x0,
1575 nir_metadata_block_index
= 0x1,
1576 nir_metadata_dominance
= 0x2,
1577 nir_metadata_live_ssa_defs
= 0x4,
1578 nir_metadata_not_properly_reset
= 0x8,
1582 nir_cf_node cf_node
;
1584 /** pointer to the function of which this is an implementation */
1585 struct nir_function
*function
;
1587 struct exec_list body
; /** < list of nir_cf_node */
1589 nir_block
*end_block
;
1591 /** list for all local variables in the function */
1592 struct exec_list locals
;
1594 /** array of variables used as parameters */
1595 unsigned num_params
;
1596 nir_variable
**params
;
1598 /** variable used to hold the result of the function */
1599 nir_variable
*return_var
;
1601 /** list of local registers in the function */
1602 struct exec_list registers
;
1604 /** next available local register index */
1607 /** next available SSA value index */
1610 /* total number of basic blocks, only valid when block_index_dirty = false */
1611 unsigned num_blocks
;
1613 nir_metadata valid_metadata
;
1614 } nir_function_impl
;
1616 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1617 nir_start_block(nir_function_impl
*impl
)
1619 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1622 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1623 nir_impl_last_block(nir_function_impl
*impl
)
1625 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1628 static inline nir_cf_node
*
1629 nir_cf_node_next(nir_cf_node
*node
)
1631 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1632 if (exec_node_is_tail_sentinel(next
))
1635 return exec_node_data(nir_cf_node
, next
, node
);
1638 static inline nir_cf_node
*
1639 nir_cf_node_prev(nir_cf_node
*node
)
1641 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1642 if (exec_node_is_head_sentinel(prev
))
1645 return exec_node_data(nir_cf_node
, prev
, node
);
1649 nir_cf_node_is_first(const nir_cf_node
*node
)
1651 return exec_node_is_head_sentinel(node
->node
.prev
);
1655 nir_cf_node_is_last(const nir_cf_node
*node
)
1657 return exec_node_is_tail_sentinel(node
->node
.next
);
1660 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1661 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1662 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1663 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1668 nir_parameter_inout
,
1669 } nir_parameter_type
;
1672 nir_parameter_type param_type
;
1673 const struct glsl_type
*type
;
1676 typedef struct nir_function
{
1677 struct exec_node node
;
1680 struct nir_shader
*shader
;
1682 unsigned num_params
;
1683 nir_parameter
*params
;
1684 const struct glsl_type
*return_type
;
1686 /** The implementation of this function.
1688 * If the function is only declared and not implemented, this is NULL.
1690 nir_function_impl
*impl
;
1693 typedef struct nir_shader_compiler_options
{
1698 /** Lowers flrp when it does not support doubles */
1705 bool lower_bitfield_extract
;
1706 bool lower_bitfield_insert
;
1707 bool lower_uadd_carry
;
1708 bool lower_usub_borrow
;
1709 /** lowers fneg and ineg to fsub and isub. */
1711 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1714 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1717 /** enables rules to lower idiv by power-of-two: */
1720 /* Does the native fdot instruction replicate its result for four
1721 * components? If so, then opt_algebraic_late will turn all fdotN
1722 * instructions into fdot_replicatedN instructions.
1724 bool fdot_replicates
;
1726 /** lowers ffract to fsub+ffloor: */
1729 bool lower_pack_half_2x16
;
1730 bool lower_pack_unorm_2x16
;
1731 bool lower_pack_snorm_2x16
;
1732 bool lower_pack_unorm_4x8
;
1733 bool lower_pack_snorm_4x8
;
1734 bool lower_unpack_half_2x16
;
1735 bool lower_unpack_unorm_2x16
;
1736 bool lower_unpack_snorm_2x16
;
1737 bool lower_unpack_unorm_4x8
;
1738 bool lower_unpack_snorm_4x8
;
1740 bool lower_extract_byte
;
1741 bool lower_extract_word
;
1744 * Does the driver support real 32-bit integers? (Otherwise, integers
1745 * are simulated by floats.)
1747 bool native_integers
;
1749 /* Indicates that the driver only has zero-based vertex id */
1750 bool vertex_id_zero_based
;
1752 bool lower_cs_local_index_from_id
;
1755 * Should nir_lower_io() create load_interpolated_input intrinsics?
1757 * If not, it generates regular load_input intrinsics and interpolation
1758 * information must be inferred from the list of input nir_variables.
1760 bool use_interpolated_input_intrinsics
;
1761 } nir_shader_compiler_options
;
1763 typedef struct nir_shader_info
{
1766 /* Descriptive name provided by the client; may be NULL */
1769 /* Number of textures used by this shader */
1770 unsigned num_textures
;
1771 /* Number of uniform buffers used by this shader */
1773 /* Number of atomic buffers used by this shader */
1775 /* Number of shader storage buffers used by this shader */
1777 /* Number of images used by this shader */
1778 unsigned num_images
;
1780 /* Which inputs are actually read */
1781 uint64_t inputs_read
;
1782 /* Which inputs are actually read and are double */
1783 uint64_t double_inputs_read
;
1784 /* Which outputs are actually written */
1785 uint64_t outputs_written
;
1786 /* Which outputs are actually read */
1787 uint64_t outputs_read
;
1788 /* Which system values are actually read */
1789 uint64_t system_values_read
;
1791 /* Which patch inputs are actually read */
1792 uint32_t patch_inputs_read
;
1793 /* Which patch outputs are actually written */
1794 uint32_t patch_outputs_written
;
1796 /* Whether or not this shader ever uses textureGather() */
1797 bool uses_texture_gather
;
1799 /* Whether or not this shader uses the gl_ClipDistance output */
1800 bool uses_clip_distance_out
;
1802 /* Whether or not separate shader objects were used */
1803 bool separate_shader
;
1805 /** Was this shader linked with any transform feedback varyings? */
1806 bool has_transform_feedback_varyings
;
1810 /** The number of vertices recieves per input primitive */
1811 unsigned vertices_in
;
1813 /** The output primitive type (GL enum value) */
1814 unsigned output_primitive
;
1816 /** The maximum number of vertices the geometry shader might write. */
1817 unsigned vertices_out
;
1819 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1820 unsigned invocations
;
1822 /** Whether or not this shader uses EndPrimitive */
1823 bool uses_end_primitive
;
1825 /** Whether or not this shader uses non-zero streams */
1833 * Whether any inputs are declared with the "sample" qualifier.
1835 bool uses_sample_qualifier
;
1838 * Whether early fragment tests are enabled as defined by
1839 * ARB_shader_image_load_store.
1841 bool early_fragment_tests
;
1843 /** gl_FragDepth layout for ARB_conservative_depth. */
1844 enum gl_frag_depth_layout depth_layout
;
1848 unsigned local_size
[3];
1852 /** The number of vertices in the TCS output patch. */
1853 unsigned vertices_out
;
1858 typedef struct nir_shader
{
1859 /** list of uniforms (nir_variable) */
1860 struct exec_list uniforms
;
1862 /** list of inputs (nir_variable) */
1863 struct exec_list inputs
;
1865 /** list of outputs (nir_variable) */
1866 struct exec_list outputs
;
1868 /** list of shared compute variables (nir_variable) */
1869 struct exec_list shared
;
1871 /** Set of driver-specific options for the shader.
1873 * The memory for the options is expected to be kept in a single static
1874 * copy by the driver.
1876 const struct nir_shader_compiler_options
*options
;
1878 /** Various bits of compile-time information about a given shader */
1879 struct nir_shader_info info
;
1881 /** list of global variables in the shader (nir_variable) */
1882 struct exec_list globals
;
1884 /** list of system value variables in the shader (nir_variable) */
1885 struct exec_list system_values
;
1887 struct exec_list functions
; /** < list of nir_function */
1889 /** list of global register in the shader */
1890 struct exec_list registers
;
1892 /** next available global register index */
1896 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1899 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1901 /** The shader stage, such as MESA_SHADER_VERTEX. */
1902 gl_shader_stage stage
;
1905 static inline nir_function
*
1906 nir_shader_get_entrypoint(nir_shader
*shader
)
1908 assert(exec_list_length(&shader
->functions
) == 1);
1909 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1910 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1911 assert(func
->return_type
== glsl_void_type());
1912 assert(func
->num_params
== 0);
1916 #define nir_foreach_function(func, shader) \
1917 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1919 nir_shader
*nir_shader_create(void *mem_ctx
,
1920 gl_shader_stage stage
,
1921 const nir_shader_compiler_options
*options
);
1923 /** creates a register, including assigning it an index and adding it to the list */
1924 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1926 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1928 void nir_reg_remove(nir_register
*reg
);
1930 /** Adds a variable to the appropreate list in nir_shader */
1931 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1934 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1936 assert(var
->data
.mode
== nir_var_local
);
1937 exec_list_push_tail(&impl
->locals
, &var
->node
);
1940 /** creates a variable, sets a few defaults, and adds it to the list */
1941 nir_variable
*nir_variable_create(nir_shader
*shader
,
1942 nir_variable_mode mode
,
1943 const struct glsl_type
*type
,
1945 /** creates a local variable and adds it to the list */
1946 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1947 const struct glsl_type
*type
,
1950 /** creates a function and adds it to the shader's list of functions */
1951 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1953 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1954 /** creates a function_impl that isn't tied to any particular function */
1955 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1957 nir_block
*nir_block_create(nir_shader
*shader
);
1958 nir_if
*nir_if_create(nir_shader
*shader
);
1959 nir_loop
*nir_loop_create(nir_shader
*shader
);
1961 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1963 /** requests that the given pieces of metadata be generated */
1964 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1965 /** dirties all but the preserved metadata */
1966 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1968 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1969 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1971 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1973 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1974 unsigned num_components
,
1977 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1978 nir_intrinsic_op op
);
1980 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1981 nir_function
*callee
);
1983 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1985 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1987 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1989 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1990 unsigned num_components
,
1993 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1994 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1995 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1997 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1999 typedef bool (*nir_deref_foreach_leaf_cb
)(nir_deref_var
*deref
, void *state
);
2000 bool nir_deref_foreach_leaf(nir_deref_var
*deref
,
2001 nir_deref_foreach_leaf_cb cb
, void *state
);
2003 nir_load_const_instr
*
2004 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
2007 * NIR Cursors and Instruction Insertion API
2010 * A tiny struct representing a point to insert/extract instructions or
2011 * control flow nodes. Helps reduce the combinatorial explosion of possible
2012 * points to insert/extract.
2014 * \sa nir_control_flow.h
2017 nir_cursor_before_block
,
2018 nir_cursor_after_block
,
2019 nir_cursor_before_instr
,
2020 nir_cursor_after_instr
,
2021 } nir_cursor_option
;
2024 nir_cursor_option option
;
2031 static inline nir_block
*
2032 nir_cursor_current_block(nir_cursor cursor
)
2034 if (cursor
.option
== nir_cursor_before_instr
||
2035 cursor
.option
== nir_cursor_after_instr
) {
2036 return cursor
.instr
->block
;
2038 return cursor
.block
;
2042 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2044 static inline nir_cursor
2045 nir_before_block(nir_block
*block
)
2048 cursor
.option
= nir_cursor_before_block
;
2049 cursor
.block
= block
;
2053 static inline nir_cursor
2054 nir_after_block(nir_block
*block
)
2057 cursor
.option
= nir_cursor_after_block
;
2058 cursor
.block
= block
;
2062 static inline nir_cursor
2063 nir_before_instr(nir_instr
*instr
)
2066 cursor
.option
= nir_cursor_before_instr
;
2067 cursor
.instr
= instr
;
2071 static inline nir_cursor
2072 nir_after_instr(nir_instr
*instr
)
2075 cursor
.option
= nir_cursor_after_instr
;
2076 cursor
.instr
= instr
;
2080 static inline nir_cursor
2081 nir_after_block_before_jump(nir_block
*block
)
2083 nir_instr
*last_instr
= nir_block_last_instr(block
);
2084 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2085 return nir_before_instr(last_instr
);
2087 return nir_after_block(block
);
2091 static inline nir_cursor
2092 nir_before_cf_node(nir_cf_node
*node
)
2094 if (node
->type
== nir_cf_node_block
)
2095 return nir_before_block(nir_cf_node_as_block(node
));
2097 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2100 static inline nir_cursor
2101 nir_after_cf_node(nir_cf_node
*node
)
2103 if (node
->type
== nir_cf_node_block
)
2104 return nir_after_block(nir_cf_node_as_block(node
));
2106 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2109 static inline nir_cursor
2110 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2112 if (node
->type
== nir_cf_node_block
)
2113 return nir_after_block(nir_cf_node_as_block(node
));
2115 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2116 assert(block
->cf_node
.type
== nir_cf_node_block
);
2118 nir_foreach_instr(instr
, block
) {
2119 if (instr
->type
!= nir_instr_type_phi
)
2120 return nir_before_instr(instr
);
2122 return nir_after_block(block
);
2125 static inline nir_cursor
2126 nir_before_cf_list(struct exec_list
*cf_list
)
2128 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2129 exec_list_get_head(cf_list
), node
);
2130 return nir_before_cf_node(first_node
);
2133 static inline nir_cursor
2134 nir_after_cf_list(struct exec_list
*cf_list
)
2136 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2137 exec_list_get_tail(cf_list
), node
);
2138 return nir_after_cf_node(last_node
);
2142 * Insert a NIR instruction at the given cursor.
2144 * Note: This does not update the cursor.
2146 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2149 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2151 nir_instr_insert(nir_before_instr(instr
), before
);
2155 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2157 nir_instr_insert(nir_after_instr(instr
), after
);
2161 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2163 nir_instr_insert(nir_before_block(block
), before
);
2167 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2169 nir_instr_insert(nir_after_block(block
), after
);
2173 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2175 nir_instr_insert(nir_before_cf_node(node
), before
);
2179 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2181 nir_instr_insert(nir_after_cf_node(node
), after
);
2185 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2187 nir_instr_insert(nir_before_cf_list(list
), before
);
2191 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2193 nir_instr_insert(nir_after_cf_list(list
), after
);
2196 void nir_instr_remove(nir_instr
*instr
);
2200 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2201 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2202 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2203 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2205 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2206 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2208 nir_const_value
*nir_src_as_const_value(nir_src src
);
2209 bool nir_src_is_dynamically_uniform(nir_src src
);
2210 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2211 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2212 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2213 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2214 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2217 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2218 unsigned num_components
, unsigned bit_size
,
2220 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2221 unsigned num_components
, unsigned bit_size
,
2223 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2224 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2225 nir_instr
*after_me
);
2227 uint8_t nir_ssa_def_components_read(nir_ssa_def
*def
);
2230 * finds the next basic block in source-code order, returns NULL if there is
2234 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2236 /* Performs the opposite of nir_block_cf_tree_next() */
2238 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2240 /* Gets the first block in a CF node in source-code order */
2242 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2244 /* Gets the last block in a CF node in source-code order */
2246 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2248 /* Gets the next block after a CF node in source-code order */
2250 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2252 /* Macros for loops that visit blocks in source-code order */
2254 #define nir_foreach_block(block, impl) \
2255 for (nir_block *block = nir_start_block(impl); block != NULL; \
2256 block = nir_block_cf_tree_next(block))
2258 #define nir_foreach_block_safe(block, impl) \
2259 for (nir_block *block = nir_start_block(impl), \
2260 *next = nir_block_cf_tree_next(block); \
2262 block = next, next = nir_block_cf_tree_next(block))
2264 #define nir_foreach_block_reverse(block, impl) \
2265 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2266 block = nir_block_cf_tree_prev(block))
2268 #define nir_foreach_block_reverse_safe(block, impl) \
2269 for (nir_block *block = nir_impl_last_block(impl), \
2270 *prev = nir_block_cf_tree_prev(block); \
2272 block = prev, prev = nir_block_cf_tree_prev(block))
2274 #define nir_foreach_block_in_cf_node(block, node) \
2275 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2276 block != nir_cf_node_cf_tree_next(node); \
2277 block = nir_block_cf_tree_next(block))
2279 /* If the following CF node is an if, this function returns that if.
2280 * Otherwise, it returns NULL.
2282 nir_if
*nir_block_get_following_if(nir_block
*block
);
2284 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2286 void nir_index_local_regs(nir_function_impl
*impl
);
2287 void nir_index_global_regs(nir_shader
*shader
);
2288 void nir_index_ssa_defs(nir_function_impl
*impl
);
2289 unsigned nir_index_instrs(nir_function_impl
*impl
);
2291 void nir_index_blocks(nir_function_impl
*impl
);
2293 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2294 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2295 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2297 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2298 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2299 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2300 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2303 void nir_validate_shader(nir_shader
*shader
);
2304 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2305 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2307 #include "util/debug.h"
2309 should_clone_nir(void)
2311 static int should_clone
= -1;
2312 if (should_clone
< 0)
2313 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2315 return should_clone
;
2318 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2319 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2320 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2321 static inline bool should_clone_nir(void) { return false; }
2324 #define _PASS(nir, do_pass) do { \
2326 nir_validate_shader(nir); \
2327 if (should_clone_nir()) { \
2328 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2334 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2335 nir_metadata_set_validation_flag(nir); \
2336 if (pass(nir, ##__VA_ARGS__)) { \
2338 nir_metadata_check_validation_flag(nir); \
2342 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2343 pass(nir, ##__VA_ARGS__); \
2346 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2347 void nir_calc_dominance(nir_shader
*shader
);
2349 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2350 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2352 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2353 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2355 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2356 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2358 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2359 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2361 int nir_gs_count_vertices(const nir_shader
*shader
);
2363 bool nir_split_var_copies(nir_shader
*shader
);
2365 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2366 bool nir_lower_returns(nir_shader
*shader
);
2368 bool nir_inline_functions(nir_shader
*shader
);
2370 bool nir_propagate_invariant(nir_shader
*shader
);
2372 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2373 void nir_lower_var_copies(nir_shader
*shader
);
2375 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2377 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2379 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2381 void nir_lower_io_to_temporaries(nir_shader
*shader
, nir_function
*entrypoint
,
2382 bool outputs
, bool inputs
);
2384 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2386 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2387 unsigned base_offset
,
2388 int (*type_size
)(const struct glsl_type
*));
2390 void nir_lower_io(nir_shader
*shader
,
2391 nir_variable_mode modes
,
2392 int (*type_size
)(const struct glsl_type
*));
2393 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2394 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2396 void nir_lower_io_types(nir_shader
*shader
);
2397 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2399 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2401 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2402 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2403 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2404 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2406 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2407 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2409 void nir_lower_samplers(nir_shader
*shader
,
2410 const struct gl_shader_program
*shader_program
);
2412 bool nir_lower_system_values(nir_shader
*shader
);
2414 typedef struct nir_lower_tex_options
{
2416 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2417 * sampler types a texture projector is lowered.
2422 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2424 bool lower_txf_offset
;
2427 * If true, lower away nir_tex_src_offset for all rect textures.
2429 bool lower_rect_offset
;
2432 * If true, lower rect textures to 2D, using txs to fetch the
2433 * texture dimensions and dividing the texture coords by the
2434 * texture dims to normalize.
2439 * If true, convert yuv to rgb.
2441 unsigned lower_y_uv_external
;
2442 unsigned lower_y_u_v_external
;
2443 unsigned lower_yx_xuxv_external
;
2446 * To emulate certain texture wrap modes, this can be used
2447 * to saturate the specified tex coord to [0.0, 1.0]. The
2448 * bits are according to sampler #, ie. if, for example:
2450 * (conf->saturate_s & (1 << n))
2452 * is true, then the s coord for sampler n is saturated.
2454 * Note that clamping must happen *after* projector lowering
2455 * so any projected texture sample instruction with a clamped
2456 * coordinate gets automatically lowered, regardless of the
2457 * 'lower_txp' setting.
2459 unsigned saturate_s
;
2460 unsigned saturate_t
;
2461 unsigned saturate_r
;
2463 /* Bitmask of textures that need swizzling.
2465 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2466 * swizzles[texture_index] is applied to the result of the texturing
2469 unsigned swizzle_result
;
2471 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2472 * while 4 and 5 represent 0 and 1 respectively.
2474 uint8_t swizzles
[32][4];
2477 * Bitmap of textures that need srgb to linear conversion. If
2478 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2479 * of the texture are lowered to linear.
2481 unsigned lower_srgb
;
2482 } nir_lower_tex_options
;
2484 bool nir_lower_tex(nir_shader
*shader
,
2485 const nir_lower_tex_options
*options
);
2487 bool nir_lower_idiv(nir_shader
*shader
);
2489 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2490 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2492 void nir_lower_two_sided_color(nir_shader
*shader
);
2494 void nir_lower_clamp_color_outputs(nir_shader
*shader
);
2496 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2498 typedef struct nir_lower_wpos_ytransform_options
{
2499 int state_tokens
[5];
2500 bool fs_coord_origin_upper_left
:1;
2501 bool fs_coord_origin_lower_left
:1;
2502 bool fs_coord_pixel_center_integer
:1;
2503 bool fs_coord_pixel_center_half_integer
:1;
2504 } nir_lower_wpos_ytransform_options
;
2506 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2507 const nir_lower_wpos_ytransform_options
*options
);
2508 bool nir_lower_wpos_center(nir_shader
*shader
);
2510 typedef struct nir_lower_drawpixels_options
{
2511 int texcoord_state_tokens
[5];
2512 int scale_state_tokens
[5];
2513 int bias_state_tokens
[5];
2514 unsigned drawpix_sampler
;
2515 unsigned pixelmap_sampler
;
2517 bool scale_and_bias
:1;
2518 } nir_lower_drawpixels_options
;
2520 void nir_lower_drawpixels(nir_shader
*shader
,
2521 const nir_lower_drawpixels_options
*options
);
2523 typedef struct nir_lower_bitmap_options
{
2526 } nir_lower_bitmap_options
;
2528 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2530 void nir_lower_atomics(nir_shader
*shader
,
2531 const struct gl_shader_program
*shader_program
);
2532 void nir_lower_to_source_mods(nir_shader
*shader
);
2534 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2537 nir_lower_drcp
= (1 << 0),
2538 nir_lower_dsqrt
= (1 << 1),
2539 nir_lower_drsq
= (1 << 2),
2540 nir_lower_dtrunc
= (1 << 3),
2541 nir_lower_dfloor
= (1 << 4),
2542 nir_lower_dceil
= (1 << 5),
2543 nir_lower_dfract
= (1 << 6),
2544 nir_lower_dround_even
= (1 << 7),
2545 nir_lower_dmod
= (1 << 8)
2546 } nir_lower_doubles_options
;
2548 void nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2549 void nir_lower_double_pack(nir_shader
*shader
);
2551 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2553 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2554 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2556 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2557 void nir_convert_to_ssa(nir_shader
*shader
);
2559 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2560 bool nir_repair_ssa(nir_shader
*shader
);
2562 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2563 * registers. If false, convert all values (even those not involved in a phi
2564 * node) to registers.
2566 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2568 bool nir_opt_algebraic(nir_shader
*shader
);
2569 bool nir_opt_algebraic_late(nir_shader
*shader
);
2570 bool nir_opt_constant_folding(nir_shader
*shader
);
2572 bool nir_opt_global_to_local(nir_shader
*shader
);
2574 bool nir_copy_prop(nir_shader
*shader
);
2576 bool nir_opt_cse(nir_shader
*shader
);
2578 bool nir_opt_dce(nir_shader
*shader
);
2580 bool nir_opt_dead_cf(nir_shader
*shader
);
2582 void nir_opt_gcm(nir_shader
*shader
);
2584 bool nir_opt_peephole_select(nir_shader
*shader
);
2586 bool nir_opt_remove_phis(nir_shader
*shader
);
2588 bool nir_opt_undef(nir_shader
*shader
);
2590 void nir_sweep(nir_shader
*shader
);
2592 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2593 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);