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)
31 #include "util/hash_table.h"
32 #include "compiler/glsl/list.h"
33 #include "GL/gl.h" /* GLenum */
34 #include "util/list.h"
35 #include "util/ralloc.h"
37 #include "util/bitscan.h"
38 #include "util/bitset.h"
39 #include "util/macros.h"
40 #include "compiler/nir_types.h"
41 #include "compiler/shader_enums.h"
42 #include "compiler/shader_info.h"
46 #include "util/debug.h"
49 #include "nir_opcodes.h"
51 #if defined(_WIN32) && !defined(snprintf)
52 #define snprintf _snprintf
60 #define NIR_TRUE (~0u)
61 #define NIR_MAX_VEC_COMPONENTS 4
62 typedef uint8_t nir_component_mask_t
;
64 /** Defines a cast function
66 * This macro defines a cast function from in_type to out_type where
67 * out_type is some structure type that contains a field of type out_type.
69 * Note that you have to be a bit careful as the generated cast function
72 #define NIR_DEFINE_CAST(name, in_type, out_type, field, \
73 type_field, type_value) \
74 static inline out_type * \
75 name(const in_type *parent) \
77 assert(parent && parent->type_field == type_value); \
78 return exec_node_data(out_type, parent, field); \
88 * Description of built-in state associated with a uniform
90 * \sa nir_variable::state_slots
93 gl_state_index16 tokens
[STATE_LENGTH
];
98 nir_var_shader_in
= (1 << 0),
99 nir_var_shader_out
= (1 << 1),
100 nir_var_global
= (1 << 2),
101 nir_var_local
= (1 << 3),
102 nir_var_uniform
= (1 << 4),
103 nir_var_shader_storage
= (1 << 5),
104 nir_var_system_value
= (1 << 6),
105 nir_var_shared
= (1 << 8),
113 nir_rounding_mode_undef
= 0,
114 nir_rounding_mode_rtne
= 1, /* round to nearest even */
115 nir_rounding_mode_ru
= 2, /* round up */
116 nir_rounding_mode_rd
= 3, /* round down */
117 nir_rounding_mode_rtz
= 4, /* round towards zero */
121 float f32
[NIR_MAX_VEC_COMPONENTS
];
122 double f64
[NIR_MAX_VEC_COMPONENTS
];
123 int8_t i8
[NIR_MAX_VEC_COMPONENTS
];
124 uint8_t u8
[NIR_MAX_VEC_COMPONENTS
];
125 int16_t i16
[NIR_MAX_VEC_COMPONENTS
];
126 uint16_t u16
[NIR_MAX_VEC_COMPONENTS
];
127 int32_t i32
[NIR_MAX_VEC_COMPONENTS
];
128 uint32_t u32
[NIR_MAX_VEC_COMPONENTS
];
129 int64_t i64
[NIR_MAX_VEC_COMPONENTS
];
130 uint64_t u64
[NIR_MAX_VEC_COMPONENTS
];
133 typedef struct nir_constant
{
135 * Value of the constant.
137 * The field used to back the values supplied by the constant is determined
138 * by the type associated with the \c nir_variable. Constants may be
139 * scalars, vectors, or matrices.
141 nir_const_value values
[NIR_MAX_VEC_COMPONENTS
];
143 /* we could get this from the var->type but makes clone *much* easier to
144 * not have to care about the type.
146 unsigned num_elements
;
148 /* Array elements / Structure Fields */
149 struct nir_constant
**elements
;
153 * \brief Layout qualifiers for gl_FragDepth.
155 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
156 * with a layout qualifier.
159 nir_depth_layout_none
, /**< No depth layout is specified. */
160 nir_depth_layout_any
,
161 nir_depth_layout_greater
,
162 nir_depth_layout_less
,
163 nir_depth_layout_unchanged
167 * Enum keeping track of how a variable was declared.
171 * Normal declaration.
173 nir_var_declared_normally
= 0,
176 * Variable is implicitly generated by the compiler and should not be
177 * visible via the API.
180 } nir_var_declaration_type
;
183 * Either a uniform, global variable, shader input, or shader output. Based on
184 * ir_variable - it should be easy to translate between the two.
187 typedef struct nir_variable
{
188 struct exec_node node
;
191 * Declared type of the variable
193 const struct glsl_type
*type
;
196 * Declared name of the variable
200 struct nir_variable_data
{
202 * Storage class of the variable.
204 * \sa nir_variable_mode
206 nir_variable_mode mode
;
209 * Is the variable read-only?
211 * This is set for variables declared as \c const, shader inputs,
214 unsigned read_only
:1;
218 unsigned invariant
:1;
221 * When separate shader programs are enabled, only input/outputs between
222 * the stages of a multi-stage separate program can be safely removed
223 * from the shader interface. Other input/outputs must remains active.
225 * This is also used to make sure xfb varyings that are unused by the
226 * fragment shader are not removed.
228 unsigned always_active_io
:1;
231 * Interpolation mode for shader inputs / outputs
233 * \sa glsl_interp_mode
235 unsigned interpolation
:2;
238 * \name ARB_fragment_coord_conventions
241 unsigned origin_upper_left
:1;
242 unsigned pixel_center_integer
:1;
246 * If non-zero, then this variable may be packed along with other variables
247 * into a single varying slot, so this offset should be applied when
248 * accessing components. For example, an offset of 1 means that the x
249 * component of this variable is actually stored in component y of the
250 * location specified by \c location.
252 unsigned location_frac
:2;
255 * If true, this variable represents an array of scalars that should
256 * be tightly packed. In other words, consecutive array elements
257 * should be stored one component apart, rather than one slot apart.
262 * Whether this is a fragment shader output implicitly initialized with
263 * the previous contents of the specified render target at the
264 * framebuffer location corresponding to this shader invocation.
266 unsigned fb_fetch_output
:1;
269 * Non-zero if this variable is considered bindless as defined by
270 * ARB_bindless_texture.
275 * Was an explicit binding set in the shader?
277 unsigned explicit_binding
:1;
280 * Was a transfer feedback buffer set in the shader?
282 unsigned explicit_xfb_buffer
:1;
285 * Was a transfer feedback stride set in the shader?
287 unsigned explicit_xfb_stride
:1;
290 * Was an explicit offset set in the shader?
292 unsigned explicit_offset
:1;
295 * \brief Layout qualifier for gl_FragDepth.
297 * This is not equal to \c ir_depth_layout_none if and only if this
298 * variable is \c gl_FragDepth and a layout qualifier is specified.
300 nir_depth_layout depth_layout
;
303 * Storage location of the base of this variable
305 * The precise meaning of this field depends on the nature of the variable.
307 * - Vertex shader input: one of the values from \c gl_vert_attrib.
308 * - Vertex shader output: one of the values from \c gl_varying_slot.
309 * - Geometry shader input: one of the values from \c gl_varying_slot.
310 * - Geometry shader output: one of the values from \c gl_varying_slot.
311 * - Fragment shader input: one of the values from \c gl_varying_slot.
312 * - Fragment shader output: one of the values from \c gl_frag_result.
313 * - Uniforms: Per-stage uniform slot number for default uniform block.
314 * - Uniforms: Index within the uniform block definition for UBO members.
315 * - Non-UBO Uniforms: uniform slot number.
316 * - Other: This field is not currently used.
318 * If the variable is a uniform, shader input, or shader output, and the
319 * slot has not been assigned, the value will be -1.
324 * The actual location of the variable in the IR. Only valid for inputs
327 unsigned int driver_location
;
330 * Vertex stream output identifier.
332 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
333 * stream of the i-th component.
338 * output index for dual source blending.
343 * Descriptor set binding for sampler or UBO.
348 * Initial binding point for a sampler or UBO.
350 * For array types, this represents the binding point for the first element.
355 * Location an atomic counter or transform feedback is stored at.
360 * Transform feedback buffer.
365 * Transform feedback stride.
370 * How the variable was declared. See nir_var_declaration_type.
372 * This is used to detect variables generated by the compiler, so should
373 * not be visible via the API.
375 unsigned how_declared
:2;
378 * ARB_shader_image_load_store qualifiers.
381 enum gl_access_qualifier access
;
383 /** Image internal format if specified explicitly, otherwise GL_NONE. */
389 * Built-in state that backs this uniform
391 * Once set at variable creation, \c state_slots must remain invariant.
392 * This is because, ideally, this array would be shared by all clones of
393 * this variable in the IR tree. In other words, we'd really like for it
394 * to be a fly-weight.
396 * If the variable is not a uniform, \c num_state_slots will be zero and
397 * \c state_slots will be \c NULL.
400 unsigned num_state_slots
; /**< Number of state slots used */
401 nir_state_slot
*state_slots
; /**< State descriptors. */
405 * Constant expression assigned in the initializer of the variable
407 * This field should only be used temporarily by creators of NIR shaders
408 * and then lower_constant_initializers can be used to get rid of them.
409 * Most of the rest of NIR ignores this field or asserts that it's NULL.
411 nir_constant
*constant_initializer
;
414 * For variables that are in an interface block or are an instance of an
415 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
417 * \sa ir_variable::location
419 const struct glsl_type
*interface_type
;
422 * Description of per-member data for per-member struct variables
424 * This is used for variables which are actually an amalgamation of
425 * multiple entities such as a struct of built-in values or a struct of
426 * inputs each with their own layout specifier. This is only allowed on
427 * variables with a struct or array of array of struct type.
429 unsigned num_members
;
430 struct nir_variable_data
*members
;
433 #define nir_foreach_variable(var, var_list) \
434 foreach_list_typed(nir_variable, var, node, var_list)
436 #define nir_foreach_variable_safe(var, var_list) \
437 foreach_list_typed_safe(nir_variable, var, node, var_list)
440 nir_variable_is_global(const nir_variable
*var
)
442 return var
->data
.mode
!= nir_var_local
;
445 typedef struct nir_register
{
446 struct exec_node node
;
448 unsigned num_components
; /** < number of vector components */
449 unsigned num_array_elems
; /** < size of array (0 for no array) */
451 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
454 /** generic register index. */
457 /** only for debug purposes, can be NULL */
460 /** whether this register is local (per-function) or global (per-shader) */
464 * If this flag is set to true, then accessing channels >= num_components
465 * is well-defined, and simply spills over to the next array element. This
466 * is useful for backends that can do per-component accessing, in
467 * particular scalar backends. By setting this flag and making
468 * num_components equal to 1, structures can be packed tightly into
469 * registers and then registers can be accessed per-component to get to
470 * each structure member, even if it crosses vec4 boundaries.
474 /** set of nir_srcs where this register is used (read from) */
475 struct list_head uses
;
477 /** set of nir_dests where this register is defined (written to) */
478 struct list_head defs
;
480 /** set of nir_ifs where this register is used as a condition */
481 struct list_head if_uses
;
484 #define nir_foreach_register(reg, reg_list) \
485 foreach_list_typed(nir_register, reg, node, reg_list)
486 #define nir_foreach_register_safe(reg, reg_list) \
487 foreach_list_typed_safe(nir_register, reg, node, reg_list)
491 nir_instr_type_deref
,
494 nir_instr_type_intrinsic
,
495 nir_instr_type_load_const
,
497 nir_instr_type_ssa_undef
,
499 nir_instr_type_parallel_copy
,
502 typedef struct nir_instr
{
503 struct exec_node node
;
505 struct nir_block
*block
;
507 /** generic instruction index. */
510 /* A temporary for optimization and analysis passes to use for storing
511 * flags. For instance, DCE uses this to store the "dead/live" info.
516 static inline nir_instr
*
517 nir_instr_next(nir_instr
*instr
)
519 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
520 if (exec_node_is_tail_sentinel(next
))
523 return exec_node_data(nir_instr
, next
, node
);
526 static inline nir_instr
*
527 nir_instr_prev(nir_instr
*instr
)
529 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
530 if (exec_node_is_head_sentinel(prev
))
533 return exec_node_data(nir_instr
, prev
, node
);
537 nir_instr_is_first(const nir_instr
*instr
)
539 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
543 nir_instr_is_last(const nir_instr
*instr
)
545 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
548 typedef struct nir_ssa_def
{
549 /** for debugging only, can be NULL */
552 /** generic SSA definition index. */
555 /** Index into the live_in and live_out bitfields */
558 /** Instruction which produces this SSA value. */
559 nir_instr
*parent_instr
;
561 /** set of nir_instrs where this register is used (read from) */
562 struct list_head uses
;
564 /** set of nir_ifs where this register is used as a condition */
565 struct list_head if_uses
;
567 uint8_t num_components
;
569 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
577 struct nir_src
*indirect
; /** < NULL for no indirect offset */
578 unsigned base_offset
;
580 /* TODO use-def chain goes here */
584 nir_instr
*parent_instr
;
585 struct list_head def_link
;
588 struct nir_src
*indirect
; /** < NULL for no indirect offset */
589 unsigned base_offset
;
591 /* TODO def-use chain goes here */
596 typedef struct nir_src
{
598 /** Instruction that consumes this value as a source. */
599 nir_instr
*parent_instr
;
600 struct nir_if
*parent_if
;
603 struct list_head use_link
;
613 static inline nir_src
616 nir_src src
= { { NULL
} };
620 #define NIR_SRC_INIT nir_src_init()
622 #define nir_foreach_use(src, reg_or_ssa_def) \
623 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
625 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
626 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
628 #define nir_foreach_if_use(src, reg_or_ssa_def) \
629 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
631 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
632 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
643 static inline nir_dest
646 nir_dest dest
= { { { NULL
} } };
650 #define NIR_DEST_INIT nir_dest_init()
652 #define nir_foreach_def(dest, reg) \
653 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
655 #define nir_foreach_def_safe(dest, reg) \
656 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
658 static inline nir_src
659 nir_src_for_ssa(nir_ssa_def
*def
)
661 nir_src src
= NIR_SRC_INIT
;
669 static inline nir_src
670 nir_src_for_reg(nir_register
*reg
)
672 nir_src src
= NIR_SRC_INIT
;
676 src
.reg
.indirect
= NULL
;
677 src
.reg
.base_offset
= 0;
682 static inline nir_dest
683 nir_dest_for_reg(nir_register
*reg
)
685 nir_dest dest
= NIR_DEST_INIT
;
692 static inline unsigned
693 nir_src_bit_size(nir_src src
)
695 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
698 static inline unsigned
699 nir_src_num_components(nir_src src
)
701 return src
.is_ssa
? src
.ssa
->num_components
: src
.reg
.reg
->num_components
;
705 nir_src_is_const(nir_src src
)
708 src
.ssa
->parent_instr
->type
== nir_instr_type_load_const
;
711 int64_t nir_src_as_int(nir_src src
);
712 uint64_t nir_src_as_uint(nir_src src
);
713 bool nir_src_as_bool(nir_src src
);
714 double nir_src_as_float(nir_src src
);
715 int64_t nir_src_comp_as_int(nir_src src
, unsigned component
);
716 uint64_t nir_src_comp_as_uint(nir_src src
, unsigned component
);
717 bool nir_src_comp_as_bool(nir_src src
, unsigned component
);
718 double nir_src_comp_as_float(nir_src src
, unsigned component
);
720 static inline unsigned
721 nir_dest_bit_size(nir_dest dest
)
723 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
726 static inline unsigned
727 nir_dest_num_components(nir_dest dest
)
729 return dest
.is_ssa
? dest
.ssa
.num_components
: dest
.reg
.reg
->num_components
;
732 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
733 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
739 * \name input modifiers
743 * For inputs interpreted as floating point, flips the sign bit. For
744 * inputs interpreted as integers, performs the two's complement negation.
749 * Clears the sign bit for floating point values, and computes the integer
750 * absolute value for integers. Note that the negate modifier acts after
751 * the absolute value modifier, therefore if both are set then all inputs
752 * will become negative.
758 * For each input component, says which component of the register it is
759 * chosen from. Note that which elements of the swizzle are used and which
760 * are ignored are based on the write mask for most opcodes - for example,
761 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
762 * a swizzle of {2, x, 1, 0} where x means "don't care."
764 uint8_t swizzle
[NIR_MAX_VEC_COMPONENTS
];
771 * \name saturate output modifier
773 * Only valid for opcodes that output floating-point numbers. Clamps the
774 * output to between 0.0 and 1.0 inclusive.
779 unsigned write_mask
: NIR_MAX_VEC_COMPONENTS
; /* ignored if dest.is_ssa is true */
783 nir_type_invalid
= 0, /* Not a valid type */
788 nir_type_bool32
= 32 | nir_type_bool
,
789 nir_type_int8
= 8 | nir_type_int
,
790 nir_type_int16
= 16 | nir_type_int
,
791 nir_type_int32
= 32 | nir_type_int
,
792 nir_type_int64
= 64 | nir_type_int
,
793 nir_type_uint8
= 8 | nir_type_uint
,
794 nir_type_uint16
= 16 | nir_type_uint
,
795 nir_type_uint32
= 32 | nir_type_uint
,
796 nir_type_uint64
= 64 | nir_type_uint
,
797 nir_type_float16
= 16 | nir_type_float
,
798 nir_type_float32
= 32 | nir_type_float
,
799 nir_type_float64
= 64 | nir_type_float
,
802 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
803 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
805 static inline unsigned
806 nir_alu_type_get_type_size(nir_alu_type type
)
808 return type
& NIR_ALU_TYPE_SIZE_MASK
;
811 static inline unsigned
812 nir_alu_type_get_base_type(nir_alu_type type
)
814 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
817 static inline nir_alu_type
818 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
822 return nir_type_bool32
;
825 return nir_type_uint32
;
828 return nir_type_int32
;
830 case GLSL_TYPE_UINT16
:
831 return nir_type_uint16
;
833 case GLSL_TYPE_INT16
:
834 return nir_type_int16
;
836 case GLSL_TYPE_UINT8
:
837 return nir_type_uint8
;
839 return nir_type_int8
;
840 case GLSL_TYPE_UINT64
:
841 return nir_type_uint64
;
843 case GLSL_TYPE_INT64
:
844 return nir_type_int64
;
846 case GLSL_TYPE_FLOAT
:
847 return nir_type_float32
;
849 case GLSL_TYPE_FLOAT16
:
850 return nir_type_float16
;
852 case GLSL_TYPE_DOUBLE
:
853 return nir_type_float64
;
856 unreachable("unknown type");
860 static inline nir_alu_type
861 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
863 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
866 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
,
867 nir_rounding_mode rnd
);
870 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
871 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
872 } nir_op_algebraic_property
;
880 * The number of components in the output
882 * If non-zero, this is the size of the output and input sizes are
883 * explicitly given; swizzle and writemask are still in effect, but if
884 * the output component is masked out, then the input component may
887 * If zero, the opcode acts in the standard, per-component manner; the
888 * operation is performed on each component (except the ones that are
889 * masked out) with the input being taken from the input swizzle for
892 * The size of some of the inputs may be given (i.e. non-zero) even
893 * though output_size is zero; in that case, the inputs with a zero
894 * size act per-component, while the inputs with non-zero size don't.
896 unsigned output_size
;
899 * The type of vector that the instruction outputs. Note that the
900 * staurate modifier is only allowed on outputs with the float type.
903 nir_alu_type output_type
;
906 * The number of components in each input
908 unsigned input_sizes
[NIR_MAX_VEC_COMPONENTS
];
911 * The type of vector that each input takes. Note that negate and
912 * absolute value are only allowed on inputs with int or float type and
913 * behave differently on the two.
915 nir_alu_type input_types
[NIR_MAX_VEC_COMPONENTS
];
917 nir_op_algebraic_property algebraic_properties
;
920 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
922 typedef struct nir_alu_instr
{
926 /** Indicates that this ALU instruction generates an exact value
928 * This is kind of a mixture of GLSL "precise" and "invariant" and not
929 * really equivalent to either. This indicates that the value generated by
930 * this operation is high-precision and any code transformations that touch
931 * it must ensure that the resulting value is bit-for-bit identical to the
940 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
941 nir_alu_instr
*instr
);
942 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
943 nir_alu_instr
*instr
);
945 /* is this source channel used? */
947 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
950 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
951 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
953 return (instr
->dest
.write_mask
>> channel
) & 1;
956 static inline nir_component_mask_t
957 nir_alu_instr_src_read_mask(const nir_alu_instr
*instr
, unsigned src
)
959 nir_component_mask_t read_mask
= 0;
960 for (unsigned c
= 0; c
< NIR_MAX_VEC_COMPONENTS
; c
++) {
961 if (!nir_alu_instr_channel_used(instr
, src
, c
))
964 read_mask
|= (1 << instr
->src
[src
].swizzle
[c
]);
970 * For instructions whose destinations are SSA, get the number of channels
973 static inline unsigned
974 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
976 assert(instr
->dest
.dest
.is_ssa
);
978 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
979 return nir_op_infos
[instr
->op
].input_sizes
[src
];
981 return instr
->dest
.dest
.ssa
.num_components
;
984 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
985 unsigned src1
, unsigned src2
);
989 nir_deref_type_array
,
990 nir_deref_type_array_wildcard
,
991 nir_deref_type_struct
,
998 /** The type of this deref instruction */
999 nir_deref_type deref_type
;
1001 /** The mode of the underlying variable */
1002 nir_variable_mode mode
;
1004 /** The dereferenced type of the resulting pointer value */
1005 const struct glsl_type
*type
;
1008 /** Variable being dereferenced if deref_type is a deref_var */
1011 /** Parent deref if deref_type is not deref_var */
1015 /** Additional deref parameters */
1026 /** Destination to store the resulting "pointer" */
1030 NIR_DEFINE_CAST(nir_instr_as_deref
, nir_instr
, nir_deref_instr
, instr
,
1031 type
, nir_instr_type_deref
)
1033 static inline nir_deref_instr
*
1034 nir_src_as_deref(nir_src src
)
1039 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_deref
)
1042 return nir_instr_as_deref(src
.ssa
->parent_instr
);
1045 static inline nir_deref_instr
*
1046 nir_deref_instr_parent(const nir_deref_instr
*instr
)
1048 if (instr
->deref_type
== nir_deref_type_var
)
1051 return nir_src_as_deref(instr
->parent
);
1054 static inline nir_variable
*
1055 nir_deref_instr_get_variable(const nir_deref_instr
*instr
)
1057 while (instr
->deref_type
!= nir_deref_type_var
) {
1058 if (instr
->deref_type
== nir_deref_type_cast
)
1061 instr
= nir_deref_instr_parent(instr
);
1067 bool nir_deref_instr_has_indirect(nir_deref_instr
*instr
);
1069 bool nir_deref_instr_remove_if_unused(nir_deref_instr
*instr
);
1074 struct nir_function
*callee
;
1076 unsigned num_params
;
1080 #include "nir_intrinsics.h"
1082 #define NIR_INTRINSIC_MAX_CONST_INDEX 4
1084 /** Represents an intrinsic
1086 * An intrinsic is an instruction type for handling things that are
1087 * more-or-less regular operations but don't just consume and produce SSA
1088 * values like ALU operations do. Intrinsics are not for things that have
1089 * special semantic meaning such as phi nodes and parallel copies.
1090 * Examples of intrinsics include variable load/store operations, system
1091 * value loads, and the like. Even though texturing more-or-less falls
1092 * under this category, texturing is its own instruction type because
1093 * trying to represent texturing with intrinsics would lead to a
1094 * combinatorial explosion of intrinsic opcodes.
1096 * By having a single instruction type for handling a lot of different
1097 * cases, optimization passes can look for intrinsics and, for the most
1098 * part, completely ignore them. Each intrinsic type also has a few
1099 * possible flags that govern whether or not they can be reordered or
1100 * eliminated. That way passes like dead code elimination can still work
1101 * on intrisics without understanding the meaning of each.
1103 * Each intrinsic has some number of constant indices, some number of
1104 * variables, and some number of sources. What these sources, variables,
1105 * and indices mean depends on the intrinsic and is documented with the
1106 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
1107 * instructions are the only types of instruction that can operate on
1113 nir_intrinsic_op intrinsic
;
1117 /** number of components if this is a vectorized intrinsic
1119 * Similarly to ALU operations, some intrinsics are vectorized.
1120 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
1121 * For vectorized intrinsics, the num_components field specifies the
1122 * number of destination components and the number of source components
1123 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
1125 uint8_t num_components
;
1127 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
1130 } nir_intrinsic_instr
;
1132 static inline nir_variable
*
1133 nir_intrinsic_get_var(nir_intrinsic_instr
*intrin
, unsigned i
)
1135 return nir_deref_instr_get_variable(nir_src_as_deref(intrin
->src
[i
]));
1139 * \name NIR intrinsics semantic flags
1141 * information about what the compiler can do with the intrinsics.
1143 * \sa nir_intrinsic_info::flags
1147 * whether the intrinsic can be safely eliminated if none of its output
1148 * value is not being used.
1150 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
1153 * Whether the intrinsic can be reordered with respect to any other
1154 * intrinsic, i.e. whether the only reordering dependencies of the
1155 * intrinsic are due to the register reads/writes.
1157 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
1158 } nir_intrinsic_semantic_flag
;
1161 * \name NIR intrinsics const-index flag
1163 * Indicates the usage of a const_index slot.
1165 * \sa nir_intrinsic_info::index_map
1169 * Generally instructions that take a offset src argument, can encode
1170 * a constant 'base' value which is added to the offset.
1172 NIR_INTRINSIC_BASE
= 1,
1175 * For store instructions, a writemask for the store.
1177 NIR_INTRINSIC_WRMASK
= 2,
1180 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1182 NIR_INTRINSIC_STREAM_ID
= 3,
1185 * The clip-plane id for load_user_clip_plane intrinsic.
1187 NIR_INTRINSIC_UCP_ID
= 4,
1190 * The amount of data, starting from BASE, that this instruction may
1191 * access. This is used to provide bounds if the offset is not constant.
1193 NIR_INTRINSIC_RANGE
= 5,
1196 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1198 NIR_INTRINSIC_DESC_SET
= 6,
1201 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1203 NIR_INTRINSIC_BINDING
= 7,
1208 NIR_INTRINSIC_COMPONENT
= 8,
1211 * Interpolation mode (only meaningful for FS inputs).
1213 NIR_INTRINSIC_INTERP_MODE
= 9,
1216 * A binary nir_op to use when performing a reduction or scan operation
1218 NIR_INTRINSIC_REDUCTION_OP
= 10,
1221 * Cluster size for reduction operations
1223 NIR_INTRINSIC_CLUSTER_SIZE
= 11,
1226 * Parameter index for a load_param intrinsic
1228 NIR_INTRINSIC_PARAM_IDX
= 12,
1231 * Image dimensionality for image intrinsics
1233 * One of GLSL_SAMPLER_DIM_*
1235 NIR_INTRINSIC_IMAGE_DIM
= 13,
1238 * Non-zero if we are accessing an array image
1240 NIR_INTRINSIC_IMAGE_ARRAY
= 14,
1243 * Image format for image intrinsics
1245 NIR_INTRINSIC_FORMAT
= 15,
1248 * Access qualifiers for image intrinsics
1250 NIR_INTRINSIC_ACCESS
= 16,
1253 * Alignment for offsets and addresses
1255 * These two parameters, specify an alignment in terms of a multiplier and
1256 * an offset. The offset or address parameter X of the intrinsic is
1257 * guaranteed to satisfy the following:
1259 * (X - align_offset) % align_mul == 0
1261 NIR_INTRINSIC_ALIGN_MUL
= 17,
1262 NIR_INTRINSIC_ALIGN_OFFSET
= 18,
1264 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1266 } nir_intrinsic_index_flag
;
1268 #define NIR_INTRINSIC_MAX_INPUTS 5
1273 unsigned num_srcs
; /** < number of register/SSA inputs */
1275 /** number of components of each input register
1277 * If this value is 0, the number of components is given by the
1278 * num_components field of nir_intrinsic_instr.
1280 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1284 /** number of components of the output register
1286 * If this value is 0, the number of components is given by the
1287 * num_components field of nir_intrinsic_instr.
1289 unsigned dest_components
;
1291 /** the number of constant indices used by the intrinsic */
1292 unsigned num_indices
;
1294 /** indicates the usage of intr->const_index[n] */
1295 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1297 /** semantic flags for calls to this intrinsic */
1298 nir_intrinsic_semantic_flag flags
;
1299 } nir_intrinsic_info
;
1301 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1303 static inline unsigned
1304 nir_intrinsic_src_components(nir_intrinsic_instr
*intr
, unsigned srcn
)
1306 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1307 assert(srcn
< info
->num_srcs
);
1308 if (info
->src_components
[srcn
])
1309 return info
->src_components
[srcn
];
1311 return intr
->num_components
;
1314 static inline unsigned
1315 nir_intrinsic_dest_components(nir_intrinsic_instr
*intr
)
1317 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1318 if (!info
->has_dest
)
1320 else if (info
->dest_components
)
1321 return info
->dest_components
;
1323 return intr
->num_components
;
1326 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1327 static inline type \
1328 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1330 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1331 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1332 return (type)instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1334 static inline void \
1335 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1337 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1338 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1339 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1342 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1343 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1344 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1345 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1346 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1347 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1348 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1349 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1350 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1351 INTRINSIC_IDX_ACCESSORS(reduction_op
, REDUCTION_OP
, unsigned)
1352 INTRINSIC_IDX_ACCESSORS(cluster_size
, CLUSTER_SIZE
, unsigned)
1353 INTRINSIC_IDX_ACCESSORS(param_idx
, PARAM_IDX
, unsigned)
1354 INTRINSIC_IDX_ACCESSORS(image_dim
, IMAGE_DIM
, enum glsl_sampler_dim
)
1355 INTRINSIC_IDX_ACCESSORS(image_array
, IMAGE_ARRAY
, bool)
1356 INTRINSIC_IDX_ACCESSORS(access
, ACCESS
, enum gl_access_qualifier
)
1357 INTRINSIC_IDX_ACCESSORS(format
, FORMAT
, unsigned)
1358 INTRINSIC_IDX_ACCESSORS(align_mul
, ALIGN_MUL
, unsigned)
1359 INTRINSIC_IDX_ACCESSORS(align_offset
, ALIGN_OFFSET
, unsigned)
1362 nir_intrinsic_set_align(nir_intrinsic_instr
*intrin
,
1363 unsigned align_mul
, unsigned align_offset
)
1365 assert(util_is_power_of_two_nonzero(align_mul
));
1366 assert(align_offset
< align_mul
);
1367 nir_intrinsic_set_align_mul(intrin
, align_mul
);
1368 nir_intrinsic_set_align_offset(intrin
, align_offset
);
1371 /** Returns a simple alignment for a load/store intrinsic offset
1373 * Instead of the full mul+offset alignment scheme provided by the ALIGN_MUL
1374 * and ALIGN_OFFSET parameters, this helper takes both into account and
1375 * provides a single simple alignment parameter. The offset X is guaranteed
1376 * to satisfy X % align == 0.
1378 static inline unsigned
1379 nir_intrinsic_align(nir_intrinsic_instr
*intrin
)
1381 const unsigned align_mul
= nir_intrinsic_align_mul(intrin
);
1382 const unsigned align_offset
= nir_intrinsic_align_offset(intrin
);
1383 assert(align_offset
< align_mul
);
1384 return align_offset
? 1 << (ffs(align_offset
) - 1) : align_mul
;
1388 * \group texture information
1390 * This gives semantic information about textures which is useful to the
1391 * frontend, the backend, and lowering passes, but not the optimizer.
1396 nir_tex_src_projector
,
1397 nir_tex_src_comparator
, /* shadow comparator */
1401 nir_tex_src_ms_index
, /* MSAA sample index */
1402 nir_tex_src_ms_mcs
, /* MSAA compression value */
1405 nir_tex_src_texture_deref
, /* < deref pointing to the texture */
1406 nir_tex_src_sampler_deref
, /* < deref pointing to the sampler */
1407 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1408 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1409 nir_tex_src_plane
, /* < selects plane for planar textures */
1410 nir_num_tex_src_types
1415 nir_tex_src_type src_type
;
1419 nir_texop_tex
, /**< Regular texture look-up */
1420 nir_texop_txb
, /**< Texture look-up with LOD bias */
1421 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1422 nir_texop_txd
, /**< Texture look-up with partial derivatives */
1423 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1424 nir_texop_txf_ms
, /**< Multisample texture fetch */
1425 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1426 nir_texop_txs
, /**< Texture size */
1427 nir_texop_lod
, /**< Texture lod query */
1428 nir_texop_tg4
, /**< Texture gather */
1429 nir_texop_query_levels
, /**< Texture levels query */
1430 nir_texop_texture_samples
, /**< Texture samples query */
1431 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1439 enum glsl_sampler_dim sampler_dim
;
1440 nir_alu_type dest_type
;
1445 unsigned num_srcs
, coord_components
;
1446 bool is_array
, is_shadow
;
1449 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1450 * components or the new-style shadow that outputs 1 component.
1452 bool is_new_style_shadow
;
1454 /* gather component selector */
1455 unsigned component
: 2;
1457 /** The texture index
1459 * If this texture instruction has a nir_tex_src_texture_offset source,
1460 * then the texture index is given by texture_index + texture_offset.
1462 unsigned texture_index
;
1464 /** The size of the texture array or 0 if it's not an array */
1465 unsigned texture_array_size
;
1467 /** The sampler index
1469 * The following operations do not require a sampler and, as such, this
1470 * field should be ignored:
1472 * - nir_texop_txf_ms
1475 * - nir_texop_query_levels
1476 * - nir_texop_texture_samples
1477 * - nir_texop_samples_identical
1479 * If this texture instruction has a nir_tex_src_sampler_offset source,
1480 * then the sampler index is given by sampler_index + sampler_offset.
1482 unsigned sampler_index
;
1485 static inline unsigned
1486 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1488 switch (instr
->op
) {
1489 case nir_texop_txs
: {
1491 switch (instr
->sampler_dim
) {
1492 case GLSL_SAMPLER_DIM_1D
:
1493 case GLSL_SAMPLER_DIM_BUF
:
1496 case GLSL_SAMPLER_DIM_2D
:
1497 case GLSL_SAMPLER_DIM_CUBE
:
1498 case GLSL_SAMPLER_DIM_MS
:
1499 case GLSL_SAMPLER_DIM_RECT
:
1500 case GLSL_SAMPLER_DIM_EXTERNAL
:
1501 case GLSL_SAMPLER_DIM_SUBPASS
:
1504 case GLSL_SAMPLER_DIM_3D
:
1508 unreachable("not reached");
1510 if (instr
->is_array
)
1518 case nir_texop_texture_samples
:
1519 case nir_texop_query_levels
:
1520 case nir_texop_samples_identical
:
1524 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1531 /* Returns true if this texture operation queries something about the texture
1532 * rather than actually sampling it.
1535 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1537 switch (instr
->op
) {
1540 case nir_texop_texture_samples
:
1541 case nir_texop_query_levels
:
1542 case nir_texop_txf_ms_mcs
:
1549 case nir_texop_txf_ms
:
1553 unreachable("Invalid texture opcode");
1558 nir_alu_instr_is_comparison(const nir_alu_instr
*instr
)
1560 switch (instr
->op
) {
1581 static inline nir_alu_type
1582 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1584 switch (instr
->src
[src
].src_type
) {
1585 case nir_tex_src_coord
:
1586 switch (instr
->op
) {
1588 case nir_texop_txf_ms
:
1589 case nir_texop_txf_ms_mcs
:
1590 case nir_texop_samples_identical
:
1591 return nir_type_int
;
1594 return nir_type_float
;
1597 case nir_tex_src_lod
:
1598 switch (instr
->op
) {
1601 return nir_type_int
;
1604 return nir_type_float
;
1607 case nir_tex_src_projector
:
1608 case nir_tex_src_comparator
:
1609 case nir_tex_src_bias
:
1610 case nir_tex_src_ddx
:
1611 case nir_tex_src_ddy
:
1612 return nir_type_float
;
1614 case nir_tex_src_offset
:
1615 case nir_tex_src_ms_index
:
1616 case nir_tex_src_texture_offset
:
1617 case nir_tex_src_sampler_offset
:
1618 return nir_type_int
;
1621 unreachable("Invalid texture source type");
1625 static inline unsigned
1626 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1628 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1629 return instr
->coord_components
;
1631 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1632 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1635 if (instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1636 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1637 if (instr
->is_array
)
1638 return instr
->coord_components
- 1;
1640 return instr
->coord_components
;
1643 /* Usual APIs don't allow cube + offset, but we allow it, with 2 coords for
1644 * the offset, since a cube maps to a single face.
1646 if (instr
->src
[src
].src_type
== nir_tex_src_offset
) {
1647 if (instr
->sampler_dim
== GLSL_SAMPLER_DIM_CUBE
)
1649 else if (instr
->is_array
)
1650 return instr
->coord_components
- 1;
1652 return instr
->coord_components
;
1659 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1661 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1662 if (instr
->src
[i
].src_type
== type
)
1668 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1669 nir_tex_src_type src_type
,
1672 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1677 nir_const_value value
;
1680 } nir_load_const_instr
;
1693 /* creates a new SSA variable in an undefined state */
1698 } nir_ssa_undef_instr
;
1701 struct exec_node node
;
1703 /* The predecessor block corresponding to this source */
1704 struct nir_block
*pred
;
1709 #define nir_foreach_phi_src(phi_src, phi) \
1710 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1711 #define nir_foreach_phi_src_safe(phi_src, phi) \
1712 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1717 struct exec_list srcs
; /** < list of nir_phi_src */
1723 struct exec_node node
;
1726 } nir_parallel_copy_entry
;
1728 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1729 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1734 /* A list of nir_parallel_copy_entrys. The sources of all of the
1735 * entries are copied to the corresponding destinations "in parallel".
1736 * In other words, if we have two entries: a -> b and b -> a, the values
1739 struct exec_list entries
;
1740 } nir_parallel_copy_instr
;
1742 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1743 type
, nir_instr_type_alu
)
1744 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1745 type
, nir_instr_type_call
)
1746 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1747 type
, nir_instr_type_jump
)
1748 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1749 type
, nir_instr_type_tex
)
1750 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1751 type
, nir_instr_type_intrinsic
)
1752 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1753 type
, nir_instr_type_load_const
)
1754 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1755 type
, nir_instr_type_ssa_undef
)
1756 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1757 type
, nir_instr_type_phi
)
1758 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1759 nir_parallel_copy_instr
, instr
,
1760 type
, nir_instr_type_parallel_copy
)
1765 * Control flow consists of a tree of control flow nodes, which include
1766 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1767 * instructions that always run start-to-finish. Each basic block also keeps
1768 * track of its successors (blocks which may run immediately after the current
1769 * block) and predecessors (blocks which could have run immediately before the
1770 * current block). Each function also has a start block and an end block which
1771 * all return statements point to (which is always empty). Together, all the
1772 * blocks with their predecessors and successors make up the control flow
1773 * graph (CFG) of the function. There are helpers that modify the tree of
1774 * control flow nodes while modifying the CFG appropriately; these should be
1775 * used instead of modifying the tree directly.
1782 nir_cf_node_function
1785 typedef struct nir_cf_node
{
1786 struct exec_node node
;
1787 nir_cf_node_type type
;
1788 struct nir_cf_node
*parent
;
1791 typedef struct nir_block
{
1792 nir_cf_node cf_node
;
1794 struct exec_list instr_list
; /** < list of nir_instr */
1796 /** generic block index; generated by nir_index_blocks */
1800 * Each block can only have up to 2 successors, so we put them in a simple
1801 * array - no need for anything more complicated.
1803 struct nir_block
*successors
[2];
1805 /* Set of nir_block predecessors in the CFG */
1806 struct set
*predecessors
;
1809 * this node's immediate dominator in the dominance tree - set to NULL for
1812 struct nir_block
*imm_dom
;
1814 /* This node's children in the dominance tree */
1815 unsigned num_dom_children
;
1816 struct nir_block
**dom_children
;
1818 /* Set of nir_blocks on the dominance frontier of this block */
1819 struct set
*dom_frontier
;
1822 * These two indices have the property that dom_{pre,post}_index for each
1823 * child of this block in the dominance tree will always be between
1824 * dom_pre_index and dom_post_index for this block, which makes testing if
1825 * a given block is dominated by another block an O(1) operation.
1827 unsigned dom_pre_index
, dom_post_index
;
1829 /* live in and out for this block; used for liveness analysis */
1830 BITSET_WORD
*live_in
;
1831 BITSET_WORD
*live_out
;
1834 static inline nir_instr
*
1835 nir_block_first_instr(nir_block
*block
)
1837 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1838 return exec_node_data(nir_instr
, head
, node
);
1841 static inline nir_instr
*
1842 nir_block_last_instr(nir_block
*block
)
1844 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1845 return exec_node_data(nir_instr
, tail
, node
);
1849 nir_block_ends_in_jump(nir_block
*block
)
1851 return !exec_list_is_empty(&block
->instr_list
) &&
1852 nir_block_last_instr(block
)->type
== nir_instr_type_jump
;
1855 #define nir_foreach_instr(instr, block) \
1856 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1857 #define nir_foreach_instr_reverse(instr, block) \
1858 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1859 #define nir_foreach_instr_safe(instr, block) \
1860 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1861 #define nir_foreach_instr_reverse_safe(instr, block) \
1862 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1864 typedef struct nir_if
{
1865 nir_cf_node cf_node
;
1868 struct exec_list then_list
; /** < list of nir_cf_node */
1869 struct exec_list else_list
; /** < list of nir_cf_node */
1875 nir_instr
*conditional_instr
;
1877 nir_block
*break_block
;
1878 nir_block
*continue_from_block
;
1880 bool continue_from_then
;
1882 struct list_head loop_terminator_link
;
1883 } nir_loop_terminator
;
1886 /* Number of instructions in the loop */
1887 unsigned num_instructions
;
1889 /* How many times the loop is run (if known) */
1890 unsigned trip_count
;
1891 bool is_trip_count_known
;
1893 /* Unroll the loop regardless of its size */
1896 /* Does the loop contain complex loop terminators, continues or other
1897 * complex behaviours? If this is true we can't rely on
1898 * loop_terminator_list to be complete or accurate.
1902 nir_loop_terminator
*limiting_terminator
;
1904 /* A list of loop_terminators terminating this loop. */
1905 struct list_head loop_terminator_list
;
1909 nir_cf_node cf_node
;
1911 struct exec_list body
; /** < list of nir_cf_node */
1913 nir_loop_info
*info
;
1917 * Various bits of metadata that can may be created or required by
1918 * optimization and analysis passes
1921 nir_metadata_none
= 0x0,
1922 nir_metadata_block_index
= 0x1,
1923 nir_metadata_dominance
= 0x2,
1924 nir_metadata_live_ssa_defs
= 0x4,
1925 nir_metadata_not_properly_reset
= 0x8,
1926 nir_metadata_loop_analysis
= 0x10,
1930 nir_cf_node cf_node
;
1932 /** pointer to the function of which this is an implementation */
1933 struct nir_function
*function
;
1935 struct exec_list body
; /** < list of nir_cf_node */
1937 nir_block
*end_block
;
1939 /** list for all local variables in the function */
1940 struct exec_list locals
;
1942 /** list of local registers in the function */
1943 struct exec_list registers
;
1945 /** next available local register index */
1948 /** next available SSA value index */
1951 /* total number of basic blocks, only valid when block_index_dirty = false */
1952 unsigned num_blocks
;
1954 nir_metadata valid_metadata
;
1955 } nir_function_impl
;
1957 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1958 nir_start_block(nir_function_impl
*impl
)
1960 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1963 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1964 nir_impl_last_block(nir_function_impl
*impl
)
1966 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1969 static inline nir_cf_node
*
1970 nir_cf_node_next(nir_cf_node
*node
)
1972 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1973 if (exec_node_is_tail_sentinel(next
))
1976 return exec_node_data(nir_cf_node
, next
, node
);
1979 static inline nir_cf_node
*
1980 nir_cf_node_prev(nir_cf_node
*node
)
1982 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1983 if (exec_node_is_head_sentinel(prev
))
1986 return exec_node_data(nir_cf_node
, prev
, node
);
1990 nir_cf_node_is_first(const nir_cf_node
*node
)
1992 return exec_node_is_head_sentinel(node
->node
.prev
);
1996 nir_cf_node_is_last(const nir_cf_node
*node
)
1998 return exec_node_is_tail_sentinel(node
->node
.next
);
2001 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
2002 type
, nir_cf_node_block
)
2003 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
2004 type
, nir_cf_node_if
)
2005 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
2006 type
, nir_cf_node_loop
)
2007 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
2008 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
2010 static inline nir_block
*
2011 nir_if_first_then_block(nir_if
*if_stmt
)
2013 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
2014 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2017 static inline nir_block
*
2018 nir_if_last_then_block(nir_if
*if_stmt
)
2020 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
2021 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2024 static inline nir_block
*
2025 nir_if_first_else_block(nir_if
*if_stmt
)
2027 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
2028 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2031 static inline nir_block
*
2032 nir_if_last_else_block(nir_if
*if_stmt
)
2034 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
2035 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2038 static inline nir_block
*
2039 nir_loop_first_block(nir_loop
*loop
)
2041 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
2042 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2045 static inline nir_block
*
2046 nir_loop_last_block(nir_loop
*loop
)
2048 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
2049 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2053 uint8_t num_components
;
2057 typedef struct nir_function
{
2058 struct exec_node node
;
2061 struct nir_shader
*shader
;
2063 unsigned num_params
;
2064 nir_parameter
*params
;
2066 /** The implementation of this function.
2068 * If the function is only declared and not implemented, this is NULL.
2070 nir_function_impl
*impl
;
2073 typedef struct nir_shader_compiler_options
{
2078 /** Lowers flrp when it does not support doubles */
2085 /** Lowers ibitfield_extract/ubitfield_extract to ibfe/ubfe. */
2086 bool lower_bitfield_extract
;
2087 /** Lowers ibitfield_extract/ubitfield_extract to bfm, compares, shifts. */
2088 bool lower_bitfield_extract_to_shifts
;
2089 /** Lowers bitfield_insert to bfi/bfm */
2090 bool lower_bitfield_insert
;
2091 /** Lowers bitfield_insert to bfm, compares, and shifts. */
2092 bool lower_bitfield_insert_to_shifts
;
2093 /** Lowers bitfield_reverse to shifts. */
2094 bool lower_bitfield_reverse
;
2095 /** Lowers bit_count to shifts. */
2096 bool lower_bit_count
;
2097 /** Lowers bfm to shifts and subtracts. */
2099 /** Lowers ifind_msb to compare and ufind_msb */
2100 bool lower_ifind_msb
;
2101 /** Lowers find_lsb to ufind_msb and logic ops */
2102 bool lower_find_lsb
;
2103 bool lower_uadd_carry
;
2104 bool lower_usub_borrow
;
2105 /** Lowers imul_high/umul_high to 16-bit multiplies and carry operations. */
2106 bool lower_mul_high
;
2107 /** lowers fneg and ineg to fsub and isub. */
2109 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
2112 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
2115 /** enables rules to lower idiv by power-of-two: */
2118 /* lower b2f to iand */
2121 /* Does the native fdot instruction replicate its result for four
2122 * components? If so, then opt_algebraic_late will turn all fdotN
2123 * instructions into fdot_replicatedN instructions.
2125 bool fdot_replicates
;
2127 /** lowers ffloor to fsub+ffract: */
2130 /** lowers ffract to fsub+ffloor: */
2133 /** lowers fceil to fneg+ffloor+fneg: */
2138 bool lower_pack_half_2x16
;
2139 bool lower_pack_unorm_2x16
;
2140 bool lower_pack_snorm_2x16
;
2141 bool lower_pack_unorm_4x8
;
2142 bool lower_pack_snorm_4x8
;
2143 bool lower_unpack_half_2x16
;
2144 bool lower_unpack_unorm_2x16
;
2145 bool lower_unpack_snorm_2x16
;
2146 bool lower_unpack_unorm_4x8
;
2147 bool lower_unpack_snorm_4x8
;
2149 bool lower_extract_byte
;
2150 bool lower_extract_word
;
2152 bool lower_all_io_to_temps
;
2155 * Does the driver support real 32-bit integers? (Otherwise, integers
2156 * are simulated by floats.)
2158 bool native_integers
;
2160 /* Indicates that the driver only has zero-based vertex id */
2161 bool vertex_id_zero_based
;
2164 * If enabled, gl_BaseVertex will be lowered as:
2165 * is_indexed_draw (~0/0) & firstvertex
2167 bool lower_base_vertex
;
2170 * If enabled, gl_HelperInvocation will be lowered as:
2172 * !((1 << sample_id) & sample_mask_in))
2174 * This depends on some possibly hw implementation details, which may
2175 * not be true for all hw. In particular that the FS is only executed
2176 * for covered samples or for helper invocations. So, do not blindly
2177 * enable this option.
2179 * Note: See also issue #22 in ARB_shader_image_load_store
2181 bool lower_helper_invocation
;
2183 bool lower_cs_local_index_from_id
;
2184 bool lower_cs_local_id_from_index
;
2186 bool lower_device_index_to_zero
;
2188 /* Set if nir_lower_wpos_ytransform() should also invert gl_PointCoord. */
2189 bool lower_wpos_pntc
;
2192 * Should nir_lower_io() create load_interpolated_input intrinsics?
2194 * If not, it generates regular load_input intrinsics and interpolation
2195 * information must be inferred from the list of input nir_variables.
2197 bool use_interpolated_input_intrinsics
;
2199 unsigned max_unroll_iterations
;
2200 } nir_shader_compiler_options
;
2202 typedef struct nir_shader
{
2203 /** list of uniforms (nir_variable) */
2204 struct exec_list uniforms
;
2206 /** list of inputs (nir_variable) */
2207 struct exec_list inputs
;
2209 /** list of outputs (nir_variable) */
2210 struct exec_list outputs
;
2212 /** list of shared compute variables (nir_variable) */
2213 struct exec_list shared
;
2215 /** Set of driver-specific options for the shader.
2217 * The memory for the options is expected to be kept in a single static
2218 * copy by the driver.
2220 const struct nir_shader_compiler_options
*options
;
2222 /** Various bits of compile-time information about a given shader */
2223 struct shader_info info
;
2225 /** list of global variables in the shader (nir_variable) */
2226 struct exec_list globals
;
2228 /** list of system value variables in the shader (nir_variable) */
2229 struct exec_list system_values
;
2231 struct exec_list functions
; /** < list of nir_function */
2233 /** list of global register in the shader */
2234 struct exec_list registers
;
2236 /** next available global register index */
2240 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
2243 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
2245 /** Constant data associated with this shader.
2247 * Constant data is loaded through load_constant intrinsics. See also
2248 * nir_opt_large_constants.
2250 void *constant_data
;
2251 unsigned constant_data_size
;
2254 static inline nir_function_impl
*
2255 nir_shader_get_entrypoint(nir_shader
*shader
)
2257 assert(exec_list_length(&shader
->functions
) == 1);
2258 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
2259 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
2260 assert(func
->num_params
== 0);
2265 #define nir_foreach_function(func, shader) \
2266 foreach_list_typed(nir_function, func, node, &(shader)->functions)
2268 nir_shader
*nir_shader_create(void *mem_ctx
,
2269 gl_shader_stage stage
,
2270 const nir_shader_compiler_options
*options
,
2273 /** creates a register, including assigning it an index and adding it to the list */
2274 nir_register
*nir_global_reg_create(nir_shader
*shader
);
2276 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
2278 void nir_reg_remove(nir_register
*reg
);
2280 /** Adds a variable to the appropriate list in nir_shader */
2281 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
2284 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
2286 assert(var
->data
.mode
== nir_var_local
);
2287 exec_list_push_tail(&impl
->locals
, &var
->node
);
2290 /** creates a variable, sets a few defaults, and adds it to the list */
2291 nir_variable
*nir_variable_create(nir_shader
*shader
,
2292 nir_variable_mode mode
,
2293 const struct glsl_type
*type
,
2295 /** creates a local variable and adds it to the list */
2296 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
2297 const struct glsl_type
*type
,
2300 /** creates a function and adds it to the shader's list of functions */
2301 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
2303 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
2304 /** creates a function_impl that isn't tied to any particular function */
2305 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
2307 nir_block
*nir_block_create(nir_shader
*shader
);
2308 nir_if
*nir_if_create(nir_shader
*shader
);
2309 nir_loop
*nir_loop_create(nir_shader
*shader
);
2311 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
2313 /** requests that the given pieces of metadata be generated */
2314 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
2315 /** dirties all but the preserved metadata */
2316 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
2318 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
2319 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
2321 nir_deref_instr
*nir_deref_instr_create(nir_shader
*shader
,
2322 nir_deref_type deref_type
);
2324 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
2326 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
2327 unsigned num_components
,
2330 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
2331 nir_intrinsic_op op
);
2333 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
2334 nir_function
*callee
);
2336 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
2338 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
2340 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
2342 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
2343 unsigned num_components
,
2346 nir_const_value
nir_alu_binop_identity(nir_op binop
, unsigned bit_size
);
2349 * NIR Cursors and Instruction Insertion API
2352 * A tiny struct representing a point to insert/extract instructions or
2353 * control flow nodes. Helps reduce the combinatorial explosion of possible
2354 * points to insert/extract.
2356 * \sa nir_control_flow.h
2359 nir_cursor_before_block
,
2360 nir_cursor_after_block
,
2361 nir_cursor_before_instr
,
2362 nir_cursor_after_instr
,
2363 } nir_cursor_option
;
2366 nir_cursor_option option
;
2373 static inline nir_block
*
2374 nir_cursor_current_block(nir_cursor cursor
)
2376 if (cursor
.option
== nir_cursor_before_instr
||
2377 cursor
.option
== nir_cursor_after_instr
) {
2378 return cursor
.instr
->block
;
2380 return cursor
.block
;
2384 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2386 static inline nir_cursor
2387 nir_before_block(nir_block
*block
)
2390 cursor
.option
= nir_cursor_before_block
;
2391 cursor
.block
= block
;
2395 static inline nir_cursor
2396 nir_after_block(nir_block
*block
)
2399 cursor
.option
= nir_cursor_after_block
;
2400 cursor
.block
= block
;
2404 static inline nir_cursor
2405 nir_before_instr(nir_instr
*instr
)
2408 cursor
.option
= nir_cursor_before_instr
;
2409 cursor
.instr
= instr
;
2413 static inline nir_cursor
2414 nir_after_instr(nir_instr
*instr
)
2417 cursor
.option
= nir_cursor_after_instr
;
2418 cursor
.instr
= instr
;
2422 static inline nir_cursor
2423 nir_after_block_before_jump(nir_block
*block
)
2425 nir_instr
*last_instr
= nir_block_last_instr(block
);
2426 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2427 return nir_before_instr(last_instr
);
2429 return nir_after_block(block
);
2433 static inline nir_cursor
2434 nir_before_src(nir_src
*src
, bool is_if_condition
)
2436 if (is_if_condition
) {
2437 nir_block
*prev_block
=
2438 nir_cf_node_as_block(nir_cf_node_prev(&src
->parent_if
->cf_node
));
2439 assert(!nir_block_ends_in_jump(prev_block
));
2440 return nir_after_block(prev_block
);
2441 } else if (src
->parent_instr
->type
== nir_instr_type_phi
) {
2443 nir_phi_instr
*cond_phi
= nir_instr_as_phi(src
->parent_instr
);
2445 nir_foreach_phi_src(phi_src
, cond_phi
) {
2446 if (phi_src
->src
.ssa
== src
->ssa
) {
2453 /* The LIST_ENTRY macro is a generic container-of macro, it just happens
2454 * to have a more specific name.
2456 nir_phi_src
*phi_src
= LIST_ENTRY(nir_phi_src
, src
, src
);
2457 return nir_after_block_before_jump(phi_src
->pred
);
2459 return nir_before_instr(src
->parent_instr
);
2463 static inline nir_cursor
2464 nir_before_cf_node(nir_cf_node
*node
)
2466 if (node
->type
== nir_cf_node_block
)
2467 return nir_before_block(nir_cf_node_as_block(node
));
2469 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2472 static inline nir_cursor
2473 nir_after_cf_node(nir_cf_node
*node
)
2475 if (node
->type
== nir_cf_node_block
)
2476 return nir_after_block(nir_cf_node_as_block(node
));
2478 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2481 static inline nir_cursor
2482 nir_after_phis(nir_block
*block
)
2484 nir_foreach_instr(instr
, block
) {
2485 if (instr
->type
!= nir_instr_type_phi
)
2486 return nir_before_instr(instr
);
2488 return nir_after_block(block
);
2491 static inline nir_cursor
2492 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2494 if (node
->type
== nir_cf_node_block
)
2495 return nir_after_block(nir_cf_node_as_block(node
));
2497 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2499 return nir_after_phis(block
);
2502 static inline nir_cursor
2503 nir_before_cf_list(struct exec_list
*cf_list
)
2505 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2506 exec_list_get_head(cf_list
), node
);
2507 return nir_before_cf_node(first_node
);
2510 static inline nir_cursor
2511 nir_after_cf_list(struct exec_list
*cf_list
)
2513 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2514 exec_list_get_tail(cf_list
), node
);
2515 return nir_after_cf_node(last_node
);
2519 * Insert a NIR instruction at the given cursor.
2521 * Note: This does not update the cursor.
2523 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2526 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2528 nir_instr_insert(nir_before_instr(instr
), before
);
2532 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2534 nir_instr_insert(nir_after_instr(instr
), after
);
2538 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2540 nir_instr_insert(nir_before_block(block
), before
);
2544 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2546 nir_instr_insert(nir_after_block(block
), after
);
2550 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2552 nir_instr_insert(nir_before_cf_node(node
), before
);
2556 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2558 nir_instr_insert(nir_after_cf_node(node
), after
);
2562 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2564 nir_instr_insert(nir_before_cf_list(list
), before
);
2568 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2570 nir_instr_insert(nir_after_cf_list(list
), after
);
2573 void nir_instr_remove_v(nir_instr
*instr
);
2575 static inline nir_cursor
2576 nir_instr_remove(nir_instr
*instr
)
2579 nir_instr
*prev
= nir_instr_prev(instr
);
2581 cursor
= nir_after_instr(prev
);
2583 cursor
= nir_before_block(instr
->block
);
2585 nir_instr_remove_v(instr
);
2591 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2592 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2593 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2594 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2596 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2597 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2599 nir_const_value
*nir_src_as_const_value(nir_src src
);
2601 static inline struct nir_instr
*
2602 nir_src_instr(const struct nir_src
*src
)
2604 return src
->is_ssa
? src
->ssa
->parent_instr
: NULL
;
2607 #define NIR_SRC_AS_(name, c_type, type_enum, cast_macro) \
2608 static inline c_type * \
2609 nir_src_as_ ## name (struct nir_src *src) \
2611 return src->is_ssa && src->ssa->parent_instr->type == type_enum \
2612 ? cast_macro(src->ssa->parent_instr) : NULL; \
2614 static inline const c_type * \
2615 nir_src_as_ ## name ## _const(const struct nir_src *src) \
2617 return src->is_ssa && src->ssa->parent_instr->type == type_enum \
2618 ? cast_macro(src->ssa->parent_instr) : NULL; \
2621 NIR_SRC_AS_(alu_instr
, nir_alu_instr
, nir_instr_type_alu
, nir_instr_as_alu
)
2623 bool nir_src_is_dynamically_uniform(nir_src src
);
2624 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2625 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2626 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2627 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2628 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2631 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2632 unsigned num_components
, unsigned bit_size
,
2634 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2635 unsigned num_components
, unsigned bit_size
,
2638 nir_ssa_dest_init_for_type(nir_instr
*instr
, nir_dest
*dest
,
2639 const struct glsl_type
*type
,
2642 assert(glsl_type_is_vector_or_scalar(type
));
2643 nir_ssa_dest_init(instr
, dest
, glsl_get_components(type
),
2644 glsl_get_bit_size(type
), name
);
2646 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2647 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2648 nir_instr
*after_me
);
2650 nir_component_mask_t
nir_ssa_def_components_read(const nir_ssa_def
*def
);
2653 * finds the next basic block in source-code order, returns NULL if there is
2657 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2659 /* Performs the opposite of nir_block_cf_tree_next() */
2661 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2663 /* Gets the first block in a CF node in source-code order */
2665 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2667 /* Gets the last block in a CF node in source-code order */
2669 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2671 /* Gets the next block after a CF node in source-code order */
2673 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2675 /* Macros for loops that visit blocks in source-code order */
2677 #define nir_foreach_block(block, impl) \
2678 for (nir_block *block = nir_start_block(impl); block != NULL; \
2679 block = nir_block_cf_tree_next(block))
2681 #define nir_foreach_block_safe(block, impl) \
2682 for (nir_block *block = nir_start_block(impl), \
2683 *next = nir_block_cf_tree_next(block); \
2685 block = next, next = nir_block_cf_tree_next(block))
2687 #define nir_foreach_block_reverse(block, impl) \
2688 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2689 block = nir_block_cf_tree_prev(block))
2691 #define nir_foreach_block_reverse_safe(block, impl) \
2692 for (nir_block *block = nir_impl_last_block(impl), \
2693 *prev = nir_block_cf_tree_prev(block); \
2695 block = prev, prev = nir_block_cf_tree_prev(block))
2697 #define nir_foreach_block_in_cf_node(block, node) \
2698 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2699 block != nir_cf_node_cf_tree_next(node); \
2700 block = nir_block_cf_tree_next(block))
2702 /* If the following CF node is an if, this function returns that if.
2703 * Otherwise, it returns NULL.
2705 nir_if
*nir_block_get_following_if(nir_block
*block
);
2707 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2709 void nir_index_local_regs(nir_function_impl
*impl
);
2710 void nir_index_global_regs(nir_shader
*shader
);
2711 void nir_index_ssa_defs(nir_function_impl
*impl
);
2712 unsigned nir_index_instrs(nir_function_impl
*impl
);
2714 void nir_index_blocks(nir_function_impl
*impl
);
2716 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2717 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2718 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2720 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2721 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2722 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2723 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2725 nir_shader
*nir_shader_serialize_deserialize(void *mem_ctx
, nir_shader
*s
);
2728 void nir_validate_shader(nir_shader
*shader
, const char *when
);
2729 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2730 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2733 should_clone_nir(void)
2735 static int should_clone
= -1;
2736 if (should_clone
< 0)
2737 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2739 return should_clone
;
2743 should_serialize_deserialize_nir(void)
2745 static int test_serialize
= -1;
2746 if (test_serialize
< 0)
2747 test_serialize
= env_var_as_boolean("NIR_TEST_SERIALIZE", false);
2749 return test_serialize
;
2753 should_print_nir(void)
2755 static int should_print
= -1;
2756 if (should_print
< 0)
2757 should_print
= env_var_as_boolean("NIR_PRINT", false);
2759 return should_print
;
2762 static inline void nir_validate_shader(nir_shader
*shader
, const char *when
) { (void) shader
; (void)when
; }
2763 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2764 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2765 static inline bool should_clone_nir(void) { return false; }
2766 static inline bool should_serialize_deserialize_nir(void) { return false; }
2767 static inline bool should_print_nir(void) { return false; }
2770 #define _PASS(pass, nir, do_pass) do { \
2772 nir_validate_shader(nir, "after " #pass); \
2773 if (should_clone_nir()) { \
2774 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2778 if (should_serialize_deserialize_nir()) { \
2779 void *mem_ctx = ralloc_parent(nir); \
2780 nir = nir_shader_serialize_deserialize(mem_ctx, nir); \
2784 #define NIR_PASS(progress, nir, pass, ...) _PASS(pass, nir, \
2785 nir_metadata_set_validation_flag(nir); \
2786 if (should_print_nir()) \
2787 printf("%s\n", #pass); \
2788 if (pass(nir, ##__VA_ARGS__)) { \
2790 if (should_print_nir()) \
2791 nir_print_shader(nir, stdout); \
2792 nir_metadata_check_validation_flag(nir); \
2796 #define NIR_PASS_V(nir, pass, ...) _PASS(pass, nir, \
2797 if (should_print_nir()) \
2798 printf("%s\n", #pass); \
2799 pass(nir, ##__VA_ARGS__); \
2800 if (should_print_nir()) \
2801 nir_print_shader(nir, stdout); \
2804 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2805 void nir_calc_dominance(nir_shader
*shader
);
2807 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2808 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2810 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2811 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2813 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2814 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2816 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2817 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2819 int nir_gs_count_vertices(const nir_shader
*shader
);
2821 bool nir_shrink_vec_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2822 bool nir_split_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2823 bool nir_split_var_copies(nir_shader
*shader
);
2824 bool nir_split_per_member_structs(nir_shader
*shader
);
2825 bool nir_split_struct_vars(nir_shader
*shader
, nir_variable_mode modes
);
2827 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2828 bool nir_lower_returns(nir_shader
*shader
);
2830 bool nir_inline_functions(nir_shader
*shader
);
2832 bool nir_propagate_invariant(nir_shader
*shader
);
2834 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2835 void nir_lower_deref_copy_instr(struct nir_builder
*b
,
2836 nir_intrinsic_instr
*copy
);
2837 bool nir_lower_var_copies(nir_shader
*shader
);
2839 void nir_fixup_deref_modes(nir_shader
*shader
);
2841 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2843 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2845 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2847 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2848 nir_function_impl
*entrypoint
,
2849 bool outputs
, bool inputs
);
2851 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2853 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2854 int (*type_size
)(const struct glsl_type
*));
2856 /* Some helpers to do very simple linking */
2857 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2858 bool nir_remove_unused_io_vars(nir_shader
*shader
, struct exec_list
*var_list
,
2859 uint64_t *used_by_other_stage
,
2860 uint64_t *used_by_other_stage_patches
);
2861 void nir_compact_varyings(nir_shader
*producer
, nir_shader
*consumer
,
2862 bool default_to_smooth_interp
);
2863 void nir_link_xfb_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2864 bool nir_link_constant_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2867 /* If set, this forces all non-flat fragment shader inputs to be
2868 * interpolated as if with the "sample" qualifier. This requires
2869 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2871 nir_lower_io_force_sample_interpolation
= (1 << 1),
2872 } nir_lower_io_options
;
2873 bool nir_lower_io(nir_shader
*shader
,
2874 nir_variable_mode modes
,
2875 int (*type_size
)(const struct glsl_type
*),
2876 nir_lower_io_options
);
2877 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2878 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2880 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2882 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2883 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2884 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2886 bool nir_remove_dead_derefs(nir_shader
*shader
);
2887 bool nir_remove_dead_derefs_impl(nir_function_impl
*impl
);
2888 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2889 bool nir_lower_constant_initializers(nir_shader
*shader
,
2890 nir_variable_mode modes
);
2892 bool nir_move_load_const(nir_shader
*shader
);
2893 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2894 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2895 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
2897 bool nir_lower_alu(nir_shader
*shader
);
2898 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2899 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2900 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
2901 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2902 void nir_lower_io_arrays_to_elements(nir_shader
*producer
, nir_shader
*consumer
);
2903 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader
*shader
,
2905 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2906 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
2908 typedef struct nir_lower_subgroups_options
{
2909 uint8_t subgroup_size
;
2910 uint8_t ballot_bit_size
;
2911 bool lower_to_scalar
:1;
2912 bool lower_vote_trivial
:1;
2913 bool lower_vote_eq_to_ballot
:1;
2914 bool lower_subgroup_masks
:1;
2915 bool lower_shuffle
:1;
2916 bool lower_shuffle_to_32bit
:1;
2918 } nir_lower_subgroups_options
;
2920 bool nir_lower_subgroups(nir_shader
*shader
,
2921 const nir_lower_subgroups_options
*options
);
2923 bool nir_lower_system_values(nir_shader
*shader
);
2925 typedef struct nir_lower_tex_options
{
2927 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2928 * sampler types a texture projector is lowered.
2933 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2935 bool lower_txf_offset
;
2938 * If true, lower away nir_tex_src_offset for all rect textures.
2940 bool lower_rect_offset
;
2943 * If true, lower rect textures to 2D, using txs to fetch the
2944 * texture dimensions and dividing the texture coords by the
2945 * texture dims to normalize.
2950 * If true, convert yuv to rgb.
2952 unsigned lower_y_uv_external
;
2953 unsigned lower_y_u_v_external
;
2954 unsigned lower_yx_xuxv_external
;
2955 unsigned lower_xy_uxvx_external
;
2956 unsigned lower_ayuv_external
;
2959 * To emulate certain texture wrap modes, this can be used
2960 * to saturate the specified tex coord to [0.0, 1.0]. The
2961 * bits are according to sampler #, ie. if, for example:
2963 * (conf->saturate_s & (1 << n))
2965 * is true, then the s coord for sampler n is saturated.
2967 * Note that clamping must happen *after* projector lowering
2968 * so any projected texture sample instruction with a clamped
2969 * coordinate gets automatically lowered, regardless of the
2970 * 'lower_txp' setting.
2972 unsigned saturate_s
;
2973 unsigned saturate_t
;
2974 unsigned saturate_r
;
2976 /* Bitmask of textures that need swizzling.
2978 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2979 * swizzles[texture_index] is applied to the result of the texturing
2982 unsigned swizzle_result
;
2984 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2985 * while 4 and 5 represent 0 and 1 respectively.
2987 uint8_t swizzles
[32][4];
2990 * Bitmap of textures that need srgb to linear conversion. If
2991 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2992 * of the texture are lowered to linear.
2994 unsigned lower_srgb
;
2997 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2999 bool lower_txd_cube_map
;
3002 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
3003 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
3004 * with lower_txd_cube_map.
3006 bool lower_txd_shadow
;
3009 * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
3010 * Implies lower_txd_cube_map and lower_txd_shadow.
3013 } nir_lower_tex_options
;
3015 bool nir_lower_tex(nir_shader
*shader
,
3016 const nir_lower_tex_options
*options
);
3018 bool nir_lower_idiv(nir_shader
*shader
);
3020 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
, bool use_vars
);
3021 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
3022 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
3024 void nir_lower_two_sided_color(nir_shader
*shader
);
3026 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
3028 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
3029 bool nir_lower_patch_vertices(nir_shader
*nir
, unsigned static_count
,
3030 const gl_state_index16
*uniform_state_tokens
);
3032 typedef struct nir_lower_wpos_ytransform_options
{
3033 gl_state_index16 state_tokens
[STATE_LENGTH
];
3034 bool fs_coord_origin_upper_left
:1;
3035 bool fs_coord_origin_lower_left
:1;
3036 bool fs_coord_pixel_center_integer
:1;
3037 bool fs_coord_pixel_center_half_integer
:1;
3038 } nir_lower_wpos_ytransform_options
;
3040 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
3041 const nir_lower_wpos_ytransform_options
*options
);
3042 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
3044 typedef struct nir_lower_drawpixels_options
{
3045 gl_state_index16 texcoord_state_tokens
[STATE_LENGTH
];
3046 gl_state_index16 scale_state_tokens
[STATE_LENGTH
];
3047 gl_state_index16 bias_state_tokens
[STATE_LENGTH
];
3048 unsigned drawpix_sampler
;
3049 unsigned pixelmap_sampler
;
3051 bool scale_and_bias
:1;
3052 } nir_lower_drawpixels_options
;
3054 void nir_lower_drawpixels(nir_shader
*shader
,
3055 const nir_lower_drawpixels_options
*options
);
3057 typedef struct nir_lower_bitmap_options
{
3060 } nir_lower_bitmap_options
;
3062 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
3064 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
3067 nir_lower_int_source_mods
= 1 << 0,
3068 nir_lower_float_source_mods
= 1 << 1,
3069 nir_lower_all_source_mods
= (1 << 2) - 1
3070 } nir_lower_to_source_mods_flags
;
3073 bool nir_lower_to_source_mods(nir_shader
*shader
, nir_lower_to_source_mods_flags options
);
3075 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
3077 typedef unsigned (*nir_lower_bit_size_callback
)(const nir_alu_instr
*, void *);
3079 bool nir_lower_bit_size(nir_shader
*shader
,
3080 nir_lower_bit_size_callback callback
,
3081 void *callback_data
);
3084 nir_lower_imul64
= (1 << 0),
3085 nir_lower_isign64
= (1 << 1),
3086 /** Lower all int64 modulus and division opcodes */
3087 nir_lower_divmod64
= (1 << 2),
3088 } nir_lower_int64_options
;
3090 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
3093 nir_lower_drcp
= (1 << 0),
3094 nir_lower_dsqrt
= (1 << 1),
3095 nir_lower_drsq
= (1 << 2),
3096 nir_lower_dtrunc
= (1 << 3),
3097 nir_lower_dfloor
= (1 << 4),
3098 nir_lower_dceil
= (1 << 5),
3099 nir_lower_dfract
= (1 << 6),
3100 nir_lower_dround_even
= (1 << 7),
3101 nir_lower_dmod
= (1 << 8)
3102 } nir_lower_doubles_options
;
3104 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
3105 bool nir_lower_pack(nir_shader
*shader
);
3107 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
3109 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
3111 void nir_loop_analyze_impl(nir_function_impl
*impl
,
3112 nir_variable_mode indirect_mask
);
3114 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
3116 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
3117 bool nir_repair_ssa(nir_shader
*shader
);
3119 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
3121 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
3122 * registers. If false, convert all values (even those not involved in a phi
3123 * node) to registers.
3125 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
3127 bool nir_lower_phis_to_regs_block(nir_block
*block
);
3128 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
3129 bool nir_rematerialize_derefs_in_use_blocks_impl(nir_function_impl
*impl
);
3131 bool nir_opt_algebraic(nir_shader
*shader
);
3132 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
3133 bool nir_opt_algebraic_late(nir_shader
*shader
);
3134 bool nir_opt_constant_folding(nir_shader
*shader
);
3136 bool nir_opt_global_to_local(nir_shader
*shader
);
3138 bool nir_copy_prop(nir_shader
*shader
);
3140 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
3142 bool nir_opt_cse(nir_shader
*shader
);
3144 bool nir_opt_dce(nir_shader
*shader
);
3146 bool nir_opt_dead_cf(nir_shader
*shader
);
3148 bool nir_opt_dead_write_vars(nir_shader
*shader
);
3150 bool nir_opt_find_array_copies(nir_shader
*shader
);
3152 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
3154 bool nir_opt_if(nir_shader
*shader
);
3156 bool nir_opt_intrinsics(nir_shader
*shader
);
3158 bool nir_opt_large_constants(nir_shader
*shader
,
3159 glsl_type_size_align_func size_align
,
3160 unsigned threshold
);
3162 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
3164 bool nir_opt_move_comparisons(nir_shader
*shader
);
3166 bool nir_opt_move_load_ubo(nir_shader
*shader
);
3168 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
3170 bool nir_opt_remove_phis_impl(nir_function_impl
*impl
);
3171 bool nir_opt_remove_phis(nir_shader
*shader
);
3173 bool nir_opt_shrink_load(nir_shader
*shader
);
3175 bool nir_opt_trivial_continues(nir_shader
*shader
);
3177 bool nir_opt_undef(nir_shader
*shader
);
3179 bool nir_opt_conditional_discard(nir_shader
*shader
);
3181 void nir_sweep(nir_shader
*shader
);
3183 void nir_remap_dual_slot_attributes(nir_shader
*shader
,
3184 uint64_t *dual_slot_inputs
);
3185 uint64_t nir_get_single_slot_attribs_mask(uint64_t attribs
, uint64_t dual_slot
);
3187 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
3188 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);