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_ubo
= (1 << 5),
104 nir_var_system_value
= (1 << 6),
105 nir_var_ssbo
= (1 << 7),
106 nir_var_shared
= (1 << 8),
114 nir_rounding_mode_undef
= 0,
115 nir_rounding_mode_rtne
= 1, /* round to nearest even */
116 nir_rounding_mode_ru
= 2, /* round up */
117 nir_rounding_mode_rd
= 3, /* round down */
118 nir_rounding_mode_rtz
= 4, /* round towards zero */
122 bool b
[NIR_MAX_VEC_COMPONENTS
];
123 float f32
[NIR_MAX_VEC_COMPONENTS
];
124 double f64
[NIR_MAX_VEC_COMPONENTS
];
125 int8_t i8
[NIR_MAX_VEC_COMPONENTS
];
126 uint8_t u8
[NIR_MAX_VEC_COMPONENTS
];
127 int16_t i16
[NIR_MAX_VEC_COMPONENTS
];
128 uint16_t u16
[NIR_MAX_VEC_COMPONENTS
];
129 int32_t i32
[NIR_MAX_VEC_COMPONENTS
];
130 uint32_t u32
[NIR_MAX_VEC_COMPONENTS
];
131 int64_t i64
[NIR_MAX_VEC_COMPONENTS
];
132 uint64_t u64
[NIR_MAX_VEC_COMPONENTS
];
135 typedef struct nir_constant
{
137 * Value of the constant.
139 * The field used to back the values supplied by the constant is determined
140 * by the type associated with the \c nir_variable. Constants may be
141 * scalars, vectors, or matrices.
143 nir_const_value values
[NIR_MAX_VEC_COMPONENTS
];
145 /* we could get this from the var->type but makes clone *much* easier to
146 * not have to care about the type.
148 unsigned num_elements
;
150 /* Array elements / Structure Fields */
151 struct nir_constant
**elements
;
155 * \brief Layout qualifiers for gl_FragDepth.
157 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
158 * with a layout qualifier.
161 nir_depth_layout_none
, /**< No depth layout is specified. */
162 nir_depth_layout_any
,
163 nir_depth_layout_greater
,
164 nir_depth_layout_less
,
165 nir_depth_layout_unchanged
169 * Enum keeping track of how a variable was declared.
173 * Normal declaration.
175 nir_var_declared_normally
= 0,
178 * Variable is implicitly generated by the compiler and should not be
179 * visible via the API.
182 } nir_var_declaration_type
;
185 * Either a uniform, global variable, shader input, or shader output. Based on
186 * ir_variable - it should be easy to translate between the two.
189 typedef struct nir_variable
{
190 struct exec_node node
;
193 * Declared type of the variable
195 const struct glsl_type
*type
;
198 * Declared name of the variable
202 struct nir_variable_data
{
204 * Storage class of the variable.
206 * \sa nir_variable_mode
208 nir_variable_mode mode
;
211 * Is the variable read-only?
213 * This is set for variables declared as \c const, shader inputs,
216 unsigned read_only
:1;
220 unsigned invariant
:1;
223 * When separate shader programs are enabled, only input/outputs between
224 * the stages of a multi-stage separate program can be safely removed
225 * from the shader interface. Other input/outputs must remains active.
227 * This is also used to make sure xfb varyings that are unused by the
228 * fragment shader are not removed.
230 unsigned always_active_io
:1;
233 * Interpolation mode for shader inputs / outputs
235 * \sa glsl_interp_mode
237 unsigned interpolation
:2;
240 * \name ARB_fragment_coord_conventions
243 unsigned origin_upper_left
:1;
244 unsigned pixel_center_integer
:1;
248 * If non-zero, then this variable may be packed along with other variables
249 * into a single varying slot, so this offset should be applied when
250 * accessing components. For example, an offset of 1 means that the x
251 * component of this variable is actually stored in component y of the
252 * location specified by \c location.
254 unsigned location_frac
:2;
257 * If true, this variable represents an array of scalars that should
258 * be tightly packed. In other words, consecutive array elements
259 * should be stored one component apart, rather than one slot apart.
264 * Whether this is a fragment shader output implicitly initialized with
265 * the previous contents of the specified render target at the
266 * framebuffer location corresponding to this shader invocation.
268 unsigned fb_fetch_output
:1;
271 * Non-zero if this variable is considered bindless as defined by
272 * ARB_bindless_texture.
277 * Was an explicit binding set in the shader?
279 unsigned explicit_binding
:1;
282 * Was a transfer feedback buffer set in the shader?
284 unsigned explicit_xfb_buffer
:1;
287 * Was a transfer feedback stride set in the shader?
289 unsigned explicit_xfb_stride
:1;
292 * Was an explicit offset set in the shader?
294 unsigned explicit_offset
:1;
297 * \brief Layout qualifier for gl_FragDepth.
299 * This is not equal to \c ir_depth_layout_none if and only if this
300 * variable is \c gl_FragDepth and a layout qualifier is specified.
302 nir_depth_layout depth_layout
;
305 * Storage location of the base of this variable
307 * The precise meaning of this field depends on the nature of the variable.
309 * - Vertex shader input: one of the values from \c gl_vert_attrib.
310 * - Vertex shader output: one of the values from \c gl_varying_slot.
311 * - Geometry shader input: one of the values from \c gl_varying_slot.
312 * - Geometry shader output: one of the values from \c gl_varying_slot.
313 * - Fragment shader input: one of the values from \c gl_varying_slot.
314 * - Fragment shader output: one of the values from \c gl_frag_result.
315 * - Uniforms: Per-stage uniform slot number for default uniform block.
316 * - Uniforms: Index within the uniform block definition for UBO members.
317 * - Non-UBO Uniforms: uniform slot number.
318 * - Other: This field is not currently used.
320 * If the variable is a uniform, shader input, or shader output, and the
321 * slot has not been assigned, the value will be -1.
326 * The actual location of the variable in the IR. Only valid for inputs
329 unsigned int driver_location
;
332 * Vertex stream output identifier.
334 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
335 * stream of the i-th component.
340 * output index for dual source blending.
345 * Descriptor set binding for sampler or UBO.
350 * Initial binding point for a sampler or UBO.
352 * For array types, this represents the binding point for the first element.
357 * Location an atomic counter or transform feedback is stored at.
362 * Transform feedback buffer.
367 * Transform feedback stride.
372 * How the variable was declared. See nir_var_declaration_type.
374 * This is used to detect variables generated by the compiler, so should
375 * not be visible via the API.
377 unsigned how_declared
:2;
380 * ARB_shader_image_load_store qualifiers.
383 enum gl_access_qualifier access
;
385 /** Image internal format if specified explicitly, otherwise GL_NONE. */
391 * Built-in state that backs this uniform
393 * Once set at variable creation, \c state_slots must remain invariant.
394 * This is because, ideally, this array would be shared by all clones of
395 * this variable in the IR tree. In other words, we'd really like for it
396 * to be a fly-weight.
398 * If the variable is not a uniform, \c num_state_slots will be zero and
399 * \c state_slots will be \c NULL.
402 unsigned num_state_slots
; /**< Number of state slots used */
403 nir_state_slot
*state_slots
; /**< State descriptors. */
407 * Constant expression assigned in the initializer of the variable
409 * This field should only be used temporarily by creators of NIR shaders
410 * and then lower_constant_initializers can be used to get rid of them.
411 * Most of the rest of NIR ignores this field or asserts that it's NULL.
413 nir_constant
*constant_initializer
;
416 * For variables that are in an interface block or are an instance of an
417 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
419 * \sa ir_variable::location
421 const struct glsl_type
*interface_type
;
424 * Description of per-member data for per-member struct variables
426 * This is used for variables which are actually an amalgamation of
427 * multiple entities such as a struct of built-in values or a struct of
428 * inputs each with their own layout specifier. This is only allowed on
429 * variables with a struct or array of array of struct type.
431 unsigned num_members
;
432 struct nir_variable_data
*members
;
435 #define nir_foreach_variable(var, var_list) \
436 foreach_list_typed(nir_variable, var, node, var_list)
438 #define nir_foreach_variable_safe(var, var_list) \
439 foreach_list_typed_safe(nir_variable, var, node, var_list)
442 nir_variable_is_global(const nir_variable
*var
)
444 return var
->data
.mode
!= nir_var_local
;
447 typedef struct nir_register
{
448 struct exec_node node
;
450 unsigned num_components
; /** < number of vector components */
451 unsigned num_array_elems
; /** < size of array (0 for no array) */
453 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
456 /** generic register index. */
459 /** only for debug purposes, can be NULL */
462 /** whether this register is local (per-function) or global (per-shader) */
466 * If this flag is set to true, then accessing channels >= num_components
467 * is well-defined, and simply spills over to the next array element. This
468 * is useful for backends that can do per-component accessing, in
469 * particular scalar backends. By setting this flag and making
470 * num_components equal to 1, structures can be packed tightly into
471 * registers and then registers can be accessed per-component to get to
472 * each structure member, even if it crosses vec4 boundaries.
476 /** set of nir_srcs where this register is used (read from) */
477 struct list_head uses
;
479 /** set of nir_dests where this register is defined (written to) */
480 struct list_head defs
;
482 /** set of nir_ifs where this register is used as a condition */
483 struct list_head if_uses
;
486 #define nir_foreach_register(reg, reg_list) \
487 foreach_list_typed(nir_register, reg, node, reg_list)
488 #define nir_foreach_register_safe(reg, reg_list) \
489 foreach_list_typed_safe(nir_register, reg, node, reg_list)
491 typedef enum PACKED
{
493 nir_instr_type_deref
,
496 nir_instr_type_intrinsic
,
497 nir_instr_type_load_const
,
499 nir_instr_type_ssa_undef
,
501 nir_instr_type_parallel_copy
,
504 typedef struct nir_instr
{
505 struct exec_node node
;
506 struct nir_block
*block
;
509 /* A temporary for optimization and analysis passes to use for storing
510 * flags. For instance, DCE uses this to store the "dead/live" info.
514 /** generic instruction index. */
518 static inline nir_instr
*
519 nir_instr_next(nir_instr
*instr
)
521 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
522 if (exec_node_is_tail_sentinel(next
))
525 return exec_node_data(nir_instr
, next
, node
);
528 static inline nir_instr
*
529 nir_instr_prev(nir_instr
*instr
)
531 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
532 if (exec_node_is_head_sentinel(prev
))
535 return exec_node_data(nir_instr
, prev
, node
);
539 nir_instr_is_first(const nir_instr
*instr
)
541 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
545 nir_instr_is_last(const nir_instr
*instr
)
547 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
550 typedef struct nir_ssa_def
{
551 /** for debugging only, can be NULL */
554 /** generic SSA definition index. */
557 /** Index into the live_in and live_out bitfields */
560 /** Instruction which produces this SSA value. */
561 nir_instr
*parent_instr
;
563 /** set of nir_instrs where this register is used (read from) */
564 struct list_head uses
;
566 /** set of nir_ifs where this register is used as a condition */
567 struct list_head if_uses
;
569 uint8_t num_components
;
571 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
579 struct nir_src
*indirect
; /** < NULL for no indirect offset */
580 unsigned base_offset
;
582 /* TODO use-def chain goes here */
586 nir_instr
*parent_instr
;
587 struct list_head def_link
;
590 struct nir_src
*indirect
; /** < NULL for no indirect offset */
591 unsigned base_offset
;
593 /* TODO def-use chain goes here */
598 typedef struct nir_src
{
600 /** Instruction that consumes this value as a source. */
601 nir_instr
*parent_instr
;
602 struct nir_if
*parent_if
;
605 struct list_head use_link
;
615 static inline nir_src
618 nir_src src
= { { NULL
} };
622 #define NIR_SRC_INIT nir_src_init()
624 #define nir_foreach_use(src, reg_or_ssa_def) \
625 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
627 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
628 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
630 #define nir_foreach_if_use(src, reg_or_ssa_def) \
631 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
633 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
634 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
645 static inline nir_dest
648 nir_dest dest
= { { { NULL
} } };
652 #define NIR_DEST_INIT nir_dest_init()
654 #define nir_foreach_def(dest, reg) \
655 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
657 #define nir_foreach_def_safe(dest, reg) \
658 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
660 static inline nir_src
661 nir_src_for_ssa(nir_ssa_def
*def
)
663 nir_src src
= NIR_SRC_INIT
;
671 static inline nir_src
672 nir_src_for_reg(nir_register
*reg
)
674 nir_src src
= NIR_SRC_INIT
;
678 src
.reg
.indirect
= NULL
;
679 src
.reg
.base_offset
= 0;
684 static inline nir_dest
685 nir_dest_for_reg(nir_register
*reg
)
687 nir_dest dest
= NIR_DEST_INIT
;
694 static inline unsigned
695 nir_src_bit_size(nir_src src
)
697 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
700 static inline unsigned
701 nir_src_num_components(nir_src src
)
703 return src
.is_ssa
? src
.ssa
->num_components
: src
.reg
.reg
->num_components
;
707 nir_src_is_const(nir_src src
)
710 src
.ssa
->parent_instr
->type
== nir_instr_type_load_const
;
713 int64_t nir_src_as_int(nir_src src
);
714 uint64_t nir_src_as_uint(nir_src src
);
715 bool nir_src_as_bool(nir_src src
);
716 double nir_src_as_float(nir_src src
);
717 int64_t nir_src_comp_as_int(nir_src src
, unsigned component
);
718 uint64_t nir_src_comp_as_uint(nir_src src
, unsigned component
);
719 bool nir_src_comp_as_bool(nir_src src
, unsigned component
);
720 double nir_src_comp_as_float(nir_src src
, unsigned component
);
722 static inline unsigned
723 nir_dest_bit_size(nir_dest dest
)
725 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
728 static inline unsigned
729 nir_dest_num_components(nir_dest dest
)
731 return dest
.is_ssa
? dest
.ssa
.num_components
: dest
.reg
.reg
->num_components
;
734 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
735 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
741 * \name input modifiers
745 * For inputs interpreted as floating point, flips the sign bit. For
746 * inputs interpreted as integers, performs the two's complement negation.
751 * Clears the sign bit for floating point values, and computes the integer
752 * absolute value for integers. Note that the negate modifier acts after
753 * the absolute value modifier, therefore if both are set then all inputs
754 * will become negative.
760 * For each input component, says which component of the register it is
761 * chosen from. Note that which elements of the swizzle are used and which
762 * are ignored are based on the write mask for most opcodes - for example,
763 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
764 * a swizzle of {2, x, 1, 0} where x means "don't care."
766 uint8_t swizzle
[NIR_MAX_VEC_COMPONENTS
];
773 * \name saturate output modifier
775 * Only valid for opcodes that output floating-point numbers. Clamps the
776 * output to between 0.0 and 1.0 inclusive.
781 unsigned write_mask
: NIR_MAX_VEC_COMPONENTS
; /* ignored if dest.is_ssa is true */
784 /** NIR sized and unsized types
786 * The values in this enum are carefully chosen so that the sized type is
787 * just the unsized type OR the number of bits.
790 nir_type_invalid
= 0, /* Not a valid type */
794 nir_type_float
= 128,
795 nir_type_bool1
= 1 | nir_type_bool
,
796 nir_type_bool32
= 32 | nir_type_bool
,
797 nir_type_int1
= 1 | nir_type_int
,
798 nir_type_int8
= 8 | nir_type_int
,
799 nir_type_int16
= 16 | nir_type_int
,
800 nir_type_int32
= 32 | nir_type_int
,
801 nir_type_int64
= 64 | nir_type_int
,
802 nir_type_uint1
= 1 | nir_type_uint
,
803 nir_type_uint8
= 8 | nir_type_uint
,
804 nir_type_uint16
= 16 | nir_type_uint
,
805 nir_type_uint32
= 32 | nir_type_uint
,
806 nir_type_uint64
= 64 | nir_type_uint
,
807 nir_type_float16
= 16 | nir_type_float
,
808 nir_type_float32
= 32 | nir_type_float
,
809 nir_type_float64
= 64 | nir_type_float
,
812 #define NIR_ALU_TYPE_SIZE_MASK 0x79
813 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x86
815 static inline unsigned
816 nir_alu_type_get_type_size(nir_alu_type type
)
818 return type
& NIR_ALU_TYPE_SIZE_MASK
;
821 static inline unsigned
822 nir_alu_type_get_base_type(nir_alu_type type
)
824 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
827 static inline nir_alu_type
828 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
832 return nir_type_bool1
;
835 return nir_type_uint32
;
838 return nir_type_int32
;
840 case GLSL_TYPE_UINT16
:
841 return nir_type_uint16
;
843 case GLSL_TYPE_INT16
:
844 return nir_type_int16
;
846 case GLSL_TYPE_UINT8
:
847 return nir_type_uint8
;
849 return nir_type_int8
;
850 case GLSL_TYPE_UINT64
:
851 return nir_type_uint64
;
853 case GLSL_TYPE_INT64
:
854 return nir_type_int64
;
856 case GLSL_TYPE_FLOAT
:
857 return nir_type_float32
;
859 case GLSL_TYPE_FLOAT16
:
860 return nir_type_float16
;
862 case GLSL_TYPE_DOUBLE
:
863 return nir_type_float64
;
866 unreachable("unknown type");
870 static inline nir_alu_type
871 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
873 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
876 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
,
877 nir_rounding_mode rnd
);
880 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
881 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
882 } nir_op_algebraic_property
;
890 * The number of components in the output
892 * If non-zero, this is the size of the output and input sizes are
893 * explicitly given; swizzle and writemask are still in effect, but if
894 * the output component is masked out, then the input component may
897 * If zero, the opcode acts in the standard, per-component manner; the
898 * operation is performed on each component (except the ones that are
899 * masked out) with the input being taken from the input swizzle for
902 * The size of some of the inputs may be given (i.e. non-zero) even
903 * though output_size is zero; in that case, the inputs with a zero
904 * size act per-component, while the inputs with non-zero size don't.
906 unsigned output_size
;
909 * The type of vector that the instruction outputs. Note that the
910 * staurate modifier is only allowed on outputs with the float type.
913 nir_alu_type output_type
;
916 * The number of components in each input
918 unsigned input_sizes
[NIR_MAX_VEC_COMPONENTS
];
921 * The type of vector that each input takes. Note that negate and
922 * absolute value are only allowed on inputs with int or float type and
923 * behave differently on the two.
925 nir_alu_type input_types
[NIR_MAX_VEC_COMPONENTS
];
927 nir_op_algebraic_property algebraic_properties
;
930 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
932 typedef struct nir_alu_instr
{
936 /** Indicates that this ALU instruction generates an exact value
938 * This is kind of a mixture of GLSL "precise" and "invariant" and not
939 * really equivalent to either. This indicates that the value generated by
940 * this operation is high-precision and any code transformations that touch
941 * it must ensure that the resulting value is bit-for-bit identical to the
950 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
951 nir_alu_instr
*instr
);
952 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
953 nir_alu_instr
*instr
);
955 /* is this source channel used? */
957 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
960 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
961 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
963 return (instr
->dest
.write_mask
>> channel
) & 1;
966 static inline nir_component_mask_t
967 nir_alu_instr_src_read_mask(const nir_alu_instr
*instr
, unsigned src
)
969 nir_component_mask_t read_mask
= 0;
970 for (unsigned c
= 0; c
< NIR_MAX_VEC_COMPONENTS
; c
++) {
971 if (!nir_alu_instr_channel_used(instr
, src
, c
))
974 read_mask
|= (1 << instr
->src
[src
].swizzle
[c
]);
980 * For instructions whose destinations are SSA, get the number of channels
983 static inline unsigned
984 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
986 assert(instr
->dest
.dest
.is_ssa
);
988 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
989 return nir_op_infos
[instr
->op
].input_sizes
[src
];
991 return instr
->dest
.dest
.ssa
.num_components
;
994 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
995 unsigned src1
, unsigned src2
);
999 nir_deref_type_array
,
1000 nir_deref_type_array_wildcard
,
1001 nir_deref_type_struct
,
1002 nir_deref_type_cast
,
1008 /** The type of this deref instruction */
1009 nir_deref_type deref_type
;
1011 /** The mode of the underlying variable */
1012 nir_variable_mode mode
;
1014 /** The dereferenced type of the resulting pointer value */
1015 const struct glsl_type
*type
;
1018 /** Variable being dereferenced if deref_type is a deref_var */
1021 /** Parent deref if deref_type is not deref_var */
1025 /** Additional deref parameters */
1036 /** Destination to store the resulting "pointer" */
1040 NIR_DEFINE_CAST(nir_instr_as_deref
, nir_instr
, nir_deref_instr
, instr
,
1041 type
, nir_instr_type_deref
)
1043 static inline nir_deref_instr
*
1044 nir_src_as_deref(nir_src src
)
1049 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_deref
)
1052 return nir_instr_as_deref(src
.ssa
->parent_instr
);
1055 static inline nir_deref_instr
*
1056 nir_deref_instr_parent(const nir_deref_instr
*instr
)
1058 if (instr
->deref_type
== nir_deref_type_var
)
1061 return nir_src_as_deref(instr
->parent
);
1064 static inline nir_variable
*
1065 nir_deref_instr_get_variable(const nir_deref_instr
*instr
)
1067 while (instr
->deref_type
!= nir_deref_type_var
) {
1068 if (instr
->deref_type
== nir_deref_type_cast
)
1071 instr
= nir_deref_instr_parent(instr
);
1077 bool nir_deref_instr_has_indirect(nir_deref_instr
*instr
);
1079 bool nir_deref_instr_remove_if_unused(nir_deref_instr
*instr
);
1084 struct nir_function
*callee
;
1086 unsigned num_params
;
1090 #include "nir_intrinsics.h"
1092 #define NIR_INTRINSIC_MAX_CONST_INDEX 4
1094 /** Represents an intrinsic
1096 * An intrinsic is an instruction type for handling things that are
1097 * more-or-less regular operations but don't just consume and produce SSA
1098 * values like ALU operations do. Intrinsics are not for things that have
1099 * special semantic meaning such as phi nodes and parallel copies.
1100 * Examples of intrinsics include variable load/store operations, system
1101 * value loads, and the like. Even though texturing more-or-less falls
1102 * under this category, texturing is its own instruction type because
1103 * trying to represent texturing with intrinsics would lead to a
1104 * combinatorial explosion of intrinsic opcodes.
1106 * By having a single instruction type for handling a lot of different
1107 * cases, optimization passes can look for intrinsics and, for the most
1108 * part, completely ignore them. Each intrinsic type also has a few
1109 * possible flags that govern whether or not they can be reordered or
1110 * eliminated. That way passes like dead code elimination can still work
1111 * on intrisics without understanding the meaning of each.
1113 * Each intrinsic has some number of constant indices, some number of
1114 * variables, and some number of sources. What these sources, variables,
1115 * and indices mean depends on the intrinsic and is documented with the
1116 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
1117 * instructions are the only types of instruction that can operate on
1123 nir_intrinsic_op intrinsic
;
1127 /** number of components if this is a vectorized intrinsic
1129 * Similarly to ALU operations, some intrinsics are vectorized.
1130 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
1131 * For vectorized intrinsics, the num_components field specifies the
1132 * number of destination components and the number of source components
1133 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
1135 uint8_t num_components
;
1137 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
1140 } nir_intrinsic_instr
;
1142 static inline nir_variable
*
1143 nir_intrinsic_get_var(nir_intrinsic_instr
*intrin
, unsigned i
)
1145 return nir_deref_instr_get_variable(nir_src_as_deref(intrin
->src
[i
]));
1149 * \name NIR intrinsics semantic flags
1151 * information about what the compiler can do with the intrinsics.
1153 * \sa nir_intrinsic_info::flags
1157 * whether the intrinsic can be safely eliminated if none of its output
1158 * value is not being used.
1160 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
1163 * Whether the intrinsic can be reordered with respect to any other
1164 * intrinsic, i.e. whether the only reordering dependencies of the
1165 * intrinsic are due to the register reads/writes.
1167 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
1168 } nir_intrinsic_semantic_flag
;
1171 * \name NIR intrinsics const-index flag
1173 * Indicates the usage of a const_index slot.
1175 * \sa nir_intrinsic_info::index_map
1179 * Generally instructions that take a offset src argument, can encode
1180 * a constant 'base' value which is added to the offset.
1182 NIR_INTRINSIC_BASE
= 1,
1185 * For store instructions, a writemask for the store.
1187 NIR_INTRINSIC_WRMASK
= 2,
1190 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1192 NIR_INTRINSIC_STREAM_ID
= 3,
1195 * The clip-plane id for load_user_clip_plane intrinsic.
1197 NIR_INTRINSIC_UCP_ID
= 4,
1200 * The amount of data, starting from BASE, that this instruction may
1201 * access. This is used to provide bounds if the offset is not constant.
1203 NIR_INTRINSIC_RANGE
= 5,
1206 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1208 NIR_INTRINSIC_DESC_SET
= 6,
1211 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1213 NIR_INTRINSIC_BINDING
= 7,
1218 NIR_INTRINSIC_COMPONENT
= 8,
1221 * Interpolation mode (only meaningful for FS inputs).
1223 NIR_INTRINSIC_INTERP_MODE
= 9,
1226 * A binary nir_op to use when performing a reduction or scan operation
1228 NIR_INTRINSIC_REDUCTION_OP
= 10,
1231 * Cluster size for reduction operations
1233 NIR_INTRINSIC_CLUSTER_SIZE
= 11,
1236 * Parameter index for a load_param intrinsic
1238 NIR_INTRINSIC_PARAM_IDX
= 12,
1241 * Image dimensionality for image intrinsics
1243 * One of GLSL_SAMPLER_DIM_*
1245 NIR_INTRINSIC_IMAGE_DIM
= 13,
1248 * Non-zero if we are accessing an array image
1250 NIR_INTRINSIC_IMAGE_ARRAY
= 14,
1253 * Image format for image intrinsics
1255 NIR_INTRINSIC_FORMAT
= 15,
1258 * Access qualifiers for image intrinsics
1260 NIR_INTRINSIC_ACCESS
= 16,
1263 * Alignment for offsets and addresses
1265 * These two parameters, specify an alignment in terms of a multiplier and
1266 * an offset. The offset or address parameter X of the intrinsic is
1267 * guaranteed to satisfy the following:
1269 * (X - align_offset) % align_mul == 0
1271 NIR_INTRINSIC_ALIGN_MUL
= 17,
1272 NIR_INTRINSIC_ALIGN_OFFSET
= 18,
1274 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1276 } nir_intrinsic_index_flag
;
1278 #define NIR_INTRINSIC_MAX_INPUTS 5
1283 unsigned num_srcs
; /** < number of register/SSA inputs */
1285 /** number of components of each input register
1287 * If this value is 0, the number of components is given by the
1288 * num_components field of nir_intrinsic_instr.
1290 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1294 /** number of components of the output register
1296 * If this value is 0, the number of components is given by the
1297 * num_components field of nir_intrinsic_instr.
1299 unsigned dest_components
;
1301 /** the number of constant indices used by the intrinsic */
1302 unsigned num_indices
;
1304 /** indicates the usage of intr->const_index[n] */
1305 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1307 /** semantic flags for calls to this intrinsic */
1308 nir_intrinsic_semantic_flag flags
;
1309 } nir_intrinsic_info
;
1311 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1313 static inline unsigned
1314 nir_intrinsic_src_components(nir_intrinsic_instr
*intr
, unsigned srcn
)
1316 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1317 assert(srcn
< info
->num_srcs
);
1318 if (info
->src_components
[srcn
])
1319 return info
->src_components
[srcn
];
1321 return intr
->num_components
;
1324 static inline unsigned
1325 nir_intrinsic_dest_components(nir_intrinsic_instr
*intr
)
1327 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1328 if (!info
->has_dest
)
1330 else if (info
->dest_components
)
1331 return info
->dest_components
;
1333 return intr
->num_components
;
1336 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1337 static inline type \
1338 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1340 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1341 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1342 return (type)instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1344 static inline void \
1345 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1347 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1348 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1349 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1352 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1353 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1354 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1355 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1356 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1357 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1358 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1359 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1360 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1361 INTRINSIC_IDX_ACCESSORS(reduction_op
, REDUCTION_OP
, unsigned)
1362 INTRINSIC_IDX_ACCESSORS(cluster_size
, CLUSTER_SIZE
, unsigned)
1363 INTRINSIC_IDX_ACCESSORS(param_idx
, PARAM_IDX
, unsigned)
1364 INTRINSIC_IDX_ACCESSORS(image_dim
, IMAGE_DIM
, enum glsl_sampler_dim
)
1365 INTRINSIC_IDX_ACCESSORS(image_array
, IMAGE_ARRAY
, bool)
1366 INTRINSIC_IDX_ACCESSORS(access
, ACCESS
, enum gl_access_qualifier
)
1367 INTRINSIC_IDX_ACCESSORS(format
, FORMAT
, unsigned)
1368 INTRINSIC_IDX_ACCESSORS(align_mul
, ALIGN_MUL
, unsigned)
1369 INTRINSIC_IDX_ACCESSORS(align_offset
, ALIGN_OFFSET
, unsigned)
1372 nir_intrinsic_set_align(nir_intrinsic_instr
*intrin
,
1373 unsigned align_mul
, unsigned align_offset
)
1375 assert(util_is_power_of_two_nonzero(align_mul
));
1376 assert(align_offset
< align_mul
);
1377 nir_intrinsic_set_align_mul(intrin
, align_mul
);
1378 nir_intrinsic_set_align_offset(intrin
, align_offset
);
1381 /** Returns a simple alignment for a load/store intrinsic offset
1383 * Instead of the full mul+offset alignment scheme provided by the ALIGN_MUL
1384 * and ALIGN_OFFSET parameters, this helper takes both into account and
1385 * provides a single simple alignment parameter. The offset X is guaranteed
1386 * to satisfy X % align == 0.
1388 static inline unsigned
1389 nir_intrinsic_align(const nir_intrinsic_instr
*intrin
)
1391 const unsigned align_mul
= nir_intrinsic_align_mul(intrin
);
1392 const unsigned align_offset
= nir_intrinsic_align_offset(intrin
);
1393 assert(align_offset
< align_mul
);
1394 return align_offset
? 1 << (ffs(align_offset
) - 1) : align_mul
;
1398 * \group texture information
1400 * This gives semantic information about textures which is useful to the
1401 * frontend, the backend, and lowering passes, but not the optimizer.
1406 nir_tex_src_projector
,
1407 nir_tex_src_comparator
, /* shadow comparator */
1411 nir_tex_src_min_lod
,
1412 nir_tex_src_ms_index
, /* MSAA sample index */
1413 nir_tex_src_ms_mcs
, /* MSAA compression value */
1416 nir_tex_src_texture_deref
, /* < deref pointing to the texture */
1417 nir_tex_src_sampler_deref
, /* < deref pointing to the sampler */
1418 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1419 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1420 nir_tex_src_plane
, /* < selects plane for planar textures */
1421 nir_num_tex_src_types
1426 nir_tex_src_type src_type
;
1430 nir_texop_tex
, /**< Regular texture look-up */
1431 nir_texop_txb
, /**< Texture look-up with LOD bias */
1432 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1433 nir_texop_txd
, /**< Texture look-up with partial derivatives */
1434 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1435 nir_texop_txf_ms
, /**< Multisample texture fetch */
1436 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1437 nir_texop_txs
, /**< Texture size */
1438 nir_texop_lod
, /**< Texture lod query */
1439 nir_texop_tg4
, /**< Texture gather */
1440 nir_texop_query_levels
, /**< Texture levels query */
1441 nir_texop_texture_samples
, /**< Texture samples query */
1442 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1450 enum glsl_sampler_dim sampler_dim
;
1451 nir_alu_type dest_type
;
1456 unsigned num_srcs
, coord_components
;
1457 bool is_array
, is_shadow
;
1460 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1461 * components or the new-style shadow that outputs 1 component.
1463 bool is_new_style_shadow
;
1465 /* gather component selector */
1466 unsigned component
: 2;
1468 /** The texture index
1470 * If this texture instruction has a nir_tex_src_texture_offset source,
1471 * then the texture index is given by texture_index + texture_offset.
1473 unsigned texture_index
;
1475 /** The size of the texture array or 0 if it's not an array */
1476 unsigned texture_array_size
;
1478 /** The sampler index
1480 * The following operations do not require a sampler and, as such, this
1481 * field should be ignored:
1483 * - nir_texop_txf_ms
1486 * - nir_texop_query_levels
1487 * - nir_texop_texture_samples
1488 * - nir_texop_samples_identical
1490 * If this texture instruction has a nir_tex_src_sampler_offset source,
1491 * then the sampler index is given by sampler_index + sampler_offset.
1493 unsigned sampler_index
;
1496 static inline unsigned
1497 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1499 switch (instr
->op
) {
1500 case nir_texop_txs
: {
1502 switch (instr
->sampler_dim
) {
1503 case GLSL_SAMPLER_DIM_1D
:
1504 case GLSL_SAMPLER_DIM_BUF
:
1507 case GLSL_SAMPLER_DIM_2D
:
1508 case GLSL_SAMPLER_DIM_CUBE
:
1509 case GLSL_SAMPLER_DIM_MS
:
1510 case GLSL_SAMPLER_DIM_RECT
:
1511 case GLSL_SAMPLER_DIM_EXTERNAL
:
1512 case GLSL_SAMPLER_DIM_SUBPASS
:
1515 case GLSL_SAMPLER_DIM_3D
:
1519 unreachable("not reached");
1521 if (instr
->is_array
)
1529 case nir_texop_texture_samples
:
1530 case nir_texop_query_levels
:
1531 case nir_texop_samples_identical
:
1535 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1542 /* Returns true if this texture operation queries something about the texture
1543 * rather than actually sampling it.
1546 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1548 switch (instr
->op
) {
1551 case nir_texop_texture_samples
:
1552 case nir_texop_query_levels
:
1553 case nir_texop_txf_ms_mcs
:
1560 case nir_texop_txf_ms
:
1564 unreachable("Invalid texture opcode");
1569 nir_alu_instr_is_comparison(const nir_alu_instr
*instr
)
1571 switch (instr
->op
) {
1592 static inline nir_alu_type
1593 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1595 switch (instr
->src
[src
].src_type
) {
1596 case nir_tex_src_coord
:
1597 switch (instr
->op
) {
1599 case nir_texop_txf_ms
:
1600 case nir_texop_txf_ms_mcs
:
1601 case nir_texop_samples_identical
:
1602 return nir_type_int
;
1605 return nir_type_float
;
1608 case nir_tex_src_lod
:
1609 switch (instr
->op
) {
1612 return nir_type_int
;
1615 return nir_type_float
;
1618 case nir_tex_src_projector
:
1619 case nir_tex_src_comparator
:
1620 case nir_tex_src_bias
:
1621 case nir_tex_src_ddx
:
1622 case nir_tex_src_ddy
:
1623 return nir_type_float
;
1625 case nir_tex_src_offset
:
1626 case nir_tex_src_ms_index
:
1627 case nir_tex_src_texture_offset
:
1628 case nir_tex_src_sampler_offset
:
1629 return nir_type_int
;
1632 unreachable("Invalid texture source type");
1636 static inline unsigned
1637 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1639 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1640 return instr
->coord_components
;
1642 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1643 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1646 if (instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1647 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1648 if (instr
->is_array
)
1649 return instr
->coord_components
- 1;
1651 return instr
->coord_components
;
1654 /* Usual APIs don't allow cube + offset, but we allow it, with 2 coords for
1655 * the offset, since a cube maps to a single face.
1657 if (instr
->src
[src
].src_type
== nir_tex_src_offset
) {
1658 if (instr
->sampler_dim
== GLSL_SAMPLER_DIM_CUBE
)
1660 else if (instr
->is_array
)
1661 return instr
->coord_components
- 1;
1663 return instr
->coord_components
;
1670 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1672 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1673 if (instr
->src
[i
].src_type
== type
)
1679 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1680 nir_tex_src_type src_type
,
1683 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1688 nir_const_value value
;
1691 } nir_load_const_instr
;
1704 /* creates a new SSA variable in an undefined state */
1709 } nir_ssa_undef_instr
;
1712 struct exec_node node
;
1714 /* The predecessor block corresponding to this source */
1715 struct nir_block
*pred
;
1720 #define nir_foreach_phi_src(phi_src, phi) \
1721 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1722 #define nir_foreach_phi_src_safe(phi_src, phi) \
1723 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1728 struct exec_list srcs
; /** < list of nir_phi_src */
1734 struct exec_node node
;
1737 } nir_parallel_copy_entry
;
1739 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1740 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1745 /* A list of nir_parallel_copy_entrys. The sources of all of the
1746 * entries are copied to the corresponding destinations "in parallel".
1747 * In other words, if we have two entries: a -> b and b -> a, the values
1750 struct exec_list entries
;
1751 } nir_parallel_copy_instr
;
1753 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1754 type
, nir_instr_type_alu
)
1755 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1756 type
, nir_instr_type_call
)
1757 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1758 type
, nir_instr_type_jump
)
1759 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1760 type
, nir_instr_type_tex
)
1761 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1762 type
, nir_instr_type_intrinsic
)
1763 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1764 type
, nir_instr_type_load_const
)
1765 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1766 type
, nir_instr_type_ssa_undef
)
1767 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1768 type
, nir_instr_type_phi
)
1769 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1770 nir_parallel_copy_instr
, instr
,
1771 type
, nir_instr_type_parallel_copy
)
1776 * Control flow consists of a tree of control flow nodes, which include
1777 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1778 * instructions that always run start-to-finish. Each basic block also keeps
1779 * track of its successors (blocks which may run immediately after the current
1780 * block) and predecessors (blocks which could have run immediately before the
1781 * current block). Each function also has a start block and an end block which
1782 * all return statements point to (which is always empty). Together, all the
1783 * blocks with their predecessors and successors make up the control flow
1784 * graph (CFG) of the function. There are helpers that modify the tree of
1785 * control flow nodes while modifying the CFG appropriately; these should be
1786 * used instead of modifying the tree directly.
1793 nir_cf_node_function
1796 typedef struct nir_cf_node
{
1797 struct exec_node node
;
1798 nir_cf_node_type type
;
1799 struct nir_cf_node
*parent
;
1802 typedef struct nir_block
{
1803 nir_cf_node cf_node
;
1805 struct exec_list instr_list
; /** < list of nir_instr */
1807 /** generic block index; generated by nir_index_blocks */
1811 * Each block can only have up to 2 successors, so we put them in a simple
1812 * array - no need for anything more complicated.
1814 struct nir_block
*successors
[2];
1816 /* Set of nir_block predecessors in the CFG */
1817 struct set
*predecessors
;
1820 * this node's immediate dominator in the dominance tree - set to NULL for
1823 struct nir_block
*imm_dom
;
1825 /* This node's children in the dominance tree */
1826 unsigned num_dom_children
;
1827 struct nir_block
**dom_children
;
1829 /* Set of nir_blocks on the dominance frontier of this block */
1830 struct set
*dom_frontier
;
1833 * These two indices have the property that dom_{pre,post}_index for each
1834 * child of this block in the dominance tree will always be between
1835 * dom_pre_index and dom_post_index for this block, which makes testing if
1836 * a given block is dominated by another block an O(1) operation.
1838 unsigned dom_pre_index
, dom_post_index
;
1840 /* live in and out for this block; used for liveness analysis */
1841 BITSET_WORD
*live_in
;
1842 BITSET_WORD
*live_out
;
1845 static inline nir_instr
*
1846 nir_block_first_instr(nir_block
*block
)
1848 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1849 return exec_node_data(nir_instr
, head
, node
);
1852 static inline nir_instr
*
1853 nir_block_last_instr(nir_block
*block
)
1855 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1856 return exec_node_data(nir_instr
, tail
, node
);
1860 nir_block_ends_in_jump(nir_block
*block
)
1862 return !exec_list_is_empty(&block
->instr_list
) &&
1863 nir_block_last_instr(block
)->type
== nir_instr_type_jump
;
1866 #define nir_foreach_instr(instr, block) \
1867 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1868 #define nir_foreach_instr_reverse(instr, block) \
1869 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1870 #define nir_foreach_instr_safe(instr, block) \
1871 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1872 #define nir_foreach_instr_reverse_safe(instr, block) \
1873 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1875 typedef struct nir_if
{
1876 nir_cf_node cf_node
;
1879 struct exec_list then_list
; /** < list of nir_cf_node */
1880 struct exec_list else_list
; /** < list of nir_cf_node */
1886 nir_instr
*conditional_instr
;
1888 nir_block
*break_block
;
1889 nir_block
*continue_from_block
;
1891 bool continue_from_then
;
1893 struct list_head loop_terminator_link
;
1894 } nir_loop_terminator
;
1897 /* Number of instructions in the loop */
1898 unsigned num_instructions
;
1900 /* Maximum number of times the loop is run (if known) */
1901 unsigned max_trip_count
;
1903 /* Do we know the exact number of times the loop will be run */
1904 bool exact_trip_count_known
;
1906 /* Unroll the loop regardless of its size */
1909 /* Does the loop contain complex loop terminators, continues or other
1910 * complex behaviours? If this is true we can't rely on
1911 * loop_terminator_list to be complete or accurate.
1915 nir_loop_terminator
*limiting_terminator
;
1917 /* A list of loop_terminators terminating this loop. */
1918 struct list_head loop_terminator_list
;
1922 nir_cf_node cf_node
;
1924 struct exec_list body
; /** < list of nir_cf_node */
1926 nir_loop_info
*info
;
1930 * Various bits of metadata that can may be created or required by
1931 * optimization and analysis passes
1934 nir_metadata_none
= 0x0,
1935 nir_metadata_block_index
= 0x1,
1936 nir_metadata_dominance
= 0x2,
1937 nir_metadata_live_ssa_defs
= 0x4,
1938 nir_metadata_not_properly_reset
= 0x8,
1939 nir_metadata_loop_analysis
= 0x10,
1943 nir_cf_node cf_node
;
1945 /** pointer to the function of which this is an implementation */
1946 struct nir_function
*function
;
1948 struct exec_list body
; /** < list of nir_cf_node */
1950 nir_block
*end_block
;
1952 /** list for all local variables in the function */
1953 struct exec_list locals
;
1955 /** list of local registers in the function */
1956 struct exec_list registers
;
1958 /** next available local register index */
1961 /** next available SSA value index */
1964 /* total number of basic blocks, only valid when block_index_dirty = false */
1965 unsigned num_blocks
;
1967 nir_metadata valid_metadata
;
1968 } nir_function_impl
;
1970 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1971 nir_start_block(nir_function_impl
*impl
)
1973 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1976 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1977 nir_impl_last_block(nir_function_impl
*impl
)
1979 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1982 static inline nir_cf_node
*
1983 nir_cf_node_next(nir_cf_node
*node
)
1985 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1986 if (exec_node_is_tail_sentinel(next
))
1989 return exec_node_data(nir_cf_node
, next
, node
);
1992 static inline nir_cf_node
*
1993 nir_cf_node_prev(nir_cf_node
*node
)
1995 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1996 if (exec_node_is_head_sentinel(prev
))
1999 return exec_node_data(nir_cf_node
, prev
, node
);
2003 nir_cf_node_is_first(const nir_cf_node
*node
)
2005 return exec_node_is_head_sentinel(node
->node
.prev
);
2009 nir_cf_node_is_last(const nir_cf_node
*node
)
2011 return exec_node_is_tail_sentinel(node
->node
.next
);
2014 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
2015 type
, nir_cf_node_block
)
2016 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
2017 type
, nir_cf_node_if
)
2018 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
2019 type
, nir_cf_node_loop
)
2020 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
2021 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
2023 static inline nir_block
*
2024 nir_if_first_then_block(nir_if
*if_stmt
)
2026 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
2027 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2030 static inline nir_block
*
2031 nir_if_last_then_block(nir_if
*if_stmt
)
2033 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
2034 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2037 static inline nir_block
*
2038 nir_if_first_else_block(nir_if
*if_stmt
)
2040 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
2041 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2044 static inline nir_block
*
2045 nir_if_last_else_block(nir_if
*if_stmt
)
2047 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
2048 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2051 static inline nir_block
*
2052 nir_loop_first_block(nir_loop
*loop
)
2054 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
2055 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2058 static inline nir_block
*
2059 nir_loop_last_block(nir_loop
*loop
)
2061 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
2062 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2066 uint8_t num_components
;
2070 typedef struct nir_function
{
2071 struct exec_node node
;
2074 struct nir_shader
*shader
;
2076 unsigned num_params
;
2077 nir_parameter
*params
;
2079 /** The implementation of this function.
2081 * If the function is only declared and not implemented, this is NULL.
2083 nir_function_impl
*impl
;
2086 typedef struct nir_shader_compiler_options
{
2091 /** Lowers flrp when it does not support doubles */
2098 /** Lowers ibitfield_extract/ubitfield_extract to ibfe/ubfe. */
2099 bool lower_bitfield_extract
;
2100 /** Lowers ibitfield_extract/ubitfield_extract to bfm, compares, shifts. */
2101 bool lower_bitfield_extract_to_shifts
;
2102 /** Lowers bitfield_insert to bfi/bfm */
2103 bool lower_bitfield_insert
;
2104 /** Lowers bitfield_insert to bfm, compares, and shifts. */
2105 bool lower_bitfield_insert_to_shifts
;
2106 /** Lowers bitfield_reverse to shifts. */
2107 bool lower_bitfield_reverse
;
2108 /** Lowers bit_count to shifts. */
2109 bool lower_bit_count
;
2110 /** Lowers bfm to shifts and subtracts. */
2112 /** Lowers ifind_msb to compare and ufind_msb */
2113 bool lower_ifind_msb
;
2114 /** Lowers find_lsb to ufind_msb and logic ops */
2115 bool lower_find_lsb
;
2116 bool lower_uadd_carry
;
2117 bool lower_usub_borrow
;
2118 /** Lowers imul_high/umul_high to 16-bit multiplies and carry operations. */
2119 bool lower_mul_high
;
2120 /** lowers fneg and ineg to fsub and isub. */
2122 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
2125 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
2128 /** enables rules to lower idiv by power-of-two: */
2131 /* Does the native fdot instruction replicate its result for four
2132 * components? If so, then opt_algebraic_late will turn all fdotN
2133 * instructions into fdot_replicatedN instructions.
2135 bool fdot_replicates
;
2137 /** lowers ffloor to fsub+ffract: */
2140 /** lowers ffract to fsub+ffloor: */
2143 /** lowers fceil to fneg+ffloor+fneg: */
2148 bool lower_pack_half_2x16
;
2149 bool lower_pack_unorm_2x16
;
2150 bool lower_pack_snorm_2x16
;
2151 bool lower_pack_unorm_4x8
;
2152 bool lower_pack_snorm_4x8
;
2153 bool lower_unpack_half_2x16
;
2154 bool lower_unpack_unorm_2x16
;
2155 bool lower_unpack_snorm_2x16
;
2156 bool lower_unpack_unorm_4x8
;
2157 bool lower_unpack_snorm_4x8
;
2159 bool lower_extract_byte
;
2160 bool lower_extract_word
;
2162 bool lower_all_io_to_temps
;
2165 * Does the driver support real 32-bit integers? (Otherwise, integers
2166 * are simulated by floats.)
2168 bool native_integers
;
2170 /* Indicates that the driver only has zero-based vertex id */
2171 bool vertex_id_zero_based
;
2174 * If enabled, gl_BaseVertex will be lowered as:
2175 * is_indexed_draw (~0/0) & firstvertex
2177 bool lower_base_vertex
;
2180 * If enabled, gl_HelperInvocation will be lowered as:
2182 * !((1 << sample_id) & sample_mask_in))
2184 * This depends on some possibly hw implementation details, which may
2185 * not be true for all hw. In particular that the FS is only executed
2186 * for covered samples or for helper invocations. So, do not blindly
2187 * enable this option.
2189 * Note: See also issue #22 in ARB_shader_image_load_store
2191 bool lower_helper_invocation
;
2193 bool lower_cs_local_index_from_id
;
2194 bool lower_cs_local_id_from_index
;
2196 bool lower_device_index_to_zero
;
2198 /* Set if nir_lower_wpos_ytransform() should also invert gl_PointCoord. */
2199 bool lower_wpos_pntc
;
2202 * Should nir_lower_io() create load_interpolated_input intrinsics?
2204 * If not, it generates regular load_input intrinsics and interpolation
2205 * information must be inferred from the list of input nir_variables.
2207 bool use_interpolated_input_intrinsics
;
2209 unsigned max_unroll_iterations
;
2210 } nir_shader_compiler_options
;
2212 typedef struct nir_shader
{
2213 /** list of uniforms (nir_variable) */
2214 struct exec_list uniforms
;
2216 /** list of inputs (nir_variable) */
2217 struct exec_list inputs
;
2219 /** list of outputs (nir_variable) */
2220 struct exec_list outputs
;
2222 /** list of shared compute variables (nir_variable) */
2223 struct exec_list shared
;
2225 /** Set of driver-specific options for the shader.
2227 * The memory for the options is expected to be kept in a single static
2228 * copy by the driver.
2230 const struct nir_shader_compiler_options
*options
;
2232 /** Various bits of compile-time information about a given shader */
2233 struct shader_info info
;
2235 /** list of global variables in the shader (nir_variable) */
2236 struct exec_list globals
;
2238 /** list of system value variables in the shader (nir_variable) */
2239 struct exec_list system_values
;
2241 struct exec_list functions
; /** < list of nir_function */
2243 /** list of global register in the shader */
2244 struct exec_list registers
;
2246 /** next available global register index */
2250 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
2253 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
2255 /** Constant data associated with this shader.
2257 * Constant data is loaded through load_constant intrinsics. See also
2258 * nir_opt_large_constants.
2260 void *constant_data
;
2261 unsigned constant_data_size
;
2264 static inline nir_function_impl
*
2265 nir_shader_get_entrypoint(nir_shader
*shader
)
2267 assert(exec_list_length(&shader
->functions
) == 1);
2268 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
2269 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
2270 assert(func
->num_params
== 0);
2275 #define nir_foreach_function(func, shader) \
2276 foreach_list_typed(nir_function, func, node, &(shader)->functions)
2278 nir_shader
*nir_shader_create(void *mem_ctx
,
2279 gl_shader_stage stage
,
2280 const nir_shader_compiler_options
*options
,
2283 /** creates a register, including assigning it an index and adding it to the list */
2284 nir_register
*nir_global_reg_create(nir_shader
*shader
);
2286 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
2288 void nir_reg_remove(nir_register
*reg
);
2290 /** Adds a variable to the appropriate list in nir_shader */
2291 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
2294 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
2296 assert(var
->data
.mode
== nir_var_local
);
2297 exec_list_push_tail(&impl
->locals
, &var
->node
);
2300 /** creates a variable, sets a few defaults, and adds it to the list */
2301 nir_variable
*nir_variable_create(nir_shader
*shader
,
2302 nir_variable_mode mode
,
2303 const struct glsl_type
*type
,
2305 /** creates a local variable and adds it to the list */
2306 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
2307 const struct glsl_type
*type
,
2310 /** creates a function and adds it to the shader's list of functions */
2311 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
2313 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
2314 /** creates a function_impl that isn't tied to any particular function */
2315 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
2317 nir_block
*nir_block_create(nir_shader
*shader
);
2318 nir_if
*nir_if_create(nir_shader
*shader
);
2319 nir_loop
*nir_loop_create(nir_shader
*shader
);
2321 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
2323 /** requests that the given pieces of metadata be generated */
2324 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
2325 /** dirties all but the preserved metadata */
2326 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
2328 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
2329 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
2331 nir_deref_instr
*nir_deref_instr_create(nir_shader
*shader
,
2332 nir_deref_type deref_type
);
2334 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
2336 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
2337 unsigned num_components
,
2340 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
2341 nir_intrinsic_op op
);
2343 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
2344 nir_function
*callee
);
2346 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
2348 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
2350 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
2352 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
2353 unsigned num_components
,
2356 nir_const_value
nir_alu_binop_identity(nir_op binop
, unsigned bit_size
);
2359 * NIR Cursors and Instruction Insertion API
2362 * A tiny struct representing a point to insert/extract instructions or
2363 * control flow nodes. Helps reduce the combinatorial explosion of possible
2364 * points to insert/extract.
2366 * \sa nir_control_flow.h
2369 nir_cursor_before_block
,
2370 nir_cursor_after_block
,
2371 nir_cursor_before_instr
,
2372 nir_cursor_after_instr
,
2373 } nir_cursor_option
;
2376 nir_cursor_option option
;
2383 static inline nir_block
*
2384 nir_cursor_current_block(nir_cursor cursor
)
2386 if (cursor
.option
== nir_cursor_before_instr
||
2387 cursor
.option
== nir_cursor_after_instr
) {
2388 return cursor
.instr
->block
;
2390 return cursor
.block
;
2394 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2396 static inline nir_cursor
2397 nir_before_block(nir_block
*block
)
2400 cursor
.option
= nir_cursor_before_block
;
2401 cursor
.block
= block
;
2405 static inline nir_cursor
2406 nir_after_block(nir_block
*block
)
2409 cursor
.option
= nir_cursor_after_block
;
2410 cursor
.block
= block
;
2414 static inline nir_cursor
2415 nir_before_instr(nir_instr
*instr
)
2418 cursor
.option
= nir_cursor_before_instr
;
2419 cursor
.instr
= instr
;
2423 static inline nir_cursor
2424 nir_after_instr(nir_instr
*instr
)
2427 cursor
.option
= nir_cursor_after_instr
;
2428 cursor
.instr
= instr
;
2432 static inline nir_cursor
2433 nir_after_block_before_jump(nir_block
*block
)
2435 nir_instr
*last_instr
= nir_block_last_instr(block
);
2436 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2437 return nir_before_instr(last_instr
);
2439 return nir_after_block(block
);
2443 static inline nir_cursor
2444 nir_before_src(nir_src
*src
, bool is_if_condition
)
2446 if (is_if_condition
) {
2447 nir_block
*prev_block
=
2448 nir_cf_node_as_block(nir_cf_node_prev(&src
->parent_if
->cf_node
));
2449 assert(!nir_block_ends_in_jump(prev_block
));
2450 return nir_after_block(prev_block
);
2451 } else if (src
->parent_instr
->type
== nir_instr_type_phi
) {
2453 nir_phi_instr
*cond_phi
= nir_instr_as_phi(src
->parent_instr
);
2455 nir_foreach_phi_src(phi_src
, cond_phi
) {
2456 if (phi_src
->src
.ssa
== src
->ssa
) {
2463 /* The LIST_ENTRY macro is a generic container-of macro, it just happens
2464 * to have a more specific name.
2466 nir_phi_src
*phi_src
= LIST_ENTRY(nir_phi_src
, src
, src
);
2467 return nir_after_block_before_jump(phi_src
->pred
);
2469 return nir_before_instr(src
->parent_instr
);
2473 static inline nir_cursor
2474 nir_before_cf_node(nir_cf_node
*node
)
2476 if (node
->type
== nir_cf_node_block
)
2477 return nir_before_block(nir_cf_node_as_block(node
));
2479 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2482 static inline nir_cursor
2483 nir_after_cf_node(nir_cf_node
*node
)
2485 if (node
->type
== nir_cf_node_block
)
2486 return nir_after_block(nir_cf_node_as_block(node
));
2488 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2491 static inline nir_cursor
2492 nir_after_phis(nir_block
*block
)
2494 nir_foreach_instr(instr
, block
) {
2495 if (instr
->type
!= nir_instr_type_phi
)
2496 return nir_before_instr(instr
);
2498 return nir_after_block(block
);
2501 static inline nir_cursor
2502 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2504 if (node
->type
== nir_cf_node_block
)
2505 return nir_after_block(nir_cf_node_as_block(node
));
2507 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2509 return nir_after_phis(block
);
2512 static inline nir_cursor
2513 nir_before_cf_list(struct exec_list
*cf_list
)
2515 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2516 exec_list_get_head(cf_list
), node
);
2517 return nir_before_cf_node(first_node
);
2520 static inline nir_cursor
2521 nir_after_cf_list(struct exec_list
*cf_list
)
2523 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2524 exec_list_get_tail(cf_list
), node
);
2525 return nir_after_cf_node(last_node
);
2529 * Insert a NIR instruction at the given cursor.
2531 * Note: This does not update the cursor.
2533 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2536 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2538 nir_instr_insert(nir_before_instr(instr
), before
);
2542 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2544 nir_instr_insert(nir_after_instr(instr
), after
);
2548 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2550 nir_instr_insert(nir_before_block(block
), before
);
2554 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2556 nir_instr_insert(nir_after_block(block
), after
);
2560 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2562 nir_instr_insert(nir_before_cf_node(node
), before
);
2566 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2568 nir_instr_insert(nir_after_cf_node(node
), after
);
2572 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2574 nir_instr_insert(nir_before_cf_list(list
), before
);
2578 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2580 nir_instr_insert(nir_after_cf_list(list
), after
);
2583 void nir_instr_remove_v(nir_instr
*instr
);
2585 static inline nir_cursor
2586 nir_instr_remove(nir_instr
*instr
)
2589 nir_instr
*prev
= nir_instr_prev(instr
);
2591 cursor
= nir_after_instr(prev
);
2593 cursor
= nir_before_block(instr
->block
);
2595 nir_instr_remove_v(instr
);
2601 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2602 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2603 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2604 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2606 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2607 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2609 nir_const_value
*nir_src_as_const_value(nir_src src
);
2611 static inline struct nir_instr
*
2612 nir_src_instr(const struct nir_src
*src
)
2614 return src
->is_ssa
? src
->ssa
->parent_instr
: NULL
;
2617 #define NIR_SRC_AS_(name, c_type, type_enum, cast_macro) \
2618 static inline c_type * \
2619 nir_src_as_ ## name (struct nir_src *src) \
2621 return src->is_ssa && src->ssa->parent_instr->type == type_enum \
2622 ? cast_macro(src->ssa->parent_instr) : NULL; \
2624 static inline const c_type * \
2625 nir_src_as_ ## name ## _const(const struct nir_src *src) \
2627 return src->is_ssa && src->ssa->parent_instr->type == type_enum \
2628 ? cast_macro(src->ssa->parent_instr) : NULL; \
2631 NIR_SRC_AS_(alu_instr
, nir_alu_instr
, nir_instr_type_alu
, nir_instr_as_alu
)
2633 bool nir_src_is_dynamically_uniform(nir_src src
);
2634 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2635 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2636 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2637 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2638 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2641 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2642 unsigned num_components
, unsigned bit_size
,
2644 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2645 unsigned num_components
, unsigned bit_size
,
2648 nir_ssa_dest_init_for_type(nir_instr
*instr
, nir_dest
*dest
,
2649 const struct glsl_type
*type
,
2652 assert(glsl_type_is_vector_or_scalar(type
));
2653 nir_ssa_dest_init(instr
, dest
, glsl_get_components(type
),
2654 glsl_get_bit_size(type
), name
);
2656 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2657 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2658 nir_instr
*after_me
);
2660 nir_component_mask_t
nir_ssa_def_components_read(const nir_ssa_def
*def
);
2663 * finds the next basic block in source-code order, returns NULL if there is
2667 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2669 /* Performs the opposite of nir_block_cf_tree_next() */
2671 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2673 /* Gets the first block in a CF node in source-code order */
2675 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2677 /* Gets the last block in a CF node in source-code order */
2679 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2681 /* Gets the next block after a CF node in source-code order */
2683 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2685 /* Macros for loops that visit blocks in source-code order */
2687 #define nir_foreach_block(block, impl) \
2688 for (nir_block *block = nir_start_block(impl); block != NULL; \
2689 block = nir_block_cf_tree_next(block))
2691 #define nir_foreach_block_safe(block, impl) \
2692 for (nir_block *block = nir_start_block(impl), \
2693 *next = nir_block_cf_tree_next(block); \
2695 block = next, next = nir_block_cf_tree_next(block))
2697 #define nir_foreach_block_reverse(block, impl) \
2698 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2699 block = nir_block_cf_tree_prev(block))
2701 #define nir_foreach_block_reverse_safe(block, impl) \
2702 for (nir_block *block = nir_impl_last_block(impl), \
2703 *prev = nir_block_cf_tree_prev(block); \
2705 block = prev, prev = nir_block_cf_tree_prev(block))
2707 #define nir_foreach_block_in_cf_node(block, node) \
2708 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2709 block != nir_cf_node_cf_tree_next(node); \
2710 block = nir_block_cf_tree_next(block))
2712 /* If the following CF node is an if, this function returns that if.
2713 * Otherwise, it returns NULL.
2715 nir_if
*nir_block_get_following_if(nir_block
*block
);
2717 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2719 void nir_index_local_regs(nir_function_impl
*impl
);
2720 void nir_index_global_regs(nir_shader
*shader
);
2721 void nir_index_ssa_defs(nir_function_impl
*impl
);
2722 unsigned nir_index_instrs(nir_function_impl
*impl
);
2724 void nir_index_blocks(nir_function_impl
*impl
);
2726 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2727 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2728 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2729 void nir_print_deref(const nir_deref_instr
*deref
, FILE *fp
);
2731 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2732 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2733 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2734 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2736 nir_shader
*nir_shader_serialize_deserialize(void *mem_ctx
, nir_shader
*s
);
2739 void nir_validate_shader(nir_shader
*shader
, const char *when
);
2740 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2741 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2744 should_clone_nir(void)
2746 static int should_clone
= -1;
2747 if (should_clone
< 0)
2748 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2750 return should_clone
;
2754 should_serialize_deserialize_nir(void)
2756 static int test_serialize
= -1;
2757 if (test_serialize
< 0)
2758 test_serialize
= env_var_as_boolean("NIR_TEST_SERIALIZE", false);
2760 return test_serialize
;
2764 should_print_nir(void)
2766 static int should_print
= -1;
2767 if (should_print
< 0)
2768 should_print
= env_var_as_boolean("NIR_PRINT", false);
2770 return should_print
;
2773 static inline void nir_validate_shader(nir_shader
*shader
, const char *when
) { (void) shader
; (void)when
; }
2774 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2775 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2776 static inline bool should_clone_nir(void) { return false; }
2777 static inline bool should_serialize_deserialize_nir(void) { return false; }
2778 static inline bool should_print_nir(void) { return false; }
2781 #define _PASS(pass, nir, do_pass) do { \
2783 nir_validate_shader(nir, "after " #pass); \
2784 if (should_clone_nir()) { \
2785 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2789 if (should_serialize_deserialize_nir()) { \
2790 void *mem_ctx = ralloc_parent(nir); \
2791 nir = nir_shader_serialize_deserialize(mem_ctx, nir); \
2795 #define NIR_PASS(progress, nir, pass, ...) _PASS(pass, nir, \
2796 nir_metadata_set_validation_flag(nir); \
2797 if (should_print_nir()) \
2798 printf("%s\n", #pass); \
2799 if (pass(nir, ##__VA_ARGS__)) { \
2801 if (should_print_nir()) \
2802 nir_print_shader(nir, stdout); \
2803 nir_metadata_check_validation_flag(nir); \
2807 #define NIR_PASS_V(nir, pass, ...) _PASS(pass, nir, \
2808 if (should_print_nir()) \
2809 printf("%s\n", #pass); \
2810 pass(nir, ##__VA_ARGS__); \
2811 if (should_print_nir()) \
2812 nir_print_shader(nir, stdout); \
2815 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2816 void nir_calc_dominance(nir_shader
*shader
);
2818 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2819 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2821 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2822 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2824 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2825 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2827 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2828 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2830 int nir_gs_count_vertices(const nir_shader
*shader
);
2832 bool nir_shrink_vec_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2833 bool nir_split_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2834 bool nir_split_var_copies(nir_shader
*shader
);
2835 bool nir_split_per_member_structs(nir_shader
*shader
);
2836 bool nir_split_struct_vars(nir_shader
*shader
, nir_variable_mode modes
);
2838 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2839 bool nir_lower_returns(nir_shader
*shader
);
2841 bool nir_inline_functions(nir_shader
*shader
);
2843 bool nir_propagate_invariant(nir_shader
*shader
);
2845 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2846 void nir_lower_deref_copy_instr(struct nir_builder
*b
,
2847 nir_intrinsic_instr
*copy
);
2848 bool nir_lower_var_copies(nir_shader
*shader
);
2850 void nir_fixup_deref_modes(nir_shader
*shader
);
2852 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2854 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2856 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2858 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2859 nir_function_impl
*entrypoint
,
2860 bool outputs
, bool inputs
);
2862 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2864 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2865 int (*type_size
)(const struct glsl_type
*));
2867 /* Some helpers to do very simple linking */
2868 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2869 bool nir_remove_unused_io_vars(nir_shader
*shader
, struct exec_list
*var_list
,
2870 uint64_t *used_by_other_stage
,
2871 uint64_t *used_by_other_stage_patches
);
2872 void nir_compact_varyings(nir_shader
*producer
, nir_shader
*consumer
,
2873 bool default_to_smooth_interp
);
2874 void nir_link_xfb_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2875 bool nir_link_opt_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2878 /* If set, this forces all non-flat fragment shader inputs to be
2879 * interpolated as if with the "sample" qualifier. This requires
2880 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2882 nir_lower_io_force_sample_interpolation
= (1 << 1),
2883 } nir_lower_io_options
;
2884 bool nir_lower_io(nir_shader
*shader
,
2885 nir_variable_mode modes
,
2886 int (*type_size
)(const struct glsl_type
*),
2887 nir_lower_io_options
);
2888 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2889 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2891 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2893 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2894 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2895 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2897 bool nir_remove_dead_derefs(nir_shader
*shader
);
2898 bool nir_remove_dead_derefs_impl(nir_function_impl
*impl
);
2899 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2900 bool nir_lower_constant_initializers(nir_shader
*shader
,
2901 nir_variable_mode modes
);
2903 bool nir_move_load_const(nir_shader
*shader
);
2904 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2905 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2906 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
2908 bool nir_lower_alu(nir_shader
*shader
);
2909 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2910 bool nir_lower_bool_to_int32(nir_shader
*shader
);
2911 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2912 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
2913 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2914 void nir_lower_io_arrays_to_elements(nir_shader
*producer
, nir_shader
*consumer
);
2915 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader
*shader
,
2917 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2918 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
2920 typedef struct nir_lower_subgroups_options
{
2921 uint8_t subgroup_size
;
2922 uint8_t ballot_bit_size
;
2923 bool lower_to_scalar
:1;
2924 bool lower_vote_trivial
:1;
2925 bool lower_vote_eq_to_ballot
:1;
2926 bool lower_subgroup_masks
:1;
2927 bool lower_shuffle
:1;
2928 bool lower_shuffle_to_32bit
:1;
2930 } nir_lower_subgroups_options
;
2932 bool nir_lower_subgroups(nir_shader
*shader
,
2933 const nir_lower_subgroups_options
*options
);
2935 bool nir_lower_system_values(nir_shader
*shader
);
2937 enum PACKED nir_lower_tex_packing
{
2938 nir_lower_tex_packing_none
= 0,
2939 /* The sampler returns up to 2 32-bit words of half floats or 16-bit signed
2940 * or unsigned ints based on the sampler type
2942 nir_lower_tex_packing_16
,
2943 /* The sampler returns 1 32-bit word of 4x8 unorm */
2944 nir_lower_tex_packing_8
,
2947 typedef struct nir_lower_tex_options
{
2949 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2950 * sampler types a texture projector is lowered.
2955 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2957 bool lower_txf_offset
;
2960 * If true, lower away nir_tex_src_offset for all rect textures.
2962 bool lower_rect_offset
;
2965 * If true, lower rect textures to 2D, using txs to fetch the
2966 * texture dimensions and dividing the texture coords by the
2967 * texture dims to normalize.
2972 * If true, convert yuv to rgb.
2974 unsigned lower_y_uv_external
;
2975 unsigned lower_y_u_v_external
;
2976 unsigned lower_yx_xuxv_external
;
2977 unsigned lower_xy_uxvx_external
;
2978 unsigned lower_ayuv_external
;
2981 * To emulate certain texture wrap modes, this can be used
2982 * to saturate the specified tex coord to [0.0, 1.0]. The
2983 * bits are according to sampler #, ie. if, for example:
2985 * (conf->saturate_s & (1 << n))
2987 * is true, then the s coord for sampler n is saturated.
2989 * Note that clamping must happen *after* projector lowering
2990 * so any projected texture sample instruction with a clamped
2991 * coordinate gets automatically lowered, regardless of the
2992 * 'lower_txp' setting.
2994 unsigned saturate_s
;
2995 unsigned saturate_t
;
2996 unsigned saturate_r
;
2998 /* Bitmask of textures that need swizzling.
3000 * If (swizzle_result & (1 << texture_index)), then the swizzle in
3001 * swizzles[texture_index] is applied to the result of the texturing
3004 unsigned swizzle_result
;
3006 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
3007 * while 4 and 5 represent 0 and 1 respectively.
3009 uint8_t swizzles
[32][4];
3012 * Bitmap of textures that need srgb to linear conversion. If
3013 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
3014 * of the texture are lowered to linear.
3016 unsigned lower_srgb
;
3019 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
3021 bool lower_txd_cube_map
;
3024 * If true, lower nir_texop_txd on 3D surfaces with nir_texop_txl.
3029 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
3030 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
3031 * with lower_txd_cube_map.
3033 bool lower_txd_shadow
;
3036 * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
3037 * Implies lower_txd_cube_map and lower_txd_shadow.
3042 * If true, lower nir_texop_txb that try to use shadow compare and min_lod
3043 * at the same time to a nir_texop_lod, some math, and nir_texop_tex.
3045 bool lower_txb_shadow_clamp
;
3048 * If true, lower nir_texop_txd on shadow samplers when it uses min_lod
3049 * with nir_texop_txl. This includes cube maps.
3051 bool lower_txd_shadow_clamp
;
3054 * If true, lower nir_texop_txd on when it uses both offset and min_lod
3055 * with nir_texop_txl. This includes cube maps.
3057 bool lower_txd_offset_clamp
;
3059 enum nir_lower_tex_packing lower_tex_packing
[32];
3060 } nir_lower_tex_options
;
3062 bool nir_lower_tex(nir_shader
*shader
,
3063 const nir_lower_tex_options
*options
);
3065 bool nir_lower_idiv(nir_shader
*shader
);
3067 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
, bool use_vars
);
3068 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
3069 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
3071 void nir_lower_two_sided_color(nir_shader
*shader
);
3073 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
3075 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
3076 bool nir_lower_patch_vertices(nir_shader
*nir
, unsigned static_count
,
3077 const gl_state_index16
*uniform_state_tokens
);
3079 typedef struct nir_lower_wpos_ytransform_options
{
3080 gl_state_index16 state_tokens
[STATE_LENGTH
];
3081 bool fs_coord_origin_upper_left
:1;
3082 bool fs_coord_origin_lower_left
:1;
3083 bool fs_coord_pixel_center_integer
:1;
3084 bool fs_coord_pixel_center_half_integer
:1;
3085 } nir_lower_wpos_ytransform_options
;
3087 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
3088 const nir_lower_wpos_ytransform_options
*options
);
3089 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
3091 typedef struct nir_lower_drawpixels_options
{
3092 gl_state_index16 texcoord_state_tokens
[STATE_LENGTH
];
3093 gl_state_index16 scale_state_tokens
[STATE_LENGTH
];
3094 gl_state_index16 bias_state_tokens
[STATE_LENGTH
];
3095 unsigned drawpix_sampler
;
3096 unsigned pixelmap_sampler
;
3098 bool scale_and_bias
:1;
3099 } nir_lower_drawpixels_options
;
3101 void nir_lower_drawpixels(nir_shader
*shader
,
3102 const nir_lower_drawpixels_options
*options
);
3104 typedef struct nir_lower_bitmap_options
{
3107 } nir_lower_bitmap_options
;
3109 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
3111 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
3114 nir_lower_int_source_mods
= 1 << 0,
3115 nir_lower_float_source_mods
= 1 << 1,
3116 nir_lower_all_source_mods
= (1 << 2) - 1
3117 } nir_lower_to_source_mods_flags
;
3120 bool nir_lower_to_source_mods(nir_shader
*shader
, nir_lower_to_source_mods_flags options
);
3122 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
3124 typedef unsigned (*nir_lower_bit_size_callback
)(const nir_alu_instr
*, void *);
3126 bool nir_lower_bit_size(nir_shader
*shader
,
3127 nir_lower_bit_size_callback callback
,
3128 void *callback_data
);
3131 nir_lower_imul64
= (1 << 0),
3132 nir_lower_isign64
= (1 << 1),
3133 /** Lower all int64 modulus and division opcodes */
3134 nir_lower_divmod64
= (1 << 2),
3135 /** Lower all 64-bit umul_high and imul_high opcodes */
3136 nir_lower_imul_high64
= (1 << 3),
3137 } nir_lower_int64_options
;
3139 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
3142 nir_lower_drcp
= (1 << 0),
3143 nir_lower_dsqrt
= (1 << 1),
3144 nir_lower_drsq
= (1 << 2),
3145 nir_lower_dtrunc
= (1 << 3),
3146 nir_lower_dfloor
= (1 << 4),
3147 nir_lower_dceil
= (1 << 5),
3148 nir_lower_dfract
= (1 << 6),
3149 nir_lower_dround_even
= (1 << 7),
3150 nir_lower_dmod
= (1 << 8)
3151 } nir_lower_doubles_options
;
3153 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
3154 bool nir_lower_pack(nir_shader
*shader
);
3156 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
3158 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
3160 void nir_loop_analyze_impl(nir_function_impl
*impl
,
3161 nir_variable_mode indirect_mask
);
3163 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
3165 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
3166 bool nir_repair_ssa(nir_shader
*shader
);
3168 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
3170 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
3171 * registers. If false, convert all values (even those not involved in a phi
3172 * node) to registers.
3174 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
3176 bool nir_lower_phis_to_regs_block(nir_block
*block
);
3177 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
3178 bool nir_rematerialize_derefs_in_use_blocks_impl(nir_function_impl
*impl
);
3180 bool nir_opt_algebraic(nir_shader
*shader
);
3181 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
3182 bool nir_opt_algebraic_late(nir_shader
*shader
);
3183 bool nir_opt_constant_folding(nir_shader
*shader
);
3185 bool nir_opt_global_to_local(nir_shader
*shader
);
3187 bool nir_copy_prop(nir_shader
*shader
);
3189 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
3191 bool nir_opt_cse(nir_shader
*shader
);
3193 bool nir_opt_dce(nir_shader
*shader
);
3195 bool nir_opt_dead_cf(nir_shader
*shader
);
3197 bool nir_opt_dead_write_vars(nir_shader
*shader
);
3199 bool nir_opt_deref(nir_shader
*shader
);
3201 bool nir_opt_find_array_copies(nir_shader
*shader
);
3203 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
3205 bool nir_opt_idiv_const(nir_shader
*shader
, unsigned min_bit_size
);
3207 bool nir_opt_if(nir_shader
*shader
);
3209 bool nir_opt_intrinsics(nir_shader
*shader
);
3211 bool nir_opt_large_constants(nir_shader
*shader
,
3212 glsl_type_size_align_func size_align
,
3213 unsigned threshold
);
3215 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
3217 bool nir_opt_move_comparisons(nir_shader
*shader
);
3219 bool nir_opt_move_load_ubo(nir_shader
*shader
);
3221 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
,
3222 bool indirect_load_ok
, bool expensive_alu_ok
);
3224 bool nir_opt_remove_phis(nir_shader
*shader
);
3226 bool nir_opt_shrink_load(nir_shader
*shader
);
3228 bool nir_opt_trivial_continues(nir_shader
*shader
);
3230 bool nir_opt_undef(nir_shader
*shader
);
3232 bool nir_opt_conditional_discard(nir_shader
*shader
);
3234 void nir_sweep(nir_shader
*shader
);
3236 void nir_remap_dual_slot_attributes(nir_shader
*shader
,
3237 uint64_t *dual_slot_inputs
);
3238 uint64_t nir_get_single_slot_attribs_mask(uint64_t attribs
, uint64_t dual_slot
);
3240 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
3241 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);