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/bitset.h"
38 #include "util/macros.h"
39 #include "compiler/nir_types.h"
40 #include "compiler/shader_enums.h"
41 #include "compiler/shader_info.h"
45 #include "util/debug.h"
48 #include "nir_opcodes.h"
50 #if defined(_WIN32) && !defined(snprintf)
51 #define snprintf _snprintf
59 #define NIR_TRUE (~0u)
60 #define NIR_MAX_VEC_COMPONENTS 4
61 typedef uint8_t nir_component_mask_t
;
63 /** Defines a cast function
65 * This macro defines a cast function from in_type to out_type where
66 * out_type is some structure type that contains a field of type out_type.
68 * Note that you have to be a bit careful as the generated cast function
71 #define NIR_DEFINE_CAST(name, in_type, out_type, field, \
72 type_field, type_value) \
73 static inline out_type * \
74 name(const in_type *parent) \
76 assert(parent && parent->type_field == type_value); \
77 return exec_node_data(out_type, parent, field); \
87 * Description of built-in state associated with a uniform
89 * \sa nir_variable::state_slots
92 gl_state_index16 tokens
[STATE_LENGTH
];
97 nir_var_shader_in
= (1 << 0),
98 nir_var_shader_out
= (1 << 1),
99 nir_var_global
= (1 << 2),
100 nir_var_local
= (1 << 3),
101 nir_var_uniform
= (1 << 4),
102 nir_var_shader_storage
= (1 << 5),
103 nir_var_system_value
= (1 << 6),
104 nir_var_shared
= (1 << 8),
112 nir_rounding_mode_undef
= 0,
113 nir_rounding_mode_rtne
= 1, /* round to nearest even */
114 nir_rounding_mode_ru
= 2, /* round up */
115 nir_rounding_mode_rd
= 3, /* round down */
116 nir_rounding_mode_rtz
= 4, /* round towards zero */
120 float f32
[NIR_MAX_VEC_COMPONENTS
];
121 double f64
[NIR_MAX_VEC_COMPONENTS
];
122 int8_t i8
[NIR_MAX_VEC_COMPONENTS
];
123 uint8_t u8
[NIR_MAX_VEC_COMPONENTS
];
124 int16_t i16
[NIR_MAX_VEC_COMPONENTS
];
125 uint16_t u16
[NIR_MAX_VEC_COMPONENTS
];
126 int32_t i32
[NIR_MAX_VEC_COMPONENTS
];
127 uint32_t u32
[NIR_MAX_VEC_COMPONENTS
];
128 int64_t i64
[NIR_MAX_VEC_COMPONENTS
];
129 uint64_t u64
[NIR_MAX_VEC_COMPONENTS
];
132 typedef struct nir_constant
{
134 * Value of the constant.
136 * The field used to back the values supplied by the constant is determined
137 * by the type associated with the \c nir_variable. Constants may be
138 * scalars, vectors, or matrices.
140 nir_const_value values
[NIR_MAX_VEC_COMPONENTS
];
142 /* we could get this from the var->type but makes clone *much* easier to
143 * not have to care about the type.
145 unsigned num_elements
;
147 /* Array elements / Structure Fields */
148 struct nir_constant
**elements
;
152 * \brief Layout qualifiers for gl_FragDepth.
154 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
155 * with a layout qualifier.
158 nir_depth_layout_none
, /**< No depth layout is specified. */
159 nir_depth_layout_any
,
160 nir_depth_layout_greater
,
161 nir_depth_layout_less
,
162 nir_depth_layout_unchanged
166 * Enum keeping track of how a variable was declared.
170 * Normal declaration.
172 nir_var_declared_normally
= 0,
175 * Variable is implicitly generated by the compiler and should not be
176 * visible via the API.
179 } nir_var_declaration_type
;
182 * Either a uniform, global variable, shader input, or shader output. Based on
183 * ir_variable - it should be easy to translate between the two.
186 typedef struct nir_variable
{
187 struct exec_node node
;
190 * Declared type of the variable
192 const struct glsl_type
*type
;
195 * Declared name of the variable
199 struct nir_variable_data
{
201 * Storage class of the variable.
203 * \sa nir_variable_mode
205 nir_variable_mode mode
;
208 * Is the variable read-only?
210 * This is set for variables declared as \c const, shader inputs,
213 unsigned read_only
:1;
217 unsigned invariant
:1;
220 * When separate shader programs are enabled, only input/outputs between
221 * the stages of a multi-stage separate program can be safely removed
222 * from the shader interface. Other input/outputs must remains active.
224 * This is also used to make sure xfb varyings that are unused by the
225 * fragment shader are not removed.
227 unsigned always_active_io
:1;
230 * Interpolation mode for shader inputs / outputs
232 * \sa glsl_interp_mode
234 unsigned interpolation
:2;
237 * \name ARB_fragment_coord_conventions
240 unsigned origin_upper_left
:1;
241 unsigned pixel_center_integer
:1;
245 * If non-zero, then this variable may be packed along with other variables
246 * into a single varying slot, so this offset should be applied when
247 * accessing components. For example, an offset of 1 means that the x
248 * component of this variable is actually stored in component y of the
249 * location specified by \c location.
251 unsigned location_frac
:2;
254 * If true, this variable represents an array of scalars that should
255 * be tightly packed. In other words, consecutive array elements
256 * should be stored one component apart, rather than one slot apart.
261 * Whether this is a fragment shader output implicitly initialized with
262 * the previous contents of the specified render target at the
263 * framebuffer location corresponding to this shader invocation.
265 unsigned fb_fetch_output
:1;
268 * Non-zero if this variable is considered bindless as defined by
269 * ARB_bindless_texture.
274 * Was an explicit binding set in the shader?
276 unsigned explicit_binding
:1;
279 * Was a transfer feedback buffer set in the shader?
281 unsigned explicit_xfb_buffer
:1;
284 * Was a transfer feedback stride set in the shader?
286 unsigned explicit_xfb_stride
:1;
289 * Was an explicit offset set in the shader?
291 unsigned explicit_offset
:1;
294 * \brief Layout qualifier for gl_FragDepth.
296 * This is not equal to \c ir_depth_layout_none if and only if this
297 * variable is \c gl_FragDepth and a layout qualifier is specified.
299 nir_depth_layout depth_layout
;
302 * Storage location of the base of this variable
304 * The precise meaning of this field depends on the nature of the variable.
306 * - Vertex shader input: one of the values from \c gl_vert_attrib.
307 * - Vertex shader output: one of the values from \c gl_varying_slot.
308 * - Geometry shader input: one of the values from \c gl_varying_slot.
309 * - Geometry shader output: one of the values from \c gl_varying_slot.
310 * - Fragment shader input: one of the values from \c gl_varying_slot.
311 * - Fragment shader output: one of the values from \c gl_frag_result.
312 * - Uniforms: Per-stage uniform slot number for default uniform block.
313 * - Uniforms: Index within the uniform block definition for UBO members.
314 * - Non-UBO Uniforms: uniform slot number.
315 * - Other: This field is not currently used.
317 * If the variable is a uniform, shader input, or shader output, and the
318 * slot has not been assigned, the value will be -1.
323 * The actual location of the variable in the IR. Only valid for inputs
326 unsigned int driver_location
;
329 * Vertex stream output identifier.
331 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
332 * stream of the i-th component.
337 * output index for dual source blending.
342 * Descriptor set binding for sampler or UBO.
347 * Initial binding point for a sampler or UBO.
349 * For array types, this represents the binding point for the first element.
354 * Location an atomic counter or transform feedback is stored at.
359 * Transform feedback buffer.
364 * Transform feedback stride.
369 * How the variable was declared. See nir_var_declaration_type.
371 * This is used to detect variables generated by the compiler, so should
372 * not be visible via the API.
374 unsigned how_declared
:2;
377 * ARB_shader_image_load_store qualifiers.
380 bool read_only
; /**< "readonly" qualifier. */
381 bool write_only
; /**< "writeonly" qualifier. */
386 /** Image internal format if specified explicitly, otherwise GL_NONE. */
392 * Built-in state that backs this uniform
394 * Once set at variable creation, \c state_slots must remain invariant.
395 * This is because, ideally, this array would be shared by all clones of
396 * this variable in the IR tree. In other words, we'd really like for it
397 * to be a fly-weight.
399 * If the variable is not a uniform, \c num_state_slots will be zero and
400 * \c state_slots will be \c NULL.
403 unsigned num_state_slots
; /**< Number of state slots used */
404 nir_state_slot
*state_slots
; /**< State descriptors. */
408 * Constant expression assigned in the initializer of the variable
410 * This field should only be used temporarily by creators of NIR shaders
411 * and then lower_constant_initializers can be used to get rid of them.
412 * Most of the rest of NIR ignores this field or asserts that it's NULL.
414 nir_constant
*constant_initializer
;
417 * For variables that are in an interface block or are an instance of an
418 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
420 * \sa ir_variable::location
422 const struct glsl_type
*interface_type
;
425 * Description of per-member data for per-member struct variables
427 * This is used for variables which are actually an amalgamation of
428 * multiple entities such as a struct of built-in values or a struct of
429 * inputs each with their own layout specifier. This is only allowed on
430 * variables with a struct or array of array of struct type.
432 unsigned num_members
;
433 struct nir_variable_data
*members
;
436 #define nir_foreach_variable(var, var_list) \
437 foreach_list_typed(nir_variable, var, node, var_list)
439 #define nir_foreach_variable_safe(var, var_list) \
440 foreach_list_typed_safe(nir_variable, var, node, var_list)
443 nir_variable_is_global(const nir_variable
*var
)
445 return var
->data
.mode
!= nir_var_local
;
448 typedef struct nir_register
{
449 struct exec_node node
;
451 unsigned num_components
; /** < number of vector components */
452 unsigned num_array_elems
; /** < size of array (0 for no array) */
454 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
457 /** generic register index. */
460 /** only for debug purposes, can be NULL */
463 /** whether this register is local (per-function) or global (per-shader) */
467 * If this flag is set to true, then accessing channels >= num_components
468 * is well-defined, and simply spills over to the next array element. This
469 * is useful for backends that can do per-component accessing, in
470 * particular scalar backends. By setting this flag and making
471 * num_components equal to 1, structures can be packed tightly into
472 * registers and then registers can be accessed per-component to get to
473 * each structure member, even if it crosses vec4 boundaries.
477 /** set of nir_srcs where this register is used (read from) */
478 struct list_head uses
;
480 /** set of nir_dests where this register is defined (written to) */
481 struct list_head defs
;
483 /** set of nir_ifs where this register is used as a condition */
484 struct list_head if_uses
;
487 #define nir_foreach_register(reg, reg_list) \
488 foreach_list_typed(nir_register, reg, node, reg_list)
489 #define nir_foreach_register_safe(reg, reg_list) \
490 foreach_list_typed_safe(nir_register, reg, node, reg_list)
494 nir_instr_type_deref
,
497 nir_instr_type_intrinsic
,
498 nir_instr_type_load_const
,
500 nir_instr_type_ssa_undef
,
502 nir_instr_type_parallel_copy
,
505 typedef struct nir_instr
{
506 struct exec_node node
;
508 struct nir_block
*block
;
510 /** generic instruction index. */
513 /* A temporary for optimization and analysis passes to use for storing
514 * flags. For instance, DCE uses this to store the "dead/live" info.
519 static inline nir_instr
*
520 nir_instr_next(nir_instr
*instr
)
522 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
523 if (exec_node_is_tail_sentinel(next
))
526 return exec_node_data(nir_instr
, next
, node
);
529 static inline nir_instr
*
530 nir_instr_prev(nir_instr
*instr
)
532 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
533 if (exec_node_is_head_sentinel(prev
))
536 return exec_node_data(nir_instr
, prev
, node
);
540 nir_instr_is_first(const nir_instr
*instr
)
542 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
546 nir_instr_is_last(const nir_instr
*instr
)
548 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
551 typedef struct nir_ssa_def
{
552 /** for debugging only, can be NULL */
555 /** generic SSA definition index. */
558 /** Index into the live_in and live_out bitfields */
561 /** Instruction which produces this SSA value. */
562 nir_instr
*parent_instr
;
564 /** set of nir_instrs where this register is used (read from) */
565 struct list_head uses
;
567 /** set of nir_ifs where this register is used as a condition */
568 struct list_head if_uses
;
570 uint8_t num_components
;
572 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
580 struct nir_src
*indirect
; /** < NULL for no indirect offset */
581 unsigned base_offset
;
583 /* TODO use-def chain goes here */
587 nir_instr
*parent_instr
;
588 struct list_head def_link
;
591 struct nir_src
*indirect
; /** < NULL for no indirect offset */
592 unsigned base_offset
;
594 /* TODO def-use chain goes here */
599 typedef struct nir_src
{
601 /** Instruction that consumes this value as a source. */
602 nir_instr
*parent_instr
;
603 struct nir_if
*parent_if
;
606 struct list_head use_link
;
616 static inline nir_src
619 nir_src src
= { { NULL
} };
623 #define NIR_SRC_INIT nir_src_init()
625 #define nir_foreach_use(src, reg_or_ssa_def) \
626 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
628 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
629 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
631 #define nir_foreach_if_use(src, reg_or_ssa_def) \
632 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
634 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
635 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
646 static inline nir_dest
649 nir_dest dest
= { { { NULL
} } };
653 #define NIR_DEST_INIT nir_dest_init()
655 #define nir_foreach_def(dest, reg) \
656 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
658 #define nir_foreach_def_safe(dest, reg) \
659 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
661 static inline nir_src
662 nir_src_for_ssa(nir_ssa_def
*def
)
664 nir_src src
= NIR_SRC_INIT
;
672 static inline nir_src
673 nir_src_for_reg(nir_register
*reg
)
675 nir_src src
= NIR_SRC_INIT
;
679 src
.reg
.indirect
= NULL
;
680 src
.reg
.base_offset
= 0;
685 static inline nir_dest
686 nir_dest_for_reg(nir_register
*reg
)
688 nir_dest dest
= NIR_DEST_INIT
;
695 static inline unsigned
696 nir_src_bit_size(nir_src src
)
698 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
701 static inline unsigned
702 nir_src_num_components(nir_src src
)
704 return src
.is_ssa
? src
.ssa
->num_components
: src
.reg
.reg
->num_components
;
707 static inline unsigned
708 nir_dest_bit_size(nir_dest dest
)
710 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
713 static inline unsigned
714 nir_dest_num_components(nir_dest dest
)
716 return dest
.is_ssa
? dest
.ssa
.num_components
: dest
.reg
.reg
->num_components
;
719 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
720 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
726 * \name input modifiers
730 * For inputs interpreted as floating point, flips the sign bit. For
731 * inputs interpreted as integers, performs the two's complement negation.
736 * Clears the sign bit for floating point values, and computes the integer
737 * absolute value for integers. Note that the negate modifier acts after
738 * the absolute value modifier, therefore if both are set then all inputs
739 * will become negative.
745 * For each input component, says which component of the register it is
746 * chosen from. Note that which elements of the swizzle are used and which
747 * are ignored are based on the write mask for most opcodes - for example,
748 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
749 * a swizzle of {2, x, 1, 0} where x means "don't care."
751 uint8_t swizzle
[NIR_MAX_VEC_COMPONENTS
];
758 * \name saturate output modifier
760 * Only valid for opcodes that output floating-point numbers. Clamps the
761 * output to between 0.0 and 1.0 inclusive.
766 unsigned write_mask
: NIR_MAX_VEC_COMPONENTS
; /* ignored if dest.is_ssa is true */
770 nir_type_invalid
= 0, /* Not a valid type */
775 nir_type_bool32
= 32 | nir_type_bool
,
776 nir_type_int8
= 8 | nir_type_int
,
777 nir_type_int16
= 16 | nir_type_int
,
778 nir_type_int32
= 32 | nir_type_int
,
779 nir_type_int64
= 64 | nir_type_int
,
780 nir_type_uint8
= 8 | nir_type_uint
,
781 nir_type_uint16
= 16 | nir_type_uint
,
782 nir_type_uint32
= 32 | nir_type_uint
,
783 nir_type_uint64
= 64 | nir_type_uint
,
784 nir_type_float16
= 16 | nir_type_float
,
785 nir_type_float32
= 32 | nir_type_float
,
786 nir_type_float64
= 64 | nir_type_float
,
789 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
790 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
792 static inline unsigned
793 nir_alu_type_get_type_size(nir_alu_type type
)
795 return type
& NIR_ALU_TYPE_SIZE_MASK
;
798 static inline unsigned
799 nir_alu_type_get_base_type(nir_alu_type type
)
801 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
804 static inline nir_alu_type
805 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
809 return nir_type_bool32
;
812 return nir_type_uint32
;
815 return nir_type_int32
;
817 case GLSL_TYPE_UINT16
:
818 return nir_type_uint16
;
820 case GLSL_TYPE_INT16
:
821 return nir_type_int16
;
823 case GLSL_TYPE_UINT8
:
824 return nir_type_uint8
;
826 return nir_type_int8
;
827 case GLSL_TYPE_UINT64
:
828 return nir_type_uint64
;
830 case GLSL_TYPE_INT64
:
831 return nir_type_int64
;
833 case GLSL_TYPE_FLOAT
:
834 return nir_type_float32
;
836 case GLSL_TYPE_FLOAT16
:
837 return nir_type_float16
;
839 case GLSL_TYPE_DOUBLE
:
840 return nir_type_float64
;
843 unreachable("unknown type");
847 static inline nir_alu_type
848 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
850 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
853 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
,
854 nir_rounding_mode rnd
);
857 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
858 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
859 } nir_op_algebraic_property
;
867 * The number of components in the output
869 * If non-zero, this is the size of the output and input sizes are
870 * explicitly given; swizzle and writemask are still in effect, but if
871 * the output component is masked out, then the input component may
874 * If zero, the opcode acts in the standard, per-component manner; the
875 * operation is performed on each component (except the ones that are
876 * masked out) with the input being taken from the input swizzle for
879 * The size of some of the inputs may be given (i.e. non-zero) even
880 * though output_size is zero; in that case, the inputs with a zero
881 * size act per-component, while the inputs with non-zero size don't.
883 unsigned output_size
;
886 * The type of vector that the instruction outputs. Note that the
887 * staurate modifier is only allowed on outputs with the float type.
890 nir_alu_type output_type
;
893 * The number of components in each input
895 unsigned input_sizes
[NIR_MAX_VEC_COMPONENTS
];
898 * The type of vector that each input takes. Note that negate and
899 * absolute value are only allowed on inputs with int or float type and
900 * behave differently on the two.
902 nir_alu_type input_types
[NIR_MAX_VEC_COMPONENTS
];
904 nir_op_algebraic_property algebraic_properties
;
907 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
909 typedef struct nir_alu_instr
{
913 /** Indicates that this ALU instruction generates an exact value
915 * This is kind of a mixture of GLSL "precise" and "invariant" and not
916 * really equivalent to either. This indicates that the value generated by
917 * this operation is high-precision and any code transformations that touch
918 * it must ensure that the resulting value is bit-for-bit identical to the
927 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
928 nir_alu_instr
*instr
);
929 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
930 nir_alu_instr
*instr
);
932 /* is this source channel used? */
934 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
937 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
938 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
940 return (instr
->dest
.write_mask
>> channel
) & 1;
944 * For instructions whose destinations are SSA, get the number of channels
947 static inline unsigned
948 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
950 assert(instr
->dest
.dest
.is_ssa
);
952 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
953 return nir_op_infos
[instr
->op
].input_sizes
[src
];
955 return instr
->dest
.dest
.ssa
.num_components
;
958 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
959 unsigned src1
, unsigned src2
);
963 nir_deref_type_array
,
964 nir_deref_type_array_wildcard
,
965 nir_deref_type_struct
,
972 /** The type of this deref instruction */
973 nir_deref_type deref_type
;
975 /** The mode of the underlying variable */
976 nir_variable_mode mode
;
978 /** The dereferenced type of the resulting pointer value */
979 const struct glsl_type
*type
;
982 /** Variable being dereferenced if deref_type is a deref_var */
985 /** Parent deref if deref_type is not deref_var */
989 /** Additional deref parameters */
1000 /** Destination to store the resulting "pointer" */
1004 NIR_DEFINE_CAST(nir_instr_as_deref
, nir_instr
, nir_deref_instr
, instr
,
1005 type
, nir_instr_type_deref
)
1007 static inline nir_deref_instr
*
1008 nir_src_as_deref(nir_src src
)
1013 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_deref
)
1016 return nir_instr_as_deref(src
.ssa
->parent_instr
);
1019 static inline nir_deref_instr
*
1020 nir_deref_instr_parent(const nir_deref_instr
*instr
)
1022 if (instr
->deref_type
== nir_deref_type_var
)
1025 return nir_src_as_deref(instr
->parent
);
1028 static inline nir_variable
*
1029 nir_deref_instr_get_variable(const nir_deref_instr
*instr
)
1031 while (instr
->deref_type
!= nir_deref_type_var
) {
1032 if (instr
->deref_type
== nir_deref_type_cast
)
1035 instr
= nir_deref_instr_parent(instr
);
1041 bool nir_deref_instr_has_indirect(nir_deref_instr
*instr
);
1043 bool nir_deref_instr_remove_if_unused(nir_deref_instr
*instr
);
1048 struct nir_function
*callee
;
1050 unsigned num_params
;
1054 #include "nir_intrinsics.h"
1056 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
1058 /** Represents an intrinsic
1060 * An intrinsic is an instruction type for handling things that are
1061 * more-or-less regular operations but don't just consume and produce SSA
1062 * values like ALU operations do. Intrinsics are not for things that have
1063 * special semantic meaning such as phi nodes and parallel copies.
1064 * Examples of intrinsics include variable load/store operations, system
1065 * value loads, and the like. Even though texturing more-or-less falls
1066 * under this category, texturing is its own instruction type because
1067 * trying to represent texturing with intrinsics would lead to a
1068 * combinatorial explosion of intrinsic opcodes.
1070 * By having a single instruction type for handling a lot of different
1071 * cases, optimization passes can look for intrinsics and, for the most
1072 * part, completely ignore them. Each intrinsic type also has a few
1073 * possible flags that govern whether or not they can be reordered or
1074 * eliminated. That way passes like dead code elimination can still work
1075 * on intrisics without understanding the meaning of each.
1077 * Each intrinsic has some number of constant indices, some number of
1078 * variables, and some number of sources. What these sources, variables,
1079 * and indices mean depends on the intrinsic and is documented with the
1080 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
1081 * instructions are the only types of instruction that can operate on
1087 nir_intrinsic_op intrinsic
;
1091 /** number of components if this is a vectorized intrinsic
1093 * Similarly to ALU operations, some intrinsics are vectorized.
1094 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
1095 * For vectorized intrinsics, the num_components field specifies the
1096 * number of destination components and the number of source components
1097 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
1099 uint8_t num_components
;
1101 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
1104 } nir_intrinsic_instr
;
1106 static inline nir_variable
*
1107 nir_intrinsic_get_var(nir_intrinsic_instr
*intrin
, unsigned i
)
1109 return nir_deref_instr_get_variable(nir_src_as_deref(intrin
->src
[i
]));
1113 * \name NIR intrinsics semantic flags
1115 * information about what the compiler can do with the intrinsics.
1117 * \sa nir_intrinsic_info::flags
1121 * whether the intrinsic can be safely eliminated if none of its output
1122 * value is not being used.
1124 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
1127 * Whether the intrinsic can be reordered with respect to any other
1128 * intrinsic, i.e. whether the only reordering dependencies of the
1129 * intrinsic are due to the register reads/writes.
1131 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
1132 } nir_intrinsic_semantic_flag
;
1135 * \name NIR intrinsics const-index flag
1137 * Indicates the usage of a const_index slot.
1139 * \sa nir_intrinsic_info::index_map
1143 * Generally instructions that take a offset src argument, can encode
1144 * a constant 'base' value which is added to the offset.
1146 NIR_INTRINSIC_BASE
= 1,
1149 * For store instructions, a writemask for the store.
1151 NIR_INTRINSIC_WRMASK
= 2,
1154 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1156 NIR_INTRINSIC_STREAM_ID
= 3,
1159 * The clip-plane id for load_user_clip_plane intrinsic.
1161 NIR_INTRINSIC_UCP_ID
= 4,
1164 * The amount of data, starting from BASE, that this instruction may
1165 * access. This is used to provide bounds if the offset is not constant.
1167 NIR_INTRINSIC_RANGE
= 5,
1170 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1172 NIR_INTRINSIC_DESC_SET
= 6,
1175 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1177 NIR_INTRINSIC_BINDING
= 7,
1182 NIR_INTRINSIC_COMPONENT
= 8,
1185 * Interpolation mode (only meaningful for FS inputs).
1187 NIR_INTRINSIC_INTERP_MODE
= 9,
1190 * A binary nir_op to use when performing a reduction or scan operation
1192 NIR_INTRINSIC_REDUCTION_OP
= 10,
1195 * Cluster size for reduction operations
1197 NIR_INTRINSIC_CLUSTER_SIZE
= 11,
1200 * Parameter index for a load_param intrinsic
1202 NIR_INTRINSIC_PARAM_IDX
= 12,
1204 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1206 } nir_intrinsic_index_flag
;
1208 #define NIR_INTRINSIC_MAX_INPUTS 5
1213 unsigned num_srcs
; /** < number of register/SSA inputs */
1215 /** number of components of each input register
1217 * If this value is 0, the number of components is given by the
1218 * num_components field of nir_intrinsic_instr.
1220 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1224 /** number of components of the output register
1226 * If this value is 0, the number of components is given by the
1227 * num_components field of nir_intrinsic_instr.
1229 unsigned dest_components
;
1231 /** the number of constant indices used by the intrinsic */
1232 unsigned num_indices
;
1234 /** indicates the usage of intr->const_index[n] */
1235 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1237 /** semantic flags for calls to this intrinsic */
1238 nir_intrinsic_semantic_flag flags
;
1239 } nir_intrinsic_info
;
1241 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1243 static inline unsigned
1244 nir_intrinsic_src_components(nir_intrinsic_instr
*intr
, unsigned srcn
)
1246 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1247 assert(srcn
< info
->num_srcs
);
1248 if (info
->src_components
[srcn
])
1249 return info
->src_components
[srcn
];
1251 return intr
->num_components
;
1254 static inline unsigned
1255 nir_intrinsic_dest_components(nir_intrinsic_instr
*intr
)
1257 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1258 if (!info
->has_dest
)
1260 else if (info
->dest_components
)
1261 return info
->dest_components
;
1263 return intr
->num_components
;
1266 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1267 static inline type \
1268 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1270 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1271 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1272 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1274 static inline void \
1275 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1277 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1278 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1279 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1282 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1283 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1284 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1285 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1286 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1287 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1288 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1289 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1290 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1291 INTRINSIC_IDX_ACCESSORS(reduction_op
, REDUCTION_OP
, unsigned)
1292 INTRINSIC_IDX_ACCESSORS(cluster_size
, CLUSTER_SIZE
, unsigned)
1293 INTRINSIC_IDX_ACCESSORS(param_idx
, PARAM_IDX
, unsigned)
1296 * \group texture information
1298 * This gives semantic information about textures which is useful to the
1299 * frontend, the backend, and lowering passes, but not the optimizer.
1304 nir_tex_src_projector
,
1305 nir_tex_src_comparator
, /* shadow comparator */
1309 nir_tex_src_ms_index
, /* MSAA sample index */
1310 nir_tex_src_ms_mcs
, /* MSAA compression value */
1313 nir_tex_src_texture_deref
, /* < deref pointing to the texture */
1314 nir_tex_src_sampler_deref
, /* < deref pointing to the sampler */
1315 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1316 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1317 nir_tex_src_plane
, /* < selects plane for planar textures */
1318 nir_num_tex_src_types
1323 nir_tex_src_type src_type
;
1327 nir_texop_tex
, /**< Regular texture look-up */
1328 nir_texop_txb
, /**< Texture look-up with LOD bias */
1329 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1330 nir_texop_txd
, /**< Texture look-up with partial derivatives */
1331 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1332 nir_texop_txf_ms
, /**< Multisample texture fetch */
1333 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1334 nir_texop_txs
, /**< Texture size */
1335 nir_texop_lod
, /**< Texture lod query */
1336 nir_texop_tg4
, /**< Texture gather */
1337 nir_texop_query_levels
, /**< Texture levels query */
1338 nir_texop_texture_samples
, /**< Texture samples query */
1339 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1347 enum glsl_sampler_dim sampler_dim
;
1348 nir_alu_type dest_type
;
1353 unsigned num_srcs
, coord_components
;
1354 bool is_array
, is_shadow
;
1357 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1358 * components or the new-style shadow that outputs 1 component.
1360 bool is_new_style_shadow
;
1362 /* gather component selector */
1363 unsigned component
: 2;
1365 /** The texture index
1367 * If this texture instruction has a nir_tex_src_texture_offset source,
1368 * then the texture index is given by texture_index + texture_offset.
1370 unsigned texture_index
;
1372 /** The size of the texture array or 0 if it's not an array */
1373 unsigned texture_array_size
;
1375 /** The sampler index
1377 * The following operations do not require a sampler and, as such, this
1378 * field should be ignored:
1380 * - nir_texop_txf_ms
1383 * - nir_texop_query_levels
1384 * - nir_texop_texture_samples
1385 * - nir_texop_samples_identical
1387 * If this texture instruction has a nir_tex_src_sampler_offset source,
1388 * then the sampler index is given by sampler_index + sampler_offset.
1390 unsigned sampler_index
;
1393 static inline unsigned
1394 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1396 switch (instr
->op
) {
1397 case nir_texop_txs
: {
1399 switch (instr
->sampler_dim
) {
1400 case GLSL_SAMPLER_DIM_1D
:
1401 case GLSL_SAMPLER_DIM_BUF
:
1404 case GLSL_SAMPLER_DIM_2D
:
1405 case GLSL_SAMPLER_DIM_CUBE
:
1406 case GLSL_SAMPLER_DIM_MS
:
1407 case GLSL_SAMPLER_DIM_RECT
:
1408 case GLSL_SAMPLER_DIM_EXTERNAL
:
1409 case GLSL_SAMPLER_DIM_SUBPASS
:
1412 case GLSL_SAMPLER_DIM_3D
:
1416 unreachable("not reached");
1418 if (instr
->is_array
)
1426 case nir_texop_texture_samples
:
1427 case nir_texop_query_levels
:
1428 case nir_texop_samples_identical
:
1432 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1439 /* Returns true if this texture operation queries something about the texture
1440 * rather than actually sampling it.
1443 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1445 switch (instr
->op
) {
1448 case nir_texop_texture_samples
:
1449 case nir_texop_query_levels
:
1450 case nir_texop_txf_ms_mcs
:
1457 case nir_texop_txf_ms
:
1461 unreachable("Invalid texture opcode");
1466 nir_alu_instr_is_comparison(const nir_alu_instr
*instr
)
1468 switch (instr
->op
) {
1489 static inline nir_alu_type
1490 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1492 switch (instr
->src
[src
].src_type
) {
1493 case nir_tex_src_coord
:
1494 switch (instr
->op
) {
1496 case nir_texop_txf_ms
:
1497 case nir_texop_txf_ms_mcs
:
1498 case nir_texop_samples_identical
:
1499 return nir_type_int
;
1502 return nir_type_float
;
1505 case nir_tex_src_lod
:
1506 switch (instr
->op
) {
1509 return nir_type_int
;
1512 return nir_type_float
;
1515 case nir_tex_src_projector
:
1516 case nir_tex_src_comparator
:
1517 case nir_tex_src_bias
:
1518 case nir_tex_src_ddx
:
1519 case nir_tex_src_ddy
:
1520 return nir_type_float
;
1522 case nir_tex_src_offset
:
1523 case nir_tex_src_ms_index
:
1524 case nir_tex_src_texture_offset
:
1525 case nir_tex_src_sampler_offset
:
1526 return nir_type_int
;
1529 unreachable("Invalid texture source type");
1533 static inline unsigned
1534 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1536 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1537 return instr
->coord_components
;
1539 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1540 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1543 if (instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1544 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1545 if (instr
->is_array
)
1546 return instr
->coord_components
- 1;
1548 return instr
->coord_components
;
1551 /* Usual APIs don't allow cube + offset, but we allow it, with 2 coords for
1552 * the offset, since a cube maps to a single face.
1554 if (instr
->src
[src
].src_type
== nir_tex_src_offset
) {
1555 if (instr
->sampler_dim
== GLSL_SAMPLER_DIM_CUBE
)
1557 else if (instr
->is_array
)
1558 return instr
->coord_components
- 1;
1560 return instr
->coord_components
;
1567 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1569 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1570 if (instr
->src
[i
].src_type
== type
)
1576 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1577 nir_tex_src_type src_type
,
1580 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1585 nir_const_value value
;
1588 } nir_load_const_instr
;
1601 /* creates a new SSA variable in an undefined state */
1606 } nir_ssa_undef_instr
;
1609 struct exec_node node
;
1611 /* The predecessor block corresponding to this source */
1612 struct nir_block
*pred
;
1617 #define nir_foreach_phi_src(phi_src, phi) \
1618 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1619 #define nir_foreach_phi_src_safe(phi_src, phi) \
1620 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1625 struct exec_list srcs
; /** < list of nir_phi_src */
1631 struct exec_node node
;
1634 } nir_parallel_copy_entry
;
1636 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1637 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1642 /* A list of nir_parallel_copy_entrys. The sources of all of the
1643 * entries are copied to the corresponding destinations "in parallel".
1644 * In other words, if we have two entries: a -> b and b -> a, the values
1647 struct exec_list entries
;
1648 } nir_parallel_copy_instr
;
1650 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1651 type
, nir_instr_type_alu
)
1652 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1653 type
, nir_instr_type_call
)
1654 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1655 type
, nir_instr_type_jump
)
1656 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1657 type
, nir_instr_type_tex
)
1658 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1659 type
, nir_instr_type_intrinsic
)
1660 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1661 type
, nir_instr_type_load_const
)
1662 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1663 type
, nir_instr_type_ssa_undef
)
1664 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1665 type
, nir_instr_type_phi
)
1666 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1667 nir_parallel_copy_instr
, instr
,
1668 type
, nir_instr_type_parallel_copy
)
1673 * Control flow consists of a tree of control flow nodes, which include
1674 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1675 * instructions that always run start-to-finish. Each basic block also keeps
1676 * track of its successors (blocks which may run immediately after the current
1677 * block) and predecessors (blocks which could have run immediately before the
1678 * current block). Each function also has a start block and an end block which
1679 * all return statements point to (which is always empty). Together, all the
1680 * blocks with their predecessors and successors make up the control flow
1681 * graph (CFG) of the function. There are helpers that modify the tree of
1682 * control flow nodes while modifying the CFG appropriately; these should be
1683 * used instead of modifying the tree directly.
1690 nir_cf_node_function
1693 typedef struct nir_cf_node
{
1694 struct exec_node node
;
1695 nir_cf_node_type type
;
1696 struct nir_cf_node
*parent
;
1699 typedef struct nir_block
{
1700 nir_cf_node cf_node
;
1702 struct exec_list instr_list
; /** < list of nir_instr */
1704 /** generic block index; generated by nir_index_blocks */
1708 * Each block can only have up to 2 successors, so we put them in a simple
1709 * array - no need for anything more complicated.
1711 struct nir_block
*successors
[2];
1713 /* Set of nir_block predecessors in the CFG */
1714 struct set
*predecessors
;
1717 * this node's immediate dominator in the dominance tree - set to NULL for
1720 struct nir_block
*imm_dom
;
1722 /* This node's children in the dominance tree */
1723 unsigned num_dom_children
;
1724 struct nir_block
**dom_children
;
1726 /* Set of nir_blocks on the dominance frontier of this block */
1727 struct set
*dom_frontier
;
1730 * These two indices have the property that dom_{pre,post}_index for each
1731 * child of this block in the dominance tree will always be between
1732 * dom_pre_index and dom_post_index for this block, which makes testing if
1733 * a given block is dominated by another block an O(1) operation.
1735 unsigned dom_pre_index
, dom_post_index
;
1737 /* live in and out for this block; used for liveness analysis */
1738 BITSET_WORD
*live_in
;
1739 BITSET_WORD
*live_out
;
1742 static inline nir_instr
*
1743 nir_block_first_instr(nir_block
*block
)
1745 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1746 return exec_node_data(nir_instr
, head
, node
);
1749 static inline nir_instr
*
1750 nir_block_last_instr(nir_block
*block
)
1752 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1753 return exec_node_data(nir_instr
, tail
, node
);
1756 #define nir_foreach_instr(instr, block) \
1757 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1758 #define nir_foreach_instr_reverse(instr, block) \
1759 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1760 #define nir_foreach_instr_safe(instr, block) \
1761 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1762 #define nir_foreach_instr_reverse_safe(instr, block) \
1763 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1765 typedef struct nir_if
{
1766 nir_cf_node cf_node
;
1769 struct exec_list then_list
; /** < list of nir_cf_node */
1770 struct exec_list else_list
; /** < list of nir_cf_node */
1776 nir_instr
*conditional_instr
;
1778 nir_block
*break_block
;
1779 nir_block
*continue_from_block
;
1781 bool continue_from_then
;
1783 struct list_head loop_terminator_link
;
1784 } nir_loop_terminator
;
1787 /* Number of instructions in the loop */
1788 unsigned num_instructions
;
1790 /* How many times the loop is run (if known) */
1791 unsigned trip_count
;
1792 bool is_trip_count_known
;
1794 /* Unroll the loop regardless of its size */
1797 nir_loop_terminator
*limiting_terminator
;
1799 /* A list of loop_terminators terminating this loop. */
1800 struct list_head loop_terminator_list
;
1804 nir_cf_node cf_node
;
1806 struct exec_list body
; /** < list of nir_cf_node */
1808 nir_loop_info
*info
;
1812 * Various bits of metadata that can may be created or required by
1813 * optimization and analysis passes
1816 nir_metadata_none
= 0x0,
1817 nir_metadata_block_index
= 0x1,
1818 nir_metadata_dominance
= 0x2,
1819 nir_metadata_live_ssa_defs
= 0x4,
1820 nir_metadata_not_properly_reset
= 0x8,
1821 nir_metadata_loop_analysis
= 0x10,
1825 nir_cf_node cf_node
;
1827 /** pointer to the function of which this is an implementation */
1828 struct nir_function
*function
;
1830 struct exec_list body
; /** < list of nir_cf_node */
1832 nir_block
*end_block
;
1834 /** list for all local variables in the function */
1835 struct exec_list locals
;
1837 /** list of local registers in the function */
1838 struct exec_list registers
;
1840 /** next available local register index */
1843 /** next available SSA value index */
1846 /* total number of basic blocks, only valid when block_index_dirty = false */
1847 unsigned num_blocks
;
1849 nir_metadata valid_metadata
;
1850 } nir_function_impl
;
1852 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1853 nir_start_block(nir_function_impl
*impl
)
1855 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1858 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1859 nir_impl_last_block(nir_function_impl
*impl
)
1861 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1864 static inline nir_cf_node
*
1865 nir_cf_node_next(nir_cf_node
*node
)
1867 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1868 if (exec_node_is_tail_sentinel(next
))
1871 return exec_node_data(nir_cf_node
, next
, node
);
1874 static inline nir_cf_node
*
1875 nir_cf_node_prev(nir_cf_node
*node
)
1877 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1878 if (exec_node_is_head_sentinel(prev
))
1881 return exec_node_data(nir_cf_node
, prev
, node
);
1885 nir_cf_node_is_first(const nir_cf_node
*node
)
1887 return exec_node_is_head_sentinel(node
->node
.prev
);
1891 nir_cf_node_is_last(const nir_cf_node
*node
)
1893 return exec_node_is_tail_sentinel(node
->node
.next
);
1896 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1897 type
, nir_cf_node_block
)
1898 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1899 type
, nir_cf_node_if
)
1900 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1901 type
, nir_cf_node_loop
)
1902 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1903 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1905 static inline nir_block
*
1906 nir_if_first_then_block(nir_if
*if_stmt
)
1908 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1909 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1912 static inline nir_block
*
1913 nir_if_last_then_block(nir_if
*if_stmt
)
1915 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1916 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1919 static inline nir_block
*
1920 nir_if_first_else_block(nir_if
*if_stmt
)
1922 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1923 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1926 static inline nir_block
*
1927 nir_if_last_else_block(nir_if
*if_stmt
)
1929 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1930 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1933 static inline nir_block
*
1934 nir_loop_first_block(nir_loop
*loop
)
1936 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1937 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1940 static inline nir_block
*
1941 nir_loop_last_block(nir_loop
*loop
)
1943 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1944 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1948 uint8_t num_components
;
1952 typedef struct nir_function
{
1953 struct exec_node node
;
1956 struct nir_shader
*shader
;
1958 unsigned num_params
;
1959 nir_parameter
*params
;
1961 /** The implementation of this function.
1963 * If the function is only declared and not implemented, this is NULL.
1965 nir_function_impl
*impl
;
1968 typedef struct nir_shader_compiler_options
{
1973 /** Lowers flrp when it does not support doubles */
1980 /** Lowers ibitfield_extract/ubitfield_extract to ibfe/ubfe. */
1981 bool lower_bitfield_extract
;
1982 /** Lowers ibitfield_extract/ubitfield_extract to bfm, compares, shifts. */
1983 bool lower_bitfield_extract_to_shifts
;
1984 /** Lowers bitfield_insert to bfi/bfm */
1985 bool lower_bitfield_insert
;
1986 /** Lowers bitfield_insert to bfm, compares, and shifts. */
1987 bool lower_bitfield_insert_to_shifts
;
1988 /** Lowers bitfield_reverse to shifts. */
1989 bool lower_bitfield_reverse
;
1990 /** Lowers bit_count to shifts. */
1991 bool lower_bit_count
;
1992 /** Lowers bfm to shifts and subtracts. */
1994 /** Lowers ifind_msb to compare and ufind_msb */
1995 bool lower_ifind_msb
;
1996 /** Lowers find_lsb to ufind_msb and logic ops */
1997 bool lower_find_lsb
;
1998 bool lower_uadd_carry
;
1999 bool lower_usub_borrow
;
2000 /** Lowers imul_high/umul_high to 16-bit multiplies and carry operations. */
2001 bool lower_mul_high
;
2002 /** lowers fneg and ineg to fsub and isub. */
2004 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
2007 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
2010 /** enables rules to lower idiv by power-of-two: */
2013 /* lower b2f to iand */
2016 /* Does the native fdot instruction replicate its result for four
2017 * components? If so, then opt_algebraic_late will turn all fdotN
2018 * instructions into fdot_replicatedN instructions.
2020 bool fdot_replicates
;
2022 /** lowers ffract to fsub+ffloor: */
2027 bool lower_pack_half_2x16
;
2028 bool lower_pack_unorm_2x16
;
2029 bool lower_pack_snorm_2x16
;
2030 bool lower_pack_unorm_4x8
;
2031 bool lower_pack_snorm_4x8
;
2032 bool lower_unpack_half_2x16
;
2033 bool lower_unpack_unorm_2x16
;
2034 bool lower_unpack_snorm_2x16
;
2035 bool lower_unpack_unorm_4x8
;
2036 bool lower_unpack_snorm_4x8
;
2038 bool lower_extract_byte
;
2039 bool lower_extract_word
;
2041 bool lower_all_io_to_temps
;
2044 * Does the driver support real 32-bit integers? (Otherwise, integers
2045 * are simulated by floats.)
2047 bool native_integers
;
2049 /* Indicates that the driver only has zero-based vertex id */
2050 bool vertex_id_zero_based
;
2053 * If enabled, gl_BaseVertex will be lowered as:
2054 * is_indexed_draw (~0/0) & firstvertex
2056 bool lower_base_vertex
;
2059 * If enabled, gl_HelperInvocation will be lowered as:
2061 * !((1 << sample_id) & sample_mask_in))
2063 * This depends on some possibly hw implementation details, which may
2064 * not be true for all hw. In particular that the FS is only executed
2065 * for covered samples or for helper invocations. So, do not blindly
2066 * enable this option.
2068 * Note: See also issue #22 in ARB_shader_image_load_store
2070 bool lower_helper_invocation
;
2072 bool lower_cs_local_index_from_id
;
2074 bool lower_device_index_to_zero
;
2076 /* Set if nir_lower_wpos_ytransform() should also invert gl_PointCoord. */
2077 bool lower_wpos_pntc
;
2080 * Should nir_lower_io() create load_interpolated_input intrinsics?
2082 * If not, it generates regular load_input intrinsics and interpolation
2083 * information must be inferred from the list of input nir_variables.
2085 bool use_interpolated_input_intrinsics
;
2088 * Do vertex shader double inputs use two locations? The Vulkan spec
2089 * requires two locations to be used, OpenGL allows a single location.
2091 bool vs_inputs_dual_locations
;
2093 unsigned max_unroll_iterations
;
2094 } nir_shader_compiler_options
;
2096 typedef struct nir_shader
{
2097 /** list of uniforms (nir_variable) */
2098 struct exec_list uniforms
;
2100 /** list of inputs (nir_variable) */
2101 struct exec_list inputs
;
2103 /** list of outputs (nir_variable) */
2104 struct exec_list outputs
;
2106 /** list of shared compute variables (nir_variable) */
2107 struct exec_list shared
;
2109 /** Set of driver-specific options for the shader.
2111 * The memory for the options is expected to be kept in a single static
2112 * copy by the driver.
2114 const struct nir_shader_compiler_options
*options
;
2116 /** Various bits of compile-time information about a given shader */
2117 struct shader_info info
;
2119 /** list of global variables in the shader (nir_variable) */
2120 struct exec_list globals
;
2122 /** list of system value variables in the shader (nir_variable) */
2123 struct exec_list system_values
;
2125 struct exec_list functions
; /** < list of nir_function */
2127 /** list of global register in the shader */
2128 struct exec_list registers
;
2130 /** next available global register index */
2134 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
2137 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
2139 /** Constant data associated with this shader.
2141 * Constant data is loaded through load_constant intrinsics. See also
2142 * nir_opt_large_constants.
2144 void *constant_data
;
2145 unsigned constant_data_size
;
2148 static inline nir_function_impl
*
2149 nir_shader_get_entrypoint(nir_shader
*shader
)
2151 assert(exec_list_length(&shader
->functions
) == 1);
2152 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
2153 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
2154 assert(func
->num_params
== 0);
2159 #define nir_foreach_function(func, shader) \
2160 foreach_list_typed(nir_function, func, node, &(shader)->functions)
2162 nir_shader
*nir_shader_create(void *mem_ctx
,
2163 gl_shader_stage stage
,
2164 const nir_shader_compiler_options
*options
,
2167 /** creates a register, including assigning it an index and adding it to the list */
2168 nir_register
*nir_global_reg_create(nir_shader
*shader
);
2170 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
2172 void nir_reg_remove(nir_register
*reg
);
2174 /** Adds a variable to the appropriate list in nir_shader */
2175 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
2178 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
2180 assert(var
->data
.mode
== nir_var_local
);
2181 exec_list_push_tail(&impl
->locals
, &var
->node
);
2184 /** creates a variable, sets a few defaults, and adds it to the list */
2185 nir_variable
*nir_variable_create(nir_shader
*shader
,
2186 nir_variable_mode mode
,
2187 const struct glsl_type
*type
,
2189 /** creates a local variable and adds it to the list */
2190 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
2191 const struct glsl_type
*type
,
2194 /** creates a function and adds it to the shader's list of functions */
2195 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
2197 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
2198 /** creates a function_impl that isn't tied to any particular function */
2199 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
2201 nir_block
*nir_block_create(nir_shader
*shader
);
2202 nir_if
*nir_if_create(nir_shader
*shader
);
2203 nir_loop
*nir_loop_create(nir_shader
*shader
);
2205 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
2207 /** requests that the given pieces of metadata be generated */
2208 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
2209 /** dirties all but the preserved metadata */
2210 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
2212 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
2213 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
2215 nir_deref_instr
*nir_deref_instr_create(nir_shader
*shader
,
2216 nir_deref_type deref_type
);
2218 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
2220 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
2221 unsigned num_components
,
2224 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
2225 nir_intrinsic_op op
);
2227 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
2228 nir_function
*callee
);
2230 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
2232 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
2234 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
2236 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
2237 unsigned num_components
,
2240 nir_const_value
nir_alu_binop_identity(nir_op binop
, unsigned bit_size
);
2243 * NIR Cursors and Instruction Insertion API
2246 * A tiny struct representing a point to insert/extract instructions or
2247 * control flow nodes. Helps reduce the combinatorial explosion of possible
2248 * points to insert/extract.
2250 * \sa nir_control_flow.h
2253 nir_cursor_before_block
,
2254 nir_cursor_after_block
,
2255 nir_cursor_before_instr
,
2256 nir_cursor_after_instr
,
2257 } nir_cursor_option
;
2260 nir_cursor_option option
;
2267 static inline nir_block
*
2268 nir_cursor_current_block(nir_cursor cursor
)
2270 if (cursor
.option
== nir_cursor_before_instr
||
2271 cursor
.option
== nir_cursor_after_instr
) {
2272 return cursor
.instr
->block
;
2274 return cursor
.block
;
2278 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2280 static inline nir_cursor
2281 nir_before_block(nir_block
*block
)
2284 cursor
.option
= nir_cursor_before_block
;
2285 cursor
.block
= block
;
2289 static inline nir_cursor
2290 nir_after_block(nir_block
*block
)
2293 cursor
.option
= nir_cursor_after_block
;
2294 cursor
.block
= block
;
2298 static inline nir_cursor
2299 nir_before_instr(nir_instr
*instr
)
2302 cursor
.option
= nir_cursor_before_instr
;
2303 cursor
.instr
= instr
;
2307 static inline nir_cursor
2308 nir_after_instr(nir_instr
*instr
)
2311 cursor
.option
= nir_cursor_after_instr
;
2312 cursor
.instr
= instr
;
2316 static inline nir_cursor
2317 nir_after_block_before_jump(nir_block
*block
)
2319 nir_instr
*last_instr
= nir_block_last_instr(block
);
2320 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2321 return nir_before_instr(last_instr
);
2323 return nir_after_block(block
);
2327 static inline nir_cursor
2328 nir_before_cf_node(nir_cf_node
*node
)
2330 if (node
->type
== nir_cf_node_block
)
2331 return nir_before_block(nir_cf_node_as_block(node
));
2333 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2336 static inline nir_cursor
2337 nir_after_cf_node(nir_cf_node
*node
)
2339 if (node
->type
== nir_cf_node_block
)
2340 return nir_after_block(nir_cf_node_as_block(node
));
2342 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2345 static inline nir_cursor
2346 nir_after_phis(nir_block
*block
)
2348 nir_foreach_instr(instr
, block
) {
2349 if (instr
->type
!= nir_instr_type_phi
)
2350 return nir_before_instr(instr
);
2352 return nir_after_block(block
);
2355 static inline nir_cursor
2356 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2358 if (node
->type
== nir_cf_node_block
)
2359 return nir_after_block(nir_cf_node_as_block(node
));
2361 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2363 return nir_after_phis(block
);
2366 static inline nir_cursor
2367 nir_before_cf_list(struct exec_list
*cf_list
)
2369 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2370 exec_list_get_head(cf_list
), node
);
2371 return nir_before_cf_node(first_node
);
2374 static inline nir_cursor
2375 nir_after_cf_list(struct exec_list
*cf_list
)
2377 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2378 exec_list_get_tail(cf_list
), node
);
2379 return nir_after_cf_node(last_node
);
2383 * Insert a NIR instruction at the given cursor.
2385 * Note: This does not update the cursor.
2387 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2390 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2392 nir_instr_insert(nir_before_instr(instr
), before
);
2396 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2398 nir_instr_insert(nir_after_instr(instr
), after
);
2402 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2404 nir_instr_insert(nir_before_block(block
), before
);
2408 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2410 nir_instr_insert(nir_after_block(block
), after
);
2414 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2416 nir_instr_insert(nir_before_cf_node(node
), before
);
2420 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2422 nir_instr_insert(nir_after_cf_node(node
), after
);
2426 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2428 nir_instr_insert(nir_before_cf_list(list
), before
);
2432 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2434 nir_instr_insert(nir_after_cf_list(list
), after
);
2437 void nir_instr_remove_v(nir_instr
*instr
);
2439 static inline nir_cursor
2440 nir_instr_remove(nir_instr
*instr
)
2443 nir_instr
*prev
= nir_instr_prev(instr
);
2445 cursor
= nir_after_instr(prev
);
2447 cursor
= nir_before_block(instr
->block
);
2449 nir_instr_remove_v(instr
);
2455 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2456 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2457 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2458 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2460 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2461 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2463 nir_const_value
*nir_src_as_const_value(nir_src src
);
2464 bool nir_src_is_dynamically_uniform(nir_src src
);
2465 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2466 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2467 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2468 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2469 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2472 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2473 unsigned num_components
, unsigned bit_size
,
2475 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2476 unsigned num_components
, unsigned bit_size
,
2479 nir_ssa_dest_init_for_type(nir_instr
*instr
, nir_dest
*dest
,
2480 const struct glsl_type
*type
,
2483 assert(glsl_type_is_vector_or_scalar(type
));
2484 nir_ssa_dest_init(instr
, dest
, glsl_get_components(type
),
2485 glsl_get_bit_size(type
), name
);
2487 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2488 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2489 nir_instr
*after_me
);
2491 nir_component_mask_t
nir_ssa_def_components_read(const nir_ssa_def
*def
);
2494 * finds the next basic block in source-code order, returns NULL if there is
2498 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2500 /* Performs the opposite of nir_block_cf_tree_next() */
2502 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2504 /* Gets the first block in a CF node in source-code order */
2506 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2508 /* Gets the last block in a CF node in source-code order */
2510 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2512 /* Gets the next block after a CF node in source-code order */
2514 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2516 /* Macros for loops that visit blocks in source-code order */
2518 #define nir_foreach_block(block, impl) \
2519 for (nir_block *block = nir_start_block(impl); block != NULL; \
2520 block = nir_block_cf_tree_next(block))
2522 #define nir_foreach_block_safe(block, impl) \
2523 for (nir_block *block = nir_start_block(impl), \
2524 *next = nir_block_cf_tree_next(block); \
2526 block = next, next = nir_block_cf_tree_next(block))
2528 #define nir_foreach_block_reverse(block, impl) \
2529 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2530 block = nir_block_cf_tree_prev(block))
2532 #define nir_foreach_block_reverse_safe(block, impl) \
2533 for (nir_block *block = nir_impl_last_block(impl), \
2534 *prev = nir_block_cf_tree_prev(block); \
2536 block = prev, prev = nir_block_cf_tree_prev(block))
2538 #define nir_foreach_block_in_cf_node(block, node) \
2539 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2540 block != nir_cf_node_cf_tree_next(node); \
2541 block = nir_block_cf_tree_next(block))
2543 /* If the following CF node is an if, this function returns that if.
2544 * Otherwise, it returns NULL.
2546 nir_if
*nir_block_get_following_if(nir_block
*block
);
2548 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2550 void nir_index_local_regs(nir_function_impl
*impl
);
2551 void nir_index_global_regs(nir_shader
*shader
);
2552 void nir_index_ssa_defs(nir_function_impl
*impl
);
2553 unsigned nir_index_instrs(nir_function_impl
*impl
);
2555 void nir_index_blocks(nir_function_impl
*impl
);
2557 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2558 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2559 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2561 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2562 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2563 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2564 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2566 nir_shader
*nir_shader_serialize_deserialize(void *mem_ctx
, nir_shader
*s
);
2569 void nir_validate_shader(nir_shader
*shader
);
2570 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2571 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2574 should_clone_nir(void)
2576 static int should_clone
= -1;
2577 if (should_clone
< 0)
2578 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2580 return should_clone
;
2584 should_serialize_deserialize_nir(void)
2586 static int test_serialize
= -1;
2587 if (test_serialize
< 0)
2588 test_serialize
= env_var_as_boolean("NIR_TEST_SERIALIZE", false);
2590 return test_serialize
;
2594 should_print_nir(void)
2596 static int should_print
= -1;
2597 if (should_print
< 0)
2598 should_print
= env_var_as_boolean("NIR_PRINT", false);
2600 return should_print
;
2603 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2604 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2605 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2606 static inline bool should_clone_nir(void) { return false; }
2607 static inline bool should_serialize_deserialize_nir(void) { return false; }
2608 static inline bool should_print_nir(void) { return false; }
2611 #define _PASS(nir, do_pass) do { \
2613 nir_validate_shader(nir); \
2614 if (should_clone_nir()) { \
2615 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2619 if (should_serialize_deserialize_nir()) { \
2620 void *mem_ctx = ralloc_parent(nir); \
2621 nir = nir_shader_serialize_deserialize(mem_ctx, nir); \
2625 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2626 nir_metadata_set_validation_flag(nir); \
2627 if (should_print_nir()) \
2628 printf("%s\n", #pass); \
2629 if (pass(nir, ##__VA_ARGS__)) { \
2631 if (should_print_nir()) \
2632 nir_print_shader(nir, stdout); \
2633 nir_metadata_check_validation_flag(nir); \
2637 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2638 if (should_print_nir()) \
2639 printf("%s\n", #pass); \
2640 pass(nir, ##__VA_ARGS__); \
2641 if (should_print_nir()) \
2642 nir_print_shader(nir, stdout); \
2645 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2646 void nir_calc_dominance(nir_shader
*shader
);
2648 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2649 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2651 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2652 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2654 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2655 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2657 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2658 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2660 int nir_gs_count_vertices(const nir_shader
*shader
);
2662 bool nir_split_var_copies(nir_shader
*shader
);
2663 bool nir_split_per_member_structs(nir_shader
*shader
);
2665 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2666 bool nir_lower_returns(nir_shader
*shader
);
2668 bool nir_inline_functions(nir_shader
*shader
);
2670 bool nir_propagate_invariant(nir_shader
*shader
);
2672 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2673 void nir_lower_deref_copy_instr(struct nir_builder
*b
,
2674 nir_intrinsic_instr
*copy
);
2675 bool nir_lower_var_copies(nir_shader
*shader
);
2677 void nir_fixup_deref_modes(nir_shader
*shader
);
2679 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2681 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2683 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2685 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2686 nir_function_impl
*entrypoint
,
2687 bool outputs
, bool inputs
);
2689 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2691 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2692 int (*type_size
)(const struct glsl_type
*));
2694 /* Some helpers to do very simple linking */
2695 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2696 void nir_compact_varyings(nir_shader
*producer
, nir_shader
*consumer
,
2697 bool default_to_smooth_interp
);
2700 /* If set, this forces all non-flat fragment shader inputs to be
2701 * interpolated as if with the "sample" qualifier. This requires
2702 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2704 nir_lower_io_force_sample_interpolation
= (1 << 1),
2705 } nir_lower_io_options
;
2706 bool nir_lower_io(nir_shader
*shader
,
2707 nir_variable_mode modes
,
2708 int (*type_size
)(const struct glsl_type
*),
2709 nir_lower_io_options
);
2710 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2711 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2713 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2715 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2716 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2717 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2719 bool nir_remove_dead_derefs(nir_shader
*shader
);
2720 bool nir_remove_dead_derefs_impl(nir_function_impl
*impl
);
2721 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2722 bool nir_lower_constant_initializers(nir_shader
*shader
,
2723 nir_variable_mode modes
);
2725 bool nir_move_load_const(nir_shader
*shader
);
2726 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2727 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2728 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
2730 bool nir_lower_alu(nir_shader
*shader
);
2731 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2732 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2733 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
2734 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2735 void nir_lower_io_arrays_to_elements(nir_shader
*producer
, nir_shader
*consumer
);
2736 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader
*shader
,
2738 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2739 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
2741 typedef struct nir_lower_subgroups_options
{
2742 uint8_t subgroup_size
;
2743 uint8_t ballot_bit_size
;
2744 bool lower_to_scalar
:1;
2745 bool lower_vote_trivial
:1;
2746 bool lower_vote_eq_to_ballot
:1;
2747 bool lower_subgroup_masks
:1;
2748 bool lower_shuffle
:1;
2749 bool lower_shuffle_to_32bit
:1;
2751 } nir_lower_subgroups_options
;
2753 bool nir_lower_subgroups(nir_shader
*shader
,
2754 const nir_lower_subgroups_options
*options
);
2756 bool nir_lower_system_values(nir_shader
*shader
);
2758 typedef struct nir_lower_tex_options
{
2760 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2761 * sampler types a texture projector is lowered.
2766 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2768 bool lower_txf_offset
;
2771 * If true, lower away nir_tex_src_offset for all rect textures.
2773 bool lower_rect_offset
;
2776 * If true, lower rect textures to 2D, using txs to fetch the
2777 * texture dimensions and dividing the texture coords by the
2778 * texture dims to normalize.
2783 * If true, convert yuv to rgb.
2785 unsigned lower_y_uv_external
;
2786 unsigned lower_y_u_v_external
;
2787 unsigned lower_yx_xuxv_external
;
2788 unsigned lower_xy_uxvx_external
;
2791 * To emulate certain texture wrap modes, this can be used
2792 * to saturate the specified tex coord to [0.0, 1.0]. The
2793 * bits are according to sampler #, ie. if, for example:
2795 * (conf->saturate_s & (1 << n))
2797 * is true, then the s coord for sampler n is saturated.
2799 * Note that clamping must happen *after* projector lowering
2800 * so any projected texture sample instruction with a clamped
2801 * coordinate gets automatically lowered, regardless of the
2802 * 'lower_txp' setting.
2804 unsigned saturate_s
;
2805 unsigned saturate_t
;
2806 unsigned saturate_r
;
2808 /* Bitmask of textures that need swizzling.
2810 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2811 * swizzles[texture_index] is applied to the result of the texturing
2814 unsigned swizzle_result
;
2816 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2817 * while 4 and 5 represent 0 and 1 respectively.
2819 uint8_t swizzles
[32][4];
2822 * Bitmap of textures that need srgb to linear conversion. If
2823 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2824 * of the texture are lowered to linear.
2826 unsigned lower_srgb
;
2829 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2831 bool lower_txd_cube_map
;
2834 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
2835 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
2836 * with lower_txd_cube_map.
2838 bool lower_txd_shadow
;
2841 * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
2842 * Implies lower_txd_cube_map and lower_txd_shadow.
2845 } nir_lower_tex_options
;
2847 bool nir_lower_tex(nir_shader
*shader
,
2848 const nir_lower_tex_options
*options
);
2850 bool nir_lower_idiv(nir_shader
*shader
);
2852 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2853 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2854 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2856 void nir_lower_two_sided_color(nir_shader
*shader
);
2858 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
2860 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2861 bool nir_lower_patch_vertices(nir_shader
*nir
, unsigned static_count
,
2862 const gl_state_index16
*uniform_state_tokens
);
2864 typedef struct nir_lower_wpos_ytransform_options
{
2865 gl_state_index16 state_tokens
[STATE_LENGTH
];
2866 bool fs_coord_origin_upper_left
:1;
2867 bool fs_coord_origin_lower_left
:1;
2868 bool fs_coord_pixel_center_integer
:1;
2869 bool fs_coord_pixel_center_half_integer
:1;
2870 } nir_lower_wpos_ytransform_options
;
2872 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2873 const nir_lower_wpos_ytransform_options
*options
);
2874 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
2876 typedef struct nir_lower_drawpixels_options
{
2877 gl_state_index16 texcoord_state_tokens
[STATE_LENGTH
];
2878 gl_state_index16 scale_state_tokens
[STATE_LENGTH
];
2879 gl_state_index16 bias_state_tokens
[STATE_LENGTH
];
2880 unsigned drawpix_sampler
;
2881 unsigned pixelmap_sampler
;
2883 bool scale_and_bias
:1;
2884 } nir_lower_drawpixels_options
;
2886 void nir_lower_drawpixels(nir_shader
*shader
,
2887 const nir_lower_drawpixels_options
*options
);
2889 typedef struct nir_lower_bitmap_options
{
2892 } nir_lower_bitmap_options
;
2894 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2896 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
2897 bool nir_lower_to_source_mods(nir_shader
*shader
);
2899 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2901 typedef unsigned (*nir_lower_bit_size_callback
)(const nir_alu_instr
*, void *);
2903 bool nir_lower_bit_size(nir_shader
*shader
,
2904 nir_lower_bit_size_callback callback
,
2905 void *callback_data
);
2908 nir_lower_imul64
= (1 << 0),
2909 nir_lower_isign64
= (1 << 1),
2910 /** Lower all int64 modulus and division opcodes */
2911 nir_lower_divmod64
= (1 << 2),
2912 } nir_lower_int64_options
;
2914 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
2917 nir_lower_drcp
= (1 << 0),
2918 nir_lower_dsqrt
= (1 << 1),
2919 nir_lower_drsq
= (1 << 2),
2920 nir_lower_dtrunc
= (1 << 3),
2921 nir_lower_dfloor
= (1 << 4),
2922 nir_lower_dceil
= (1 << 5),
2923 nir_lower_dfract
= (1 << 6),
2924 nir_lower_dround_even
= (1 << 7),
2925 nir_lower_dmod
= (1 << 8)
2926 } nir_lower_doubles_options
;
2928 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2929 bool nir_lower_pack(nir_shader
*shader
);
2931 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2933 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2935 void nir_loop_analyze_impl(nir_function_impl
*impl
,
2936 nir_variable_mode indirect_mask
);
2938 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2940 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2941 bool nir_repair_ssa(nir_shader
*shader
);
2943 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
2945 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2946 * registers. If false, convert all values (even those not involved in a phi
2947 * node) to registers.
2949 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2951 bool nir_lower_phis_to_regs_block(nir_block
*block
);
2952 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
2954 bool nir_opt_algebraic(nir_shader
*shader
);
2955 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
2956 bool nir_opt_algebraic_late(nir_shader
*shader
);
2957 bool nir_opt_constant_folding(nir_shader
*shader
);
2959 bool nir_opt_global_to_local(nir_shader
*shader
);
2961 bool nir_copy_prop(nir_shader
*shader
);
2963 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
2965 bool nir_opt_cse(nir_shader
*shader
);
2967 bool nir_opt_dce(nir_shader
*shader
);
2969 bool nir_opt_dead_cf(nir_shader
*shader
);
2971 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2973 bool nir_opt_if(nir_shader
*shader
);
2975 bool nir_opt_intrinsics(nir_shader
*shader
);
2977 bool nir_opt_large_constants(nir_shader
*shader
,
2978 glsl_type_size_align_func size_align
,
2979 unsigned threshold
);
2981 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
2983 bool nir_opt_move_comparisons(nir_shader
*shader
);
2985 bool nir_opt_move_load_ubo(nir_shader
*shader
);
2987 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2989 bool nir_opt_remove_phis_impl(nir_function_impl
*impl
);
2990 bool nir_opt_remove_phis(nir_shader
*shader
);
2992 bool nir_opt_shrink_load(nir_shader
*shader
);
2994 bool nir_opt_trivial_continues(nir_shader
*shader
);
2996 bool nir_opt_undef(nir_shader
*shader
);
2998 bool nir_opt_conditional_discard(nir_shader
*shader
);
3000 void nir_sweep(nir_shader
*shader
);
3002 void nir_remap_attributes(nir_shader
*shader
,
3003 const nir_shader_compiler_options
*options
);
3005 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
3006 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);