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 enum gl_access_qualifier access
;
382 /** Image internal format if specified explicitly, otherwise GL_NONE. */
388 * Built-in state that backs this uniform
390 * Once set at variable creation, \c state_slots must remain invariant.
391 * This is because, ideally, this array would be shared by all clones of
392 * this variable in the IR tree. In other words, we'd really like for it
393 * to be a fly-weight.
395 * If the variable is not a uniform, \c num_state_slots will be zero and
396 * \c state_slots will be \c NULL.
399 unsigned num_state_slots
; /**< Number of state slots used */
400 nir_state_slot
*state_slots
; /**< State descriptors. */
404 * Constant expression assigned in the initializer of the variable
406 * This field should only be used temporarily by creators of NIR shaders
407 * and then lower_constant_initializers can be used to get rid of them.
408 * Most of the rest of NIR ignores this field or asserts that it's NULL.
410 nir_constant
*constant_initializer
;
413 * For variables that are in an interface block or are an instance of an
414 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
416 * \sa ir_variable::location
418 const struct glsl_type
*interface_type
;
421 * Description of per-member data for per-member struct variables
423 * This is used for variables which are actually an amalgamation of
424 * multiple entities such as a struct of built-in values or a struct of
425 * inputs each with their own layout specifier. This is only allowed on
426 * variables with a struct or array of array of struct type.
428 unsigned num_members
;
429 struct nir_variable_data
*members
;
432 #define nir_foreach_variable(var, var_list) \
433 foreach_list_typed(nir_variable, var, node, var_list)
435 #define nir_foreach_variable_safe(var, var_list) \
436 foreach_list_typed_safe(nir_variable, var, node, var_list)
439 nir_variable_is_global(const nir_variable
*var
)
441 return var
->data
.mode
!= nir_var_local
;
444 typedef struct nir_register
{
445 struct exec_node node
;
447 unsigned num_components
; /** < number of vector components */
448 unsigned num_array_elems
; /** < size of array (0 for no array) */
450 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
453 /** generic register index. */
456 /** only for debug purposes, can be NULL */
459 /** whether this register is local (per-function) or global (per-shader) */
463 * If this flag is set to true, then accessing channels >= num_components
464 * is well-defined, and simply spills over to the next array element. This
465 * is useful for backends that can do per-component accessing, in
466 * particular scalar backends. By setting this flag and making
467 * num_components equal to 1, structures can be packed tightly into
468 * registers and then registers can be accessed per-component to get to
469 * each structure member, even if it crosses vec4 boundaries.
473 /** set of nir_srcs where this register is used (read from) */
474 struct list_head uses
;
476 /** set of nir_dests where this register is defined (written to) */
477 struct list_head defs
;
479 /** set of nir_ifs where this register is used as a condition */
480 struct list_head if_uses
;
483 #define nir_foreach_register(reg, reg_list) \
484 foreach_list_typed(nir_register, reg, node, reg_list)
485 #define nir_foreach_register_safe(reg, reg_list) \
486 foreach_list_typed_safe(nir_register, reg, node, reg_list)
490 nir_instr_type_deref
,
493 nir_instr_type_intrinsic
,
494 nir_instr_type_load_const
,
496 nir_instr_type_ssa_undef
,
498 nir_instr_type_parallel_copy
,
501 typedef struct nir_instr
{
502 struct exec_node node
;
504 struct nir_block
*block
;
506 /** generic instruction index. */
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.
515 static inline nir_instr
*
516 nir_instr_next(nir_instr
*instr
)
518 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
519 if (exec_node_is_tail_sentinel(next
))
522 return exec_node_data(nir_instr
, next
, node
);
525 static inline nir_instr
*
526 nir_instr_prev(nir_instr
*instr
)
528 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
529 if (exec_node_is_head_sentinel(prev
))
532 return exec_node_data(nir_instr
, prev
, node
);
536 nir_instr_is_first(const nir_instr
*instr
)
538 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
542 nir_instr_is_last(const nir_instr
*instr
)
544 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
547 typedef struct nir_ssa_def
{
548 /** for debugging only, can be NULL */
551 /** generic SSA definition index. */
554 /** Index into the live_in and live_out bitfields */
557 /** Instruction which produces this SSA value. */
558 nir_instr
*parent_instr
;
560 /** set of nir_instrs where this register is used (read from) */
561 struct list_head uses
;
563 /** set of nir_ifs where this register is used as a condition */
564 struct list_head if_uses
;
566 uint8_t num_components
;
568 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
576 struct nir_src
*indirect
; /** < NULL for no indirect offset */
577 unsigned base_offset
;
579 /* TODO use-def chain goes here */
583 nir_instr
*parent_instr
;
584 struct list_head def_link
;
587 struct nir_src
*indirect
; /** < NULL for no indirect offset */
588 unsigned base_offset
;
590 /* TODO def-use chain goes here */
595 typedef struct nir_src
{
597 /** Instruction that consumes this value as a source. */
598 nir_instr
*parent_instr
;
599 struct nir_if
*parent_if
;
602 struct list_head use_link
;
612 static inline nir_src
615 nir_src src
= { { NULL
} };
619 #define NIR_SRC_INIT nir_src_init()
621 #define nir_foreach_use(src, reg_or_ssa_def) \
622 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
624 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
625 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
627 #define nir_foreach_if_use(src, reg_or_ssa_def) \
628 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
630 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
631 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
642 static inline nir_dest
645 nir_dest dest
= { { { NULL
} } };
649 #define NIR_DEST_INIT nir_dest_init()
651 #define nir_foreach_def(dest, reg) \
652 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
654 #define nir_foreach_def_safe(dest, reg) \
655 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
657 static inline nir_src
658 nir_src_for_ssa(nir_ssa_def
*def
)
660 nir_src src
= NIR_SRC_INIT
;
668 static inline nir_src
669 nir_src_for_reg(nir_register
*reg
)
671 nir_src src
= NIR_SRC_INIT
;
675 src
.reg
.indirect
= NULL
;
676 src
.reg
.base_offset
= 0;
681 static inline nir_dest
682 nir_dest_for_reg(nir_register
*reg
)
684 nir_dest dest
= NIR_DEST_INIT
;
691 static inline unsigned
692 nir_src_bit_size(nir_src src
)
694 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
697 static inline unsigned
698 nir_src_num_components(nir_src src
)
700 return src
.is_ssa
? src
.ssa
->num_components
: src
.reg
.reg
->num_components
;
704 nir_src_is_const(nir_src src
)
707 src
.ssa
->parent_instr
->type
== nir_instr_type_load_const
;
710 int64_t nir_src_as_int(nir_src src
);
711 uint64_t nir_src_as_uint(nir_src src
);
712 bool nir_src_as_bool(nir_src src
);
713 double nir_src_as_float(nir_src src
);
714 int64_t nir_src_comp_as_int(nir_src src
, unsigned component
);
715 uint64_t nir_src_comp_as_uint(nir_src src
, unsigned component
);
716 bool nir_src_comp_as_bool(nir_src src
, unsigned component
);
717 double nir_src_comp_as_float(nir_src src
, unsigned component
);
719 static inline unsigned
720 nir_dest_bit_size(nir_dest dest
)
722 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
725 static inline unsigned
726 nir_dest_num_components(nir_dest dest
)
728 return dest
.is_ssa
? dest
.ssa
.num_components
: dest
.reg
.reg
->num_components
;
731 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
732 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
738 * \name input modifiers
742 * For inputs interpreted as floating point, flips the sign bit. For
743 * inputs interpreted as integers, performs the two's complement negation.
748 * Clears the sign bit for floating point values, and computes the integer
749 * absolute value for integers. Note that the negate modifier acts after
750 * the absolute value modifier, therefore if both are set then all inputs
751 * will become negative.
757 * For each input component, says which component of the register it is
758 * chosen from. Note that which elements of the swizzle are used and which
759 * are ignored are based on the write mask for most opcodes - for example,
760 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
761 * a swizzle of {2, x, 1, 0} where x means "don't care."
763 uint8_t swizzle
[NIR_MAX_VEC_COMPONENTS
];
770 * \name saturate output modifier
772 * Only valid for opcodes that output floating-point numbers. Clamps the
773 * output to between 0.0 and 1.0 inclusive.
778 unsigned write_mask
: NIR_MAX_VEC_COMPONENTS
; /* ignored if dest.is_ssa is true */
782 nir_type_invalid
= 0, /* Not a valid type */
787 nir_type_bool32
= 32 | nir_type_bool
,
788 nir_type_int8
= 8 | nir_type_int
,
789 nir_type_int16
= 16 | nir_type_int
,
790 nir_type_int32
= 32 | nir_type_int
,
791 nir_type_int64
= 64 | nir_type_int
,
792 nir_type_uint8
= 8 | nir_type_uint
,
793 nir_type_uint16
= 16 | nir_type_uint
,
794 nir_type_uint32
= 32 | nir_type_uint
,
795 nir_type_uint64
= 64 | nir_type_uint
,
796 nir_type_float16
= 16 | nir_type_float
,
797 nir_type_float32
= 32 | nir_type_float
,
798 nir_type_float64
= 64 | nir_type_float
,
801 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
802 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
804 static inline unsigned
805 nir_alu_type_get_type_size(nir_alu_type type
)
807 return type
& NIR_ALU_TYPE_SIZE_MASK
;
810 static inline unsigned
811 nir_alu_type_get_base_type(nir_alu_type type
)
813 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
816 static inline nir_alu_type
817 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
821 return nir_type_bool32
;
824 return nir_type_uint32
;
827 return nir_type_int32
;
829 case GLSL_TYPE_UINT16
:
830 return nir_type_uint16
;
832 case GLSL_TYPE_INT16
:
833 return nir_type_int16
;
835 case GLSL_TYPE_UINT8
:
836 return nir_type_uint8
;
838 return nir_type_int8
;
839 case GLSL_TYPE_UINT64
:
840 return nir_type_uint64
;
842 case GLSL_TYPE_INT64
:
843 return nir_type_int64
;
845 case GLSL_TYPE_FLOAT
:
846 return nir_type_float32
;
848 case GLSL_TYPE_FLOAT16
:
849 return nir_type_float16
;
851 case GLSL_TYPE_DOUBLE
:
852 return nir_type_float64
;
855 unreachable("unknown type");
859 static inline nir_alu_type
860 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
862 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
865 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
,
866 nir_rounding_mode rnd
);
869 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
870 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
871 } nir_op_algebraic_property
;
879 * The number of components in the output
881 * If non-zero, this is the size of the output and input sizes are
882 * explicitly given; swizzle and writemask are still in effect, but if
883 * the output component is masked out, then the input component may
886 * If zero, the opcode acts in the standard, per-component manner; the
887 * operation is performed on each component (except the ones that are
888 * masked out) with the input being taken from the input swizzle for
891 * The size of some of the inputs may be given (i.e. non-zero) even
892 * though output_size is zero; in that case, the inputs with a zero
893 * size act per-component, while the inputs with non-zero size don't.
895 unsigned output_size
;
898 * The type of vector that the instruction outputs. Note that the
899 * staurate modifier is only allowed on outputs with the float type.
902 nir_alu_type output_type
;
905 * The number of components in each input
907 unsigned input_sizes
[NIR_MAX_VEC_COMPONENTS
];
910 * The type of vector that each input takes. Note that negate and
911 * absolute value are only allowed on inputs with int or float type and
912 * behave differently on the two.
914 nir_alu_type input_types
[NIR_MAX_VEC_COMPONENTS
];
916 nir_op_algebraic_property algebraic_properties
;
919 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
921 typedef struct nir_alu_instr
{
925 /** Indicates that this ALU instruction generates an exact value
927 * This is kind of a mixture of GLSL "precise" and "invariant" and not
928 * really equivalent to either. This indicates that the value generated by
929 * this operation is high-precision and any code transformations that touch
930 * it must ensure that the resulting value is bit-for-bit identical to the
939 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
940 nir_alu_instr
*instr
);
941 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
942 nir_alu_instr
*instr
);
944 /* is this source channel used? */
946 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
949 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
950 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
952 return (instr
->dest
.write_mask
>> channel
) & 1;
956 * For instructions whose destinations are SSA, get the number of channels
959 static inline unsigned
960 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
962 assert(instr
->dest
.dest
.is_ssa
);
964 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
965 return nir_op_infos
[instr
->op
].input_sizes
[src
];
967 return instr
->dest
.dest
.ssa
.num_components
;
970 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
971 unsigned src1
, unsigned src2
);
975 nir_deref_type_array
,
976 nir_deref_type_array_wildcard
,
977 nir_deref_type_struct
,
984 /** The type of this deref instruction */
985 nir_deref_type deref_type
;
987 /** The mode of the underlying variable */
988 nir_variable_mode mode
;
990 /** The dereferenced type of the resulting pointer value */
991 const struct glsl_type
*type
;
994 /** Variable being dereferenced if deref_type is a deref_var */
997 /** Parent deref if deref_type is not deref_var */
1001 /** Additional deref parameters */
1012 /** Destination to store the resulting "pointer" */
1016 NIR_DEFINE_CAST(nir_instr_as_deref
, nir_instr
, nir_deref_instr
, instr
,
1017 type
, nir_instr_type_deref
)
1019 static inline nir_deref_instr
*
1020 nir_src_as_deref(nir_src src
)
1025 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_deref
)
1028 return nir_instr_as_deref(src
.ssa
->parent_instr
);
1031 static inline nir_deref_instr
*
1032 nir_deref_instr_parent(const nir_deref_instr
*instr
)
1034 if (instr
->deref_type
== nir_deref_type_var
)
1037 return nir_src_as_deref(instr
->parent
);
1040 static inline nir_variable
*
1041 nir_deref_instr_get_variable(const nir_deref_instr
*instr
)
1043 while (instr
->deref_type
!= nir_deref_type_var
) {
1044 if (instr
->deref_type
== nir_deref_type_cast
)
1047 instr
= nir_deref_instr_parent(instr
);
1053 bool nir_deref_instr_has_indirect(nir_deref_instr
*instr
);
1055 bool nir_deref_instr_remove_if_unused(nir_deref_instr
*instr
);
1060 struct nir_function
*callee
;
1062 unsigned num_params
;
1066 #include "nir_intrinsics.h"
1068 #define NIR_INTRINSIC_MAX_CONST_INDEX 4
1070 /** Represents an intrinsic
1072 * An intrinsic is an instruction type for handling things that are
1073 * more-or-less regular operations but don't just consume and produce SSA
1074 * values like ALU operations do. Intrinsics are not for things that have
1075 * special semantic meaning such as phi nodes and parallel copies.
1076 * Examples of intrinsics include variable load/store operations, system
1077 * value loads, and the like. Even though texturing more-or-less falls
1078 * under this category, texturing is its own instruction type because
1079 * trying to represent texturing with intrinsics would lead to a
1080 * combinatorial explosion of intrinsic opcodes.
1082 * By having a single instruction type for handling a lot of different
1083 * cases, optimization passes can look for intrinsics and, for the most
1084 * part, completely ignore them. Each intrinsic type also has a few
1085 * possible flags that govern whether or not they can be reordered or
1086 * eliminated. That way passes like dead code elimination can still work
1087 * on intrisics without understanding the meaning of each.
1089 * Each intrinsic has some number of constant indices, some number of
1090 * variables, and some number of sources. What these sources, variables,
1091 * and indices mean depends on the intrinsic and is documented with the
1092 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
1093 * instructions are the only types of instruction that can operate on
1099 nir_intrinsic_op intrinsic
;
1103 /** number of components if this is a vectorized intrinsic
1105 * Similarly to ALU operations, some intrinsics are vectorized.
1106 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
1107 * For vectorized intrinsics, the num_components field specifies the
1108 * number of destination components and the number of source components
1109 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
1111 uint8_t num_components
;
1113 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
1116 } nir_intrinsic_instr
;
1118 static inline nir_variable
*
1119 nir_intrinsic_get_var(nir_intrinsic_instr
*intrin
, unsigned i
)
1121 return nir_deref_instr_get_variable(nir_src_as_deref(intrin
->src
[i
]));
1125 * \name NIR intrinsics semantic flags
1127 * information about what the compiler can do with the intrinsics.
1129 * \sa nir_intrinsic_info::flags
1133 * whether the intrinsic can be safely eliminated if none of its output
1134 * value is not being used.
1136 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
1139 * Whether the intrinsic can be reordered with respect to any other
1140 * intrinsic, i.e. whether the only reordering dependencies of the
1141 * intrinsic are due to the register reads/writes.
1143 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
1144 } nir_intrinsic_semantic_flag
;
1147 * \name NIR intrinsics const-index flag
1149 * Indicates the usage of a const_index slot.
1151 * \sa nir_intrinsic_info::index_map
1155 * Generally instructions that take a offset src argument, can encode
1156 * a constant 'base' value which is added to the offset.
1158 NIR_INTRINSIC_BASE
= 1,
1161 * For store instructions, a writemask for the store.
1163 NIR_INTRINSIC_WRMASK
= 2,
1166 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1168 NIR_INTRINSIC_STREAM_ID
= 3,
1171 * The clip-plane id for load_user_clip_plane intrinsic.
1173 NIR_INTRINSIC_UCP_ID
= 4,
1176 * The amount of data, starting from BASE, that this instruction may
1177 * access. This is used to provide bounds if the offset is not constant.
1179 NIR_INTRINSIC_RANGE
= 5,
1182 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1184 NIR_INTRINSIC_DESC_SET
= 6,
1187 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1189 NIR_INTRINSIC_BINDING
= 7,
1194 NIR_INTRINSIC_COMPONENT
= 8,
1197 * Interpolation mode (only meaningful for FS inputs).
1199 NIR_INTRINSIC_INTERP_MODE
= 9,
1202 * A binary nir_op to use when performing a reduction or scan operation
1204 NIR_INTRINSIC_REDUCTION_OP
= 10,
1207 * Cluster size for reduction operations
1209 NIR_INTRINSIC_CLUSTER_SIZE
= 11,
1212 * Parameter index for a load_param intrinsic
1214 NIR_INTRINSIC_PARAM_IDX
= 12,
1217 * Image dimensionality for image intrinsics
1219 * One of GLSL_SAMPLER_DIM_*
1221 NIR_INTRINSIC_IMAGE_DIM
= 13,
1224 * Non-zero if we are accessing an array image
1226 NIR_INTRINSIC_IMAGE_ARRAY
= 14,
1229 * Image format for image intrinsics
1231 NIR_INTRINSIC_FORMAT
= 15,
1234 * Access qualifiers for image intrinsics
1236 NIR_INTRINSIC_ACCESS
= 16,
1238 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1240 } nir_intrinsic_index_flag
;
1242 #define NIR_INTRINSIC_MAX_INPUTS 5
1247 unsigned num_srcs
; /** < number of register/SSA inputs */
1249 /** number of components of each input register
1251 * If this value is 0, the number of components is given by the
1252 * num_components field of nir_intrinsic_instr.
1254 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1258 /** number of components of the output register
1260 * If this value is 0, the number of components is given by the
1261 * num_components field of nir_intrinsic_instr.
1263 unsigned dest_components
;
1265 /** the number of constant indices used by the intrinsic */
1266 unsigned num_indices
;
1268 /** indicates the usage of intr->const_index[n] */
1269 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1271 /** semantic flags for calls to this intrinsic */
1272 nir_intrinsic_semantic_flag flags
;
1273 } nir_intrinsic_info
;
1275 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1277 static inline unsigned
1278 nir_intrinsic_src_components(nir_intrinsic_instr
*intr
, unsigned srcn
)
1280 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1281 assert(srcn
< info
->num_srcs
);
1282 if (info
->src_components
[srcn
])
1283 return info
->src_components
[srcn
];
1285 return intr
->num_components
;
1288 static inline unsigned
1289 nir_intrinsic_dest_components(nir_intrinsic_instr
*intr
)
1291 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1292 if (!info
->has_dest
)
1294 else if (info
->dest_components
)
1295 return info
->dest_components
;
1297 return intr
->num_components
;
1300 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1301 static inline type \
1302 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1304 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1305 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1306 return (type)instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1308 static inline void \
1309 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1311 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1312 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1313 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1316 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1317 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1318 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1319 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1320 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1321 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1322 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1323 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1324 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1325 INTRINSIC_IDX_ACCESSORS(reduction_op
, REDUCTION_OP
, unsigned)
1326 INTRINSIC_IDX_ACCESSORS(cluster_size
, CLUSTER_SIZE
, unsigned)
1327 INTRINSIC_IDX_ACCESSORS(param_idx
, PARAM_IDX
, unsigned)
1328 INTRINSIC_IDX_ACCESSORS(image_dim
, IMAGE_DIM
, enum glsl_sampler_dim
)
1329 INTRINSIC_IDX_ACCESSORS(image_array
, IMAGE_ARRAY
, bool)
1330 INTRINSIC_IDX_ACCESSORS(access
, ACCESS
, enum gl_access_qualifier
)
1331 INTRINSIC_IDX_ACCESSORS(format
, FORMAT
, unsigned)
1334 * \group texture information
1336 * This gives semantic information about textures which is useful to the
1337 * frontend, the backend, and lowering passes, but not the optimizer.
1342 nir_tex_src_projector
,
1343 nir_tex_src_comparator
, /* shadow comparator */
1347 nir_tex_src_ms_index
, /* MSAA sample index */
1348 nir_tex_src_ms_mcs
, /* MSAA compression value */
1351 nir_tex_src_texture_deref
, /* < deref pointing to the texture */
1352 nir_tex_src_sampler_deref
, /* < deref pointing to the sampler */
1353 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1354 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1355 nir_tex_src_plane
, /* < selects plane for planar textures */
1356 nir_num_tex_src_types
1361 nir_tex_src_type src_type
;
1365 nir_texop_tex
, /**< Regular texture look-up */
1366 nir_texop_txb
, /**< Texture look-up with LOD bias */
1367 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1368 nir_texop_txd
, /**< Texture look-up with partial derivatives */
1369 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1370 nir_texop_txf_ms
, /**< Multisample texture fetch */
1371 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1372 nir_texop_txs
, /**< Texture size */
1373 nir_texop_lod
, /**< Texture lod query */
1374 nir_texop_tg4
, /**< Texture gather */
1375 nir_texop_query_levels
, /**< Texture levels query */
1376 nir_texop_texture_samples
, /**< Texture samples query */
1377 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1385 enum glsl_sampler_dim sampler_dim
;
1386 nir_alu_type dest_type
;
1391 unsigned num_srcs
, coord_components
;
1392 bool is_array
, is_shadow
;
1395 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1396 * components or the new-style shadow that outputs 1 component.
1398 bool is_new_style_shadow
;
1400 /* gather component selector */
1401 unsigned component
: 2;
1403 /** The texture index
1405 * If this texture instruction has a nir_tex_src_texture_offset source,
1406 * then the texture index is given by texture_index + texture_offset.
1408 unsigned texture_index
;
1410 /** The size of the texture array or 0 if it's not an array */
1411 unsigned texture_array_size
;
1413 /** The sampler index
1415 * The following operations do not require a sampler and, as such, this
1416 * field should be ignored:
1418 * - nir_texop_txf_ms
1421 * - nir_texop_query_levels
1422 * - nir_texop_texture_samples
1423 * - nir_texop_samples_identical
1425 * If this texture instruction has a nir_tex_src_sampler_offset source,
1426 * then the sampler index is given by sampler_index + sampler_offset.
1428 unsigned sampler_index
;
1431 static inline unsigned
1432 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1434 switch (instr
->op
) {
1435 case nir_texop_txs
: {
1437 switch (instr
->sampler_dim
) {
1438 case GLSL_SAMPLER_DIM_1D
:
1439 case GLSL_SAMPLER_DIM_BUF
:
1442 case GLSL_SAMPLER_DIM_2D
:
1443 case GLSL_SAMPLER_DIM_CUBE
:
1444 case GLSL_SAMPLER_DIM_MS
:
1445 case GLSL_SAMPLER_DIM_RECT
:
1446 case GLSL_SAMPLER_DIM_EXTERNAL
:
1447 case GLSL_SAMPLER_DIM_SUBPASS
:
1450 case GLSL_SAMPLER_DIM_3D
:
1454 unreachable("not reached");
1456 if (instr
->is_array
)
1464 case nir_texop_texture_samples
:
1465 case nir_texop_query_levels
:
1466 case nir_texop_samples_identical
:
1470 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1477 /* Returns true if this texture operation queries something about the texture
1478 * rather than actually sampling it.
1481 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1483 switch (instr
->op
) {
1486 case nir_texop_texture_samples
:
1487 case nir_texop_query_levels
:
1488 case nir_texop_txf_ms_mcs
:
1495 case nir_texop_txf_ms
:
1499 unreachable("Invalid texture opcode");
1504 nir_alu_instr_is_comparison(const nir_alu_instr
*instr
)
1506 switch (instr
->op
) {
1527 static inline nir_alu_type
1528 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1530 switch (instr
->src
[src
].src_type
) {
1531 case nir_tex_src_coord
:
1532 switch (instr
->op
) {
1534 case nir_texop_txf_ms
:
1535 case nir_texop_txf_ms_mcs
:
1536 case nir_texop_samples_identical
:
1537 return nir_type_int
;
1540 return nir_type_float
;
1543 case nir_tex_src_lod
:
1544 switch (instr
->op
) {
1547 return nir_type_int
;
1550 return nir_type_float
;
1553 case nir_tex_src_projector
:
1554 case nir_tex_src_comparator
:
1555 case nir_tex_src_bias
:
1556 case nir_tex_src_ddx
:
1557 case nir_tex_src_ddy
:
1558 return nir_type_float
;
1560 case nir_tex_src_offset
:
1561 case nir_tex_src_ms_index
:
1562 case nir_tex_src_texture_offset
:
1563 case nir_tex_src_sampler_offset
:
1564 return nir_type_int
;
1567 unreachable("Invalid texture source type");
1571 static inline unsigned
1572 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1574 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1575 return instr
->coord_components
;
1577 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1578 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1581 if (instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1582 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1583 if (instr
->is_array
)
1584 return instr
->coord_components
- 1;
1586 return instr
->coord_components
;
1589 /* Usual APIs don't allow cube + offset, but we allow it, with 2 coords for
1590 * the offset, since a cube maps to a single face.
1592 if (instr
->src
[src
].src_type
== nir_tex_src_offset
) {
1593 if (instr
->sampler_dim
== GLSL_SAMPLER_DIM_CUBE
)
1595 else if (instr
->is_array
)
1596 return instr
->coord_components
- 1;
1598 return instr
->coord_components
;
1605 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1607 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1608 if (instr
->src
[i
].src_type
== type
)
1614 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1615 nir_tex_src_type src_type
,
1618 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1623 nir_const_value value
;
1626 } nir_load_const_instr
;
1639 /* creates a new SSA variable in an undefined state */
1644 } nir_ssa_undef_instr
;
1647 struct exec_node node
;
1649 /* The predecessor block corresponding to this source */
1650 struct nir_block
*pred
;
1655 #define nir_foreach_phi_src(phi_src, phi) \
1656 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1657 #define nir_foreach_phi_src_safe(phi_src, phi) \
1658 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1663 struct exec_list srcs
; /** < list of nir_phi_src */
1669 struct exec_node node
;
1672 } nir_parallel_copy_entry
;
1674 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1675 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1680 /* A list of nir_parallel_copy_entrys. The sources of all of the
1681 * entries are copied to the corresponding destinations "in parallel".
1682 * In other words, if we have two entries: a -> b and b -> a, the values
1685 struct exec_list entries
;
1686 } nir_parallel_copy_instr
;
1688 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1689 type
, nir_instr_type_alu
)
1690 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1691 type
, nir_instr_type_call
)
1692 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1693 type
, nir_instr_type_jump
)
1694 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1695 type
, nir_instr_type_tex
)
1696 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1697 type
, nir_instr_type_intrinsic
)
1698 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1699 type
, nir_instr_type_load_const
)
1700 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1701 type
, nir_instr_type_ssa_undef
)
1702 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1703 type
, nir_instr_type_phi
)
1704 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1705 nir_parallel_copy_instr
, instr
,
1706 type
, nir_instr_type_parallel_copy
)
1711 * Control flow consists of a tree of control flow nodes, which include
1712 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1713 * instructions that always run start-to-finish. Each basic block also keeps
1714 * track of its successors (blocks which may run immediately after the current
1715 * block) and predecessors (blocks which could have run immediately before the
1716 * current block). Each function also has a start block and an end block which
1717 * all return statements point to (which is always empty). Together, all the
1718 * blocks with their predecessors and successors make up the control flow
1719 * graph (CFG) of the function. There are helpers that modify the tree of
1720 * control flow nodes while modifying the CFG appropriately; these should be
1721 * used instead of modifying the tree directly.
1728 nir_cf_node_function
1731 typedef struct nir_cf_node
{
1732 struct exec_node node
;
1733 nir_cf_node_type type
;
1734 struct nir_cf_node
*parent
;
1737 typedef struct nir_block
{
1738 nir_cf_node cf_node
;
1740 struct exec_list instr_list
; /** < list of nir_instr */
1742 /** generic block index; generated by nir_index_blocks */
1746 * Each block can only have up to 2 successors, so we put them in a simple
1747 * array - no need for anything more complicated.
1749 struct nir_block
*successors
[2];
1751 /* Set of nir_block predecessors in the CFG */
1752 struct set
*predecessors
;
1755 * this node's immediate dominator in the dominance tree - set to NULL for
1758 struct nir_block
*imm_dom
;
1760 /* This node's children in the dominance tree */
1761 unsigned num_dom_children
;
1762 struct nir_block
**dom_children
;
1764 /* Set of nir_blocks on the dominance frontier of this block */
1765 struct set
*dom_frontier
;
1768 * These two indices have the property that dom_{pre,post}_index for each
1769 * child of this block in the dominance tree will always be between
1770 * dom_pre_index and dom_post_index for this block, which makes testing if
1771 * a given block is dominated by another block an O(1) operation.
1773 unsigned dom_pre_index
, dom_post_index
;
1775 /* live in and out for this block; used for liveness analysis */
1776 BITSET_WORD
*live_in
;
1777 BITSET_WORD
*live_out
;
1780 static inline nir_instr
*
1781 nir_block_first_instr(nir_block
*block
)
1783 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1784 return exec_node_data(nir_instr
, head
, node
);
1787 static inline nir_instr
*
1788 nir_block_last_instr(nir_block
*block
)
1790 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1791 return exec_node_data(nir_instr
, tail
, node
);
1795 nir_block_ends_in_jump(nir_block
*block
)
1797 return !exec_list_is_empty(&block
->instr_list
) &&
1798 nir_block_last_instr(block
)->type
== nir_instr_type_jump
;
1801 #define nir_foreach_instr(instr, block) \
1802 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1803 #define nir_foreach_instr_reverse(instr, block) \
1804 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1805 #define nir_foreach_instr_safe(instr, block) \
1806 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1807 #define nir_foreach_instr_reverse_safe(instr, block) \
1808 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1810 typedef struct nir_if
{
1811 nir_cf_node cf_node
;
1814 struct exec_list then_list
; /** < list of nir_cf_node */
1815 struct exec_list else_list
; /** < list of nir_cf_node */
1821 nir_instr
*conditional_instr
;
1823 nir_block
*break_block
;
1824 nir_block
*continue_from_block
;
1826 bool continue_from_then
;
1828 struct list_head loop_terminator_link
;
1829 } nir_loop_terminator
;
1832 /* Number of instructions in the loop */
1833 unsigned num_instructions
;
1835 /* How many times the loop is run (if known) */
1836 unsigned trip_count
;
1837 bool is_trip_count_known
;
1839 /* Unroll the loop regardless of its size */
1842 /* Does the loop contain complex loop terminators, continues or other
1843 * complex behaviours? If this is true we can't rely on
1844 * loop_terminator_list to be complete or accurate.
1848 nir_loop_terminator
*limiting_terminator
;
1850 /* A list of loop_terminators terminating this loop. */
1851 struct list_head loop_terminator_list
;
1855 nir_cf_node cf_node
;
1857 struct exec_list body
; /** < list of nir_cf_node */
1859 nir_loop_info
*info
;
1863 * Various bits of metadata that can may be created or required by
1864 * optimization and analysis passes
1867 nir_metadata_none
= 0x0,
1868 nir_metadata_block_index
= 0x1,
1869 nir_metadata_dominance
= 0x2,
1870 nir_metadata_live_ssa_defs
= 0x4,
1871 nir_metadata_not_properly_reset
= 0x8,
1872 nir_metadata_loop_analysis
= 0x10,
1876 nir_cf_node cf_node
;
1878 /** pointer to the function of which this is an implementation */
1879 struct nir_function
*function
;
1881 struct exec_list body
; /** < list of nir_cf_node */
1883 nir_block
*end_block
;
1885 /** list for all local variables in the function */
1886 struct exec_list locals
;
1888 /** list of local registers in the function */
1889 struct exec_list registers
;
1891 /** next available local register index */
1894 /** next available SSA value index */
1897 /* total number of basic blocks, only valid when block_index_dirty = false */
1898 unsigned num_blocks
;
1900 nir_metadata valid_metadata
;
1901 } nir_function_impl
;
1903 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1904 nir_start_block(nir_function_impl
*impl
)
1906 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1909 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1910 nir_impl_last_block(nir_function_impl
*impl
)
1912 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1915 static inline nir_cf_node
*
1916 nir_cf_node_next(nir_cf_node
*node
)
1918 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1919 if (exec_node_is_tail_sentinel(next
))
1922 return exec_node_data(nir_cf_node
, next
, node
);
1925 static inline nir_cf_node
*
1926 nir_cf_node_prev(nir_cf_node
*node
)
1928 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1929 if (exec_node_is_head_sentinel(prev
))
1932 return exec_node_data(nir_cf_node
, prev
, node
);
1936 nir_cf_node_is_first(const nir_cf_node
*node
)
1938 return exec_node_is_head_sentinel(node
->node
.prev
);
1942 nir_cf_node_is_last(const nir_cf_node
*node
)
1944 return exec_node_is_tail_sentinel(node
->node
.next
);
1947 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1948 type
, nir_cf_node_block
)
1949 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1950 type
, nir_cf_node_if
)
1951 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1952 type
, nir_cf_node_loop
)
1953 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1954 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1956 static inline nir_block
*
1957 nir_if_first_then_block(nir_if
*if_stmt
)
1959 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1960 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1963 static inline nir_block
*
1964 nir_if_last_then_block(nir_if
*if_stmt
)
1966 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1967 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1970 static inline nir_block
*
1971 nir_if_first_else_block(nir_if
*if_stmt
)
1973 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1974 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1977 static inline nir_block
*
1978 nir_if_last_else_block(nir_if
*if_stmt
)
1980 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1981 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1984 static inline nir_block
*
1985 nir_loop_first_block(nir_loop
*loop
)
1987 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1988 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1991 static inline nir_block
*
1992 nir_loop_last_block(nir_loop
*loop
)
1994 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1995 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1999 uint8_t num_components
;
2003 typedef struct nir_function
{
2004 struct exec_node node
;
2007 struct nir_shader
*shader
;
2009 unsigned num_params
;
2010 nir_parameter
*params
;
2012 /** The implementation of this function.
2014 * If the function is only declared and not implemented, this is NULL.
2016 nir_function_impl
*impl
;
2019 typedef struct nir_shader_compiler_options
{
2024 /** Lowers flrp when it does not support doubles */
2031 /** Lowers ibitfield_extract/ubitfield_extract to ibfe/ubfe. */
2032 bool lower_bitfield_extract
;
2033 /** Lowers ibitfield_extract/ubitfield_extract to bfm, compares, shifts. */
2034 bool lower_bitfield_extract_to_shifts
;
2035 /** Lowers bitfield_insert to bfi/bfm */
2036 bool lower_bitfield_insert
;
2037 /** Lowers bitfield_insert to bfm, compares, and shifts. */
2038 bool lower_bitfield_insert_to_shifts
;
2039 /** Lowers bitfield_reverse to shifts. */
2040 bool lower_bitfield_reverse
;
2041 /** Lowers bit_count to shifts. */
2042 bool lower_bit_count
;
2043 /** Lowers bfm to shifts and subtracts. */
2045 /** Lowers ifind_msb to compare and ufind_msb */
2046 bool lower_ifind_msb
;
2047 /** Lowers find_lsb to ufind_msb and logic ops */
2048 bool lower_find_lsb
;
2049 bool lower_uadd_carry
;
2050 bool lower_usub_borrow
;
2051 /** Lowers imul_high/umul_high to 16-bit multiplies and carry operations. */
2052 bool lower_mul_high
;
2053 /** lowers fneg and ineg to fsub and isub. */
2055 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
2058 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
2061 /** enables rules to lower idiv by power-of-two: */
2064 /* lower b2f to iand */
2067 /* Does the native fdot instruction replicate its result for four
2068 * components? If so, then opt_algebraic_late will turn all fdotN
2069 * instructions into fdot_replicatedN instructions.
2071 bool fdot_replicates
;
2073 /** lowers ffract to fsub+ffloor: */
2078 bool lower_pack_half_2x16
;
2079 bool lower_pack_unorm_2x16
;
2080 bool lower_pack_snorm_2x16
;
2081 bool lower_pack_unorm_4x8
;
2082 bool lower_pack_snorm_4x8
;
2083 bool lower_unpack_half_2x16
;
2084 bool lower_unpack_unorm_2x16
;
2085 bool lower_unpack_snorm_2x16
;
2086 bool lower_unpack_unorm_4x8
;
2087 bool lower_unpack_snorm_4x8
;
2089 bool lower_extract_byte
;
2090 bool lower_extract_word
;
2092 bool lower_all_io_to_temps
;
2095 * Does the driver support real 32-bit integers? (Otherwise, integers
2096 * are simulated by floats.)
2098 bool native_integers
;
2100 /* Indicates that the driver only has zero-based vertex id */
2101 bool vertex_id_zero_based
;
2104 * If enabled, gl_BaseVertex will be lowered as:
2105 * is_indexed_draw (~0/0) & firstvertex
2107 bool lower_base_vertex
;
2110 * If enabled, gl_HelperInvocation will be lowered as:
2112 * !((1 << sample_id) & sample_mask_in))
2114 * This depends on some possibly hw implementation details, which may
2115 * not be true for all hw. In particular that the FS is only executed
2116 * for covered samples or for helper invocations. So, do not blindly
2117 * enable this option.
2119 * Note: See also issue #22 in ARB_shader_image_load_store
2121 bool lower_helper_invocation
;
2123 bool lower_cs_local_index_from_id
;
2125 bool lower_device_index_to_zero
;
2127 /* Set if nir_lower_wpos_ytransform() should also invert gl_PointCoord. */
2128 bool lower_wpos_pntc
;
2131 * Should nir_lower_io() create load_interpolated_input intrinsics?
2133 * If not, it generates regular load_input intrinsics and interpolation
2134 * information must be inferred from the list of input nir_variables.
2136 bool use_interpolated_input_intrinsics
;
2138 unsigned max_unroll_iterations
;
2139 } nir_shader_compiler_options
;
2141 typedef struct nir_shader
{
2142 /** list of uniforms (nir_variable) */
2143 struct exec_list uniforms
;
2145 /** list of inputs (nir_variable) */
2146 struct exec_list inputs
;
2148 /** list of outputs (nir_variable) */
2149 struct exec_list outputs
;
2151 /** list of shared compute variables (nir_variable) */
2152 struct exec_list shared
;
2154 /** Set of driver-specific options for the shader.
2156 * The memory for the options is expected to be kept in a single static
2157 * copy by the driver.
2159 const struct nir_shader_compiler_options
*options
;
2161 /** Various bits of compile-time information about a given shader */
2162 struct shader_info info
;
2164 /** list of global variables in the shader (nir_variable) */
2165 struct exec_list globals
;
2167 /** list of system value variables in the shader (nir_variable) */
2168 struct exec_list system_values
;
2170 struct exec_list functions
; /** < list of nir_function */
2172 /** list of global register in the shader */
2173 struct exec_list registers
;
2175 /** next available global register index */
2179 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
2182 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
2184 /** Constant data associated with this shader.
2186 * Constant data is loaded through load_constant intrinsics. See also
2187 * nir_opt_large_constants.
2189 void *constant_data
;
2190 unsigned constant_data_size
;
2193 static inline nir_function_impl
*
2194 nir_shader_get_entrypoint(nir_shader
*shader
)
2196 assert(exec_list_length(&shader
->functions
) == 1);
2197 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
2198 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
2199 assert(func
->num_params
== 0);
2204 #define nir_foreach_function(func, shader) \
2205 foreach_list_typed(nir_function, func, node, &(shader)->functions)
2207 nir_shader
*nir_shader_create(void *mem_ctx
,
2208 gl_shader_stage stage
,
2209 const nir_shader_compiler_options
*options
,
2212 /** creates a register, including assigning it an index and adding it to the list */
2213 nir_register
*nir_global_reg_create(nir_shader
*shader
);
2215 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
2217 void nir_reg_remove(nir_register
*reg
);
2219 /** Adds a variable to the appropriate list in nir_shader */
2220 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
2223 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
2225 assert(var
->data
.mode
== nir_var_local
);
2226 exec_list_push_tail(&impl
->locals
, &var
->node
);
2229 /** creates a variable, sets a few defaults, and adds it to the list */
2230 nir_variable
*nir_variable_create(nir_shader
*shader
,
2231 nir_variable_mode mode
,
2232 const struct glsl_type
*type
,
2234 /** creates a local variable and adds it to the list */
2235 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
2236 const struct glsl_type
*type
,
2239 /** creates a function and adds it to the shader's list of functions */
2240 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
2242 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
2243 /** creates a function_impl that isn't tied to any particular function */
2244 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
2246 nir_block
*nir_block_create(nir_shader
*shader
);
2247 nir_if
*nir_if_create(nir_shader
*shader
);
2248 nir_loop
*nir_loop_create(nir_shader
*shader
);
2250 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
2252 /** requests that the given pieces of metadata be generated */
2253 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
2254 /** dirties all but the preserved metadata */
2255 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
2257 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
2258 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
2260 nir_deref_instr
*nir_deref_instr_create(nir_shader
*shader
,
2261 nir_deref_type deref_type
);
2263 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
2265 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
2266 unsigned num_components
,
2269 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
2270 nir_intrinsic_op op
);
2272 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
2273 nir_function
*callee
);
2275 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
2277 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
2279 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
2281 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
2282 unsigned num_components
,
2285 nir_const_value
nir_alu_binop_identity(nir_op binop
, unsigned bit_size
);
2288 * NIR Cursors and Instruction Insertion API
2291 * A tiny struct representing a point to insert/extract instructions or
2292 * control flow nodes. Helps reduce the combinatorial explosion of possible
2293 * points to insert/extract.
2295 * \sa nir_control_flow.h
2298 nir_cursor_before_block
,
2299 nir_cursor_after_block
,
2300 nir_cursor_before_instr
,
2301 nir_cursor_after_instr
,
2302 } nir_cursor_option
;
2305 nir_cursor_option option
;
2312 static inline nir_block
*
2313 nir_cursor_current_block(nir_cursor cursor
)
2315 if (cursor
.option
== nir_cursor_before_instr
||
2316 cursor
.option
== nir_cursor_after_instr
) {
2317 return cursor
.instr
->block
;
2319 return cursor
.block
;
2323 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2325 static inline nir_cursor
2326 nir_before_block(nir_block
*block
)
2329 cursor
.option
= nir_cursor_before_block
;
2330 cursor
.block
= block
;
2334 static inline nir_cursor
2335 nir_after_block(nir_block
*block
)
2338 cursor
.option
= nir_cursor_after_block
;
2339 cursor
.block
= block
;
2343 static inline nir_cursor
2344 nir_before_instr(nir_instr
*instr
)
2347 cursor
.option
= nir_cursor_before_instr
;
2348 cursor
.instr
= instr
;
2352 static inline nir_cursor
2353 nir_after_instr(nir_instr
*instr
)
2356 cursor
.option
= nir_cursor_after_instr
;
2357 cursor
.instr
= instr
;
2361 static inline nir_cursor
2362 nir_after_block_before_jump(nir_block
*block
)
2364 nir_instr
*last_instr
= nir_block_last_instr(block
);
2365 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2366 return nir_before_instr(last_instr
);
2368 return nir_after_block(block
);
2372 static inline nir_cursor
2373 nir_before_src(nir_src
*src
, bool is_if_condition
)
2375 if (is_if_condition
) {
2376 nir_block
*prev_block
=
2377 nir_cf_node_as_block(nir_cf_node_prev(&src
->parent_if
->cf_node
));
2378 assert(!nir_block_ends_in_jump(prev_block
));
2379 return nir_after_block(prev_block
);
2380 } else if (src
->parent_instr
->type
== nir_instr_type_phi
) {
2382 nir_phi_instr
*cond_phi
= nir_instr_as_phi(src
->parent_instr
);
2384 nir_foreach_phi_src(phi_src
, cond_phi
) {
2385 if (phi_src
->src
.ssa
== src
->ssa
) {
2392 /* The LIST_ENTRY macro is a generic container-of macro, it just happens
2393 * to have a more specific name.
2395 nir_phi_src
*phi_src
= LIST_ENTRY(nir_phi_src
, src
, src
);
2396 return nir_after_block_before_jump(phi_src
->pred
);
2398 return nir_before_instr(src
->parent_instr
);
2402 static inline nir_cursor
2403 nir_before_cf_node(nir_cf_node
*node
)
2405 if (node
->type
== nir_cf_node_block
)
2406 return nir_before_block(nir_cf_node_as_block(node
));
2408 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2411 static inline nir_cursor
2412 nir_after_cf_node(nir_cf_node
*node
)
2414 if (node
->type
== nir_cf_node_block
)
2415 return nir_after_block(nir_cf_node_as_block(node
));
2417 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2420 static inline nir_cursor
2421 nir_after_phis(nir_block
*block
)
2423 nir_foreach_instr(instr
, block
) {
2424 if (instr
->type
!= nir_instr_type_phi
)
2425 return nir_before_instr(instr
);
2427 return nir_after_block(block
);
2430 static inline nir_cursor
2431 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2433 if (node
->type
== nir_cf_node_block
)
2434 return nir_after_block(nir_cf_node_as_block(node
));
2436 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2438 return nir_after_phis(block
);
2441 static inline nir_cursor
2442 nir_before_cf_list(struct exec_list
*cf_list
)
2444 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2445 exec_list_get_head(cf_list
), node
);
2446 return nir_before_cf_node(first_node
);
2449 static inline nir_cursor
2450 nir_after_cf_list(struct exec_list
*cf_list
)
2452 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2453 exec_list_get_tail(cf_list
), node
);
2454 return nir_after_cf_node(last_node
);
2458 * Insert a NIR instruction at the given cursor.
2460 * Note: This does not update the cursor.
2462 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2465 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2467 nir_instr_insert(nir_before_instr(instr
), before
);
2471 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2473 nir_instr_insert(nir_after_instr(instr
), after
);
2477 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2479 nir_instr_insert(nir_before_block(block
), before
);
2483 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2485 nir_instr_insert(nir_after_block(block
), after
);
2489 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2491 nir_instr_insert(nir_before_cf_node(node
), before
);
2495 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2497 nir_instr_insert(nir_after_cf_node(node
), after
);
2501 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2503 nir_instr_insert(nir_before_cf_list(list
), before
);
2507 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2509 nir_instr_insert(nir_after_cf_list(list
), after
);
2512 void nir_instr_remove_v(nir_instr
*instr
);
2514 static inline nir_cursor
2515 nir_instr_remove(nir_instr
*instr
)
2518 nir_instr
*prev
= nir_instr_prev(instr
);
2520 cursor
= nir_after_instr(prev
);
2522 cursor
= nir_before_block(instr
->block
);
2524 nir_instr_remove_v(instr
);
2530 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2531 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2532 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2533 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2535 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2536 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2538 nir_const_value
*nir_src_as_const_value(nir_src src
);
2540 static inline struct nir_instr
*
2541 nir_src_instr(const struct nir_src
*src
)
2543 return src
->is_ssa
? src
->ssa
->parent_instr
: NULL
;
2546 #define NIR_SRC_AS_(name, c_type, type_enum, cast_macro) \
2547 static inline c_type * \
2548 nir_src_as_ ## name (struct nir_src *src) \
2550 return src->is_ssa && src->ssa->parent_instr->type == type_enum \
2551 ? cast_macro(src->ssa->parent_instr) : NULL; \
2553 static inline const c_type * \
2554 nir_src_as_ ## name ## _const(const struct nir_src *src) \
2556 return src->is_ssa && src->ssa->parent_instr->type == type_enum \
2557 ? cast_macro(src->ssa->parent_instr) : NULL; \
2560 NIR_SRC_AS_(alu_instr
, nir_alu_instr
, nir_instr_type_alu
, nir_instr_as_alu
)
2562 bool nir_src_is_dynamically_uniform(nir_src src
);
2563 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2564 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2565 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2566 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2567 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2570 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2571 unsigned num_components
, unsigned bit_size
,
2573 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2574 unsigned num_components
, unsigned bit_size
,
2577 nir_ssa_dest_init_for_type(nir_instr
*instr
, nir_dest
*dest
,
2578 const struct glsl_type
*type
,
2581 assert(glsl_type_is_vector_or_scalar(type
));
2582 nir_ssa_dest_init(instr
, dest
, glsl_get_components(type
),
2583 glsl_get_bit_size(type
), name
);
2585 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2586 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2587 nir_instr
*after_me
);
2589 nir_component_mask_t
nir_ssa_def_components_read(const nir_ssa_def
*def
);
2592 * finds the next basic block in source-code order, returns NULL if there is
2596 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2598 /* Performs the opposite of nir_block_cf_tree_next() */
2600 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2602 /* Gets the first block in a CF node in source-code order */
2604 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2606 /* Gets the last block in a CF node in source-code order */
2608 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2610 /* Gets the next block after a CF node in source-code order */
2612 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2614 /* Macros for loops that visit blocks in source-code order */
2616 #define nir_foreach_block(block, impl) \
2617 for (nir_block *block = nir_start_block(impl); block != NULL; \
2618 block = nir_block_cf_tree_next(block))
2620 #define nir_foreach_block_safe(block, impl) \
2621 for (nir_block *block = nir_start_block(impl), \
2622 *next = nir_block_cf_tree_next(block); \
2624 block = next, next = nir_block_cf_tree_next(block))
2626 #define nir_foreach_block_reverse(block, impl) \
2627 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2628 block = nir_block_cf_tree_prev(block))
2630 #define nir_foreach_block_reverse_safe(block, impl) \
2631 for (nir_block *block = nir_impl_last_block(impl), \
2632 *prev = nir_block_cf_tree_prev(block); \
2634 block = prev, prev = nir_block_cf_tree_prev(block))
2636 #define nir_foreach_block_in_cf_node(block, node) \
2637 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2638 block != nir_cf_node_cf_tree_next(node); \
2639 block = nir_block_cf_tree_next(block))
2641 /* If the following CF node is an if, this function returns that if.
2642 * Otherwise, it returns NULL.
2644 nir_if
*nir_block_get_following_if(nir_block
*block
);
2646 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2648 void nir_index_local_regs(nir_function_impl
*impl
);
2649 void nir_index_global_regs(nir_shader
*shader
);
2650 void nir_index_ssa_defs(nir_function_impl
*impl
);
2651 unsigned nir_index_instrs(nir_function_impl
*impl
);
2653 void nir_index_blocks(nir_function_impl
*impl
);
2655 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2656 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2657 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2659 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2660 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2661 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2662 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2664 nir_shader
*nir_shader_serialize_deserialize(void *mem_ctx
, nir_shader
*s
);
2667 void nir_validate_shader(nir_shader
*shader
);
2668 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2669 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2672 should_clone_nir(void)
2674 static int should_clone
= -1;
2675 if (should_clone
< 0)
2676 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2678 return should_clone
;
2682 should_serialize_deserialize_nir(void)
2684 static int test_serialize
= -1;
2685 if (test_serialize
< 0)
2686 test_serialize
= env_var_as_boolean("NIR_TEST_SERIALIZE", false);
2688 return test_serialize
;
2692 should_print_nir(void)
2694 static int should_print
= -1;
2695 if (should_print
< 0)
2696 should_print
= env_var_as_boolean("NIR_PRINT", false);
2698 return should_print
;
2701 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2702 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2703 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2704 static inline bool should_clone_nir(void) { return false; }
2705 static inline bool should_serialize_deserialize_nir(void) { return false; }
2706 static inline bool should_print_nir(void) { return false; }
2709 #define _PASS(nir, do_pass) do { \
2711 nir_validate_shader(nir); \
2712 if (should_clone_nir()) { \
2713 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2717 if (should_serialize_deserialize_nir()) { \
2718 void *mem_ctx = ralloc_parent(nir); \
2719 nir = nir_shader_serialize_deserialize(mem_ctx, nir); \
2723 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2724 nir_metadata_set_validation_flag(nir); \
2725 if (should_print_nir()) \
2726 printf("%s\n", #pass); \
2727 if (pass(nir, ##__VA_ARGS__)) { \
2729 if (should_print_nir()) \
2730 nir_print_shader(nir, stdout); \
2731 nir_metadata_check_validation_flag(nir); \
2735 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2736 if (should_print_nir()) \
2737 printf("%s\n", #pass); \
2738 pass(nir, ##__VA_ARGS__); \
2739 if (should_print_nir()) \
2740 nir_print_shader(nir, stdout); \
2743 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2744 void nir_calc_dominance(nir_shader
*shader
);
2746 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2747 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2749 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2750 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2752 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2753 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2755 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2756 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2758 int nir_gs_count_vertices(const nir_shader
*shader
);
2760 bool nir_shrink_vec_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2761 bool nir_split_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2762 bool nir_split_var_copies(nir_shader
*shader
);
2763 bool nir_split_per_member_structs(nir_shader
*shader
);
2764 bool nir_split_struct_vars(nir_shader
*shader
, nir_variable_mode modes
);
2766 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2767 bool nir_lower_returns(nir_shader
*shader
);
2769 bool nir_inline_functions(nir_shader
*shader
);
2771 bool nir_propagate_invariant(nir_shader
*shader
);
2773 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2774 void nir_lower_deref_copy_instr(struct nir_builder
*b
,
2775 nir_intrinsic_instr
*copy
);
2776 bool nir_lower_var_copies(nir_shader
*shader
);
2778 void nir_fixup_deref_modes(nir_shader
*shader
);
2780 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2782 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2784 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2786 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2787 nir_function_impl
*entrypoint
,
2788 bool outputs
, bool inputs
);
2790 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2792 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2793 int (*type_size
)(const struct glsl_type
*));
2795 /* Some helpers to do very simple linking */
2796 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2797 bool nir_remove_unused_io_vars(nir_shader
*shader
, struct exec_list
*var_list
,
2798 uint64_t *used_by_other_stage
,
2799 uint64_t *used_by_other_stage_patches
);
2800 void nir_compact_varyings(nir_shader
*producer
, nir_shader
*consumer
,
2801 bool default_to_smooth_interp
);
2804 /* If set, this forces all non-flat fragment shader inputs to be
2805 * interpolated as if with the "sample" qualifier. This requires
2806 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2808 nir_lower_io_force_sample_interpolation
= (1 << 1),
2809 } nir_lower_io_options
;
2810 bool nir_lower_io(nir_shader
*shader
,
2811 nir_variable_mode modes
,
2812 int (*type_size
)(const struct glsl_type
*),
2813 nir_lower_io_options
);
2814 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2815 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2817 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2819 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2820 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2821 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2823 bool nir_remove_dead_derefs(nir_shader
*shader
);
2824 bool nir_remove_dead_derefs_impl(nir_function_impl
*impl
);
2825 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2826 bool nir_lower_constant_initializers(nir_shader
*shader
,
2827 nir_variable_mode modes
);
2829 bool nir_move_load_const(nir_shader
*shader
);
2830 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2831 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2832 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
2834 bool nir_lower_alu(nir_shader
*shader
);
2835 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2836 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2837 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
2838 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2839 void nir_lower_io_arrays_to_elements(nir_shader
*producer
, nir_shader
*consumer
);
2840 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader
*shader
,
2842 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2843 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
2845 typedef struct nir_lower_subgroups_options
{
2846 uint8_t subgroup_size
;
2847 uint8_t ballot_bit_size
;
2848 bool lower_to_scalar
:1;
2849 bool lower_vote_trivial
:1;
2850 bool lower_vote_eq_to_ballot
:1;
2851 bool lower_subgroup_masks
:1;
2852 bool lower_shuffle
:1;
2853 bool lower_shuffle_to_32bit
:1;
2855 } nir_lower_subgroups_options
;
2857 bool nir_lower_subgroups(nir_shader
*shader
,
2858 const nir_lower_subgroups_options
*options
);
2860 bool nir_lower_system_values(nir_shader
*shader
);
2862 typedef struct nir_lower_tex_options
{
2864 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2865 * sampler types a texture projector is lowered.
2870 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2872 bool lower_txf_offset
;
2875 * If true, lower away nir_tex_src_offset for all rect textures.
2877 bool lower_rect_offset
;
2880 * If true, lower rect textures to 2D, using txs to fetch the
2881 * texture dimensions and dividing the texture coords by the
2882 * texture dims to normalize.
2887 * If true, convert yuv to rgb.
2889 unsigned lower_y_uv_external
;
2890 unsigned lower_y_u_v_external
;
2891 unsigned lower_yx_xuxv_external
;
2892 unsigned lower_xy_uxvx_external
;
2895 * To emulate certain texture wrap modes, this can be used
2896 * to saturate the specified tex coord to [0.0, 1.0]. The
2897 * bits are according to sampler #, ie. if, for example:
2899 * (conf->saturate_s & (1 << n))
2901 * is true, then the s coord for sampler n is saturated.
2903 * Note that clamping must happen *after* projector lowering
2904 * so any projected texture sample instruction with a clamped
2905 * coordinate gets automatically lowered, regardless of the
2906 * 'lower_txp' setting.
2908 unsigned saturate_s
;
2909 unsigned saturate_t
;
2910 unsigned saturate_r
;
2912 /* Bitmask of textures that need swizzling.
2914 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2915 * swizzles[texture_index] is applied to the result of the texturing
2918 unsigned swizzle_result
;
2920 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2921 * while 4 and 5 represent 0 and 1 respectively.
2923 uint8_t swizzles
[32][4];
2926 * Bitmap of textures that need srgb to linear conversion. If
2927 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2928 * of the texture are lowered to linear.
2930 unsigned lower_srgb
;
2933 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2935 bool lower_txd_cube_map
;
2938 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
2939 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
2940 * with lower_txd_cube_map.
2942 bool lower_txd_shadow
;
2945 * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
2946 * Implies lower_txd_cube_map and lower_txd_shadow.
2949 } nir_lower_tex_options
;
2951 bool nir_lower_tex(nir_shader
*shader
,
2952 const nir_lower_tex_options
*options
);
2954 bool nir_lower_idiv(nir_shader
*shader
);
2956 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2957 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2958 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2960 void nir_lower_two_sided_color(nir_shader
*shader
);
2962 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
2964 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2965 bool nir_lower_patch_vertices(nir_shader
*nir
, unsigned static_count
,
2966 const gl_state_index16
*uniform_state_tokens
);
2968 typedef struct nir_lower_wpos_ytransform_options
{
2969 gl_state_index16 state_tokens
[STATE_LENGTH
];
2970 bool fs_coord_origin_upper_left
:1;
2971 bool fs_coord_origin_lower_left
:1;
2972 bool fs_coord_pixel_center_integer
:1;
2973 bool fs_coord_pixel_center_half_integer
:1;
2974 } nir_lower_wpos_ytransform_options
;
2976 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2977 const nir_lower_wpos_ytransform_options
*options
);
2978 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
2980 typedef struct nir_lower_drawpixels_options
{
2981 gl_state_index16 texcoord_state_tokens
[STATE_LENGTH
];
2982 gl_state_index16 scale_state_tokens
[STATE_LENGTH
];
2983 gl_state_index16 bias_state_tokens
[STATE_LENGTH
];
2984 unsigned drawpix_sampler
;
2985 unsigned pixelmap_sampler
;
2987 bool scale_and_bias
:1;
2988 } nir_lower_drawpixels_options
;
2990 void nir_lower_drawpixels(nir_shader
*shader
,
2991 const nir_lower_drawpixels_options
*options
);
2993 typedef struct nir_lower_bitmap_options
{
2996 } nir_lower_bitmap_options
;
2998 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
3000 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
3001 bool nir_lower_to_source_mods(nir_shader
*shader
);
3003 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
3005 typedef unsigned (*nir_lower_bit_size_callback
)(const nir_alu_instr
*, void *);
3007 bool nir_lower_bit_size(nir_shader
*shader
,
3008 nir_lower_bit_size_callback callback
,
3009 void *callback_data
);
3012 nir_lower_imul64
= (1 << 0),
3013 nir_lower_isign64
= (1 << 1),
3014 /** Lower all int64 modulus and division opcodes */
3015 nir_lower_divmod64
= (1 << 2),
3016 } nir_lower_int64_options
;
3018 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
3021 nir_lower_drcp
= (1 << 0),
3022 nir_lower_dsqrt
= (1 << 1),
3023 nir_lower_drsq
= (1 << 2),
3024 nir_lower_dtrunc
= (1 << 3),
3025 nir_lower_dfloor
= (1 << 4),
3026 nir_lower_dceil
= (1 << 5),
3027 nir_lower_dfract
= (1 << 6),
3028 nir_lower_dround_even
= (1 << 7),
3029 nir_lower_dmod
= (1 << 8)
3030 } nir_lower_doubles_options
;
3032 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
3033 bool nir_lower_pack(nir_shader
*shader
);
3035 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
3037 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
3039 void nir_loop_analyze_impl(nir_function_impl
*impl
,
3040 nir_variable_mode indirect_mask
);
3042 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
3044 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
3045 bool nir_repair_ssa(nir_shader
*shader
);
3047 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
3049 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
3050 * registers. If false, convert all values (even those not involved in a phi
3051 * node) to registers.
3053 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
3055 bool nir_lower_phis_to_regs_block(nir_block
*block
);
3056 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
3057 bool nir_rematerialize_derefs_in_use_blocks_impl(nir_function_impl
*impl
);
3059 bool nir_opt_algebraic(nir_shader
*shader
);
3060 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
3061 bool nir_opt_algebraic_late(nir_shader
*shader
);
3062 bool nir_opt_constant_folding(nir_shader
*shader
);
3064 bool nir_opt_global_to_local(nir_shader
*shader
);
3066 bool nir_copy_prop(nir_shader
*shader
);
3068 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
3070 bool nir_opt_cse(nir_shader
*shader
);
3072 bool nir_opt_dce(nir_shader
*shader
);
3074 bool nir_opt_dead_cf(nir_shader
*shader
);
3076 bool nir_opt_dead_write_vars(nir_shader
*shader
);
3078 bool nir_opt_find_array_copies(nir_shader
*shader
);
3080 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
3082 bool nir_opt_if(nir_shader
*shader
);
3084 bool nir_opt_intrinsics(nir_shader
*shader
);
3086 bool nir_opt_large_constants(nir_shader
*shader
,
3087 glsl_type_size_align_func size_align
,
3088 unsigned threshold
);
3090 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
3092 bool nir_opt_move_comparisons(nir_shader
*shader
);
3094 bool nir_opt_move_load_ubo(nir_shader
*shader
);
3096 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
3098 bool nir_opt_remove_phis_impl(nir_function_impl
*impl
);
3099 bool nir_opt_remove_phis(nir_shader
*shader
);
3101 bool nir_opt_shrink_load(nir_shader
*shader
);
3103 bool nir_opt_trivial_continues(nir_shader
*shader
);
3105 bool nir_opt_undef(nir_shader
*shader
);
3107 bool nir_opt_conditional_discard(nir_shader
*shader
);
3109 void nir_sweep(nir_shader
*shader
);
3111 void nir_remap_dual_slot_attributes(nir_shader
*shader
,
3112 uint64_t *dual_slot_inputs
);
3113 uint64_t nir_get_single_slot_attribs_mask(uint64_t attribs
, uint64_t dual_slot
);
3115 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
3116 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);