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
;
703 static inline unsigned
704 nir_dest_bit_size(nir_dest dest
)
706 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
709 static inline unsigned
710 nir_dest_num_components(nir_dest dest
)
712 return dest
.is_ssa
? dest
.ssa
.num_components
: dest
.reg
.reg
->num_components
;
715 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
716 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
722 * \name input modifiers
726 * For inputs interpreted as floating point, flips the sign bit. For
727 * inputs interpreted as integers, performs the two's complement negation.
732 * Clears the sign bit for floating point values, and computes the integer
733 * absolute value for integers. Note that the negate modifier acts after
734 * the absolute value modifier, therefore if both are set then all inputs
735 * will become negative.
741 * For each input component, says which component of the register it is
742 * chosen from. Note that which elements of the swizzle are used and which
743 * are ignored are based on the write mask for most opcodes - for example,
744 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
745 * a swizzle of {2, x, 1, 0} where x means "don't care."
747 uint8_t swizzle
[NIR_MAX_VEC_COMPONENTS
];
754 * \name saturate output modifier
756 * Only valid for opcodes that output floating-point numbers. Clamps the
757 * output to between 0.0 and 1.0 inclusive.
762 unsigned write_mask
: NIR_MAX_VEC_COMPONENTS
; /* ignored if dest.is_ssa is true */
766 nir_type_invalid
= 0, /* Not a valid type */
771 nir_type_bool32
= 32 | nir_type_bool
,
772 nir_type_int8
= 8 | nir_type_int
,
773 nir_type_int16
= 16 | nir_type_int
,
774 nir_type_int32
= 32 | nir_type_int
,
775 nir_type_int64
= 64 | nir_type_int
,
776 nir_type_uint8
= 8 | nir_type_uint
,
777 nir_type_uint16
= 16 | nir_type_uint
,
778 nir_type_uint32
= 32 | nir_type_uint
,
779 nir_type_uint64
= 64 | nir_type_uint
,
780 nir_type_float16
= 16 | nir_type_float
,
781 nir_type_float32
= 32 | nir_type_float
,
782 nir_type_float64
= 64 | nir_type_float
,
785 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
786 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
788 static inline unsigned
789 nir_alu_type_get_type_size(nir_alu_type type
)
791 return type
& NIR_ALU_TYPE_SIZE_MASK
;
794 static inline unsigned
795 nir_alu_type_get_base_type(nir_alu_type type
)
797 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
800 static inline nir_alu_type
801 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
805 return nir_type_bool32
;
808 return nir_type_uint32
;
811 return nir_type_int32
;
813 case GLSL_TYPE_UINT16
:
814 return nir_type_uint16
;
816 case GLSL_TYPE_INT16
:
817 return nir_type_int16
;
819 case GLSL_TYPE_UINT8
:
820 return nir_type_uint8
;
822 return nir_type_int8
;
823 case GLSL_TYPE_UINT64
:
824 return nir_type_uint64
;
826 case GLSL_TYPE_INT64
:
827 return nir_type_int64
;
829 case GLSL_TYPE_FLOAT
:
830 return nir_type_float32
;
832 case GLSL_TYPE_FLOAT16
:
833 return nir_type_float16
;
835 case GLSL_TYPE_DOUBLE
:
836 return nir_type_float64
;
839 unreachable("unknown type");
843 static inline nir_alu_type
844 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
846 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
849 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
,
850 nir_rounding_mode rnd
);
853 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
854 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
855 } nir_op_algebraic_property
;
863 * The number of components in the output
865 * If non-zero, this is the size of the output and input sizes are
866 * explicitly given; swizzle and writemask are still in effect, but if
867 * the output component is masked out, then the input component may
870 * If zero, the opcode acts in the standard, per-component manner; the
871 * operation is performed on each component (except the ones that are
872 * masked out) with the input being taken from the input swizzle for
875 * The size of some of the inputs may be given (i.e. non-zero) even
876 * though output_size is zero; in that case, the inputs with a zero
877 * size act per-component, while the inputs with non-zero size don't.
879 unsigned output_size
;
882 * The type of vector that the instruction outputs. Note that the
883 * staurate modifier is only allowed on outputs with the float type.
886 nir_alu_type output_type
;
889 * The number of components in each input
891 unsigned input_sizes
[NIR_MAX_VEC_COMPONENTS
];
894 * The type of vector that each input takes. Note that negate and
895 * absolute value are only allowed on inputs with int or float type and
896 * behave differently on the two.
898 nir_alu_type input_types
[NIR_MAX_VEC_COMPONENTS
];
900 nir_op_algebraic_property algebraic_properties
;
903 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
905 typedef struct nir_alu_instr
{
909 /** Indicates that this ALU instruction generates an exact value
911 * This is kind of a mixture of GLSL "precise" and "invariant" and not
912 * really equivalent to either. This indicates that the value generated by
913 * this operation is high-precision and any code transformations that touch
914 * it must ensure that the resulting value is bit-for-bit identical to the
923 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
924 nir_alu_instr
*instr
);
925 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
926 nir_alu_instr
*instr
);
928 /* is this source channel used? */
930 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
933 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
934 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
936 return (instr
->dest
.write_mask
>> channel
) & 1;
940 * For instructions whose destinations are SSA, get the number of channels
943 static inline unsigned
944 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
946 assert(instr
->dest
.dest
.is_ssa
);
948 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
949 return nir_op_infos
[instr
->op
].input_sizes
[src
];
951 return instr
->dest
.dest
.ssa
.num_components
;
954 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
955 unsigned src1
, unsigned src2
);
959 nir_deref_type_array
,
960 nir_deref_type_array_wildcard
,
961 nir_deref_type_struct
,
968 /** The type of this deref instruction */
969 nir_deref_type deref_type
;
971 /** The mode of the underlying variable */
972 nir_variable_mode mode
;
974 /** The dereferenced type of the resulting pointer value */
975 const struct glsl_type
*type
;
978 /** Variable being dereferenced if deref_type is a deref_var */
981 /** Parent deref if deref_type is not deref_var */
985 /** Additional deref parameters */
996 /** Destination to store the resulting "pointer" */
1000 NIR_DEFINE_CAST(nir_instr_as_deref
, nir_instr
, nir_deref_instr
, instr
,
1001 type
, nir_instr_type_deref
)
1003 static inline nir_deref_instr
*
1004 nir_src_as_deref(nir_src src
)
1009 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_deref
)
1012 return nir_instr_as_deref(src
.ssa
->parent_instr
);
1015 static inline nir_deref_instr
*
1016 nir_deref_instr_parent(const nir_deref_instr
*instr
)
1018 if (instr
->deref_type
== nir_deref_type_var
)
1021 return nir_src_as_deref(instr
->parent
);
1024 static inline nir_variable
*
1025 nir_deref_instr_get_variable(const nir_deref_instr
*instr
)
1027 while (instr
->deref_type
!= nir_deref_type_var
) {
1028 if (instr
->deref_type
== nir_deref_type_cast
)
1031 instr
= nir_deref_instr_parent(instr
);
1037 bool nir_deref_instr_has_indirect(nir_deref_instr
*instr
);
1039 bool nir_deref_instr_remove_if_unused(nir_deref_instr
*instr
);
1044 struct nir_function
*callee
;
1046 unsigned num_params
;
1050 #include "nir_intrinsics.h"
1052 #define NIR_INTRINSIC_MAX_CONST_INDEX 4
1054 /** Represents an intrinsic
1056 * An intrinsic is an instruction type for handling things that are
1057 * more-or-less regular operations but don't just consume and produce SSA
1058 * values like ALU operations do. Intrinsics are not for things that have
1059 * special semantic meaning such as phi nodes and parallel copies.
1060 * Examples of intrinsics include variable load/store operations, system
1061 * value loads, and the like. Even though texturing more-or-less falls
1062 * under this category, texturing is its own instruction type because
1063 * trying to represent texturing with intrinsics would lead to a
1064 * combinatorial explosion of intrinsic opcodes.
1066 * By having a single instruction type for handling a lot of different
1067 * cases, optimization passes can look for intrinsics and, for the most
1068 * part, completely ignore them. Each intrinsic type also has a few
1069 * possible flags that govern whether or not they can be reordered or
1070 * eliminated. That way passes like dead code elimination can still work
1071 * on intrisics without understanding the meaning of each.
1073 * Each intrinsic has some number of constant indices, some number of
1074 * variables, and some number of sources. What these sources, variables,
1075 * and indices mean depends on the intrinsic and is documented with the
1076 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
1077 * instructions are the only types of instruction that can operate on
1083 nir_intrinsic_op intrinsic
;
1087 /** number of components if this is a vectorized intrinsic
1089 * Similarly to ALU operations, some intrinsics are vectorized.
1090 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
1091 * For vectorized intrinsics, the num_components field specifies the
1092 * number of destination components and the number of source components
1093 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
1095 uint8_t num_components
;
1097 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
1100 } nir_intrinsic_instr
;
1102 static inline nir_variable
*
1103 nir_intrinsic_get_var(nir_intrinsic_instr
*intrin
, unsigned i
)
1105 return nir_deref_instr_get_variable(nir_src_as_deref(intrin
->src
[i
]));
1109 * \name NIR intrinsics semantic flags
1111 * information about what the compiler can do with the intrinsics.
1113 * \sa nir_intrinsic_info::flags
1117 * whether the intrinsic can be safely eliminated if none of its output
1118 * value is not being used.
1120 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
1123 * Whether the intrinsic can be reordered with respect to any other
1124 * intrinsic, i.e. whether the only reordering dependencies of the
1125 * intrinsic are due to the register reads/writes.
1127 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
1128 } nir_intrinsic_semantic_flag
;
1131 * \name NIR intrinsics const-index flag
1133 * Indicates the usage of a const_index slot.
1135 * \sa nir_intrinsic_info::index_map
1139 * Generally instructions that take a offset src argument, can encode
1140 * a constant 'base' value which is added to the offset.
1142 NIR_INTRINSIC_BASE
= 1,
1145 * For store instructions, a writemask for the store.
1147 NIR_INTRINSIC_WRMASK
= 2,
1150 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1152 NIR_INTRINSIC_STREAM_ID
= 3,
1155 * The clip-plane id for load_user_clip_plane intrinsic.
1157 NIR_INTRINSIC_UCP_ID
= 4,
1160 * The amount of data, starting from BASE, that this instruction may
1161 * access. This is used to provide bounds if the offset is not constant.
1163 NIR_INTRINSIC_RANGE
= 5,
1166 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1168 NIR_INTRINSIC_DESC_SET
= 6,
1171 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1173 NIR_INTRINSIC_BINDING
= 7,
1178 NIR_INTRINSIC_COMPONENT
= 8,
1181 * Interpolation mode (only meaningful for FS inputs).
1183 NIR_INTRINSIC_INTERP_MODE
= 9,
1186 * A binary nir_op to use when performing a reduction or scan operation
1188 NIR_INTRINSIC_REDUCTION_OP
= 10,
1191 * Cluster size for reduction operations
1193 NIR_INTRINSIC_CLUSTER_SIZE
= 11,
1196 * Parameter index for a load_param intrinsic
1198 NIR_INTRINSIC_PARAM_IDX
= 12,
1201 * Image dimensionality for image intrinsics
1203 * One of GLSL_SAMPLER_DIM_*
1205 NIR_INTRINSIC_IMAGE_DIM
= 13,
1208 * Non-zero if we are accessing an array image
1210 NIR_INTRINSIC_IMAGE_ARRAY
= 14,
1213 * Image format for image intrinsics
1215 NIR_INTRINSIC_FORMAT
= 15,
1218 * Access qualifiers for image intrinsics
1220 NIR_INTRINSIC_ACCESS
= 16,
1222 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1224 } nir_intrinsic_index_flag
;
1226 #define NIR_INTRINSIC_MAX_INPUTS 5
1231 unsigned num_srcs
; /** < number of register/SSA inputs */
1233 /** number of components of each input register
1235 * If this value is 0, the number of components is given by the
1236 * num_components field of nir_intrinsic_instr.
1238 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1242 /** number of components of the output register
1244 * If this value is 0, the number of components is given by the
1245 * num_components field of nir_intrinsic_instr.
1247 unsigned dest_components
;
1249 /** the number of constant indices used by the intrinsic */
1250 unsigned num_indices
;
1252 /** indicates the usage of intr->const_index[n] */
1253 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1255 /** semantic flags for calls to this intrinsic */
1256 nir_intrinsic_semantic_flag flags
;
1257 } nir_intrinsic_info
;
1259 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1261 static inline unsigned
1262 nir_intrinsic_src_components(nir_intrinsic_instr
*intr
, unsigned srcn
)
1264 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1265 assert(srcn
< info
->num_srcs
);
1266 if (info
->src_components
[srcn
])
1267 return info
->src_components
[srcn
];
1269 return intr
->num_components
;
1272 static inline unsigned
1273 nir_intrinsic_dest_components(nir_intrinsic_instr
*intr
)
1275 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1276 if (!info
->has_dest
)
1278 else if (info
->dest_components
)
1279 return info
->dest_components
;
1281 return intr
->num_components
;
1284 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1285 static inline type \
1286 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1288 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1289 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1290 return (type)instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1292 static inline void \
1293 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1295 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1296 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1297 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1300 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1301 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1302 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1303 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1304 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1305 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1306 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1307 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1308 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1309 INTRINSIC_IDX_ACCESSORS(reduction_op
, REDUCTION_OP
, unsigned)
1310 INTRINSIC_IDX_ACCESSORS(cluster_size
, CLUSTER_SIZE
, unsigned)
1311 INTRINSIC_IDX_ACCESSORS(param_idx
, PARAM_IDX
, unsigned)
1312 INTRINSIC_IDX_ACCESSORS(image_dim
, IMAGE_DIM
, enum glsl_sampler_dim
)
1313 INTRINSIC_IDX_ACCESSORS(image_array
, IMAGE_ARRAY
, bool)
1314 INTRINSIC_IDX_ACCESSORS(access
, ACCESS
, enum gl_access_qualifier
)
1315 INTRINSIC_IDX_ACCESSORS(format
, FORMAT
, unsigned)
1318 * \group texture information
1320 * This gives semantic information about textures which is useful to the
1321 * frontend, the backend, and lowering passes, but not the optimizer.
1326 nir_tex_src_projector
,
1327 nir_tex_src_comparator
, /* shadow comparator */
1331 nir_tex_src_ms_index
, /* MSAA sample index */
1332 nir_tex_src_ms_mcs
, /* MSAA compression value */
1335 nir_tex_src_texture_deref
, /* < deref pointing to the texture */
1336 nir_tex_src_sampler_deref
, /* < deref pointing to the sampler */
1337 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1338 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1339 nir_tex_src_plane
, /* < selects plane for planar textures */
1340 nir_num_tex_src_types
1345 nir_tex_src_type src_type
;
1349 nir_texop_tex
, /**< Regular texture look-up */
1350 nir_texop_txb
, /**< Texture look-up with LOD bias */
1351 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1352 nir_texop_txd
, /**< Texture look-up with partial derivatives */
1353 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1354 nir_texop_txf_ms
, /**< Multisample texture fetch */
1355 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1356 nir_texop_txs
, /**< Texture size */
1357 nir_texop_lod
, /**< Texture lod query */
1358 nir_texop_tg4
, /**< Texture gather */
1359 nir_texop_query_levels
, /**< Texture levels query */
1360 nir_texop_texture_samples
, /**< Texture samples query */
1361 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1369 enum glsl_sampler_dim sampler_dim
;
1370 nir_alu_type dest_type
;
1375 unsigned num_srcs
, coord_components
;
1376 bool is_array
, is_shadow
;
1379 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1380 * components or the new-style shadow that outputs 1 component.
1382 bool is_new_style_shadow
;
1384 /* gather component selector */
1385 unsigned component
: 2;
1387 /** The texture index
1389 * If this texture instruction has a nir_tex_src_texture_offset source,
1390 * then the texture index is given by texture_index + texture_offset.
1392 unsigned texture_index
;
1394 /** The size of the texture array or 0 if it's not an array */
1395 unsigned texture_array_size
;
1397 /** The sampler index
1399 * The following operations do not require a sampler and, as such, this
1400 * field should be ignored:
1402 * - nir_texop_txf_ms
1405 * - nir_texop_query_levels
1406 * - nir_texop_texture_samples
1407 * - nir_texop_samples_identical
1409 * If this texture instruction has a nir_tex_src_sampler_offset source,
1410 * then the sampler index is given by sampler_index + sampler_offset.
1412 unsigned sampler_index
;
1415 static inline unsigned
1416 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1418 switch (instr
->op
) {
1419 case nir_texop_txs
: {
1421 switch (instr
->sampler_dim
) {
1422 case GLSL_SAMPLER_DIM_1D
:
1423 case GLSL_SAMPLER_DIM_BUF
:
1426 case GLSL_SAMPLER_DIM_2D
:
1427 case GLSL_SAMPLER_DIM_CUBE
:
1428 case GLSL_SAMPLER_DIM_MS
:
1429 case GLSL_SAMPLER_DIM_RECT
:
1430 case GLSL_SAMPLER_DIM_EXTERNAL
:
1431 case GLSL_SAMPLER_DIM_SUBPASS
:
1434 case GLSL_SAMPLER_DIM_3D
:
1438 unreachable("not reached");
1440 if (instr
->is_array
)
1448 case nir_texop_texture_samples
:
1449 case nir_texop_query_levels
:
1450 case nir_texop_samples_identical
:
1454 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1461 /* Returns true if this texture operation queries something about the texture
1462 * rather than actually sampling it.
1465 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1467 switch (instr
->op
) {
1470 case nir_texop_texture_samples
:
1471 case nir_texop_query_levels
:
1472 case nir_texop_txf_ms_mcs
:
1479 case nir_texop_txf_ms
:
1483 unreachable("Invalid texture opcode");
1488 nir_alu_instr_is_comparison(const nir_alu_instr
*instr
)
1490 switch (instr
->op
) {
1511 static inline nir_alu_type
1512 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1514 switch (instr
->src
[src
].src_type
) {
1515 case nir_tex_src_coord
:
1516 switch (instr
->op
) {
1518 case nir_texop_txf_ms
:
1519 case nir_texop_txf_ms_mcs
:
1520 case nir_texop_samples_identical
:
1521 return nir_type_int
;
1524 return nir_type_float
;
1527 case nir_tex_src_lod
:
1528 switch (instr
->op
) {
1531 return nir_type_int
;
1534 return nir_type_float
;
1537 case nir_tex_src_projector
:
1538 case nir_tex_src_comparator
:
1539 case nir_tex_src_bias
:
1540 case nir_tex_src_ddx
:
1541 case nir_tex_src_ddy
:
1542 return nir_type_float
;
1544 case nir_tex_src_offset
:
1545 case nir_tex_src_ms_index
:
1546 case nir_tex_src_texture_offset
:
1547 case nir_tex_src_sampler_offset
:
1548 return nir_type_int
;
1551 unreachable("Invalid texture source type");
1555 static inline unsigned
1556 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1558 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1559 return instr
->coord_components
;
1561 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1562 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1565 if (instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1566 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1567 if (instr
->is_array
)
1568 return instr
->coord_components
- 1;
1570 return instr
->coord_components
;
1573 /* Usual APIs don't allow cube + offset, but we allow it, with 2 coords for
1574 * the offset, since a cube maps to a single face.
1576 if (instr
->src
[src
].src_type
== nir_tex_src_offset
) {
1577 if (instr
->sampler_dim
== GLSL_SAMPLER_DIM_CUBE
)
1579 else if (instr
->is_array
)
1580 return instr
->coord_components
- 1;
1582 return instr
->coord_components
;
1589 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1591 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1592 if (instr
->src
[i
].src_type
== type
)
1598 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1599 nir_tex_src_type src_type
,
1602 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1607 nir_const_value value
;
1610 } nir_load_const_instr
;
1623 /* creates a new SSA variable in an undefined state */
1628 } nir_ssa_undef_instr
;
1631 struct exec_node node
;
1633 /* The predecessor block corresponding to this source */
1634 struct nir_block
*pred
;
1639 #define nir_foreach_phi_src(phi_src, phi) \
1640 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1641 #define nir_foreach_phi_src_safe(phi_src, phi) \
1642 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1647 struct exec_list srcs
; /** < list of nir_phi_src */
1653 struct exec_node node
;
1656 } nir_parallel_copy_entry
;
1658 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1659 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1664 /* A list of nir_parallel_copy_entrys. The sources of all of the
1665 * entries are copied to the corresponding destinations "in parallel".
1666 * In other words, if we have two entries: a -> b and b -> a, the values
1669 struct exec_list entries
;
1670 } nir_parallel_copy_instr
;
1672 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1673 type
, nir_instr_type_alu
)
1674 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1675 type
, nir_instr_type_call
)
1676 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1677 type
, nir_instr_type_jump
)
1678 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1679 type
, nir_instr_type_tex
)
1680 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1681 type
, nir_instr_type_intrinsic
)
1682 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1683 type
, nir_instr_type_load_const
)
1684 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1685 type
, nir_instr_type_ssa_undef
)
1686 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1687 type
, nir_instr_type_phi
)
1688 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1689 nir_parallel_copy_instr
, instr
,
1690 type
, nir_instr_type_parallel_copy
)
1695 * Control flow consists of a tree of control flow nodes, which include
1696 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1697 * instructions that always run start-to-finish. Each basic block also keeps
1698 * track of its successors (blocks which may run immediately after the current
1699 * block) and predecessors (blocks which could have run immediately before the
1700 * current block). Each function also has a start block and an end block which
1701 * all return statements point to (which is always empty). Together, all the
1702 * blocks with their predecessors and successors make up the control flow
1703 * graph (CFG) of the function. There are helpers that modify the tree of
1704 * control flow nodes while modifying the CFG appropriately; these should be
1705 * used instead of modifying the tree directly.
1712 nir_cf_node_function
1715 typedef struct nir_cf_node
{
1716 struct exec_node node
;
1717 nir_cf_node_type type
;
1718 struct nir_cf_node
*parent
;
1721 typedef struct nir_block
{
1722 nir_cf_node cf_node
;
1724 struct exec_list instr_list
; /** < list of nir_instr */
1726 /** generic block index; generated by nir_index_blocks */
1730 * Each block can only have up to 2 successors, so we put them in a simple
1731 * array - no need for anything more complicated.
1733 struct nir_block
*successors
[2];
1735 /* Set of nir_block predecessors in the CFG */
1736 struct set
*predecessors
;
1739 * this node's immediate dominator in the dominance tree - set to NULL for
1742 struct nir_block
*imm_dom
;
1744 /* This node's children in the dominance tree */
1745 unsigned num_dom_children
;
1746 struct nir_block
**dom_children
;
1748 /* Set of nir_blocks on the dominance frontier of this block */
1749 struct set
*dom_frontier
;
1752 * These two indices have the property that dom_{pre,post}_index for each
1753 * child of this block in the dominance tree will always be between
1754 * dom_pre_index and dom_post_index for this block, which makes testing if
1755 * a given block is dominated by another block an O(1) operation.
1757 unsigned dom_pre_index
, dom_post_index
;
1759 /* live in and out for this block; used for liveness analysis */
1760 BITSET_WORD
*live_in
;
1761 BITSET_WORD
*live_out
;
1764 static inline nir_instr
*
1765 nir_block_first_instr(nir_block
*block
)
1767 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1768 return exec_node_data(nir_instr
, head
, node
);
1771 static inline nir_instr
*
1772 nir_block_last_instr(nir_block
*block
)
1774 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1775 return exec_node_data(nir_instr
, tail
, node
);
1779 nir_block_ends_in_jump(nir_block
*block
)
1781 return !exec_list_is_empty(&block
->instr_list
) &&
1782 nir_block_last_instr(block
)->type
== nir_instr_type_jump
;
1785 #define nir_foreach_instr(instr, block) \
1786 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1787 #define nir_foreach_instr_reverse(instr, block) \
1788 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1789 #define nir_foreach_instr_safe(instr, block) \
1790 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1791 #define nir_foreach_instr_reverse_safe(instr, block) \
1792 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1794 typedef struct nir_if
{
1795 nir_cf_node cf_node
;
1798 struct exec_list then_list
; /** < list of nir_cf_node */
1799 struct exec_list else_list
; /** < list of nir_cf_node */
1805 nir_instr
*conditional_instr
;
1807 nir_block
*break_block
;
1808 nir_block
*continue_from_block
;
1810 bool continue_from_then
;
1812 struct list_head loop_terminator_link
;
1813 } nir_loop_terminator
;
1816 /* Number of instructions in the loop */
1817 unsigned num_instructions
;
1819 /* How many times the loop is run (if known) */
1820 unsigned trip_count
;
1821 bool is_trip_count_known
;
1823 /* Unroll the loop regardless of its size */
1826 /* Does the loop contain complex loop terminators, continues or other
1827 * complex behaviours? If this is true we can't rely on
1828 * loop_terminator_list to be complete or accurate.
1832 nir_loop_terminator
*limiting_terminator
;
1834 /* A list of loop_terminators terminating this loop. */
1835 struct list_head loop_terminator_list
;
1839 nir_cf_node cf_node
;
1841 struct exec_list body
; /** < list of nir_cf_node */
1843 nir_loop_info
*info
;
1847 * Various bits of metadata that can may be created or required by
1848 * optimization and analysis passes
1851 nir_metadata_none
= 0x0,
1852 nir_metadata_block_index
= 0x1,
1853 nir_metadata_dominance
= 0x2,
1854 nir_metadata_live_ssa_defs
= 0x4,
1855 nir_metadata_not_properly_reset
= 0x8,
1856 nir_metadata_loop_analysis
= 0x10,
1860 nir_cf_node cf_node
;
1862 /** pointer to the function of which this is an implementation */
1863 struct nir_function
*function
;
1865 struct exec_list body
; /** < list of nir_cf_node */
1867 nir_block
*end_block
;
1869 /** list for all local variables in the function */
1870 struct exec_list locals
;
1872 /** list of local registers in the function */
1873 struct exec_list registers
;
1875 /** next available local register index */
1878 /** next available SSA value index */
1881 /* total number of basic blocks, only valid when block_index_dirty = false */
1882 unsigned num_blocks
;
1884 nir_metadata valid_metadata
;
1885 } nir_function_impl
;
1887 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1888 nir_start_block(nir_function_impl
*impl
)
1890 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1893 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1894 nir_impl_last_block(nir_function_impl
*impl
)
1896 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1899 static inline nir_cf_node
*
1900 nir_cf_node_next(nir_cf_node
*node
)
1902 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1903 if (exec_node_is_tail_sentinel(next
))
1906 return exec_node_data(nir_cf_node
, next
, node
);
1909 static inline nir_cf_node
*
1910 nir_cf_node_prev(nir_cf_node
*node
)
1912 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1913 if (exec_node_is_head_sentinel(prev
))
1916 return exec_node_data(nir_cf_node
, prev
, node
);
1920 nir_cf_node_is_first(const nir_cf_node
*node
)
1922 return exec_node_is_head_sentinel(node
->node
.prev
);
1926 nir_cf_node_is_last(const nir_cf_node
*node
)
1928 return exec_node_is_tail_sentinel(node
->node
.next
);
1931 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1932 type
, nir_cf_node_block
)
1933 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1934 type
, nir_cf_node_if
)
1935 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1936 type
, nir_cf_node_loop
)
1937 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1938 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1940 static inline nir_block
*
1941 nir_if_first_then_block(nir_if
*if_stmt
)
1943 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1944 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1947 static inline nir_block
*
1948 nir_if_last_then_block(nir_if
*if_stmt
)
1950 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1951 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1954 static inline nir_block
*
1955 nir_if_first_else_block(nir_if
*if_stmt
)
1957 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1958 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1961 static inline nir_block
*
1962 nir_if_last_else_block(nir_if
*if_stmt
)
1964 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1965 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1968 static inline nir_block
*
1969 nir_loop_first_block(nir_loop
*loop
)
1971 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1972 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1975 static inline nir_block
*
1976 nir_loop_last_block(nir_loop
*loop
)
1978 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1979 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1983 uint8_t num_components
;
1987 typedef struct nir_function
{
1988 struct exec_node node
;
1991 struct nir_shader
*shader
;
1993 unsigned num_params
;
1994 nir_parameter
*params
;
1996 /** The implementation of this function.
1998 * If the function is only declared and not implemented, this is NULL.
2000 nir_function_impl
*impl
;
2003 typedef struct nir_shader_compiler_options
{
2008 /** Lowers flrp when it does not support doubles */
2015 /** Lowers ibitfield_extract/ubitfield_extract to ibfe/ubfe. */
2016 bool lower_bitfield_extract
;
2017 /** Lowers ibitfield_extract/ubitfield_extract to bfm, compares, shifts. */
2018 bool lower_bitfield_extract_to_shifts
;
2019 /** Lowers bitfield_insert to bfi/bfm */
2020 bool lower_bitfield_insert
;
2021 /** Lowers bitfield_insert to bfm, compares, and shifts. */
2022 bool lower_bitfield_insert_to_shifts
;
2023 /** Lowers bitfield_reverse to shifts. */
2024 bool lower_bitfield_reverse
;
2025 /** Lowers bit_count to shifts. */
2026 bool lower_bit_count
;
2027 /** Lowers bfm to shifts and subtracts. */
2029 /** Lowers ifind_msb to compare and ufind_msb */
2030 bool lower_ifind_msb
;
2031 /** Lowers find_lsb to ufind_msb and logic ops */
2032 bool lower_find_lsb
;
2033 bool lower_uadd_carry
;
2034 bool lower_usub_borrow
;
2035 /** Lowers imul_high/umul_high to 16-bit multiplies and carry operations. */
2036 bool lower_mul_high
;
2037 /** lowers fneg and ineg to fsub and isub. */
2039 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
2042 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
2045 /** enables rules to lower idiv by power-of-two: */
2048 /* lower b2f to iand */
2051 /* Does the native fdot instruction replicate its result for four
2052 * components? If so, then opt_algebraic_late will turn all fdotN
2053 * instructions into fdot_replicatedN instructions.
2055 bool fdot_replicates
;
2057 /** lowers ffract to fsub+ffloor: */
2062 bool lower_pack_half_2x16
;
2063 bool lower_pack_unorm_2x16
;
2064 bool lower_pack_snorm_2x16
;
2065 bool lower_pack_unorm_4x8
;
2066 bool lower_pack_snorm_4x8
;
2067 bool lower_unpack_half_2x16
;
2068 bool lower_unpack_unorm_2x16
;
2069 bool lower_unpack_snorm_2x16
;
2070 bool lower_unpack_unorm_4x8
;
2071 bool lower_unpack_snorm_4x8
;
2073 bool lower_extract_byte
;
2074 bool lower_extract_word
;
2076 bool lower_all_io_to_temps
;
2079 * Does the driver support real 32-bit integers? (Otherwise, integers
2080 * are simulated by floats.)
2082 bool native_integers
;
2084 /* Indicates that the driver only has zero-based vertex id */
2085 bool vertex_id_zero_based
;
2088 * If enabled, gl_BaseVertex will be lowered as:
2089 * is_indexed_draw (~0/0) & firstvertex
2091 bool lower_base_vertex
;
2094 * If enabled, gl_HelperInvocation will be lowered as:
2096 * !((1 << sample_id) & sample_mask_in))
2098 * This depends on some possibly hw implementation details, which may
2099 * not be true for all hw. In particular that the FS is only executed
2100 * for covered samples or for helper invocations. So, do not blindly
2101 * enable this option.
2103 * Note: See also issue #22 in ARB_shader_image_load_store
2105 bool lower_helper_invocation
;
2107 bool lower_cs_local_index_from_id
;
2109 bool lower_device_index_to_zero
;
2111 /* Set if nir_lower_wpos_ytransform() should also invert gl_PointCoord. */
2112 bool lower_wpos_pntc
;
2115 * Should nir_lower_io() create load_interpolated_input intrinsics?
2117 * If not, it generates regular load_input intrinsics and interpolation
2118 * information must be inferred from the list of input nir_variables.
2120 bool use_interpolated_input_intrinsics
;
2122 unsigned max_unroll_iterations
;
2123 } nir_shader_compiler_options
;
2125 typedef struct nir_shader
{
2126 /** list of uniforms (nir_variable) */
2127 struct exec_list uniforms
;
2129 /** list of inputs (nir_variable) */
2130 struct exec_list inputs
;
2132 /** list of outputs (nir_variable) */
2133 struct exec_list outputs
;
2135 /** list of shared compute variables (nir_variable) */
2136 struct exec_list shared
;
2138 /** Set of driver-specific options for the shader.
2140 * The memory for the options is expected to be kept in a single static
2141 * copy by the driver.
2143 const struct nir_shader_compiler_options
*options
;
2145 /** Various bits of compile-time information about a given shader */
2146 struct shader_info info
;
2148 /** list of global variables in the shader (nir_variable) */
2149 struct exec_list globals
;
2151 /** list of system value variables in the shader (nir_variable) */
2152 struct exec_list system_values
;
2154 struct exec_list functions
; /** < list of nir_function */
2156 /** list of global register in the shader */
2157 struct exec_list registers
;
2159 /** next available global register index */
2163 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
2166 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
2168 /** Constant data associated with this shader.
2170 * Constant data is loaded through load_constant intrinsics. See also
2171 * nir_opt_large_constants.
2173 void *constant_data
;
2174 unsigned constant_data_size
;
2177 static inline nir_function_impl
*
2178 nir_shader_get_entrypoint(nir_shader
*shader
)
2180 assert(exec_list_length(&shader
->functions
) == 1);
2181 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
2182 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
2183 assert(func
->num_params
== 0);
2188 #define nir_foreach_function(func, shader) \
2189 foreach_list_typed(nir_function, func, node, &(shader)->functions)
2191 nir_shader
*nir_shader_create(void *mem_ctx
,
2192 gl_shader_stage stage
,
2193 const nir_shader_compiler_options
*options
,
2196 /** creates a register, including assigning it an index and adding it to the list */
2197 nir_register
*nir_global_reg_create(nir_shader
*shader
);
2199 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
2201 void nir_reg_remove(nir_register
*reg
);
2203 /** Adds a variable to the appropriate list in nir_shader */
2204 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
2207 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
2209 assert(var
->data
.mode
== nir_var_local
);
2210 exec_list_push_tail(&impl
->locals
, &var
->node
);
2213 /** creates a variable, sets a few defaults, and adds it to the list */
2214 nir_variable
*nir_variable_create(nir_shader
*shader
,
2215 nir_variable_mode mode
,
2216 const struct glsl_type
*type
,
2218 /** creates a local variable and adds it to the list */
2219 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
2220 const struct glsl_type
*type
,
2223 /** creates a function and adds it to the shader's list of functions */
2224 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
2226 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
2227 /** creates a function_impl that isn't tied to any particular function */
2228 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
2230 nir_block
*nir_block_create(nir_shader
*shader
);
2231 nir_if
*nir_if_create(nir_shader
*shader
);
2232 nir_loop
*nir_loop_create(nir_shader
*shader
);
2234 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
2236 /** requests that the given pieces of metadata be generated */
2237 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
2238 /** dirties all but the preserved metadata */
2239 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
2241 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
2242 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
2244 nir_deref_instr
*nir_deref_instr_create(nir_shader
*shader
,
2245 nir_deref_type deref_type
);
2247 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
2249 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
2250 unsigned num_components
,
2253 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
2254 nir_intrinsic_op op
);
2256 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
2257 nir_function
*callee
);
2259 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
2261 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
2263 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
2265 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
2266 unsigned num_components
,
2269 nir_const_value
nir_alu_binop_identity(nir_op binop
, unsigned bit_size
);
2272 * NIR Cursors and Instruction Insertion API
2275 * A tiny struct representing a point to insert/extract instructions or
2276 * control flow nodes. Helps reduce the combinatorial explosion of possible
2277 * points to insert/extract.
2279 * \sa nir_control_flow.h
2282 nir_cursor_before_block
,
2283 nir_cursor_after_block
,
2284 nir_cursor_before_instr
,
2285 nir_cursor_after_instr
,
2286 } nir_cursor_option
;
2289 nir_cursor_option option
;
2296 static inline nir_block
*
2297 nir_cursor_current_block(nir_cursor cursor
)
2299 if (cursor
.option
== nir_cursor_before_instr
||
2300 cursor
.option
== nir_cursor_after_instr
) {
2301 return cursor
.instr
->block
;
2303 return cursor
.block
;
2307 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2309 static inline nir_cursor
2310 nir_before_block(nir_block
*block
)
2313 cursor
.option
= nir_cursor_before_block
;
2314 cursor
.block
= block
;
2318 static inline nir_cursor
2319 nir_after_block(nir_block
*block
)
2322 cursor
.option
= nir_cursor_after_block
;
2323 cursor
.block
= block
;
2327 static inline nir_cursor
2328 nir_before_instr(nir_instr
*instr
)
2331 cursor
.option
= nir_cursor_before_instr
;
2332 cursor
.instr
= instr
;
2336 static inline nir_cursor
2337 nir_after_instr(nir_instr
*instr
)
2340 cursor
.option
= nir_cursor_after_instr
;
2341 cursor
.instr
= instr
;
2345 static inline nir_cursor
2346 nir_after_block_before_jump(nir_block
*block
)
2348 nir_instr
*last_instr
= nir_block_last_instr(block
);
2349 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2350 return nir_before_instr(last_instr
);
2352 return nir_after_block(block
);
2356 static inline nir_cursor
2357 nir_before_src(nir_src
*src
, bool is_if_condition
)
2359 if (is_if_condition
) {
2360 nir_block
*prev_block
=
2361 nir_cf_node_as_block(nir_cf_node_prev(&src
->parent_if
->cf_node
));
2362 assert(!nir_block_ends_in_jump(prev_block
));
2363 return nir_after_block(prev_block
);
2364 } else if (src
->parent_instr
->type
== nir_instr_type_phi
) {
2366 nir_phi_instr
*cond_phi
= nir_instr_as_phi(src
->parent_instr
);
2368 nir_foreach_phi_src(phi_src
, cond_phi
) {
2369 if (phi_src
->src
.ssa
== src
->ssa
) {
2376 /* The LIST_ENTRY macro is a generic container-of macro, it just happens
2377 * to have a more specific name.
2379 nir_phi_src
*phi_src
= LIST_ENTRY(nir_phi_src
, src
, src
);
2380 return nir_after_block_before_jump(phi_src
->pred
);
2382 return nir_before_instr(src
->parent_instr
);
2386 static inline nir_cursor
2387 nir_before_cf_node(nir_cf_node
*node
)
2389 if (node
->type
== nir_cf_node_block
)
2390 return nir_before_block(nir_cf_node_as_block(node
));
2392 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2395 static inline nir_cursor
2396 nir_after_cf_node(nir_cf_node
*node
)
2398 if (node
->type
== nir_cf_node_block
)
2399 return nir_after_block(nir_cf_node_as_block(node
));
2401 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2404 static inline nir_cursor
2405 nir_after_phis(nir_block
*block
)
2407 nir_foreach_instr(instr
, block
) {
2408 if (instr
->type
!= nir_instr_type_phi
)
2409 return nir_before_instr(instr
);
2411 return nir_after_block(block
);
2414 static inline nir_cursor
2415 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2417 if (node
->type
== nir_cf_node_block
)
2418 return nir_after_block(nir_cf_node_as_block(node
));
2420 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2422 return nir_after_phis(block
);
2425 static inline nir_cursor
2426 nir_before_cf_list(struct exec_list
*cf_list
)
2428 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2429 exec_list_get_head(cf_list
), node
);
2430 return nir_before_cf_node(first_node
);
2433 static inline nir_cursor
2434 nir_after_cf_list(struct exec_list
*cf_list
)
2436 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2437 exec_list_get_tail(cf_list
), node
);
2438 return nir_after_cf_node(last_node
);
2442 * Insert a NIR instruction at the given cursor.
2444 * Note: This does not update the cursor.
2446 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2449 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2451 nir_instr_insert(nir_before_instr(instr
), before
);
2455 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2457 nir_instr_insert(nir_after_instr(instr
), after
);
2461 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2463 nir_instr_insert(nir_before_block(block
), before
);
2467 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2469 nir_instr_insert(nir_after_block(block
), after
);
2473 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2475 nir_instr_insert(nir_before_cf_node(node
), before
);
2479 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2481 nir_instr_insert(nir_after_cf_node(node
), after
);
2485 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2487 nir_instr_insert(nir_before_cf_list(list
), before
);
2491 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2493 nir_instr_insert(nir_after_cf_list(list
), after
);
2496 void nir_instr_remove_v(nir_instr
*instr
);
2498 static inline nir_cursor
2499 nir_instr_remove(nir_instr
*instr
)
2502 nir_instr
*prev
= nir_instr_prev(instr
);
2504 cursor
= nir_after_instr(prev
);
2506 cursor
= nir_before_block(instr
->block
);
2508 nir_instr_remove_v(instr
);
2514 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2515 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2516 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2517 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2519 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2520 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2522 nir_const_value
*nir_src_as_const_value(nir_src src
);
2523 bool nir_src_is_dynamically_uniform(nir_src src
);
2524 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2525 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2526 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2527 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2528 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2531 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2532 unsigned num_components
, unsigned bit_size
,
2534 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2535 unsigned num_components
, unsigned bit_size
,
2538 nir_ssa_dest_init_for_type(nir_instr
*instr
, nir_dest
*dest
,
2539 const struct glsl_type
*type
,
2542 assert(glsl_type_is_vector_or_scalar(type
));
2543 nir_ssa_dest_init(instr
, dest
, glsl_get_components(type
),
2544 glsl_get_bit_size(type
), name
);
2546 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2547 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2548 nir_instr
*after_me
);
2550 nir_component_mask_t
nir_ssa_def_components_read(const nir_ssa_def
*def
);
2553 * finds the next basic block in source-code order, returns NULL if there is
2557 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2559 /* Performs the opposite of nir_block_cf_tree_next() */
2561 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2563 /* Gets the first block in a CF node in source-code order */
2565 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2567 /* Gets the last block in a CF node in source-code order */
2569 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2571 /* Gets the next block after a CF node in source-code order */
2573 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2575 /* Macros for loops that visit blocks in source-code order */
2577 #define nir_foreach_block(block, impl) \
2578 for (nir_block *block = nir_start_block(impl); block != NULL; \
2579 block = nir_block_cf_tree_next(block))
2581 #define nir_foreach_block_safe(block, impl) \
2582 for (nir_block *block = nir_start_block(impl), \
2583 *next = nir_block_cf_tree_next(block); \
2585 block = next, next = nir_block_cf_tree_next(block))
2587 #define nir_foreach_block_reverse(block, impl) \
2588 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2589 block = nir_block_cf_tree_prev(block))
2591 #define nir_foreach_block_reverse_safe(block, impl) \
2592 for (nir_block *block = nir_impl_last_block(impl), \
2593 *prev = nir_block_cf_tree_prev(block); \
2595 block = prev, prev = nir_block_cf_tree_prev(block))
2597 #define nir_foreach_block_in_cf_node(block, node) \
2598 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2599 block != nir_cf_node_cf_tree_next(node); \
2600 block = nir_block_cf_tree_next(block))
2602 /* If the following CF node is an if, this function returns that if.
2603 * Otherwise, it returns NULL.
2605 nir_if
*nir_block_get_following_if(nir_block
*block
);
2607 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2609 void nir_index_local_regs(nir_function_impl
*impl
);
2610 void nir_index_global_regs(nir_shader
*shader
);
2611 void nir_index_ssa_defs(nir_function_impl
*impl
);
2612 unsigned nir_index_instrs(nir_function_impl
*impl
);
2614 void nir_index_blocks(nir_function_impl
*impl
);
2616 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2617 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2618 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2620 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2621 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2622 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2623 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2625 nir_shader
*nir_shader_serialize_deserialize(void *mem_ctx
, nir_shader
*s
);
2628 void nir_validate_shader(nir_shader
*shader
);
2629 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2630 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2633 should_clone_nir(void)
2635 static int should_clone
= -1;
2636 if (should_clone
< 0)
2637 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2639 return should_clone
;
2643 should_serialize_deserialize_nir(void)
2645 static int test_serialize
= -1;
2646 if (test_serialize
< 0)
2647 test_serialize
= env_var_as_boolean("NIR_TEST_SERIALIZE", false);
2649 return test_serialize
;
2653 should_print_nir(void)
2655 static int should_print
= -1;
2656 if (should_print
< 0)
2657 should_print
= env_var_as_boolean("NIR_PRINT", false);
2659 return should_print
;
2662 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2663 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2664 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2665 static inline bool should_clone_nir(void) { return false; }
2666 static inline bool should_serialize_deserialize_nir(void) { return false; }
2667 static inline bool should_print_nir(void) { return false; }
2670 #define _PASS(nir, do_pass) do { \
2672 nir_validate_shader(nir); \
2673 if (should_clone_nir()) { \
2674 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2678 if (should_serialize_deserialize_nir()) { \
2679 void *mem_ctx = ralloc_parent(nir); \
2680 nir = nir_shader_serialize_deserialize(mem_ctx, nir); \
2684 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2685 nir_metadata_set_validation_flag(nir); \
2686 if (should_print_nir()) \
2687 printf("%s\n", #pass); \
2688 if (pass(nir, ##__VA_ARGS__)) { \
2690 if (should_print_nir()) \
2691 nir_print_shader(nir, stdout); \
2692 nir_metadata_check_validation_flag(nir); \
2696 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2697 if (should_print_nir()) \
2698 printf("%s\n", #pass); \
2699 pass(nir, ##__VA_ARGS__); \
2700 if (should_print_nir()) \
2701 nir_print_shader(nir, stdout); \
2704 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2705 void nir_calc_dominance(nir_shader
*shader
);
2707 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2708 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2710 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2711 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2713 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2714 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2716 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2717 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2719 int nir_gs_count_vertices(const nir_shader
*shader
);
2721 bool nir_shrink_vec_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2722 bool nir_split_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2723 bool nir_split_var_copies(nir_shader
*shader
);
2724 bool nir_split_per_member_structs(nir_shader
*shader
);
2725 bool nir_split_struct_vars(nir_shader
*shader
, nir_variable_mode modes
);
2727 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2728 bool nir_lower_returns(nir_shader
*shader
);
2730 bool nir_inline_functions(nir_shader
*shader
);
2732 bool nir_propagate_invariant(nir_shader
*shader
);
2734 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2735 void nir_lower_deref_copy_instr(struct nir_builder
*b
,
2736 nir_intrinsic_instr
*copy
);
2737 bool nir_lower_var_copies(nir_shader
*shader
);
2739 void nir_fixup_deref_modes(nir_shader
*shader
);
2741 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2743 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2745 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2747 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2748 nir_function_impl
*entrypoint
,
2749 bool outputs
, bool inputs
);
2751 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2753 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2754 int (*type_size
)(const struct glsl_type
*));
2756 /* Some helpers to do very simple linking */
2757 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2758 void nir_compact_varyings(nir_shader
*producer
, nir_shader
*consumer
,
2759 bool default_to_smooth_interp
);
2762 /* If set, this forces all non-flat fragment shader inputs to be
2763 * interpolated as if with the "sample" qualifier. This requires
2764 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2766 nir_lower_io_force_sample_interpolation
= (1 << 1),
2767 } nir_lower_io_options
;
2768 bool nir_lower_io(nir_shader
*shader
,
2769 nir_variable_mode modes
,
2770 int (*type_size
)(const struct glsl_type
*),
2771 nir_lower_io_options
);
2772 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2773 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2775 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2777 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2778 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2779 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2781 bool nir_remove_dead_derefs(nir_shader
*shader
);
2782 bool nir_remove_dead_derefs_impl(nir_function_impl
*impl
);
2783 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2784 bool nir_lower_constant_initializers(nir_shader
*shader
,
2785 nir_variable_mode modes
);
2787 bool nir_move_load_const(nir_shader
*shader
);
2788 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2789 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2790 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
2792 bool nir_lower_alu(nir_shader
*shader
);
2793 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2794 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2795 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
2796 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2797 void nir_lower_io_arrays_to_elements(nir_shader
*producer
, nir_shader
*consumer
);
2798 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader
*shader
,
2800 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2801 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
2803 typedef struct nir_lower_subgroups_options
{
2804 uint8_t subgroup_size
;
2805 uint8_t ballot_bit_size
;
2806 bool lower_to_scalar
:1;
2807 bool lower_vote_trivial
:1;
2808 bool lower_vote_eq_to_ballot
:1;
2809 bool lower_subgroup_masks
:1;
2810 bool lower_shuffle
:1;
2811 bool lower_shuffle_to_32bit
:1;
2813 } nir_lower_subgroups_options
;
2815 bool nir_lower_subgroups(nir_shader
*shader
,
2816 const nir_lower_subgroups_options
*options
);
2818 bool nir_lower_system_values(nir_shader
*shader
);
2820 typedef struct nir_lower_tex_options
{
2822 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2823 * sampler types a texture projector is lowered.
2828 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2830 bool lower_txf_offset
;
2833 * If true, lower away nir_tex_src_offset for all rect textures.
2835 bool lower_rect_offset
;
2838 * If true, lower rect textures to 2D, using txs to fetch the
2839 * texture dimensions and dividing the texture coords by the
2840 * texture dims to normalize.
2845 * If true, convert yuv to rgb.
2847 unsigned lower_y_uv_external
;
2848 unsigned lower_y_u_v_external
;
2849 unsigned lower_yx_xuxv_external
;
2850 unsigned lower_xy_uxvx_external
;
2853 * To emulate certain texture wrap modes, this can be used
2854 * to saturate the specified tex coord to [0.0, 1.0]. The
2855 * bits are according to sampler #, ie. if, for example:
2857 * (conf->saturate_s & (1 << n))
2859 * is true, then the s coord for sampler n is saturated.
2861 * Note that clamping must happen *after* projector lowering
2862 * so any projected texture sample instruction with a clamped
2863 * coordinate gets automatically lowered, regardless of the
2864 * 'lower_txp' setting.
2866 unsigned saturate_s
;
2867 unsigned saturate_t
;
2868 unsigned saturate_r
;
2870 /* Bitmask of textures that need swizzling.
2872 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2873 * swizzles[texture_index] is applied to the result of the texturing
2876 unsigned swizzle_result
;
2878 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2879 * while 4 and 5 represent 0 and 1 respectively.
2881 uint8_t swizzles
[32][4];
2884 * Bitmap of textures that need srgb to linear conversion. If
2885 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2886 * of the texture are lowered to linear.
2888 unsigned lower_srgb
;
2891 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2893 bool lower_txd_cube_map
;
2896 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
2897 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
2898 * with lower_txd_cube_map.
2900 bool lower_txd_shadow
;
2903 * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
2904 * Implies lower_txd_cube_map and lower_txd_shadow.
2907 } nir_lower_tex_options
;
2909 bool nir_lower_tex(nir_shader
*shader
,
2910 const nir_lower_tex_options
*options
);
2912 bool nir_lower_idiv(nir_shader
*shader
);
2914 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2915 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2916 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2918 void nir_lower_two_sided_color(nir_shader
*shader
);
2920 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
2922 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2923 bool nir_lower_patch_vertices(nir_shader
*nir
, unsigned static_count
,
2924 const gl_state_index16
*uniform_state_tokens
);
2926 typedef struct nir_lower_wpos_ytransform_options
{
2927 gl_state_index16 state_tokens
[STATE_LENGTH
];
2928 bool fs_coord_origin_upper_left
:1;
2929 bool fs_coord_origin_lower_left
:1;
2930 bool fs_coord_pixel_center_integer
:1;
2931 bool fs_coord_pixel_center_half_integer
:1;
2932 } nir_lower_wpos_ytransform_options
;
2934 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2935 const nir_lower_wpos_ytransform_options
*options
);
2936 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
2938 typedef struct nir_lower_drawpixels_options
{
2939 gl_state_index16 texcoord_state_tokens
[STATE_LENGTH
];
2940 gl_state_index16 scale_state_tokens
[STATE_LENGTH
];
2941 gl_state_index16 bias_state_tokens
[STATE_LENGTH
];
2942 unsigned drawpix_sampler
;
2943 unsigned pixelmap_sampler
;
2945 bool scale_and_bias
:1;
2946 } nir_lower_drawpixels_options
;
2948 void nir_lower_drawpixels(nir_shader
*shader
,
2949 const nir_lower_drawpixels_options
*options
);
2951 typedef struct nir_lower_bitmap_options
{
2954 } nir_lower_bitmap_options
;
2956 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2958 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
2959 bool nir_lower_to_source_mods(nir_shader
*shader
);
2961 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2963 typedef unsigned (*nir_lower_bit_size_callback
)(const nir_alu_instr
*, void *);
2965 bool nir_lower_bit_size(nir_shader
*shader
,
2966 nir_lower_bit_size_callback callback
,
2967 void *callback_data
);
2970 nir_lower_imul64
= (1 << 0),
2971 nir_lower_isign64
= (1 << 1),
2972 /** Lower all int64 modulus and division opcodes */
2973 nir_lower_divmod64
= (1 << 2),
2974 } nir_lower_int64_options
;
2976 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
2979 nir_lower_drcp
= (1 << 0),
2980 nir_lower_dsqrt
= (1 << 1),
2981 nir_lower_drsq
= (1 << 2),
2982 nir_lower_dtrunc
= (1 << 3),
2983 nir_lower_dfloor
= (1 << 4),
2984 nir_lower_dceil
= (1 << 5),
2985 nir_lower_dfract
= (1 << 6),
2986 nir_lower_dround_even
= (1 << 7),
2987 nir_lower_dmod
= (1 << 8)
2988 } nir_lower_doubles_options
;
2990 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2991 bool nir_lower_pack(nir_shader
*shader
);
2993 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2995 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2997 void nir_loop_analyze_impl(nir_function_impl
*impl
,
2998 nir_variable_mode indirect_mask
);
3000 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
3002 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
3003 bool nir_repair_ssa(nir_shader
*shader
);
3005 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
3007 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
3008 * registers. If false, convert all values (even those not involved in a phi
3009 * node) to registers.
3011 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
3013 bool nir_lower_phis_to_regs_block(nir_block
*block
);
3014 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
3016 bool nir_opt_algebraic(nir_shader
*shader
);
3017 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
3018 bool nir_opt_algebraic_late(nir_shader
*shader
);
3019 bool nir_opt_constant_folding(nir_shader
*shader
);
3021 bool nir_opt_global_to_local(nir_shader
*shader
);
3023 bool nir_copy_prop(nir_shader
*shader
);
3025 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
3027 bool nir_opt_cse(nir_shader
*shader
);
3029 bool nir_opt_dce(nir_shader
*shader
);
3031 bool nir_opt_dead_cf(nir_shader
*shader
);
3033 bool nir_opt_find_array_copies(nir_shader
*shader
);
3035 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
3037 bool nir_opt_if(nir_shader
*shader
);
3039 bool nir_opt_intrinsics(nir_shader
*shader
);
3041 bool nir_opt_large_constants(nir_shader
*shader
,
3042 glsl_type_size_align_func size_align
,
3043 unsigned threshold
);
3045 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
3047 bool nir_opt_move_comparisons(nir_shader
*shader
);
3049 bool nir_opt_move_load_ubo(nir_shader
*shader
);
3051 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
3053 bool nir_opt_remove_phis_impl(nir_function_impl
*impl
);
3054 bool nir_opt_remove_phis(nir_shader
*shader
);
3056 bool nir_opt_shrink_load(nir_shader
*shader
);
3058 bool nir_opt_trivial_continues(nir_shader
*shader
);
3060 bool nir_opt_undef(nir_shader
*shader
);
3062 bool nir_opt_conditional_discard(nir_shader
*shader
);
3064 void nir_sweep(nir_shader
*shader
);
3066 void nir_remap_dual_slot_attributes(nir_shader
*shader
,
3067 uint64_t *dual_slot_inputs
);
3068 uint64_t nir_get_single_slot_attribs_mask(uint64_t attribs
, uint64_t dual_slot
);
3070 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
3071 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);