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)
30 #include "util/hash_table.h"
31 #include "compiler/glsl/list.h"
32 #include "GL/gl.h" /* GLenum */
33 #include "util/list.h"
34 #include "util/ralloc.h"
36 #include "util/bitset.h"
37 #include "util/macros.h"
38 #include "compiler/nir_types.h"
39 #include "compiler/shader_enums.h"
40 #include "compiler/shader_info.h"
44 #include "util/debug.h"
47 #include "nir_opcodes.h"
54 struct gl_shader_program
;
57 #define NIR_TRUE (~0u)
59 /** Defines a cast function
61 * This macro defines a cast function from in_type to out_type where
62 * out_type is some structure type that contains a field of type out_type.
64 * Note that you have to be a bit careful as the generated cast function
67 #define NIR_DEFINE_CAST(name, in_type, out_type, field, \
68 type_field, type_value) \
69 static inline out_type * \
70 name(const in_type *parent) \
72 assert(parent && parent->type_field == type_value); \
73 return exec_node_data(out_type, parent, field); \
82 * Description of built-in state associated with a uniform
84 * \sa nir_variable::state_slots
92 nir_var_shader_in
= (1 << 0),
93 nir_var_shader_out
= (1 << 1),
94 nir_var_global
= (1 << 2),
95 nir_var_local
= (1 << 3),
96 nir_var_uniform
= (1 << 4),
97 nir_var_shader_storage
= (1 << 5),
98 nir_var_system_value
= (1 << 6),
99 nir_var_param
= (1 << 7),
100 nir_var_shared
= (1 << 8),
114 typedef struct nir_constant
{
116 * Value of the constant.
118 * The field used to back the values supplied by the constant is determined
119 * by the type associated with the \c nir_variable. Constants may be
120 * scalars, vectors, or matrices.
122 nir_const_value values
[4];
124 /* we could get this from the var->type but makes clone *much* easier to
125 * not have to care about the type.
127 unsigned num_elements
;
129 /* Array elements / Structure Fields */
130 struct nir_constant
**elements
;
134 * \brief Layout qualifiers for gl_FragDepth.
136 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
137 * with a layout qualifier.
140 nir_depth_layout_none
, /**< No depth layout is specified. */
141 nir_depth_layout_any
,
142 nir_depth_layout_greater
,
143 nir_depth_layout_less
,
144 nir_depth_layout_unchanged
148 * Either a uniform, global variable, shader input, or shader output. Based on
149 * ir_variable - it should be easy to translate between the two.
152 typedef struct nir_variable
{
153 struct exec_node node
;
156 * Declared type of the variable
158 const struct glsl_type
*type
;
161 * Declared name of the variable
165 struct nir_variable_data
{
167 * Storage class of the variable.
169 * \sa nir_variable_mode
171 nir_variable_mode mode
;
174 * Is the variable read-only?
176 * This is set for variables declared as \c const, shader inputs,
179 unsigned read_only
:1;
183 unsigned invariant
:1;
186 * Interpolation mode for shader inputs / outputs
188 * \sa glsl_interp_mode
190 unsigned interpolation
:2;
193 * \name ARB_fragment_coord_conventions
196 unsigned origin_upper_left
:1;
197 unsigned pixel_center_integer
:1;
201 * If non-zero, then this variable may be packed along with other variables
202 * into a single varying slot, so this offset should be applied when
203 * accessing components. For example, an offset of 1 means that the x
204 * component of this variable is actually stored in component y of the
205 * location specified by \c location.
207 unsigned location_frac
:2;
210 * If true, this variable represents an array of scalars that should
211 * be tightly packed. In other words, consecutive array elements
212 * should be stored one component apart, rather than one slot apart.
217 * Whether this is a fragment shader output implicitly initialized with
218 * the previous contents of the specified render target at the
219 * framebuffer location corresponding to this shader invocation.
221 unsigned fb_fetch_output
:1;
224 * \brief Layout qualifier for gl_FragDepth.
226 * This is not equal to \c ir_depth_layout_none if and only if this
227 * variable is \c gl_FragDepth and a layout qualifier is specified.
229 nir_depth_layout depth_layout
;
232 * Storage location of the base of this variable
234 * The precise meaning of this field depends on the nature of the variable.
236 * - Vertex shader input: one of the values from \c gl_vert_attrib.
237 * - Vertex shader output: one of the values from \c gl_varying_slot.
238 * - Geometry shader input: one of the values from \c gl_varying_slot.
239 * - Geometry shader output: one of the values from \c gl_varying_slot.
240 * - Fragment shader input: one of the values from \c gl_varying_slot.
241 * - Fragment shader output: one of the values from \c gl_frag_result.
242 * - Uniforms: Per-stage uniform slot number for default uniform block.
243 * - Uniforms: Index within the uniform block definition for UBO members.
244 * - Non-UBO Uniforms: uniform slot number.
245 * - Other: This field is not currently used.
247 * If the variable is a uniform, shader input, or shader output, and the
248 * slot has not been assigned, the value will be -1.
253 * The actual location of the variable in the IR. Only valid for inputs
256 unsigned int driver_location
;
259 * output index for dual source blending.
264 * Descriptor set binding for sampler or UBO.
269 * Initial binding point for a sampler or UBO.
271 * For array types, this represents the binding point for the first element.
276 * Location an atomic counter is stored at.
281 * ARB_shader_image_load_store qualifiers.
284 bool read_only
; /**< "readonly" qualifier. */
285 bool write_only
; /**< "writeonly" qualifier. */
290 /** Image internal format if specified explicitly, otherwise GL_NONE. */
296 * Built-in state that backs this uniform
298 * Once set at variable creation, \c state_slots must remain invariant.
299 * This is because, ideally, this array would be shared by all clones of
300 * this variable in the IR tree. In other words, we'd really like for it
301 * to be a fly-weight.
303 * If the variable is not a uniform, \c num_state_slots will be zero and
304 * \c state_slots will be \c NULL.
307 unsigned num_state_slots
; /**< Number of state slots used */
308 nir_state_slot
*state_slots
; /**< State descriptors. */
312 * Constant expression assigned in the initializer of the variable
314 * This field should only be used temporarily by creators of NIR shaders
315 * and then lower_constant_initializers can be used to get rid of them.
316 * Most of the rest of NIR ignores this field or asserts that it's NULL.
318 nir_constant
*constant_initializer
;
321 * For variables that are in an interface block or are an instance of an
322 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
324 * \sa ir_variable::location
326 const struct glsl_type
*interface_type
;
329 #define nir_foreach_variable(var, var_list) \
330 foreach_list_typed(nir_variable, var, node, var_list)
332 #define nir_foreach_variable_safe(var, var_list) \
333 foreach_list_typed_safe(nir_variable, var, node, var_list)
336 nir_variable_is_global(const nir_variable
*var
)
338 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
341 typedef struct nir_register
{
342 struct exec_node node
;
344 unsigned num_components
; /** < number of vector components */
345 unsigned num_array_elems
; /** < size of array (0 for no array) */
347 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
350 /** generic register index. */
353 /** only for debug purposes, can be NULL */
356 /** whether this register is local (per-function) or global (per-shader) */
360 * If this flag is set to true, then accessing channels >= num_components
361 * is well-defined, and simply spills over to the next array element. This
362 * is useful for backends that can do per-component accessing, in
363 * particular scalar backends. By setting this flag and making
364 * num_components equal to 1, structures can be packed tightly into
365 * registers and then registers can be accessed per-component to get to
366 * each structure member, even if it crosses vec4 boundaries.
370 /** set of nir_src's where this register is used (read from) */
371 struct list_head uses
;
373 /** set of nir_dest's where this register is defined (written to) */
374 struct list_head defs
;
376 /** set of nir_if's where this register is used as a condition */
377 struct list_head if_uses
;
380 #define nir_foreach_register(reg, reg_list) \
381 foreach_list_typed(nir_register, reg, node, reg_list)
382 #define nir_foreach_register_safe(reg, reg_list) \
383 foreach_list_typed_safe(nir_register, reg, node, reg_list)
389 nir_instr_type_intrinsic
,
390 nir_instr_type_load_const
,
392 nir_instr_type_ssa_undef
,
394 nir_instr_type_parallel_copy
,
397 typedef struct nir_instr
{
398 struct exec_node node
;
400 struct nir_block
*block
;
402 /** generic instruction index. */
405 /* A temporary for optimization and analysis passes to use for storing
406 * flags. For instance, DCE uses this to store the "dead/live" info.
411 static inline nir_instr
*
412 nir_instr_next(nir_instr
*instr
)
414 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
415 if (exec_node_is_tail_sentinel(next
))
418 return exec_node_data(nir_instr
, next
, node
);
421 static inline nir_instr
*
422 nir_instr_prev(nir_instr
*instr
)
424 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
425 if (exec_node_is_head_sentinel(prev
))
428 return exec_node_data(nir_instr
, prev
, node
);
432 nir_instr_is_first(nir_instr
*instr
)
434 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
438 nir_instr_is_last(nir_instr
*instr
)
440 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
443 typedef struct nir_ssa_def
{
444 /** for debugging only, can be NULL */
447 /** generic SSA definition index. */
450 /** Index into the live_in and live_out bitfields */
453 nir_instr
*parent_instr
;
455 /** set of nir_instr's where this register is used (read from) */
456 struct list_head uses
;
458 /** set of nir_if's where this register is used as a condition */
459 struct list_head if_uses
;
461 uint8_t num_components
;
463 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
471 struct nir_src
*indirect
; /** < NULL for no indirect offset */
472 unsigned base_offset
;
474 /* TODO use-def chain goes here */
478 nir_instr
*parent_instr
;
479 struct list_head def_link
;
482 struct nir_src
*indirect
; /** < NULL for no indirect offset */
483 unsigned base_offset
;
485 /* TODO def-use chain goes here */
490 typedef struct nir_src
{
492 nir_instr
*parent_instr
;
493 struct nir_if
*parent_if
;
496 struct list_head use_link
;
506 static inline nir_src
509 nir_src src
= { { NULL
} };
513 #define NIR_SRC_INIT nir_src_init()
515 #define nir_foreach_use(src, reg_or_ssa_def) \
516 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
518 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
519 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
521 #define nir_foreach_if_use(src, reg_or_ssa_def) \
522 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
524 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
525 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
536 static inline nir_dest
539 nir_dest dest
= { { { NULL
} } };
543 #define NIR_DEST_INIT nir_dest_init()
545 #define nir_foreach_def(dest, reg) \
546 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
548 #define nir_foreach_def_safe(dest, reg) \
549 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
551 static inline nir_src
552 nir_src_for_ssa(nir_ssa_def
*def
)
554 nir_src src
= NIR_SRC_INIT
;
562 static inline nir_src
563 nir_src_for_reg(nir_register
*reg
)
565 nir_src src
= NIR_SRC_INIT
;
569 src
.reg
.indirect
= NULL
;
570 src
.reg
.base_offset
= 0;
575 static inline nir_dest
576 nir_dest_for_reg(nir_register
*reg
)
578 nir_dest dest
= NIR_DEST_INIT
;
585 static inline unsigned
586 nir_src_bit_size(nir_src src
)
588 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
591 static inline unsigned
592 nir_dest_bit_size(nir_dest dest
)
594 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
597 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
598 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
604 * \name input modifiers
608 * For inputs interpreted as floating point, flips the sign bit. For
609 * inputs interpreted as integers, performs the two's complement negation.
614 * Clears the sign bit for floating point values, and computes the integer
615 * absolute value for integers. Note that the negate modifier acts after
616 * the absolute value modifier, therefore if both are set then all inputs
617 * will become negative.
623 * For each input component, says which component of the register it is
624 * chosen from. Note that which elements of the swizzle are used and which
625 * are ignored are based on the write mask for most opcodes - for example,
626 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
627 * a swizzle of {2, x, 1, 0} where x means "don't care."
636 * \name saturate output modifier
638 * Only valid for opcodes that output floating-point numbers. Clamps the
639 * output to between 0.0 and 1.0 inclusive.
644 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
648 nir_type_invalid
= 0, /* Not a valid type */
653 nir_type_bool32
= 32 | nir_type_bool
,
654 nir_type_int8
= 8 | nir_type_int
,
655 nir_type_int16
= 16 | nir_type_int
,
656 nir_type_int32
= 32 | nir_type_int
,
657 nir_type_int64
= 64 | nir_type_int
,
658 nir_type_uint8
= 8 | nir_type_uint
,
659 nir_type_uint16
= 16 | nir_type_uint
,
660 nir_type_uint32
= 32 | nir_type_uint
,
661 nir_type_uint64
= 64 | nir_type_uint
,
662 nir_type_float16
= 16 | nir_type_float
,
663 nir_type_float32
= 32 | nir_type_float
,
664 nir_type_float64
= 64 | nir_type_float
,
667 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
668 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
670 static inline unsigned
671 nir_alu_type_get_type_size(nir_alu_type type
)
673 return type
& NIR_ALU_TYPE_SIZE_MASK
;
676 static inline unsigned
677 nir_alu_type_get_base_type(nir_alu_type type
)
679 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
682 static inline nir_alu_type
683 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
685 switch (glsl_get_base_type(type
)) {
687 return nir_type_bool32
;
690 return nir_type_uint32
;
693 return nir_type_int32
;
695 case GLSL_TYPE_UINT64
:
696 return nir_type_uint64
;
698 case GLSL_TYPE_INT64
:
699 return nir_type_int64
;
701 case GLSL_TYPE_FLOAT
:
702 return nir_type_float32
;
704 case GLSL_TYPE_DOUBLE
:
705 return nir_type_float64
;
708 unreachable("unknown type");
712 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
);
715 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
716 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
717 } nir_op_algebraic_property
;
725 * The number of components in the output
727 * If non-zero, this is the size of the output and input sizes are
728 * explicitly given; swizzle and writemask are still in effect, but if
729 * the output component is masked out, then the input component may
732 * If zero, the opcode acts in the standard, per-component manner; the
733 * operation is performed on each component (except the ones that are
734 * masked out) with the input being taken from the input swizzle for
737 * The size of some of the inputs may be given (i.e. non-zero) even
738 * though output_size is zero; in that case, the inputs with a zero
739 * size act per-component, while the inputs with non-zero size don't.
741 unsigned output_size
;
744 * The type of vector that the instruction outputs. Note that the
745 * staurate modifier is only allowed on outputs with the float type.
748 nir_alu_type output_type
;
751 * The number of components in each input
753 unsigned input_sizes
[4];
756 * The type of vector that each input takes. Note that negate and
757 * absolute value are only allowed on inputs with int or float type and
758 * behave differently on the two.
760 nir_alu_type input_types
[4];
762 nir_op_algebraic_property algebraic_properties
;
765 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
767 typedef struct nir_alu_instr
{
771 /** Indicates that this ALU instruction generates an exact value
773 * This is kind of a mixture of GLSL "precise" and "invariant" and not
774 * really equivalent to either. This indicates that the value generated by
775 * this operation is high-precision and any code transformations that touch
776 * it must ensure that the resulting value is bit-for-bit identical to the
785 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
786 nir_alu_instr
*instr
);
787 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
788 nir_alu_instr
*instr
);
790 /* is this source channel used? */
792 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
794 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
795 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
797 return (instr
->dest
.write_mask
>> channel
) & 1;
801 * For instructions whose destinations are SSA, get the number of channels
804 static inline unsigned
805 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
807 assert(instr
->dest
.dest
.is_ssa
);
809 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
810 return nir_op_infos
[instr
->op
].input_sizes
[src
];
812 return instr
->dest
.dest
.ssa
.num_components
;
815 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
816 unsigned src1
, unsigned src2
);
820 nir_deref_type_array
,
821 nir_deref_type_struct
824 typedef struct nir_deref
{
825 nir_deref_type deref_type
;
826 struct nir_deref
*child
;
827 const struct glsl_type
*type
;
836 /* This enum describes how the array is referenced. If the deref is
837 * direct then the base_offset is used. If the deref is indirect then
838 * offset is given by base_offset + indirect. If the deref is a wildcard
839 * then the deref refers to all of the elements of the array at the same
840 * time. Wildcard dereferences are only ever allowed in copy_var
841 * intrinsics and the source and destination derefs must have matching
845 nir_deref_array_type_direct
,
846 nir_deref_array_type_indirect
,
847 nir_deref_array_type_wildcard
,
848 } nir_deref_array_type
;
853 nir_deref_array_type deref_array_type
;
854 unsigned base_offset
;
864 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
,
865 deref_type
, nir_deref_type_var
)
866 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
,
867 deref_type
, nir_deref_type_array
)
868 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
,
869 deref_type
, nir_deref_type_struct
)
871 /* Returns the last deref in the chain. */
872 static inline nir_deref
*
873 nir_deref_tail(nir_deref
*deref
)
876 deref
= deref
->child
;
884 nir_deref_var
**params
;
885 nir_deref_var
*return_deref
;
887 struct nir_function
*callee
;
890 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
891 num_variables, num_indices, idx0, idx1, idx2, flags) \
892 nir_intrinsic_##name,
894 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
897 #include "nir_intrinsics.h"
898 nir_num_intrinsics
= nir_last_intrinsic
+ 1
901 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
903 /** Represents an intrinsic
905 * An intrinsic is an instruction type for handling things that are
906 * more-or-less regular operations but don't just consume and produce SSA
907 * values like ALU operations do. Intrinsics are not for things that have
908 * special semantic meaning such as phi nodes and parallel copies.
909 * Examples of intrinsics include variable load/store operations, system
910 * value loads, and the like. Even though texturing more-or-less falls
911 * under this category, texturing is its own instruction type because
912 * trying to represent texturing with intrinsics would lead to a
913 * combinatorial explosion of intrinsic opcodes.
915 * By having a single instruction type for handling a lot of different
916 * cases, optimization passes can look for intrinsics and, for the most
917 * part, completely ignore them. Each intrinsic type also has a few
918 * possible flags that govern whether or not they can be reordered or
919 * eliminated. That way passes like dead code elimination can still work
920 * on intrisics without understanding the meaning of each.
922 * Each intrinsic has some number of constant indices, some number of
923 * variables, and some number of sources. What these sources, variables,
924 * and indices mean depends on the intrinsic and is documented with the
925 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
926 * instructions are the only types of instruction that can operate on
932 nir_intrinsic_op intrinsic
;
936 /** number of components if this is a vectorized intrinsic
938 * Similarly to ALU operations, some intrinsics are vectorized.
939 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
940 * For vectorized intrinsics, the num_components field specifies the
941 * number of destination components and the number of source components
942 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
944 uint8_t num_components
;
946 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
948 nir_deref_var
*variables
[2];
951 } nir_intrinsic_instr
;
954 * \name NIR intrinsics semantic flags
956 * information about what the compiler can do with the intrinsics.
958 * \sa nir_intrinsic_info::flags
962 * whether the intrinsic can be safely eliminated if none of its output
963 * value is not being used.
965 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
968 * Whether the intrinsic can be reordered with respect to any other
969 * intrinsic, i.e. whether the only reordering dependencies of the
970 * intrinsic are due to the register reads/writes.
972 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
973 } nir_intrinsic_semantic_flag
;
976 * \name NIR intrinsics const-index flag
978 * Indicates the usage of a const_index slot.
980 * \sa nir_intrinsic_info::index_map
984 * Generally instructions that take a offset src argument, can encode
985 * a constant 'base' value which is added to the offset.
987 NIR_INTRINSIC_BASE
= 1,
990 * For store instructions, a writemask for the store.
992 NIR_INTRINSIC_WRMASK
= 2,
995 * The stream-id for GS emit_vertex/end_primitive intrinsics.
997 NIR_INTRINSIC_STREAM_ID
= 3,
1000 * The clip-plane id for load_user_clip_plane intrinsic.
1002 NIR_INTRINSIC_UCP_ID
= 4,
1005 * The amount of data, starting from BASE, that this instruction may
1006 * access. This is used to provide bounds if the offset is not constant.
1008 NIR_INTRINSIC_RANGE
= 5,
1011 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1013 NIR_INTRINSIC_DESC_SET
= 6,
1016 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1018 NIR_INTRINSIC_BINDING
= 7,
1023 NIR_INTRINSIC_COMPONENT
= 8,
1026 * Interpolation mode (only meaningful for FS inputs).
1028 NIR_INTRINSIC_INTERP_MODE
= 9,
1030 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1032 } nir_intrinsic_index_flag
;
1034 #define NIR_INTRINSIC_MAX_INPUTS 4
1039 unsigned num_srcs
; /** < number of register/SSA inputs */
1041 /** number of components of each input register
1043 * If this value is 0, the number of components is given by the
1044 * num_components field of nir_intrinsic_instr.
1046 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1050 /** number of components of the output register
1052 * If this value is 0, the number of components is given by the
1053 * num_components field of nir_intrinsic_instr.
1055 unsigned dest_components
;
1057 /** the number of inputs/outputs that are variables */
1058 unsigned num_variables
;
1060 /** the number of constant indices used by the intrinsic */
1061 unsigned num_indices
;
1063 /** indicates the usage of intr->const_index[n] */
1064 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1066 /** semantic flags for calls to this intrinsic */
1067 nir_intrinsic_semantic_flag flags
;
1068 } nir_intrinsic_info
;
1070 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1073 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1074 static inline type \
1075 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
1077 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1078 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1079 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1081 static inline void \
1082 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1084 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1085 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1086 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1089 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1090 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1091 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1092 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1093 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1094 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1095 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1096 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1097 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1100 * \group texture information
1102 * This gives semantic information about textures which is useful to the
1103 * frontend, the backend, and lowering passes, but not the optimizer.
1108 nir_tex_src_projector
,
1109 nir_tex_src_comparator
, /* shadow comparator */
1113 nir_tex_src_ms_index
, /* MSAA sample index */
1114 nir_tex_src_ms_mcs
, /* MSAA compression value */
1117 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1118 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1119 nir_tex_src_plane
, /* < selects plane for planar textures */
1120 nir_num_tex_src_types
1125 nir_tex_src_type src_type
;
1129 nir_texop_tex
, /**< Regular texture look-up */
1130 nir_texop_txb
, /**< Texture look-up with LOD bias */
1131 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1132 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1133 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1134 nir_texop_txf_ms
, /**< Multisample texture fetch */
1135 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1136 nir_texop_txs
, /**< Texture size */
1137 nir_texop_lod
, /**< Texture lod query */
1138 nir_texop_tg4
, /**< Texture gather */
1139 nir_texop_query_levels
, /**< Texture levels query */
1140 nir_texop_texture_samples
, /**< Texture samples query */
1141 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1149 enum glsl_sampler_dim sampler_dim
;
1150 nir_alu_type dest_type
;
1155 unsigned num_srcs
, coord_components
;
1156 bool is_array
, is_shadow
;
1159 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1160 * components or the new-style shadow that outputs 1 component.
1162 bool is_new_style_shadow
;
1164 /* gather component selector */
1165 unsigned component
: 2;
1167 /** The texture index
1169 * If this texture instruction has a nir_tex_src_texture_offset source,
1170 * then the texture index is given by texture_index + texture_offset.
1172 unsigned texture_index
;
1174 /** The size of the texture array or 0 if it's not an array */
1175 unsigned texture_array_size
;
1177 /** The texture deref
1179 * If this is null, use texture_index instead.
1181 nir_deref_var
*texture
;
1183 /** The sampler index
1185 * The following operations do not require a sampler and, as such, this
1186 * field should be ignored:
1188 * - nir_texop_txf_ms
1192 * - nir_texop_query_levels
1193 * - nir_texop_texture_samples
1194 * - nir_texop_samples_identical
1196 * If this texture instruction has a nir_tex_src_sampler_offset source,
1197 * then the sampler index is given by sampler_index + sampler_offset.
1199 unsigned sampler_index
;
1201 /** The sampler deref
1203 * If this is null, use sampler_index instead.
1205 nir_deref_var
*sampler
;
1208 static inline unsigned
1209 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1211 switch (instr
->op
) {
1212 case nir_texop_txs
: {
1214 switch (instr
->sampler_dim
) {
1215 case GLSL_SAMPLER_DIM_1D
:
1216 case GLSL_SAMPLER_DIM_BUF
:
1219 case GLSL_SAMPLER_DIM_2D
:
1220 case GLSL_SAMPLER_DIM_CUBE
:
1221 case GLSL_SAMPLER_DIM_MS
:
1222 case GLSL_SAMPLER_DIM_RECT
:
1223 case GLSL_SAMPLER_DIM_EXTERNAL
:
1224 case GLSL_SAMPLER_DIM_SUBPASS
:
1227 case GLSL_SAMPLER_DIM_3D
:
1231 unreachable("not reached");
1233 if (instr
->is_array
)
1241 case nir_texop_texture_samples
:
1242 case nir_texop_query_levels
:
1243 case nir_texop_samples_identical
:
1247 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1254 /* Returns true if this texture operation queries something about the texture
1255 * rather than actually sampling it.
1258 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1260 switch (instr
->op
) {
1263 case nir_texop_texture_samples
:
1264 case nir_texop_query_levels
:
1265 case nir_texop_txf_ms_mcs
:
1272 case nir_texop_txf_ms
:
1276 unreachable("Invalid texture opcode");
1280 static inline nir_alu_type
1281 nir_tex_instr_src_type(nir_tex_instr
*instr
, unsigned src
)
1283 switch (instr
->src
[src
].src_type
) {
1284 case nir_tex_src_coord
:
1285 switch (instr
->op
) {
1287 case nir_texop_txf_ms
:
1288 case nir_texop_txf_ms_mcs
:
1289 case nir_texop_samples_identical
:
1290 return nir_type_int
;
1293 return nir_type_float
;
1296 case nir_tex_src_lod
:
1297 switch (instr
->op
) {
1300 return nir_type_int
;
1303 return nir_type_float
;
1306 case nir_tex_src_projector
:
1307 case nir_tex_src_comparator
:
1308 case nir_tex_src_bias
:
1309 case nir_tex_src_ddx
:
1310 case nir_tex_src_ddy
:
1311 return nir_type_float
;
1313 case nir_tex_src_offset
:
1314 case nir_tex_src_ms_index
:
1315 case nir_tex_src_texture_offset
:
1316 case nir_tex_src_sampler_offset
:
1317 return nir_type_int
;
1320 unreachable("Invalid texture source type");
1324 static inline unsigned
1325 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1327 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1328 return instr
->coord_components
;
1330 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1331 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1334 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1335 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1336 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1337 if (instr
->is_array
)
1338 return instr
->coord_components
- 1;
1340 return instr
->coord_components
;
1347 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1349 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1350 if (instr
->src
[i
].src_type
== type
)
1356 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1361 nir_const_value value
;
1364 } nir_load_const_instr
;
1377 /* creates a new SSA variable in an undefined state */
1382 } nir_ssa_undef_instr
;
1385 struct exec_node node
;
1387 /* The predecessor block corresponding to this source */
1388 struct nir_block
*pred
;
1393 #define nir_foreach_phi_src(phi_src, phi) \
1394 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1395 #define nir_foreach_phi_src_safe(phi_src, phi) \
1396 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1401 struct exec_list srcs
; /** < list of nir_phi_src */
1407 struct exec_node node
;
1410 } nir_parallel_copy_entry
;
1412 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1413 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1418 /* A list of nir_parallel_copy_entry's. The sources of all of the
1419 * entries are copied to the corresponding destinations "in parallel".
1420 * In other words, if we have two entries: a -> b and b -> a, the values
1423 struct exec_list entries
;
1424 } nir_parallel_copy_instr
;
1426 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1427 type
, nir_instr_type_alu
)
1428 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1429 type
, nir_instr_type_call
)
1430 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1431 type
, nir_instr_type_jump
)
1432 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1433 type
, nir_instr_type_tex
)
1434 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1435 type
, nir_instr_type_intrinsic
)
1436 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1437 type
, nir_instr_type_load_const
)
1438 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1439 type
, nir_instr_type_ssa_undef
)
1440 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1441 type
, nir_instr_type_phi
)
1442 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1443 nir_parallel_copy_instr
, instr
,
1444 type
, nir_instr_type_parallel_copy
)
1449 * Control flow consists of a tree of control flow nodes, which include
1450 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1451 * instructions that always run start-to-finish. Each basic block also keeps
1452 * track of its successors (blocks which may run immediately after the current
1453 * block) and predecessors (blocks which could have run immediately before the
1454 * current block). Each function also has a start block and an end block which
1455 * all return statements point to (which is always empty). Together, all the
1456 * blocks with their predecessors and successors make up the control flow
1457 * graph (CFG) of the function. There are helpers that modify the tree of
1458 * control flow nodes while modifying the CFG appropriately; these should be
1459 * used instead of modifying the tree directly.
1466 nir_cf_node_function
1469 typedef struct nir_cf_node
{
1470 struct exec_node node
;
1471 nir_cf_node_type type
;
1472 struct nir_cf_node
*parent
;
1475 typedef struct nir_block
{
1476 nir_cf_node cf_node
;
1478 struct exec_list instr_list
; /** < list of nir_instr */
1480 /** generic block index; generated by nir_index_blocks */
1484 * Each block can only have up to 2 successors, so we put them in a simple
1485 * array - no need for anything more complicated.
1487 struct nir_block
*successors
[2];
1489 /* Set of nir_block predecessors in the CFG */
1490 struct set
*predecessors
;
1493 * this node's immediate dominator in the dominance tree - set to NULL for
1496 struct nir_block
*imm_dom
;
1498 /* This node's children in the dominance tree */
1499 unsigned num_dom_children
;
1500 struct nir_block
**dom_children
;
1502 /* Set of nir_block's on the dominance frontier of this block */
1503 struct set
*dom_frontier
;
1506 * These two indices have the property that dom_{pre,post}_index for each
1507 * child of this block in the dominance tree will always be between
1508 * dom_pre_index and dom_post_index for this block, which makes testing if
1509 * a given block is dominated by another block an O(1) operation.
1511 unsigned dom_pre_index
, dom_post_index
;
1513 /* live in and out for this block; used for liveness analysis */
1514 BITSET_WORD
*live_in
;
1515 BITSET_WORD
*live_out
;
1518 static inline nir_instr
*
1519 nir_block_first_instr(nir_block
*block
)
1521 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1522 return exec_node_data(nir_instr
, head
, node
);
1525 static inline nir_instr
*
1526 nir_block_last_instr(nir_block
*block
)
1528 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1529 return exec_node_data(nir_instr
, tail
, node
);
1532 #define nir_foreach_instr(instr, block) \
1533 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1534 #define nir_foreach_instr_reverse(instr, block) \
1535 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1536 #define nir_foreach_instr_safe(instr, block) \
1537 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1538 #define nir_foreach_instr_reverse_safe(instr, block) \
1539 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1541 typedef struct nir_if
{
1542 nir_cf_node cf_node
;
1545 struct exec_list then_list
; /** < list of nir_cf_node */
1546 struct exec_list else_list
; /** < list of nir_cf_node */
1552 nir_instr
*conditional_instr
;
1554 nir_block
*break_block
;
1555 nir_block
*continue_from_block
;
1557 bool continue_from_then
;
1559 struct list_head loop_terminator_link
;
1560 } nir_loop_terminator
;
1563 /* Number of instructions in the loop */
1564 unsigned num_instructions
;
1566 /* How many times the loop is run (if known) */
1567 unsigned trip_count
;
1568 bool is_trip_count_known
;
1570 /* Unroll the loop regardless of its size */
1573 nir_loop_terminator
*limiting_terminator
;
1575 /* A list of loop_terminators terminating this loop. */
1576 struct list_head loop_terminator_list
;
1580 nir_cf_node cf_node
;
1582 struct exec_list body
; /** < list of nir_cf_node */
1584 nir_loop_info
*info
;
1588 * Various bits of metadata that can may be created or required by
1589 * optimization and analysis passes
1592 nir_metadata_none
= 0x0,
1593 nir_metadata_block_index
= 0x1,
1594 nir_metadata_dominance
= 0x2,
1595 nir_metadata_live_ssa_defs
= 0x4,
1596 nir_metadata_not_properly_reset
= 0x8,
1597 nir_metadata_loop_analysis
= 0x10,
1601 nir_cf_node cf_node
;
1603 /** pointer to the function of which this is an implementation */
1604 struct nir_function
*function
;
1606 struct exec_list body
; /** < list of nir_cf_node */
1608 nir_block
*end_block
;
1610 /** list for all local variables in the function */
1611 struct exec_list locals
;
1613 /** array of variables used as parameters */
1614 unsigned num_params
;
1615 nir_variable
**params
;
1617 /** variable used to hold the result of the function */
1618 nir_variable
*return_var
;
1620 /** list of local registers in the function */
1621 struct exec_list registers
;
1623 /** next available local register index */
1626 /** next available SSA value index */
1629 /* total number of basic blocks, only valid when block_index_dirty = false */
1630 unsigned num_blocks
;
1632 nir_metadata valid_metadata
;
1633 } nir_function_impl
;
1635 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1636 nir_start_block(nir_function_impl
*impl
)
1638 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1641 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1642 nir_impl_last_block(nir_function_impl
*impl
)
1644 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1647 static inline nir_cf_node
*
1648 nir_cf_node_next(nir_cf_node
*node
)
1650 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1651 if (exec_node_is_tail_sentinel(next
))
1654 return exec_node_data(nir_cf_node
, next
, node
);
1657 static inline nir_cf_node
*
1658 nir_cf_node_prev(nir_cf_node
*node
)
1660 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1661 if (exec_node_is_head_sentinel(prev
))
1664 return exec_node_data(nir_cf_node
, prev
, node
);
1668 nir_cf_node_is_first(const nir_cf_node
*node
)
1670 return exec_node_is_head_sentinel(node
->node
.prev
);
1674 nir_cf_node_is_last(const nir_cf_node
*node
)
1676 return exec_node_is_tail_sentinel(node
->node
.next
);
1679 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1680 type
, nir_cf_node_block
)
1681 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1682 type
, nir_cf_node_if
)
1683 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1684 type
, nir_cf_node_loop
)
1685 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1686 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1688 static inline nir_block
*
1689 nir_if_first_then_block(nir_if
*if_stmt
)
1691 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1692 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1695 static inline nir_block
*
1696 nir_if_last_then_block(nir_if
*if_stmt
)
1698 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1699 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1702 static inline nir_block
*
1703 nir_if_first_else_block(nir_if
*if_stmt
)
1705 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1706 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1709 static inline nir_block
*
1710 nir_if_last_else_block(nir_if
*if_stmt
)
1712 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1713 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1716 static inline nir_block
*
1717 nir_loop_first_block(nir_loop
*loop
)
1719 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1720 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1723 static inline nir_block
*
1724 nir_loop_last_block(nir_loop
*loop
)
1726 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1727 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1733 nir_parameter_inout
,
1734 } nir_parameter_type
;
1737 nir_parameter_type param_type
;
1738 const struct glsl_type
*type
;
1741 typedef struct nir_function
{
1742 struct exec_node node
;
1745 struct nir_shader
*shader
;
1747 unsigned num_params
;
1748 nir_parameter
*params
;
1749 const struct glsl_type
*return_type
;
1751 /** The implementation of this function.
1753 * If the function is only declared and not implemented, this is NULL.
1755 nir_function_impl
*impl
;
1758 typedef struct nir_shader_compiler_options
{
1763 /** Lowers flrp when it does not support doubles */
1770 bool lower_bitfield_extract
;
1771 bool lower_bitfield_insert
;
1772 bool lower_uadd_carry
;
1773 bool lower_usub_borrow
;
1774 /** lowers fneg and ineg to fsub and isub. */
1776 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1779 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1782 /** enables rules to lower idiv by power-of-two: */
1785 /* Does the native fdot instruction replicate its result for four
1786 * components? If so, then opt_algebraic_late will turn all fdotN
1787 * instructions into fdot_replicatedN instructions.
1789 bool fdot_replicates
;
1791 /** lowers ffract to fsub+ffloor: */
1794 bool lower_pack_half_2x16
;
1795 bool lower_pack_unorm_2x16
;
1796 bool lower_pack_snorm_2x16
;
1797 bool lower_pack_unorm_4x8
;
1798 bool lower_pack_snorm_4x8
;
1799 bool lower_unpack_half_2x16
;
1800 bool lower_unpack_unorm_2x16
;
1801 bool lower_unpack_snorm_2x16
;
1802 bool lower_unpack_unorm_4x8
;
1803 bool lower_unpack_snorm_4x8
;
1805 bool lower_extract_byte
;
1806 bool lower_extract_word
;
1809 * Does the driver support real 32-bit integers? (Otherwise, integers
1810 * are simulated by floats.)
1812 bool native_integers
;
1814 /* Indicates that the driver only has zero-based vertex id */
1815 bool vertex_id_zero_based
;
1817 bool lower_cs_local_index_from_id
;
1820 * Should nir_lower_io() create load_interpolated_input intrinsics?
1822 * If not, it generates regular load_input intrinsics and interpolation
1823 * information must be inferred from the list of input nir_variables.
1825 bool use_interpolated_input_intrinsics
;
1827 unsigned max_unroll_iterations
;
1828 } nir_shader_compiler_options
;
1830 typedef struct nir_shader
{
1831 /** list of uniforms (nir_variable) */
1832 struct exec_list uniforms
;
1834 /** list of inputs (nir_variable) */
1835 struct exec_list inputs
;
1837 /** list of outputs (nir_variable) */
1838 struct exec_list outputs
;
1840 /** list of shared compute variables (nir_variable) */
1841 struct exec_list shared
;
1843 /** Set of driver-specific options for the shader.
1845 * The memory for the options is expected to be kept in a single static
1846 * copy by the driver.
1848 const struct nir_shader_compiler_options
*options
;
1850 /** Various bits of compile-time information about a given shader */
1851 struct shader_info
*info
;
1853 /** list of global variables in the shader (nir_variable) */
1854 struct exec_list globals
;
1856 /** list of system value variables in the shader (nir_variable) */
1857 struct exec_list system_values
;
1859 struct exec_list functions
; /** < list of nir_function */
1861 /** list of global register in the shader */
1862 struct exec_list registers
;
1864 /** next available global register index */
1868 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1871 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1873 /** The shader stage, such as MESA_SHADER_VERTEX. */
1874 gl_shader_stage stage
;
1877 static inline nir_function_impl
*
1878 nir_shader_get_entrypoint(nir_shader
*shader
)
1880 assert(exec_list_length(&shader
->functions
) == 1);
1881 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1882 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1883 assert(func
->return_type
== glsl_void_type());
1884 assert(func
->num_params
== 0);
1889 #define nir_foreach_function(func, shader) \
1890 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1892 nir_shader
*nir_shader_create(void *mem_ctx
,
1893 gl_shader_stage stage
,
1894 const nir_shader_compiler_options
*options
,
1897 /** creates a register, including assigning it an index and adding it to the list */
1898 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1900 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1902 void nir_reg_remove(nir_register
*reg
);
1904 /** Adds a variable to the appropreate list in nir_shader */
1905 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1908 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1910 assert(var
->data
.mode
== nir_var_local
);
1911 exec_list_push_tail(&impl
->locals
, &var
->node
);
1914 /** creates a variable, sets a few defaults, and adds it to the list */
1915 nir_variable
*nir_variable_create(nir_shader
*shader
,
1916 nir_variable_mode mode
,
1917 const struct glsl_type
*type
,
1919 /** creates a local variable and adds it to the list */
1920 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1921 const struct glsl_type
*type
,
1924 /** creates a function and adds it to the shader's list of functions */
1925 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1927 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1928 /** creates a function_impl that isn't tied to any particular function */
1929 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1931 nir_block
*nir_block_create(nir_shader
*shader
);
1932 nir_if
*nir_if_create(nir_shader
*shader
);
1933 nir_loop
*nir_loop_create(nir_shader
*shader
);
1935 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1937 /** requests that the given pieces of metadata be generated */
1938 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
1939 /** dirties all but the preserved metadata */
1940 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1942 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1943 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1945 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1947 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1948 unsigned num_components
,
1951 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1952 nir_intrinsic_op op
);
1954 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1955 nir_function
*callee
);
1957 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1959 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1961 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1963 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1964 unsigned num_components
,
1967 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1968 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1969 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1971 typedef bool (*nir_deref_foreach_leaf_cb
)(nir_deref_var
*deref
, void *state
);
1972 bool nir_deref_foreach_leaf(nir_deref_var
*deref
,
1973 nir_deref_foreach_leaf_cb cb
, void *state
);
1975 nir_load_const_instr
*
1976 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1979 * NIR Cursors and Instruction Insertion API
1982 * A tiny struct representing a point to insert/extract instructions or
1983 * control flow nodes. Helps reduce the combinatorial explosion of possible
1984 * points to insert/extract.
1986 * \sa nir_control_flow.h
1989 nir_cursor_before_block
,
1990 nir_cursor_after_block
,
1991 nir_cursor_before_instr
,
1992 nir_cursor_after_instr
,
1993 } nir_cursor_option
;
1996 nir_cursor_option option
;
2003 static inline nir_block
*
2004 nir_cursor_current_block(nir_cursor cursor
)
2006 if (cursor
.option
== nir_cursor_before_instr
||
2007 cursor
.option
== nir_cursor_after_instr
) {
2008 return cursor
.instr
->block
;
2010 return cursor
.block
;
2014 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2016 static inline nir_cursor
2017 nir_before_block(nir_block
*block
)
2020 cursor
.option
= nir_cursor_before_block
;
2021 cursor
.block
= block
;
2025 static inline nir_cursor
2026 nir_after_block(nir_block
*block
)
2029 cursor
.option
= nir_cursor_after_block
;
2030 cursor
.block
= block
;
2034 static inline nir_cursor
2035 nir_before_instr(nir_instr
*instr
)
2038 cursor
.option
= nir_cursor_before_instr
;
2039 cursor
.instr
= instr
;
2043 static inline nir_cursor
2044 nir_after_instr(nir_instr
*instr
)
2047 cursor
.option
= nir_cursor_after_instr
;
2048 cursor
.instr
= instr
;
2052 static inline nir_cursor
2053 nir_after_block_before_jump(nir_block
*block
)
2055 nir_instr
*last_instr
= nir_block_last_instr(block
);
2056 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2057 return nir_before_instr(last_instr
);
2059 return nir_after_block(block
);
2063 static inline nir_cursor
2064 nir_before_cf_node(nir_cf_node
*node
)
2066 if (node
->type
== nir_cf_node_block
)
2067 return nir_before_block(nir_cf_node_as_block(node
));
2069 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2072 static inline nir_cursor
2073 nir_after_cf_node(nir_cf_node
*node
)
2075 if (node
->type
== nir_cf_node_block
)
2076 return nir_after_block(nir_cf_node_as_block(node
));
2078 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2081 static inline nir_cursor
2082 nir_after_phis(nir_block
*block
)
2084 nir_foreach_instr(instr
, block
) {
2085 if (instr
->type
!= nir_instr_type_phi
)
2086 return nir_before_instr(instr
);
2088 return nir_after_block(block
);
2091 static inline nir_cursor
2092 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2094 if (node
->type
== nir_cf_node_block
)
2095 return nir_after_block(nir_cf_node_as_block(node
));
2097 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2099 return nir_after_phis(block
);
2102 static inline nir_cursor
2103 nir_before_cf_list(struct exec_list
*cf_list
)
2105 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2106 exec_list_get_head(cf_list
), node
);
2107 return nir_before_cf_node(first_node
);
2110 static inline nir_cursor
2111 nir_after_cf_list(struct exec_list
*cf_list
)
2113 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2114 exec_list_get_tail(cf_list
), node
);
2115 return nir_after_cf_node(last_node
);
2119 * Insert a NIR instruction at the given cursor.
2121 * Note: This does not update the cursor.
2123 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2126 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2128 nir_instr_insert(nir_before_instr(instr
), before
);
2132 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2134 nir_instr_insert(nir_after_instr(instr
), after
);
2138 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2140 nir_instr_insert(nir_before_block(block
), before
);
2144 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2146 nir_instr_insert(nir_after_block(block
), after
);
2150 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2152 nir_instr_insert(nir_before_cf_node(node
), before
);
2156 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2158 nir_instr_insert(nir_after_cf_node(node
), after
);
2162 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2164 nir_instr_insert(nir_before_cf_list(list
), before
);
2168 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2170 nir_instr_insert(nir_after_cf_list(list
), after
);
2173 void nir_instr_remove(nir_instr
*instr
);
2177 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2178 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2179 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2180 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2182 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2183 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2185 nir_const_value
*nir_src_as_const_value(nir_src src
);
2186 bool nir_src_is_dynamically_uniform(nir_src src
);
2187 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2188 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2189 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2190 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2191 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2194 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2195 unsigned num_components
, unsigned bit_size
,
2197 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2198 unsigned num_components
, unsigned bit_size
,
2200 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2201 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2202 nir_instr
*after_me
);
2204 uint8_t nir_ssa_def_components_read(nir_ssa_def
*def
);
2207 * finds the next basic block in source-code order, returns NULL if there is
2211 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2213 /* Performs the opposite of nir_block_cf_tree_next() */
2215 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2217 /* Gets the first block in a CF node in source-code order */
2219 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2221 /* Gets the last block in a CF node in source-code order */
2223 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2225 /* Gets the next block after a CF node in source-code order */
2227 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2229 /* Macros for loops that visit blocks in source-code order */
2231 #define nir_foreach_block(block, impl) \
2232 for (nir_block *block = nir_start_block(impl); block != NULL; \
2233 block = nir_block_cf_tree_next(block))
2235 #define nir_foreach_block_safe(block, impl) \
2236 for (nir_block *block = nir_start_block(impl), \
2237 *next = nir_block_cf_tree_next(block); \
2239 block = next, next = nir_block_cf_tree_next(block))
2241 #define nir_foreach_block_reverse(block, impl) \
2242 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2243 block = nir_block_cf_tree_prev(block))
2245 #define nir_foreach_block_reverse_safe(block, impl) \
2246 for (nir_block *block = nir_impl_last_block(impl), \
2247 *prev = nir_block_cf_tree_prev(block); \
2249 block = prev, prev = nir_block_cf_tree_prev(block))
2251 #define nir_foreach_block_in_cf_node(block, node) \
2252 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2253 block != nir_cf_node_cf_tree_next(node); \
2254 block = nir_block_cf_tree_next(block))
2256 /* If the following CF node is an if, this function returns that if.
2257 * Otherwise, it returns NULL.
2259 nir_if
*nir_block_get_following_if(nir_block
*block
);
2261 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2263 void nir_index_local_regs(nir_function_impl
*impl
);
2264 void nir_index_global_regs(nir_shader
*shader
);
2265 void nir_index_ssa_defs(nir_function_impl
*impl
);
2266 unsigned nir_index_instrs(nir_function_impl
*impl
);
2268 void nir_index_blocks(nir_function_impl
*impl
);
2270 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2271 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2272 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2274 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2275 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2276 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2277 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2278 nir_deref
*nir_deref_clone(const nir_deref
*deref
, void *mem_ctx
);
2279 nir_deref_var
*nir_deref_var_clone(const nir_deref_var
*deref
, void *mem_ctx
);
2282 void nir_validate_shader(nir_shader
*shader
);
2283 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2284 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2287 should_clone_nir(void)
2289 static int should_clone
= -1;
2290 if (should_clone
< 0)
2291 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2293 return should_clone
;
2296 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2297 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2298 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2299 static inline bool should_clone_nir(void) { return false; }
2302 #define _PASS(nir, do_pass) do { \
2304 nir_validate_shader(nir); \
2305 if (should_clone_nir()) { \
2306 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2312 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2313 nir_metadata_set_validation_flag(nir); \
2314 if (pass(nir, ##__VA_ARGS__)) { \
2316 nir_metadata_check_validation_flag(nir); \
2320 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2321 pass(nir, ##__VA_ARGS__); \
2324 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2325 void nir_calc_dominance(nir_shader
*shader
);
2327 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2328 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2330 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2331 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2333 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2334 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2336 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2337 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2339 int nir_gs_count_vertices(const nir_shader
*shader
);
2341 bool nir_split_var_copies(nir_shader
*shader
);
2343 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2344 bool nir_lower_returns(nir_shader
*shader
);
2346 bool nir_inline_functions(nir_shader
*shader
);
2348 bool nir_propagate_invariant(nir_shader
*shader
);
2350 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2351 void nir_lower_var_copies(nir_shader
*shader
);
2353 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2355 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2357 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2359 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2360 nir_function_impl
*entrypoint
,
2361 bool outputs
, bool inputs
);
2363 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2365 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2366 int (*type_size
)(const struct glsl_type
*));
2369 /* If set, this forces all non-flat fragment shader inputs to be
2370 * interpolated as if with the "sample" qualifier. This requires
2371 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2373 nir_lower_io_force_sample_interpolation
= (1 << 1),
2374 } nir_lower_io_options
;
2375 void nir_lower_io(nir_shader
*shader
,
2376 nir_variable_mode modes
,
2377 int (*type_size
)(const struct glsl_type
*),
2378 nir_lower_io_options
);
2379 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2380 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2382 bool nir_is_per_vertex_io(nir_variable
*var
, gl_shader_stage stage
);
2384 void nir_lower_io_types(nir_shader
*shader
);
2385 void nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2386 void nir_lower_regs_to_ssa(nir_shader
*shader
);
2387 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2389 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2390 bool nir_lower_constant_initializers(nir_shader
*shader
,
2391 nir_variable_mode modes
);
2393 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2394 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2395 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2396 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2398 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2399 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2401 void nir_lower_samplers(nir_shader
*shader
,
2402 const struct gl_shader_program
*shader_program
);
2404 bool nir_lower_system_values(nir_shader
*shader
);
2406 typedef struct nir_lower_tex_options
{
2408 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2409 * sampler types a texture projector is lowered.
2414 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2416 bool lower_txf_offset
;
2419 * If true, lower away nir_tex_src_offset for all rect textures.
2421 bool lower_rect_offset
;
2424 * If true, lower rect textures to 2D, using txs to fetch the
2425 * texture dimensions and dividing the texture coords by the
2426 * texture dims to normalize.
2431 * If true, convert yuv to rgb.
2433 unsigned lower_y_uv_external
;
2434 unsigned lower_y_u_v_external
;
2435 unsigned lower_yx_xuxv_external
;
2438 * To emulate certain texture wrap modes, this can be used
2439 * to saturate the specified tex coord to [0.0, 1.0]. The
2440 * bits are according to sampler #, ie. if, for example:
2442 * (conf->saturate_s & (1 << n))
2444 * is true, then the s coord for sampler n is saturated.
2446 * Note that clamping must happen *after* projector lowering
2447 * so any projected texture sample instruction with a clamped
2448 * coordinate gets automatically lowered, regardless of the
2449 * 'lower_txp' setting.
2451 unsigned saturate_s
;
2452 unsigned saturate_t
;
2453 unsigned saturate_r
;
2455 /* Bitmask of textures that need swizzling.
2457 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2458 * swizzles[texture_index] is applied to the result of the texturing
2461 unsigned swizzle_result
;
2463 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2464 * while 4 and 5 represent 0 and 1 respectively.
2466 uint8_t swizzles
[32][4];
2469 * Bitmap of textures that need srgb to linear conversion. If
2470 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2471 * of the texture are lowered to linear.
2473 unsigned lower_srgb
;
2476 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2478 bool lower_txd_cube_map
;
2481 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
2482 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
2483 * with lower_txd_cube_map.
2485 bool lower_txd_shadow
;
2486 } nir_lower_tex_options
;
2488 bool nir_lower_tex(nir_shader
*shader
,
2489 const nir_lower_tex_options
*options
);
2491 bool nir_lower_idiv(nir_shader
*shader
);
2493 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2494 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2495 void nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2497 void nir_lower_two_sided_color(nir_shader
*shader
);
2499 void nir_lower_clamp_color_outputs(nir_shader
*shader
);
2501 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2502 void nir_lower_tes_patch_vertices(nir_shader
*tes
, unsigned patch_vertices
);
2504 typedef struct nir_lower_wpos_ytransform_options
{
2505 int state_tokens
[5];
2506 bool fs_coord_origin_upper_left
:1;
2507 bool fs_coord_origin_lower_left
:1;
2508 bool fs_coord_pixel_center_integer
:1;
2509 bool fs_coord_pixel_center_half_integer
:1;
2510 } nir_lower_wpos_ytransform_options
;
2512 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2513 const nir_lower_wpos_ytransform_options
*options
);
2514 bool nir_lower_wpos_center(nir_shader
*shader
);
2516 typedef struct nir_lower_drawpixels_options
{
2517 int texcoord_state_tokens
[5];
2518 int scale_state_tokens
[5];
2519 int bias_state_tokens
[5];
2520 unsigned drawpix_sampler
;
2521 unsigned pixelmap_sampler
;
2523 bool scale_and_bias
:1;
2524 } nir_lower_drawpixels_options
;
2526 void nir_lower_drawpixels(nir_shader
*shader
,
2527 const nir_lower_drawpixels_options
*options
);
2529 typedef struct nir_lower_bitmap_options
{
2532 } nir_lower_bitmap_options
;
2534 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2536 void nir_lower_atomics(nir_shader
*shader
,
2537 const struct gl_shader_program
*shader_program
);
2538 void nir_lower_to_source_mods(nir_shader
*shader
);
2540 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2543 nir_lower_imul64
= (1 << 0),
2544 nir_lower_isign64
= (1 << 1),
2545 /** Lower all int64 modulus and division opcodes */
2546 nir_lower_divmod64
= (1 << 2),
2547 } nir_lower_int64_options
;
2549 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
2552 nir_lower_drcp
= (1 << 0),
2553 nir_lower_dsqrt
= (1 << 1),
2554 nir_lower_drsq
= (1 << 2),
2555 nir_lower_dtrunc
= (1 << 3),
2556 nir_lower_dfloor
= (1 << 4),
2557 nir_lower_dceil
= (1 << 5),
2558 nir_lower_dfract
= (1 << 6),
2559 nir_lower_dround_even
= (1 << 7),
2560 nir_lower_dmod
= (1 << 8)
2561 } nir_lower_doubles_options
;
2563 void nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2564 void nir_lower_64bit_pack(nir_shader
*shader
);
2566 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2568 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2570 void nir_loop_analyze_impl(nir_function_impl
*impl
,
2571 nir_variable_mode indirect_mask
);
2573 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2575 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2576 bool nir_repair_ssa(nir_shader
*shader
);
2578 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
2580 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2581 * registers. If false, convert all values (even those not involved in a phi
2582 * node) to registers.
2584 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2586 bool nir_lower_phis_to_regs_block(nir_block
*block
);
2587 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
2589 bool nir_opt_algebraic(nir_shader
*shader
);
2590 bool nir_opt_algebraic_late(nir_shader
*shader
);
2591 bool nir_opt_constant_folding(nir_shader
*shader
);
2593 bool nir_opt_global_to_local(nir_shader
*shader
);
2595 bool nir_copy_prop(nir_shader
*shader
);
2597 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
2599 bool nir_opt_cse(nir_shader
*shader
);
2601 bool nir_opt_dce(nir_shader
*shader
);
2603 bool nir_opt_dead_cf(nir_shader
*shader
);
2605 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2607 bool nir_opt_if(nir_shader
*shader
);
2609 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
2611 bool nir_opt_move_comparisons(nir_shader
*shader
);
2613 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2615 bool nir_opt_remove_phis(nir_shader
*shader
);
2617 bool nir_opt_trivial_continues(nir_shader
*shader
);
2619 bool nir_opt_undef(nir_shader
*shader
);
2621 bool nir_opt_conditional_discard(nir_shader
*shader
);
2623 void nir_sweep(nir_shader
*shader
);
2625 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2626 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);