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 struct gl_shader_program
;
62 #define NIR_TRUE (~0u)
64 /** Defines a cast function
66 * This macro defines a cast function from in_type to out_type where
67 * out_type is some structure type that contains a field of type out_type.
69 * Note that you have to be a bit careful as the generated cast function
72 #define NIR_DEFINE_CAST(name, in_type, out_type, field, \
73 type_field, type_value) \
74 static inline out_type * \
75 name(const in_type *parent) \
77 assert(parent && parent->type_field == type_value); \
78 return exec_node_data(out_type, parent, field); \
87 * Description of built-in state associated with a uniform
89 * \sa nir_variable::state_slots
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_param
= (1 << 7),
105 nir_var_shared
= (1 << 8),
123 typedef struct nir_constant
{
125 * Value of the constant.
127 * The field used to back the values supplied by the constant is determined
128 * by the type associated with the \c nir_variable. Constants may be
129 * scalars, vectors, or matrices.
131 nir_const_value values
[4];
133 /* we could get this from the var->type but makes clone *much* easier to
134 * not have to care about the type.
136 unsigned num_elements
;
138 /* Array elements / Structure Fields */
139 struct nir_constant
**elements
;
143 * \brief Layout qualifiers for gl_FragDepth.
145 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
146 * with a layout qualifier.
149 nir_depth_layout_none
, /**< No depth layout is specified. */
150 nir_depth_layout_any
,
151 nir_depth_layout_greater
,
152 nir_depth_layout_less
,
153 nir_depth_layout_unchanged
157 * Either a uniform, global variable, shader input, or shader output. Based on
158 * ir_variable - it should be easy to translate between the two.
161 typedef struct nir_variable
{
162 struct exec_node node
;
165 * Declared type of the variable
167 const struct glsl_type
*type
;
170 * Declared name of the variable
174 struct nir_variable_data
{
176 * Storage class of the variable.
178 * \sa nir_variable_mode
180 nir_variable_mode mode
;
183 * Is the variable read-only?
185 * This is set for variables declared as \c const, shader inputs,
188 unsigned read_only
:1;
192 unsigned invariant
:1;
195 * Interpolation mode for shader inputs / outputs
197 * \sa glsl_interp_mode
199 unsigned interpolation
:2;
202 * \name ARB_fragment_coord_conventions
205 unsigned origin_upper_left
:1;
206 unsigned pixel_center_integer
:1;
210 * If non-zero, then this variable may be packed along with other variables
211 * into a single varying slot, so this offset should be applied when
212 * accessing components. For example, an offset of 1 means that the x
213 * component of this variable is actually stored in component y of the
214 * location specified by \c location.
216 unsigned location_frac
:2;
219 * If true, this variable represents an array of scalars that should
220 * be tightly packed. In other words, consecutive array elements
221 * should be stored one component apart, rather than one slot apart.
226 * Whether this is a fragment shader output implicitly initialized with
227 * the previous contents of the specified render target at the
228 * framebuffer location corresponding to this shader invocation.
230 unsigned fb_fetch_output
:1;
233 * \brief Layout qualifier for gl_FragDepth.
235 * This is not equal to \c ir_depth_layout_none if and only if this
236 * variable is \c gl_FragDepth and a layout qualifier is specified.
238 nir_depth_layout depth_layout
;
241 * Storage location of the base of this variable
243 * The precise meaning of this field depends on the nature of the variable.
245 * - Vertex shader input: one of the values from \c gl_vert_attrib.
246 * - Vertex shader output: one of the values from \c gl_varying_slot.
247 * - Geometry shader input: one of the values from \c gl_varying_slot.
248 * - Geometry shader output: one of the values from \c gl_varying_slot.
249 * - Fragment shader input: one of the values from \c gl_varying_slot.
250 * - Fragment shader output: one of the values from \c gl_frag_result.
251 * - Uniforms: Per-stage uniform slot number for default uniform block.
252 * - Uniforms: Index within the uniform block definition for UBO members.
253 * - Non-UBO Uniforms: uniform slot number.
254 * - Other: This field is not currently used.
256 * If the variable is a uniform, shader input, or shader output, and the
257 * slot has not been assigned, the value will be -1.
262 * The actual location of the variable in the IR. Only valid for inputs
265 unsigned int driver_location
;
268 * output index for dual source blending.
273 * Descriptor set binding for sampler or UBO.
278 * Initial binding point for a sampler or UBO.
280 * For array types, this represents the binding point for the first element.
285 * Location an atomic counter is stored at.
290 * ARB_shader_image_load_store qualifiers.
293 bool read_only
; /**< "readonly" qualifier. */
294 bool write_only
; /**< "writeonly" qualifier. */
299 /** Image internal format if specified explicitly, otherwise GL_NONE. */
305 * Built-in state that backs this uniform
307 * Once set at variable creation, \c state_slots must remain invariant.
308 * This is because, ideally, this array would be shared by all clones of
309 * this variable in the IR tree. In other words, we'd really like for it
310 * to be a fly-weight.
312 * If the variable is not a uniform, \c num_state_slots will be zero and
313 * \c state_slots will be \c NULL.
316 unsigned num_state_slots
; /**< Number of state slots used */
317 nir_state_slot
*state_slots
; /**< State descriptors. */
321 * Constant expression assigned in the initializer of the variable
323 * This field should only be used temporarily by creators of NIR shaders
324 * and then lower_constant_initializers can be used to get rid of them.
325 * Most of the rest of NIR ignores this field or asserts that it's NULL.
327 nir_constant
*constant_initializer
;
330 * For variables that are in an interface block or are an instance of an
331 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
333 * \sa ir_variable::location
335 const struct glsl_type
*interface_type
;
338 #define nir_foreach_variable(var, var_list) \
339 foreach_list_typed(nir_variable, var, node, var_list)
341 #define nir_foreach_variable_safe(var, var_list) \
342 foreach_list_typed_safe(nir_variable, var, node, var_list)
345 nir_variable_is_global(const nir_variable
*var
)
347 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
350 typedef struct nir_register
{
351 struct exec_node node
;
353 unsigned num_components
; /** < number of vector components */
354 unsigned num_array_elems
; /** < size of array (0 for no array) */
356 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
359 /** generic register index. */
362 /** only for debug purposes, can be NULL */
365 /** whether this register is local (per-function) or global (per-shader) */
369 * If this flag is set to true, then accessing channels >= num_components
370 * is well-defined, and simply spills over to the next array element. This
371 * is useful for backends that can do per-component accessing, in
372 * particular scalar backends. By setting this flag and making
373 * num_components equal to 1, structures can be packed tightly into
374 * registers and then registers can be accessed per-component to get to
375 * each structure member, even if it crosses vec4 boundaries.
379 /** set of nir_srcs where this register is used (read from) */
380 struct list_head uses
;
382 /** set of nir_dests where this register is defined (written to) */
383 struct list_head defs
;
385 /** set of nir_ifs where this register is used as a condition */
386 struct list_head if_uses
;
389 #define nir_foreach_register(reg, reg_list) \
390 foreach_list_typed(nir_register, reg, node, reg_list)
391 #define nir_foreach_register_safe(reg, reg_list) \
392 foreach_list_typed_safe(nir_register, reg, node, reg_list)
398 nir_instr_type_intrinsic
,
399 nir_instr_type_load_const
,
401 nir_instr_type_ssa_undef
,
403 nir_instr_type_parallel_copy
,
406 typedef struct nir_instr
{
407 struct exec_node node
;
409 struct nir_block
*block
;
411 /** generic instruction index. */
414 /* A temporary for optimization and analysis passes to use for storing
415 * flags. For instance, DCE uses this to store the "dead/live" info.
420 static inline nir_instr
*
421 nir_instr_next(nir_instr
*instr
)
423 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
424 if (exec_node_is_tail_sentinel(next
))
427 return exec_node_data(nir_instr
, next
, node
);
430 static inline nir_instr
*
431 nir_instr_prev(nir_instr
*instr
)
433 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
434 if (exec_node_is_head_sentinel(prev
))
437 return exec_node_data(nir_instr
, prev
, node
);
441 nir_instr_is_first(const nir_instr
*instr
)
443 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
447 nir_instr_is_last(const nir_instr
*instr
)
449 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
452 typedef struct nir_ssa_def
{
453 /** for debugging only, can be NULL */
456 /** generic SSA definition index. */
459 /** Index into the live_in and live_out bitfields */
462 nir_instr
*parent_instr
;
464 /** set of nir_instrs where this register is used (read from) */
465 struct list_head uses
;
467 /** set of nir_ifs where this register is used as a condition */
468 struct list_head if_uses
;
470 uint8_t num_components
;
472 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
480 struct nir_src
*indirect
; /** < NULL for no indirect offset */
481 unsigned base_offset
;
483 /* TODO use-def chain goes here */
487 nir_instr
*parent_instr
;
488 struct list_head def_link
;
491 struct nir_src
*indirect
; /** < NULL for no indirect offset */
492 unsigned base_offset
;
494 /* TODO def-use chain goes here */
499 typedef struct nir_src
{
501 nir_instr
*parent_instr
;
502 struct nir_if
*parent_if
;
505 struct list_head use_link
;
515 static inline nir_src
518 nir_src src
= { { NULL
} };
522 #define NIR_SRC_INIT nir_src_init()
524 #define nir_foreach_use(src, reg_or_ssa_def) \
525 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
527 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
528 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
530 #define nir_foreach_if_use(src, reg_or_ssa_def) \
531 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
533 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
534 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
545 static inline nir_dest
548 nir_dest dest
= { { { NULL
} } };
552 #define NIR_DEST_INIT nir_dest_init()
554 #define nir_foreach_def(dest, reg) \
555 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
557 #define nir_foreach_def_safe(dest, reg) \
558 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
560 static inline nir_src
561 nir_src_for_ssa(nir_ssa_def
*def
)
563 nir_src src
= NIR_SRC_INIT
;
571 static inline nir_src
572 nir_src_for_reg(nir_register
*reg
)
574 nir_src src
= NIR_SRC_INIT
;
578 src
.reg
.indirect
= NULL
;
579 src
.reg
.base_offset
= 0;
584 static inline nir_dest
585 nir_dest_for_reg(nir_register
*reg
)
587 nir_dest dest
= NIR_DEST_INIT
;
594 static inline unsigned
595 nir_src_bit_size(nir_src src
)
597 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
600 static inline unsigned
601 nir_dest_bit_size(nir_dest dest
)
603 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
606 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
607 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
613 * \name input modifiers
617 * For inputs interpreted as floating point, flips the sign bit. For
618 * inputs interpreted as integers, performs the two's complement negation.
623 * Clears the sign bit for floating point values, and computes the integer
624 * absolute value for integers. Note that the negate modifier acts after
625 * the absolute value modifier, therefore if both are set then all inputs
626 * will become negative.
632 * For each input component, says which component of the register it is
633 * chosen from. Note that which elements of the swizzle are used and which
634 * are ignored are based on the write mask for most opcodes - for example,
635 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
636 * a swizzle of {2, x, 1, 0} where x means "don't care."
645 * \name saturate output modifier
647 * Only valid for opcodes that output floating-point numbers. Clamps the
648 * output to between 0.0 and 1.0 inclusive.
653 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
657 nir_type_invalid
= 0, /* Not a valid type */
662 nir_type_bool32
= 32 | nir_type_bool
,
663 nir_type_int8
= 8 | nir_type_int
,
664 nir_type_int16
= 16 | nir_type_int
,
665 nir_type_int32
= 32 | nir_type_int
,
666 nir_type_int64
= 64 | nir_type_int
,
667 nir_type_uint8
= 8 | nir_type_uint
,
668 nir_type_uint16
= 16 | nir_type_uint
,
669 nir_type_uint32
= 32 | nir_type_uint
,
670 nir_type_uint64
= 64 | nir_type_uint
,
671 nir_type_float16
= 16 | nir_type_float
,
672 nir_type_float32
= 32 | nir_type_float
,
673 nir_type_float64
= 64 | nir_type_float
,
676 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
677 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
679 static inline unsigned
680 nir_alu_type_get_type_size(nir_alu_type type
)
682 return type
& NIR_ALU_TYPE_SIZE_MASK
;
685 static inline unsigned
686 nir_alu_type_get_base_type(nir_alu_type type
)
688 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
691 static inline nir_alu_type
692 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
696 return nir_type_bool32
;
699 return nir_type_uint32
;
702 return nir_type_int32
;
704 case GLSL_TYPE_UINT64
:
705 return nir_type_uint64
;
707 case GLSL_TYPE_INT64
:
708 return nir_type_int64
;
710 case GLSL_TYPE_FLOAT
:
711 return nir_type_float32
;
713 case GLSL_TYPE_DOUBLE
:
714 return nir_type_float64
;
717 unreachable("unknown type");
721 static inline nir_alu_type
722 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
724 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
727 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
);
730 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
731 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
732 } nir_op_algebraic_property
;
740 * The number of components in the output
742 * If non-zero, this is the size of the output and input sizes are
743 * explicitly given; swizzle and writemask are still in effect, but if
744 * the output component is masked out, then the input component may
747 * If zero, the opcode acts in the standard, per-component manner; the
748 * operation is performed on each component (except the ones that are
749 * masked out) with the input being taken from the input swizzle for
752 * The size of some of the inputs may be given (i.e. non-zero) even
753 * though output_size is zero; in that case, the inputs with a zero
754 * size act per-component, while the inputs with non-zero size don't.
756 unsigned output_size
;
759 * The type of vector that the instruction outputs. Note that the
760 * staurate modifier is only allowed on outputs with the float type.
763 nir_alu_type output_type
;
766 * The number of components in each input
768 unsigned input_sizes
[4];
771 * The type of vector that each input takes. Note that negate and
772 * absolute value are only allowed on inputs with int or float type and
773 * behave differently on the two.
775 nir_alu_type input_types
[4];
777 nir_op_algebraic_property algebraic_properties
;
780 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
782 typedef struct nir_alu_instr
{
786 /** Indicates that this ALU instruction generates an exact value
788 * This is kind of a mixture of GLSL "precise" and "invariant" and not
789 * really equivalent to either. This indicates that the value generated by
790 * this operation is high-precision and any code transformations that touch
791 * it must ensure that the resulting value is bit-for-bit identical to the
800 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
801 nir_alu_instr
*instr
);
802 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
803 nir_alu_instr
*instr
);
805 /* is this source channel used? */
807 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
810 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
811 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
813 return (instr
->dest
.write_mask
>> channel
) & 1;
817 * For instructions whose destinations are SSA, get the number of channels
820 static inline unsigned
821 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
823 assert(instr
->dest
.dest
.is_ssa
);
825 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
826 return nir_op_infos
[instr
->op
].input_sizes
[src
];
828 return instr
->dest
.dest
.ssa
.num_components
;
831 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
832 unsigned src1
, unsigned src2
);
836 nir_deref_type_array
,
837 nir_deref_type_struct
840 typedef struct nir_deref
{
841 nir_deref_type deref_type
;
842 struct nir_deref
*child
;
843 const struct glsl_type
*type
;
852 /* This enum describes how the array is referenced. If the deref is
853 * direct then the base_offset is used. If the deref is indirect then
854 * offset is given by base_offset + indirect. If the deref is a wildcard
855 * then the deref refers to all of the elements of the array at the same
856 * time. Wildcard dereferences are only ever allowed in copy_var
857 * intrinsics and the source and destination derefs must have matching
861 nir_deref_array_type_direct
,
862 nir_deref_array_type_indirect
,
863 nir_deref_array_type_wildcard
,
864 } nir_deref_array_type
;
869 nir_deref_array_type deref_array_type
;
870 unsigned base_offset
;
880 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
,
881 deref_type
, nir_deref_type_var
)
882 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
,
883 deref_type
, nir_deref_type_array
)
884 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
,
885 deref_type
, nir_deref_type_struct
)
887 /* Returns the last deref in the chain. */
888 static inline nir_deref
*
889 nir_deref_tail(nir_deref
*deref
)
892 deref
= deref
->child
;
900 nir_deref_var
**params
;
901 nir_deref_var
*return_deref
;
903 struct nir_function
*callee
;
906 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
907 num_variables, num_indices, idx0, idx1, idx2, flags) \
908 nir_intrinsic_##name,
910 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
913 #include "nir_intrinsics.h"
914 nir_num_intrinsics
= nir_last_intrinsic
+ 1
917 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
919 /** Represents an intrinsic
921 * An intrinsic is an instruction type for handling things that are
922 * more-or-less regular operations but don't just consume and produce SSA
923 * values like ALU operations do. Intrinsics are not for things that have
924 * special semantic meaning such as phi nodes and parallel copies.
925 * Examples of intrinsics include variable load/store operations, system
926 * value loads, and the like. Even though texturing more-or-less falls
927 * under this category, texturing is its own instruction type because
928 * trying to represent texturing with intrinsics would lead to a
929 * combinatorial explosion of intrinsic opcodes.
931 * By having a single instruction type for handling a lot of different
932 * cases, optimization passes can look for intrinsics and, for the most
933 * part, completely ignore them. Each intrinsic type also has a few
934 * possible flags that govern whether or not they can be reordered or
935 * eliminated. That way passes like dead code elimination can still work
936 * on intrisics without understanding the meaning of each.
938 * Each intrinsic has some number of constant indices, some number of
939 * variables, and some number of sources. What these sources, variables,
940 * and indices mean depends on the intrinsic and is documented with the
941 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
942 * instructions are the only types of instruction that can operate on
948 nir_intrinsic_op intrinsic
;
952 /** number of components if this is a vectorized intrinsic
954 * Similarly to ALU operations, some intrinsics are vectorized.
955 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
956 * For vectorized intrinsics, the num_components field specifies the
957 * number of destination components and the number of source components
958 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
960 uint8_t num_components
;
962 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
964 nir_deref_var
*variables
[2];
967 } nir_intrinsic_instr
;
970 * \name NIR intrinsics semantic flags
972 * information about what the compiler can do with the intrinsics.
974 * \sa nir_intrinsic_info::flags
978 * whether the intrinsic can be safely eliminated if none of its output
979 * value is not being used.
981 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
984 * Whether the intrinsic can be reordered with respect to any other
985 * intrinsic, i.e. whether the only reordering dependencies of the
986 * intrinsic are due to the register reads/writes.
988 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
989 } nir_intrinsic_semantic_flag
;
992 * \name NIR intrinsics const-index flag
994 * Indicates the usage of a const_index slot.
996 * \sa nir_intrinsic_info::index_map
1000 * Generally instructions that take a offset src argument, can encode
1001 * a constant 'base' value which is added to the offset.
1003 NIR_INTRINSIC_BASE
= 1,
1006 * For store instructions, a writemask for the store.
1008 NIR_INTRINSIC_WRMASK
= 2,
1011 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1013 NIR_INTRINSIC_STREAM_ID
= 3,
1016 * The clip-plane id for load_user_clip_plane intrinsic.
1018 NIR_INTRINSIC_UCP_ID
= 4,
1021 * The amount of data, starting from BASE, that this instruction may
1022 * access. This is used to provide bounds if the offset is not constant.
1024 NIR_INTRINSIC_RANGE
= 5,
1027 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1029 NIR_INTRINSIC_DESC_SET
= 6,
1032 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1034 NIR_INTRINSIC_BINDING
= 7,
1039 NIR_INTRINSIC_COMPONENT
= 8,
1042 * Interpolation mode (only meaningful for FS inputs).
1044 NIR_INTRINSIC_INTERP_MODE
= 9,
1046 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1048 } nir_intrinsic_index_flag
;
1050 #define NIR_INTRINSIC_MAX_INPUTS 4
1055 unsigned num_srcs
; /** < number of register/SSA inputs */
1057 /** number of components of each input register
1059 * If this value is 0, the number of components is given by the
1060 * num_components field of nir_intrinsic_instr.
1062 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1066 /** number of components of the output register
1068 * If this value is 0, the number of components is given by the
1069 * num_components field of nir_intrinsic_instr.
1071 unsigned dest_components
;
1073 /** the number of inputs/outputs that are variables */
1074 unsigned num_variables
;
1076 /** the number of constant indices used by the intrinsic */
1077 unsigned num_indices
;
1079 /** indicates the usage of intr->const_index[n] */
1080 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1082 /** semantic flags for calls to this intrinsic */
1083 nir_intrinsic_semantic_flag flags
;
1084 } nir_intrinsic_info
;
1086 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1089 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1090 static inline type \
1091 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1093 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1094 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1095 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1097 static inline void \
1098 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1100 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1101 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1102 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1105 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1106 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1107 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1108 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1109 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1110 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1111 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1112 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1113 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1116 * \group texture information
1118 * This gives semantic information about textures which is useful to the
1119 * frontend, the backend, and lowering passes, but not the optimizer.
1124 nir_tex_src_projector
,
1125 nir_tex_src_comparator
, /* shadow comparator */
1129 nir_tex_src_ms_index
, /* MSAA sample index */
1130 nir_tex_src_ms_mcs
, /* MSAA compression value */
1133 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1134 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1135 nir_tex_src_plane
, /* < selects plane for planar textures */
1136 nir_num_tex_src_types
1141 nir_tex_src_type src_type
;
1145 nir_texop_tex
, /**< Regular texture look-up */
1146 nir_texop_txb
, /**< Texture look-up with LOD bias */
1147 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1148 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1149 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1150 nir_texop_txf_ms
, /**< Multisample texture fetch */
1151 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1152 nir_texop_txs
, /**< Texture size */
1153 nir_texop_lod
, /**< Texture lod query */
1154 nir_texop_tg4
, /**< Texture gather */
1155 nir_texop_query_levels
, /**< Texture levels query */
1156 nir_texop_texture_samples
, /**< Texture samples query */
1157 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1165 enum glsl_sampler_dim sampler_dim
;
1166 nir_alu_type dest_type
;
1171 unsigned num_srcs
, coord_components
;
1172 bool is_array
, is_shadow
;
1175 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1176 * components or the new-style shadow that outputs 1 component.
1178 bool is_new_style_shadow
;
1180 /* gather component selector */
1181 unsigned component
: 2;
1183 /** The texture index
1185 * If this texture instruction has a nir_tex_src_texture_offset source,
1186 * then the texture index is given by texture_index + texture_offset.
1188 unsigned texture_index
;
1190 /** The size of the texture array or 0 if it's not an array */
1191 unsigned texture_array_size
;
1193 /** The texture deref
1195 * If this is null, use texture_index instead.
1197 nir_deref_var
*texture
;
1199 /** The sampler index
1201 * The following operations do not require a sampler and, as such, this
1202 * field should be ignored:
1204 * - nir_texop_txf_ms
1208 * - nir_texop_query_levels
1209 * - nir_texop_texture_samples
1210 * - nir_texop_samples_identical
1212 * If this texture instruction has a nir_tex_src_sampler_offset source,
1213 * then the sampler index is given by sampler_index + sampler_offset.
1215 unsigned sampler_index
;
1217 /** The sampler deref
1219 * If this is null, use sampler_index instead.
1221 nir_deref_var
*sampler
;
1224 static inline unsigned
1225 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1227 switch (instr
->op
) {
1228 case nir_texop_txs
: {
1230 switch (instr
->sampler_dim
) {
1231 case GLSL_SAMPLER_DIM_1D
:
1232 case GLSL_SAMPLER_DIM_BUF
:
1235 case GLSL_SAMPLER_DIM_2D
:
1236 case GLSL_SAMPLER_DIM_CUBE
:
1237 case GLSL_SAMPLER_DIM_MS
:
1238 case GLSL_SAMPLER_DIM_RECT
:
1239 case GLSL_SAMPLER_DIM_EXTERNAL
:
1240 case GLSL_SAMPLER_DIM_SUBPASS
:
1243 case GLSL_SAMPLER_DIM_3D
:
1247 unreachable("not reached");
1249 if (instr
->is_array
)
1257 case nir_texop_texture_samples
:
1258 case nir_texop_query_levels
:
1259 case nir_texop_samples_identical
:
1263 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1270 /* Returns true if this texture operation queries something about the texture
1271 * rather than actually sampling it.
1274 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1276 switch (instr
->op
) {
1279 case nir_texop_texture_samples
:
1280 case nir_texop_query_levels
:
1281 case nir_texop_txf_ms_mcs
:
1288 case nir_texop_txf_ms
:
1292 unreachable("Invalid texture opcode");
1296 static inline nir_alu_type
1297 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1299 switch (instr
->src
[src
].src_type
) {
1300 case nir_tex_src_coord
:
1301 switch (instr
->op
) {
1303 case nir_texop_txf_ms
:
1304 case nir_texop_txf_ms_mcs
:
1305 case nir_texop_samples_identical
:
1306 return nir_type_int
;
1309 return nir_type_float
;
1312 case nir_tex_src_lod
:
1313 switch (instr
->op
) {
1316 return nir_type_int
;
1319 return nir_type_float
;
1322 case nir_tex_src_projector
:
1323 case nir_tex_src_comparator
:
1324 case nir_tex_src_bias
:
1325 case nir_tex_src_ddx
:
1326 case nir_tex_src_ddy
:
1327 return nir_type_float
;
1329 case nir_tex_src_offset
:
1330 case nir_tex_src_ms_index
:
1331 case nir_tex_src_texture_offset
:
1332 case nir_tex_src_sampler_offset
:
1333 return nir_type_int
;
1336 unreachable("Invalid texture source type");
1340 static inline unsigned
1341 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1343 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1344 return instr
->coord_components
;
1346 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1347 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1350 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1351 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1352 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1353 if (instr
->is_array
)
1354 return instr
->coord_components
- 1;
1356 return instr
->coord_components
;
1363 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1365 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1366 if (instr
->src
[i
].src_type
== type
)
1372 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1377 nir_const_value value
;
1380 } nir_load_const_instr
;
1393 /* creates a new SSA variable in an undefined state */
1398 } nir_ssa_undef_instr
;
1401 struct exec_node node
;
1403 /* The predecessor block corresponding to this source */
1404 struct nir_block
*pred
;
1409 #define nir_foreach_phi_src(phi_src, phi) \
1410 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1411 #define nir_foreach_phi_src_safe(phi_src, phi) \
1412 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1417 struct exec_list srcs
; /** < list of nir_phi_src */
1423 struct exec_node node
;
1426 } nir_parallel_copy_entry
;
1428 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1429 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1434 /* A list of nir_parallel_copy_entrys. The sources of all of the
1435 * entries are copied to the corresponding destinations "in parallel".
1436 * In other words, if we have two entries: a -> b and b -> a, the values
1439 struct exec_list entries
;
1440 } nir_parallel_copy_instr
;
1442 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1443 type
, nir_instr_type_alu
)
1444 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1445 type
, nir_instr_type_call
)
1446 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1447 type
, nir_instr_type_jump
)
1448 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1449 type
, nir_instr_type_tex
)
1450 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1451 type
, nir_instr_type_intrinsic
)
1452 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1453 type
, nir_instr_type_load_const
)
1454 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1455 type
, nir_instr_type_ssa_undef
)
1456 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1457 type
, nir_instr_type_phi
)
1458 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1459 nir_parallel_copy_instr
, instr
,
1460 type
, nir_instr_type_parallel_copy
)
1465 * Control flow consists of a tree of control flow nodes, which include
1466 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1467 * instructions that always run start-to-finish. Each basic block also keeps
1468 * track of its successors (blocks which may run immediately after the current
1469 * block) and predecessors (blocks which could have run immediately before the
1470 * current block). Each function also has a start block and an end block which
1471 * all return statements point to (which is always empty). Together, all the
1472 * blocks with their predecessors and successors make up the control flow
1473 * graph (CFG) of the function. There are helpers that modify the tree of
1474 * control flow nodes while modifying the CFG appropriately; these should be
1475 * used instead of modifying the tree directly.
1482 nir_cf_node_function
1485 typedef struct nir_cf_node
{
1486 struct exec_node node
;
1487 nir_cf_node_type type
;
1488 struct nir_cf_node
*parent
;
1491 typedef struct nir_block
{
1492 nir_cf_node cf_node
;
1494 struct exec_list instr_list
; /** < list of nir_instr */
1496 /** generic block index; generated by nir_index_blocks */
1500 * Each block can only have up to 2 successors, so we put them in a simple
1501 * array - no need for anything more complicated.
1503 struct nir_block
*successors
[2];
1505 /* Set of nir_block predecessors in the CFG */
1506 struct set
*predecessors
;
1509 * this node's immediate dominator in the dominance tree - set to NULL for
1512 struct nir_block
*imm_dom
;
1514 /* This node's children in the dominance tree */
1515 unsigned num_dom_children
;
1516 struct nir_block
**dom_children
;
1518 /* Set of nir_blocks on the dominance frontier of this block */
1519 struct set
*dom_frontier
;
1522 * These two indices have the property that dom_{pre,post}_index for each
1523 * child of this block in the dominance tree will always be between
1524 * dom_pre_index and dom_post_index for this block, which makes testing if
1525 * a given block is dominated by another block an O(1) operation.
1527 unsigned dom_pre_index
, dom_post_index
;
1529 /* live in and out for this block; used for liveness analysis */
1530 BITSET_WORD
*live_in
;
1531 BITSET_WORD
*live_out
;
1534 static inline nir_instr
*
1535 nir_block_first_instr(nir_block
*block
)
1537 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1538 return exec_node_data(nir_instr
, head
, node
);
1541 static inline nir_instr
*
1542 nir_block_last_instr(nir_block
*block
)
1544 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1545 return exec_node_data(nir_instr
, tail
, node
);
1548 #define nir_foreach_instr(instr, block) \
1549 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1550 #define nir_foreach_instr_reverse(instr, block) \
1551 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1552 #define nir_foreach_instr_safe(instr, block) \
1553 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1554 #define nir_foreach_instr_reverse_safe(instr, block) \
1555 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1557 typedef struct nir_if
{
1558 nir_cf_node cf_node
;
1561 struct exec_list then_list
; /** < list of nir_cf_node */
1562 struct exec_list else_list
; /** < list of nir_cf_node */
1568 nir_instr
*conditional_instr
;
1570 nir_block
*break_block
;
1571 nir_block
*continue_from_block
;
1573 bool continue_from_then
;
1575 struct list_head loop_terminator_link
;
1576 } nir_loop_terminator
;
1579 /* Number of instructions in the loop */
1580 unsigned num_instructions
;
1582 /* How many times the loop is run (if known) */
1583 unsigned trip_count
;
1584 bool is_trip_count_known
;
1586 /* Unroll the loop regardless of its size */
1589 nir_loop_terminator
*limiting_terminator
;
1591 /* A list of loop_terminators terminating this loop. */
1592 struct list_head loop_terminator_list
;
1596 nir_cf_node cf_node
;
1598 struct exec_list body
; /** < list of nir_cf_node */
1600 nir_loop_info
*info
;
1604 * Various bits of metadata that can may be created or required by
1605 * optimization and analysis passes
1608 nir_metadata_none
= 0x0,
1609 nir_metadata_block_index
= 0x1,
1610 nir_metadata_dominance
= 0x2,
1611 nir_metadata_live_ssa_defs
= 0x4,
1612 nir_metadata_not_properly_reset
= 0x8,
1613 nir_metadata_loop_analysis
= 0x10,
1617 nir_cf_node cf_node
;
1619 /** pointer to the function of which this is an implementation */
1620 struct nir_function
*function
;
1622 struct exec_list body
; /** < list of nir_cf_node */
1624 nir_block
*end_block
;
1626 /** list for all local variables in the function */
1627 struct exec_list locals
;
1629 /** array of variables used as parameters */
1630 unsigned num_params
;
1631 nir_variable
**params
;
1633 /** variable used to hold the result of the function */
1634 nir_variable
*return_var
;
1636 /** list of local registers in the function */
1637 struct exec_list registers
;
1639 /** next available local register index */
1642 /** next available SSA value index */
1645 /* total number of basic blocks, only valid when block_index_dirty = false */
1646 unsigned num_blocks
;
1648 nir_metadata valid_metadata
;
1649 } nir_function_impl
;
1651 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1652 nir_start_block(nir_function_impl
*impl
)
1654 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1657 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1658 nir_impl_last_block(nir_function_impl
*impl
)
1660 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1663 static inline nir_cf_node
*
1664 nir_cf_node_next(nir_cf_node
*node
)
1666 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1667 if (exec_node_is_tail_sentinel(next
))
1670 return exec_node_data(nir_cf_node
, next
, node
);
1673 static inline nir_cf_node
*
1674 nir_cf_node_prev(nir_cf_node
*node
)
1676 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1677 if (exec_node_is_head_sentinel(prev
))
1680 return exec_node_data(nir_cf_node
, prev
, node
);
1684 nir_cf_node_is_first(const nir_cf_node
*node
)
1686 return exec_node_is_head_sentinel(node
->node
.prev
);
1690 nir_cf_node_is_last(const nir_cf_node
*node
)
1692 return exec_node_is_tail_sentinel(node
->node
.next
);
1695 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1696 type
, nir_cf_node_block
)
1697 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1698 type
, nir_cf_node_if
)
1699 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1700 type
, nir_cf_node_loop
)
1701 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1702 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1704 static inline nir_block
*
1705 nir_if_first_then_block(nir_if
*if_stmt
)
1707 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1708 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1711 static inline nir_block
*
1712 nir_if_last_then_block(nir_if
*if_stmt
)
1714 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1715 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1718 static inline nir_block
*
1719 nir_if_first_else_block(nir_if
*if_stmt
)
1721 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1722 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1725 static inline nir_block
*
1726 nir_if_last_else_block(nir_if
*if_stmt
)
1728 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1729 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1732 static inline nir_block
*
1733 nir_loop_first_block(nir_loop
*loop
)
1735 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1736 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1739 static inline nir_block
*
1740 nir_loop_last_block(nir_loop
*loop
)
1742 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1743 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1749 nir_parameter_inout
,
1750 } nir_parameter_type
;
1753 nir_parameter_type param_type
;
1754 const struct glsl_type
*type
;
1757 typedef struct nir_function
{
1758 struct exec_node node
;
1761 struct nir_shader
*shader
;
1763 unsigned num_params
;
1764 nir_parameter
*params
;
1765 const struct glsl_type
*return_type
;
1767 /** The implementation of this function.
1769 * If the function is only declared and not implemented, this is NULL.
1771 nir_function_impl
*impl
;
1774 typedef struct nir_shader_compiler_options
{
1779 /** Lowers flrp when it does not support doubles */
1786 bool lower_bitfield_extract
;
1787 bool lower_bitfield_insert
;
1788 bool lower_uadd_carry
;
1789 bool lower_usub_borrow
;
1790 /** lowers fneg and ineg to fsub and isub. */
1792 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1795 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1798 /** enables rules to lower idiv by power-of-two: */
1801 /* Does the native fdot instruction replicate its result for four
1802 * components? If so, then opt_algebraic_late will turn all fdotN
1803 * instructions into fdot_replicatedN instructions.
1805 bool fdot_replicates
;
1807 /** lowers ffract to fsub+ffloor: */
1810 bool lower_pack_half_2x16
;
1811 bool lower_pack_unorm_2x16
;
1812 bool lower_pack_snorm_2x16
;
1813 bool lower_pack_unorm_4x8
;
1814 bool lower_pack_snorm_4x8
;
1815 bool lower_unpack_half_2x16
;
1816 bool lower_unpack_unorm_2x16
;
1817 bool lower_unpack_snorm_2x16
;
1818 bool lower_unpack_unorm_4x8
;
1819 bool lower_unpack_snorm_4x8
;
1821 bool lower_extract_byte
;
1822 bool lower_extract_word
;
1825 * Does the driver support real 32-bit integers? (Otherwise, integers
1826 * are simulated by floats.)
1828 bool native_integers
;
1830 /* Indicates that the driver only has zero-based vertex id */
1831 bool vertex_id_zero_based
;
1833 bool lower_cs_local_index_from_id
;
1836 * Should nir_lower_io() create load_interpolated_input intrinsics?
1838 * If not, it generates regular load_input intrinsics and interpolation
1839 * information must be inferred from the list of input nir_variables.
1841 bool use_interpolated_input_intrinsics
;
1843 unsigned max_unroll_iterations
;
1844 } nir_shader_compiler_options
;
1846 typedef struct nir_shader
{
1847 /** list of uniforms (nir_variable) */
1848 struct exec_list uniforms
;
1850 /** list of inputs (nir_variable) */
1851 struct exec_list inputs
;
1853 /** list of outputs (nir_variable) */
1854 struct exec_list outputs
;
1856 /** list of shared compute variables (nir_variable) */
1857 struct exec_list shared
;
1859 /** Set of driver-specific options for the shader.
1861 * The memory for the options is expected to be kept in a single static
1862 * copy by the driver.
1864 const struct nir_shader_compiler_options
*options
;
1866 /** Various bits of compile-time information about a given shader */
1867 struct shader_info info
;
1869 /** list of global variables in the shader (nir_variable) */
1870 struct exec_list globals
;
1872 /** list of system value variables in the shader (nir_variable) */
1873 struct exec_list system_values
;
1875 struct exec_list functions
; /** < list of nir_function */
1877 /** list of global register in the shader */
1878 struct exec_list registers
;
1880 /** next available global register index */
1884 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1887 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1889 /** The shader stage, such as MESA_SHADER_VERTEX. */
1890 gl_shader_stage stage
;
1893 static inline nir_function_impl
*
1894 nir_shader_get_entrypoint(nir_shader
*shader
)
1896 assert(exec_list_length(&shader
->functions
) == 1);
1897 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1898 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1899 assert(func
->return_type
== glsl_void_type());
1900 assert(func
->num_params
== 0);
1905 #define nir_foreach_function(func, shader) \
1906 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1908 nir_shader
*nir_shader_create(void *mem_ctx
,
1909 gl_shader_stage stage
,
1910 const nir_shader_compiler_options
*options
,
1913 /** creates a register, including assigning it an index and adding it to the list */
1914 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1916 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1918 void nir_reg_remove(nir_register
*reg
);
1920 /** Adds a variable to the appropriate list in nir_shader */
1921 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1924 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1926 assert(var
->data
.mode
== nir_var_local
);
1927 exec_list_push_tail(&impl
->locals
, &var
->node
);
1930 /** creates a variable, sets a few defaults, and adds it to the list */
1931 nir_variable
*nir_variable_create(nir_shader
*shader
,
1932 nir_variable_mode mode
,
1933 const struct glsl_type
*type
,
1935 /** creates a local variable and adds it to the list */
1936 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1937 const struct glsl_type
*type
,
1940 /** creates a function and adds it to the shader's list of functions */
1941 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1943 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1944 /** creates a function_impl that isn't tied to any particular function */
1945 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1947 nir_block
*nir_block_create(nir_shader
*shader
);
1948 nir_if
*nir_if_create(nir_shader
*shader
);
1949 nir_loop
*nir_loop_create(nir_shader
*shader
);
1951 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1953 /** requests that the given pieces of metadata be generated */
1954 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
1955 /** dirties all but the preserved metadata */
1956 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1958 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1959 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1961 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1963 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1964 unsigned num_components
,
1967 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1968 nir_intrinsic_op op
);
1970 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1971 nir_function
*callee
);
1973 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1975 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1977 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1979 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1980 unsigned num_components
,
1983 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1984 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1985 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1987 typedef bool (*nir_deref_foreach_leaf_cb
)(nir_deref_var
*deref
, void *state
);
1988 bool nir_deref_foreach_leaf(nir_deref_var
*deref
,
1989 nir_deref_foreach_leaf_cb cb
, void *state
);
1991 nir_load_const_instr
*
1992 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1995 * NIR Cursors and Instruction Insertion API
1998 * A tiny struct representing a point to insert/extract instructions or
1999 * control flow nodes. Helps reduce the combinatorial explosion of possible
2000 * points to insert/extract.
2002 * \sa nir_control_flow.h
2005 nir_cursor_before_block
,
2006 nir_cursor_after_block
,
2007 nir_cursor_before_instr
,
2008 nir_cursor_after_instr
,
2009 } nir_cursor_option
;
2012 nir_cursor_option option
;
2019 static inline nir_block
*
2020 nir_cursor_current_block(nir_cursor cursor
)
2022 if (cursor
.option
== nir_cursor_before_instr
||
2023 cursor
.option
== nir_cursor_after_instr
) {
2024 return cursor
.instr
->block
;
2026 return cursor
.block
;
2030 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2032 static inline nir_cursor
2033 nir_before_block(nir_block
*block
)
2036 cursor
.option
= nir_cursor_before_block
;
2037 cursor
.block
= block
;
2041 static inline nir_cursor
2042 nir_after_block(nir_block
*block
)
2045 cursor
.option
= nir_cursor_after_block
;
2046 cursor
.block
= block
;
2050 static inline nir_cursor
2051 nir_before_instr(nir_instr
*instr
)
2054 cursor
.option
= nir_cursor_before_instr
;
2055 cursor
.instr
= instr
;
2059 static inline nir_cursor
2060 nir_after_instr(nir_instr
*instr
)
2063 cursor
.option
= nir_cursor_after_instr
;
2064 cursor
.instr
= instr
;
2068 static inline nir_cursor
2069 nir_after_block_before_jump(nir_block
*block
)
2071 nir_instr
*last_instr
= nir_block_last_instr(block
);
2072 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2073 return nir_before_instr(last_instr
);
2075 return nir_after_block(block
);
2079 static inline nir_cursor
2080 nir_before_cf_node(nir_cf_node
*node
)
2082 if (node
->type
== nir_cf_node_block
)
2083 return nir_before_block(nir_cf_node_as_block(node
));
2085 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2088 static inline nir_cursor
2089 nir_after_cf_node(nir_cf_node
*node
)
2091 if (node
->type
== nir_cf_node_block
)
2092 return nir_after_block(nir_cf_node_as_block(node
));
2094 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2097 static inline nir_cursor
2098 nir_after_phis(nir_block
*block
)
2100 nir_foreach_instr(instr
, block
) {
2101 if (instr
->type
!= nir_instr_type_phi
)
2102 return nir_before_instr(instr
);
2104 return nir_after_block(block
);
2107 static inline nir_cursor
2108 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2110 if (node
->type
== nir_cf_node_block
)
2111 return nir_after_block(nir_cf_node_as_block(node
));
2113 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2115 return nir_after_phis(block
);
2118 static inline nir_cursor
2119 nir_before_cf_list(struct exec_list
*cf_list
)
2121 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2122 exec_list_get_head(cf_list
), node
);
2123 return nir_before_cf_node(first_node
);
2126 static inline nir_cursor
2127 nir_after_cf_list(struct exec_list
*cf_list
)
2129 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2130 exec_list_get_tail(cf_list
), node
);
2131 return nir_after_cf_node(last_node
);
2135 * Insert a NIR instruction at the given cursor.
2137 * Note: This does not update the cursor.
2139 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2142 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2144 nir_instr_insert(nir_before_instr(instr
), before
);
2148 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2150 nir_instr_insert(nir_after_instr(instr
), after
);
2154 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2156 nir_instr_insert(nir_before_block(block
), before
);
2160 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2162 nir_instr_insert(nir_after_block(block
), after
);
2166 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2168 nir_instr_insert(nir_before_cf_node(node
), before
);
2172 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2174 nir_instr_insert(nir_after_cf_node(node
), after
);
2178 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2180 nir_instr_insert(nir_before_cf_list(list
), before
);
2184 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2186 nir_instr_insert(nir_after_cf_list(list
), after
);
2189 void nir_instr_remove(nir_instr
*instr
);
2193 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2194 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2195 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2196 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2198 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2199 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2201 nir_const_value
*nir_src_as_const_value(nir_src src
);
2202 bool nir_src_is_dynamically_uniform(nir_src src
);
2203 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2204 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2205 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2206 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2207 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2210 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2211 unsigned num_components
, unsigned bit_size
,
2213 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2214 unsigned num_components
, unsigned bit_size
,
2216 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2217 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2218 nir_instr
*after_me
);
2220 uint8_t nir_ssa_def_components_read(nir_ssa_def
*def
);
2223 * finds the next basic block in source-code order, returns NULL if there is
2227 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2229 /* Performs the opposite of nir_block_cf_tree_next() */
2231 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2233 /* Gets the first block in a CF node in source-code order */
2235 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2237 /* Gets the last block in a CF node in source-code order */
2239 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2241 /* Gets the next block after a CF node in source-code order */
2243 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2245 /* Macros for loops that visit blocks in source-code order */
2247 #define nir_foreach_block(block, impl) \
2248 for (nir_block *block = nir_start_block(impl); block != NULL; \
2249 block = nir_block_cf_tree_next(block))
2251 #define nir_foreach_block_safe(block, impl) \
2252 for (nir_block *block = nir_start_block(impl), \
2253 *next = nir_block_cf_tree_next(block); \
2255 block = next, next = nir_block_cf_tree_next(block))
2257 #define nir_foreach_block_reverse(block, impl) \
2258 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2259 block = nir_block_cf_tree_prev(block))
2261 #define nir_foreach_block_reverse_safe(block, impl) \
2262 for (nir_block *block = nir_impl_last_block(impl), \
2263 *prev = nir_block_cf_tree_prev(block); \
2265 block = prev, prev = nir_block_cf_tree_prev(block))
2267 #define nir_foreach_block_in_cf_node(block, node) \
2268 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2269 block != nir_cf_node_cf_tree_next(node); \
2270 block = nir_block_cf_tree_next(block))
2272 /* If the following CF node is an if, this function returns that if.
2273 * Otherwise, it returns NULL.
2275 nir_if
*nir_block_get_following_if(nir_block
*block
);
2277 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2279 void nir_index_local_regs(nir_function_impl
*impl
);
2280 void nir_index_global_regs(nir_shader
*shader
);
2281 void nir_index_ssa_defs(nir_function_impl
*impl
);
2282 unsigned nir_index_instrs(nir_function_impl
*impl
);
2284 void nir_index_blocks(nir_function_impl
*impl
);
2286 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2287 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2288 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2290 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2291 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2292 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2293 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2294 nir_deref
*nir_deref_clone(const nir_deref
*deref
, void *mem_ctx
);
2295 nir_deref_var
*nir_deref_var_clone(const nir_deref_var
*deref
, void *mem_ctx
);
2298 void nir_validate_shader(nir_shader
*shader
);
2299 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2300 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2303 should_clone_nir(void)
2305 static int should_clone
= -1;
2306 if (should_clone
< 0)
2307 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2309 return should_clone
;
2313 should_print_nir(void)
2315 static int should_print
= -1;
2316 if (should_print
< 0)
2317 should_print
= env_var_as_boolean("NIR_PRINT", false);
2319 return should_print
;
2322 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2323 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2324 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2325 static inline bool should_clone_nir(void) { return false; }
2326 static inline bool should_print_nir(void) { return false; }
2329 #define _PASS(nir, do_pass) do { \
2331 nir_validate_shader(nir); \
2332 if (should_clone_nir()) { \
2333 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2339 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2340 nir_metadata_set_validation_flag(nir); \
2341 if (should_print_nir()) \
2342 printf("%s\n", #pass); \
2343 if (pass(nir, ##__VA_ARGS__)) { \
2345 if (should_print_nir()) \
2346 nir_print_shader(nir, stdout); \
2347 nir_metadata_check_validation_flag(nir); \
2351 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2352 if (should_print_nir()) \
2353 printf("%s\n", #pass); \
2354 pass(nir, ##__VA_ARGS__); \
2355 if (should_print_nir()) \
2356 nir_print_shader(nir, stdout); \
2359 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2360 void nir_calc_dominance(nir_shader
*shader
);
2362 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2363 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2365 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2366 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2368 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2369 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2371 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2372 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2374 int nir_gs_count_vertices(const nir_shader
*shader
);
2376 bool nir_split_var_copies(nir_shader
*shader
);
2378 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2379 bool nir_lower_returns(nir_shader
*shader
);
2381 bool nir_inline_functions(nir_shader
*shader
);
2383 bool nir_propagate_invariant(nir_shader
*shader
);
2385 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2386 bool nir_lower_var_copies(nir_shader
*shader
);
2388 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2390 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2392 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2394 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2395 nir_function_impl
*entrypoint
,
2396 bool outputs
, bool inputs
);
2398 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2400 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2401 int (*type_size
)(const struct glsl_type
*));
2404 /* If set, this forces all non-flat fragment shader inputs to be
2405 * interpolated as if with the "sample" qualifier. This requires
2406 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2408 nir_lower_io_force_sample_interpolation
= (1 << 1),
2409 } nir_lower_io_options
;
2410 bool nir_lower_io(nir_shader
*shader
,
2411 nir_variable_mode modes
,
2412 int (*type_size
)(const struct glsl_type
*),
2413 nir_lower_io_options
);
2414 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2415 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2417 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2419 void nir_lower_io_types(nir_shader
*shader
);
2420 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2421 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2422 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2424 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2425 bool nir_lower_constant_initializers(nir_shader
*shader
,
2426 nir_variable_mode modes
);
2428 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2429 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2430 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2431 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2433 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2434 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2436 bool nir_lower_samplers(nir_shader
*shader
,
2437 const struct gl_shader_program
*shader_program
);
2439 bool nir_lower_system_values(nir_shader
*shader
);
2441 typedef struct nir_lower_tex_options
{
2443 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2444 * sampler types a texture projector is lowered.
2449 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2451 bool lower_txf_offset
;
2454 * If true, lower away nir_tex_src_offset for all rect textures.
2456 bool lower_rect_offset
;
2459 * If true, lower rect textures to 2D, using txs to fetch the
2460 * texture dimensions and dividing the texture coords by the
2461 * texture dims to normalize.
2466 * If true, convert yuv to rgb.
2468 unsigned lower_y_uv_external
;
2469 unsigned lower_y_u_v_external
;
2470 unsigned lower_yx_xuxv_external
;
2471 unsigned lower_xy_uxvx_external
;
2474 * To emulate certain texture wrap modes, this can be used
2475 * to saturate the specified tex coord to [0.0, 1.0]. The
2476 * bits are according to sampler #, ie. if, for example:
2478 * (conf->saturate_s & (1 << n))
2480 * is true, then the s coord for sampler n is saturated.
2482 * Note that clamping must happen *after* projector lowering
2483 * so any projected texture sample instruction with a clamped
2484 * coordinate gets automatically lowered, regardless of the
2485 * 'lower_txp' setting.
2487 unsigned saturate_s
;
2488 unsigned saturate_t
;
2489 unsigned saturate_r
;
2491 /* Bitmask of textures that need swizzling.
2493 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2494 * swizzles[texture_index] is applied to the result of the texturing
2497 unsigned swizzle_result
;
2499 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2500 * while 4 and 5 represent 0 and 1 respectively.
2502 uint8_t swizzles
[32][4];
2505 * Bitmap of textures that need srgb to linear conversion. If
2506 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2507 * of the texture are lowered to linear.
2509 unsigned lower_srgb
;
2512 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2514 bool lower_txd_cube_map
;
2517 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
2518 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
2519 * with lower_txd_cube_map.
2521 bool lower_txd_shadow
;
2522 } nir_lower_tex_options
;
2524 bool nir_lower_tex(nir_shader
*shader
,
2525 const nir_lower_tex_options
*options
);
2527 bool nir_lower_idiv(nir_shader
*shader
);
2529 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2530 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2531 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2533 void nir_lower_two_sided_color(nir_shader
*shader
);
2535 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
2537 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2538 void nir_lower_tes_patch_vertices(nir_shader
*tes
, unsigned patch_vertices
);
2540 typedef struct nir_lower_wpos_ytransform_options
{
2541 int state_tokens
[5];
2542 bool fs_coord_origin_upper_left
:1;
2543 bool fs_coord_origin_lower_left
:1;
2544 bool fs_coord_pixel_center_integer
:1;
2545 bool fs_coord_pixel_center_half_integer
:1;
2546 } nir_lower_wpos_ytransform_options
;
2548 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2549 const nir_lower_wpos_ytransform_options
*options
);
2550 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
2552 typedef struct nir_lower_drawpixels_options
{
2553 int texcoord_state_tokens
[5];
2554 int scale_state_tokens
[5];
2555 int bias_state_tokens
[5];
2556 unsigned drawpix_sampler
;
2557 unsigned pixelmap_sampler
;
2559 bool scale_and_bias
:1;
2560 } nir_lower_drawpixels_options
;
2562 void nir_lower_drawpixels(nir_shader
*shader
,
2563 const nir_lower_drawpixels_options
*options
);
2565 typedef struct nir_lower_bitmap_options
{
2568 } nir_lower_bitmap_options
;
2570 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2572 bool nir_lower_atomics(nir_shader
*shader
,
2573 const struct gl_shader_program
*shader_program
);
2574 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
2575 bool nir_lower_to_source_mods(nir_shader
*shader
);
2577 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2580 nir_lower_imul64
= (1 << 0),
2581 nir_lower_isign64
= (1 << 1),
2582 /** Lower all int64 modulus and division opcodes */
2583 nir_lower_divmod64
= (1 << 2),
2584 } nir_lower_int64_options
;
2586 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
2589 nir_lower_drcp
= (1 << 0),
2590 nir_lower_dsqrt
= (1 << 1),
2591 nir_lower_drsq
= (1 << 2),
2592 nir_lower_dtrunc
= (1 << 3),
2593 nir_lower_dfloor
= (1 << 4),
2594 nir_lower_dceil
= (1 << 5),
2595 nir_lower_dfract
= (1 << 6),
2596 nir_lower_dround_even
= (1 << 7),
2597 nir_lower_dmod
= (1 << 8)
2598 } nir_lower_doubles_options
;
2600 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2601 bool nir_lower_64bit_pack(nir_shader
*shader
);
2603 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2605 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2607 void nir_loop_analyze_impl(nir_function_impl
*impl
,
2608 nir_variable_mode indirect_mask
);
2610 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2612 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2613 bool nir_repair_ssa(nir_shader
*shader
);
2615 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
2617 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2618 * registers. If false, convert all values (even those not involved in a phi
2619 * node) to registers.
2621 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2623 bool nir_lower_phis_to_regs_block(nir_block
*block
);
2624 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
2626 bool nir_opt_algebraic(nir_shader
*shader
);
2627 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
2628 bool nir_opt_algebraic_late(nir_shader
*shader
);
2629 bool nir_opt_constant_folding(nir_shader
*shader
);
2631 bool nir_opt_global_to_local(nir_shader
*shader
);
2633 bool nir_copy_prop(nir_shader
*shader
);
2635 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
2637 bool nir_opt_cse(nir_shader
*shader
);
2639 bool nir_opt_dce(nir_shader
*shader
);
2641 bool nir_opt_dead_cf(nir_shader
*shader
);
2643 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2645 bool nir_opt_if(nir_shader
*shader
);
2647 bool nir_opt_intrinsics(nir_shader
*shader
);
2649 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
2651 bool nir_opt_move_comparisons(nir_shader
*shader
);
2653 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2655 bool nir_opt_remove_phis(nir_shader
*shader
);
2657 bool nir_opt_trivial_continues(nir_shader
*shader
);
2659 bool nir_opt_undef(nir_shader
*shader
);
2661 bool nir_opt_conditional_discard(nir_shader
*shader
);
2663 void nir_sweep(nir_shader
*shader
);
2665 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2666 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);