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"
55 struct gl_shader_program
;
58 #define NIR_TRUE (~0u)
60 /** Defines a cast function
62 * This macro defines a cast function from in_type to out_type where
63 * out_type is some structure type that contains a field of type out_type.
65 * Note that you have to be a bit careful as the generated cast function
68 #define NIR_DEFINE_CAST(name, in_type, out_type, field, \
69 type_field, type_value) \
70 static inline out_type * \
71 name(const in_type *parent) \
73 assert(parent && parent->type_field == type_value); \
74 return exec_node_data(out_type, parent, field); \
83 * Description of built-in state associated with a uniform
85 * \sa nir_variable::state_slots
93 nir_var_shader_in
= (1 << 0),
94 nir_var_shader_out
= (1 << 1),
95 nir_var_global
= (1 << 2),
96 nir_var_local
= (1 << 3),
97 nir_var_uniform
= (1 << 4),
98 nir_var_shader_storage
= (1 << 5),
99 nir_var_system_value
= (1 << 6),
100 nir_var_param
= (1 << 7),
101 nir_var_shared
= (1 << 8),
115 typedef struct nir_constant
{
117 * Value of the constant.
119 * The field used to back the values supplied by the constant is determined
120 * by the type associated with the \c nir_variable. Constants may be
121 * scalars, vectors, or matrices.
123 nir_const_value values
[4];
125 /* we could get this from the var->type but makes clone *much* easier to
126 * not have to care about the type.
128 unsigned num_elements
;
130 /* Array elements / Structure Fields */
131 struct nir_constant
**elements
;
135 * \brief Layout qualifiers for gl_FragDepth.
137 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
138 * with a layout qualifier.
141 nir_depth_layout_none
, /**< No depth layout is specified. */
142 nir_depth_layout_any
,
143 nir_depth_layout_greater
,
144 nir_depth_layout_less
,
145 nir_depth_layout_unchanged
149 * Either a uniform, global variable, shader input, or shader output. Based on
150 * ir_variable - it should be easy to translate between the two.
153 typedef struct nir_variable
{
154 struct exec_node node
;
157 * Declared type of the variable
159 const struct glsl_type
*type
;
162 * Declared name of the variable
166 struct nir_variable_data
{
168 * Storage class of the variable.
170 * \sa nir_variable_mode
172 nir_variable_mode mode
;
175 * Is the variable read-only?
177 * This is set for variables declared as \c const, shader inputs,
180 unsigned read_only
:1;
184 unsigned invariant
:1;
187 * Interpolation mode for shader inputs / outputs
189 * \sa glsl_interp_mode
191 unsigned interpolation
:2;
194 * \name ARB_fragment_coord_conventions
197 unsigned origin_upper_left
:1;
198 unsigned pixel_center_integer
:1;
202 * If non-zero, then this variable may be packed along with other variables
203 * into a single varying slot, so this offset should be applied when
204 * accessing components. For example, an offset of 1 means that the x
205 * component of this variable is actually stored in component y of the
206 * location specified by \c location.
208 unsigned location_frac
:2;
211 * If true, this variable represents an array of scalars that should
212 * be tightly packed. In other words, consecutive array elements
213 * should be stored one component apart, rather than one slot apart.
218 * Whether this is a fragment shader output implicitly initialized with
219 * the previous contents of the specified render target at the
220 * framebuffer location corresponding to this shader invocation.
222 unsigned fb_fetch_output
:1;
225 * \brief Layout qualifier for gl_FragDepth.
227 * This is not equal to \c ir_depth_layout_none if and only if this
228 * variable is \c gl_FragDepth and a layout qualifier is specified.
230 nir_depth_layout depth_layout
;
233 * Storage location of the base of this variable
235 * The precise meaning of this field depends on the nature of the variable.
237 * - Vertex shader input: one of the values from \c gl_vert_attrib.
238 * - Vertex shader output: one of the values from \c gl_varying_slot.
239 * - Geometry shader input: one of the values from \c gl_varying_slot.
240 * - Geometry shader output: one of the values from \c gl_varying_slot.
241 * - Fragment shader input: one of the values from \c gl_varying_slot.
242 * - Fragment shader output: one of the values from \c gl_frag_result.
243 * - Uniforms: Per-stage uniform slot number for default uniform block.
244 * - Uniforms: Index within the uniform block definition for UBO members.
245 * - Non-UBO Uniforms: uniform slot number.
246 * - Other: This field is not currently used.
248 * If the variable is a uniform, shader input, or shader output, and the
249 * slot has not been assigned, the value will be -1.
254 * The actual location of the variable in the IR. Only valid for inputs
257 unsigned int driver_location
;
260 * output index for dual source blending.
265 * Descriptor set binding for sampler or UBO.
270 * Initial binding point for a sampler or UBO.
272 * For array types, this represents the binding point for the first element.
277 * Location an atomic counter is stored at.
282 * ARB_shader_image_load_store qualifiers.
285 bool read_only
; /**< "readonly" qualifier. */
286 bool write_only
; /**< "writeonly" qualifier. */
291 /** Image internal format if specified explicitly, otherwise GL_NONE. */
297 * Built-in state that backs this uniform
299 * Once set at variable creation, \c state_slots must remain invariant.
300 * This is because, ideally, this array would be shared by all clones of
301 * this variable in the IR tree. In other words, we'd really like for it
302 * to be a fly-weight.
304 * If the variable is not a uniform, \c num_state_slots will be zero and
305 * \c state_slots will be \c NULL.
308 unsigned num_state_slots
; /**< Number of state slots used */
309 nir_state_slot
*state_slots
; /**< State descriptors. */
313 * Constant expression assigned in the initializer of the variable
315 * This field should only be used temporarily by creators of NIR shaders
316 * and then lower_constant_initializers can be used to get rid of them.
317 * Most of the rest of NIR ignores this field or asserts that it's NULL.
319 nir_constant
*constant_initializer
;
322 * For variables that are in an interface block or are an instance of an
323 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
325 * \sa ir_variable::location
327 const struct glsl_type
*interface_type
;
330 #define nir_foreach_variable(var, var_list) \
331 foreach_list_typed(nir_variable, var, node, var_list)
333 #define nir_foreach_variable_safe(var, var_list) \
334 foreach_list_typed_safe(nir_variable, var, node, var_list)
337 nir_variable_is_global(const nir_variable
*var
)
339 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
342 typedef struct nir_register
{
343 struct exec_node node
;
345 unsigned num_components
; /** < number of vector components */
346 unsigned num_array_elems
; /** < size of array (0 for no array) */
348 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
351 /** generic register index. */
354 /** only for debug purposes, can be NULL */
357 /** whether this register is local (per-function) or global (per-shader) */
361 * If this flag is set to true, then accessing channels >= num_components
362 * is well-defined, and simply spills over to the next array element. This
363 * is useful for backends that can do per-component accessing, in
364 * particular scalar backends. By setting this flag and making
365 * num_components equal to 1, structures can be packed tightly into
366 * registers and then registers can be accessed per-component to get to
367 * each structure member, even if it crosses vec4 boundaries.
371 /** set of nir_srcs where this register is used (read from) */
372 struct list_head uses
;
374 /** set of nir_dests where this register is defined (written to) */
375 struct list_head defs
;
377 /** set of nir_ifs where this register is used as a condition */
378 struct list_head if_uses
;
381 #define nir_foreach_register(reg, reg_list) \
382 foreach_list_typed(nir_register, reg, node, reg_list)
383 #define nir_foreach_register_safe(reg, reg_list) \
384 foreach_list_typed_safe(nir_register, reg, node, reg_list)
390 nir_instr_type_intrinsic
,
391 nir_instr_type_load_const
,
393 nir_instr_type_ssa_undef
,
395 nir_instr_type_parallel_copy
,
398 typedef struct nir_instr
{
399 struct exec_node node
;
401 struct nir_block
*block
;
403 /** generic instruction index. */
406 /* A temporary for optimization and analysis passes to use for storing
407 * flags. For instance, DCE uses this to store the "dead/live" info.
412 static inline nir_instr
*
413 nir_instr_next(nir_instr
*instr
)
415 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
416 if (exec_node_is_tail_sentinel(next
))
419 return exec_node_data(nir_instr
, next
, node
);
422 static inline nir_instr
*
423 nir_instr_prev(nir_instr
*instr
)
425 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
426 if (exec_node_is_head_sentinel(prev
))
429 return exec_node_data(nir_instr
, prev
, node
);
433 nir_instr_is_first(nir_instr
*instr
)
435 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
439 nir_instr_is_last(nir_instr
*instr
)
441 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
444 typedef struct nir_ssa_def
{
445 /** for debugging only, can be NULL */
448 /** generic SSA definition index. */
451 /** Index into the live_in and live_out bitfields */
454 nir_instr
*parent_instr
;
456 /** set of nir_instrs where this register is used (read from) */
457 struct list_head uses
;
459 /** set of nir_ifs where this register is used as a condition */
460 struct list_head if_uses
;
462 uint8_t num_components
;
464 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
472 struct nir_src
*indirect
; /** < NULL for no indirect offset */
473 unsigned base_offset
;
475 /* TODO use-def chain goes here */
479 nir_instr
*parent_instr
;
480 struct list_head def_link
;
483 struct nir_src
*indirect
; /** < NULL for no indirect offset */
484 unsigned base_offset
;
486 /* TODO def-use chain goes here */
491 typedef struct nir_src
{
493 nir_instr
*parent_instr
;
494 struct nir_if
*parent_if
;
497 struct list_head use_link
;
507 static inline nir_src
510 nir_src src
= { { NULL
} };
514 #define NIR_SRC_INIT nir_src_init()
516 #define nir_foreach_use(src, reg_or_ssa_def) \
517 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
519 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
520 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
522 #define nir_foreach_if_use(src, reg_or_ssa_def) \
523 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
525 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
526 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
537 static inline nir_dest
540 nir_dest dest
= { { { NULL
} } };
544 #define NIR_DEST_INIT nir_dest_init()
546 #define nir_foreach_def(dest, reg) \
547 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
549 #define nir_foreach_def_safe(dest, reg) \
550 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
552 static inline nir_src
553 nir_src_for_ssa(nir_ssa_def
*def
)
555 nir_src src
= NIR_SRC_INIT
;
563 static inline nir_src
564 nir_src_for_reg(nir_register
*reg
)
566 nir_src src
= NIR_SRC_INIT
;
570 src
.reg
.indirect
= NULL
;
571 src
.reg
.base_offset
= 0;
576 static inline nir_dest
577 nir_dest_for_reg(nir_register
*reg
)
579 nir_dest dest
= NIR_DEST_INIT
;
586 static inline unsigned
587 nir_src_bit_size(nir_src src
)
589 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
592 static inline unsigned
593 nir_dest_bit_size(nir_dest dest
)
595 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
598 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
599 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
605 * \name input modifiers
609 * For inputs interpreted as floating point, flips the sign bit. For
610 * inputs interpreted as integers, performs the two's complement negation.
615 * Clears the sign bit for floating point values, and computes the integer
616 * absolute value for integers. Note that the negate modifier acts after
617 * the absolute value modifier, therefore if both are set then all inputs
618 * will become negative.
624 * For each input component, says which component of the register it is
625 * chosen from. Note that which elements of the swizzle are used and which
626 * are ignored are based on the write mask for most opcodes - for example,
627 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
628 * a swizzle of {2, x, 1, 0} where x means "don't care."
637 * \name saturate output modifier
639 * Only valid for opcodes that output floating-point numbers. Clamps the
640 * output to between 0.0 and 1.0 inclusive.
645 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
649 nir_type_invalid
= 0, /* Not a valid type */
654 nir_type_bool32
= 32 | nir_type_bool
,
655 nir_type_int8
= 8 | nir_type_int
,
656 nir_type_int16
= 16 | nir_type_int
,
657 nir_type_int32
= 32 | nir_type_int
,
658 nir_type_int64
= 64 | nir_type_int
,
659 nir_type_uint8
= 8 | nir_type_uint
,
660 nir_type_uint16
= 16 | nir_type_uint
,
661 nir_type_uint32
= 32 | nir_type_uint
,
662 nir_type_uint64
= 64 | nir_type_uint
,
663 nir_type_float16
= 16 | nir_type_float
,
664 nir_type_float32
= 32 | nir_type_float
,
665 nir_type_float64
= 64 | nir_type_float
,
668 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
669 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
671 static inline unsigned
672 nir_alu_type_get_type_size(nir_alu_type type
)
674 return type
& NIR_ALU_TYPE_SIZE_MASK
;
677 static inline unsigned
678 nir_alu_type_get_base_type(nir_alu_type type
)
680 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
683 static inline nir_alu_type
684 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
688 return nir_type_bool32
;
691 return nir_type_uint32
;
694 return nir_type_int32
;
696 case GLSL_TYPE_UINT64
:
697 return nir_type_uint64
;
699 case GLSL_TYPE_INT64
:
700 return nir_type_int64
;
702 case GLSL_TYPE_FLOAT
:
703 return nir_type_float32
;
705 case GLSL_TYPE_DOUBLE
:
706 return nir_type_float64
;
709 unreachable("unknown type");
713 static inline nir_alu_type
714 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
716 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
719 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
);
722 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
723 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
724 } nir_op_algebraic_property
;
732 * The number of components in the output
734 * If non-zero, this is the size of the output and input sizes are
735 * explicitly given; swizzle and writemask are still in effect, but if
736 * the output component is masked out, then the input component may
739 * If zero, the opcode acts in the standard, per-component manner; the
740 * operation is performed on each component (except the ones that are
741 * masked out) with the input being taken from the input swizzle for
744 * The size of some of the inputs may be given (i.e. non-zero) even
745 * though output_size is zero; in that case, the inputs with a zero
746 * size act per-component, while the inputs with non-zero size don't.
748 unsigned output_size
;
751 * The type of vector that the instruction outputs. Note that the
752 * staurate modifier is only allowed on outputs with the float type.
755 nir_alu_type output_type
;
758 * The number of components in each input
760 unsigned input_sizes
[4];
763 * The type of vector that each input takes. Note that negate and
764 * absolute value are only allowed on inputs with int or float type and
765 * behave differently on the two.
767 nir_alu_type input_types
[4];
769 nir_op_algebraic_property algebraic_properties
;
772 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
774 typedef struct nir_alu_instr
{
778 /** Indicates that this ALU instruction generates an exact value
780 * This is kind of a mixture of GLSL "precise" and "invariant" and not
781 * really equivalent to either. This indicates that the value generated by
782 * this operation is high-precision and any code transformations that touch
783 * it must ensure that the resulting value is bit-for-bit identical to the
792 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
793 nir_alu_instr
*instr
);
794 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
795 nir_alu_instr
*instr
);
797 /* is this source channel used? */
799 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
801 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
802 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
804 return (instr
->dest
.write_mask
>> channel
) & 1;
808 * For instructions whose destinations are SSA, get the number of channels
811 static inline unsigned
812 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
814 assert(instr
->dest
.dest
.is_ssa
);
816 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
817 return nir_op_infos
[instr
->op
].input_sizes
[src
];
819 return instr
->dest
.dest
.ssa
.num_components
;
822 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
823 unsigned src1
, unsigned src2
);
827 nir_deref_type_array
,
828 nir_deref_type_struct
831 typedef struct nir_deref
{
832 nir_deref_type deref_type
;
833 struct nir_deref
*child
;
834 const struct glsl_type
*type
;
843 /* This enum describes how the array is referenced. If the deref is
844 * direct then the base_offset is used. If the deref is indirect then
845 * offset is given by base_offset + indirect. If the deref is a wildcard
846 * then the deref refers to all of the elements of the array at the same
847 * time. Wildcard dereferences are only ever allowed in copy_var
848 * intrinsics and the source and destination derefs must have matching
852 nir_deref_array_type_direct
,
853 nir_deref_array_type_indirect
,
854 nir_deref_array_type_wildcard
,
855 } nir_deref_array_type
;
860 nir_deref_array_type deref_array_type
;
861 unsigned base_offset
;
871 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
,
872 deref_type
, nir_deref_type_var
)
873 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
,
874 deref_type
, nir_deref_type_array
)
875 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
,
876 deref_type
, nir_deref_type_struct
)
878 /* Returns the last deref in the chain. */
879 static inline nir_deref
*
880 nir_deref_tail(nir_deref
*deref
)
883 deref
= deref
->child
;
891 nir_deref_var
**params
;
892 nir_deref_var
*return_deref
;
894 struct nir_function
*callee
;
897 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
898 num_variables, num_indices, idx0, idx1, idx2, flags) \
899 nir_intrinsic_##name,
901 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
904 #include "nir_intrinsics.h"
905 nir_num_intrinsics
= nir_last_intrinsic
+ 1
908 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
910 /** Represents an intrinsic
912 * An intrinsic is an instruction type for handling things that are
913 * more-or-less regular operations but don't just consume and produce SSA
914 * values like ALU operations do. Intrinsics are not for things that have
915 * special semantic meaning such as phi nodes and parallel copies.
916 * Examples of intrinsics include variable load/store operations, system
917 * value loads, and the like. Even though texturing more-or-less falls
918 * under this category, texturing is its own instruction type because
919 * trying to represent texturing with intrinsics would lead to a
920 * combinatorial explosion of intrinsic opcodes.
922 * By having a single instruction type for handling a lot of different
923 * cases, optimization passes can look for intrinsics and, for the most
924 * part, completely ignore them. Each intrinsic type also has a few
925 * possible flags that govern whether or not they can be reordered or
926 * eliminated. That way passes like dead code elimination can still work
927 * on intrisics without understanding the meaning of each.
929 * Each intrinsic has some number of constant indices, some number of
930 * variables, and some number of sources. What these sources, variables,
931 * and indices mean depends on the intrinsic and is documented with the
932 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
933 * instructions are the only types of instruction that can operate on
939 nir_intrinsic_op intrinsic
;
943 /** number of components if this is a vectorized intrinsic
945 * Similarly to ALU operations, some intrinsics are vectorized.
946 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
947 * For vectorized intrinsics, the num_components field specifies the
948 * number of destination components and the number of source components
949 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
951 uint8_t num_components
;
953 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
955 nir_deref_var
*variables
[2];
958 } nir_intrinsic_instr
;
961 * \name NIR intrinsics semantic flags
963 * information about what the compiler can do with the intrinsics.
965 * \sa nir_intrinsic_info::flags
969 * whether the intrinsic can be safely eliminated if none of its output
970 * value is not being used.
972 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
975 * Whether the intrinsic can be reordered with respect to any other
976 * intrinsic, i.e. whether the only reordering dependencies of the
977 * intrinsic are due to the register reads/writes.
979 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
980 } nir_intrinsic_semantic_flag
;
983 * \name NIR intrinsics const-index flag
985 * Indicates the usage of a const_index slot.
987 * \sa nir_intrinsic_info::index_map
991 * Generally instructions that take a offset src argument, can encode
992 * a constant 'base' value which is added to the offset.
994 NIR_INTRINSIC_BASE
= 1,
997 * For store instructions, a writemask for the store.
999 NIR_INTRINSIC_WRMASK
= 2,
1002 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1004 NIR_INTRINSIC_STREAM_ID
= 3,
1007 * The clip-plane id for load_user_clip_plane intrinsic.
1009 NIR_INTRINSIC_UCP_ID
= 4,
1012 * The amount of data, starting from BASE, that this instruction may
1013 * access. This is used to provide bounds if the offset is not constant.
1015 NIR_INTRINSIC_RANGE
= 5,
1018 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1020 NIR_INTRINSIC_DESC_SET
= 6,
1023 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1025 NIR_INTRINSIC_BINDING
= 7,
1030 NIR_INTRINSIC_COMPONENT
= 8,
1033 * Interpolation mode (only meaningful for FS inputs).
1035 NIR_INTRINSIC_INTERP_MODE
= 9,
1037 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1039 } nir_intrinsic_index_flag
;
1041 #define NIR_INTRINSIC_MAX_INPUTS 4
1046 unsigned num_srcs
; /** < number of register/SSA inputs */
1048 /** number of components of each input register
1050 * If this value is 0, the number of components is given by the
1051 * num_components field of nir_intrinsic_instr.
1053 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1057 /** number of components of the output 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 dest_components
;
1064 /** the number of inputs/outputs that are variables */
1065 unsigned num_variables
;
1067 /** the number of constant indices used by the intrinsic */
1068 unsigned num_indices
;
1070 /** indicates the usage of intr->const_index[n] */
1071 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1073 /** semantic flags for calls to this intrinsic */
1074 nir_intrinsic_semantic_flag flags
;
1075 } nir_intrinsic_info
;
1077 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1080 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1081 static inline type \
1082 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
1084 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1085 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1086 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1088 static inline void \
1089 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1091 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1092 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1093 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1096 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1097 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1098 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1099 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1100 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1101 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1102 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1103 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1104 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1107 * \group texture information
1109 * This gives semantic information about textures which is useful to the
1110 * frontend, the backend, and lowering passes, but not the optimizer.
1115 nir_tex_src_projector
,
1116 nir_tex_src_comparator
, /* shadow comparator */
1120 nir_tex_src_ms_index
, /* MSAA sample index */
1121 nir_tex_src_ms_mcs
, /* MSAA compression value */
1124 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1125 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1126 nir_tex_src_plane
, /* < selects plane for planar textures */
1127 nir_num_tex_src_types
1132 nir_tex_src_type src_type
;
1136 nir_texop_tex
, /**< Regular texture look-up */
1137 nir_texop_txb
, /**< Texture look-up with LOD bias */
1138 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1139 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1140 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1141 nir_texop_txf_ms
, /**< Multisample texture fetch */
1142 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1143 nir_texop_txs
, /**< Texture size */
1144 nir_texop_lod
, /**< Texture lod query */
1145 nir_texop_tg4
, /**< Texture gather */
1146 nir_texop_query_levels
, /**< Texture levels query */
1147 nir_texop_texture_samples
, /**< Texture samples query */
1148 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1156 enum glsl_sampler_dim sampler_dim
;
1157 nir_alu_type dest_type
;
1162 unsigned num_srcs
, coord_components
;
1163 bool is_array
, is_shadow
;
1166 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1167 * components or the new-style shadow that outputs 1 component.
1169 bool is_new_style_shadow
;
1171 /* gather component selector */
1172 unsigned component
: 2;
1174 /** The texture index
1176 * If this texture instruction has a nir_tex_src_texture_offset source,
1177 * then the texture index is given by texture_index + texture_offset.
1179 unsigned texture_index
;
1181 /** The size of the texture array or 0 if it's not an array */
1182 unsigned texture_array_size
;
1184 /** The texture deref
1186 * If this is null, use texture_index instead.
1188 nir_deref_var
*texture
;
1190 /** The sampler index
1192 * The following operations do not require a sampler and, as such, this
1193 * field should be ignored:
1195 * - nir_texop_txf_ms
1199 * - nir_texop_query_levels
1200 * - nir_texop_texture_samples
1201 * - nir_texop_samples_identical
1203 * If this texture instruction has a nir_tex_src_sampler_offset source,
1204 * then the sampler index is given by sampler_index + sampler_offset.
1206 unsigned sampler_index
;
1208 /** The sampler deref
1210 * If this is null, use sampler_index instead.
1212 nir_deref_var
*sampler
;
1215 static inline unsigned
1216 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1218 switch (instr
->op
) {
1219 case nir_texop_txs
: {
1221 switch (instr
->sampler_dim
) {
1222 case GLSL_SAMPLER_DIM_1D
:
1223 case GLSL_SAMPLER_DIM_BUF
:
1226 case GLSL_SAMPLER_DIM_2D
:
1227 case GLSL_SAMPLER_DIM_CUBE
:
1228 case GLSL_SAMPLER_DIM_MS
:
1229 case GLSL_SAMPLER_DIM_RECT
:
1230 case GLSL_SAMPLER_DIM_EXTERNAL
:
1231 case GLSL_SAMPLER_DIM_SUBPASS
:
1234 case GLSL_SAMPLER_DIM_3D
:
1238 unreachable("not reached");
1240 if (instr
->is_array
)
1248 case nir_texop_texture_samples
:
1249 case nir_texop_query_levels
:
1250 case nir_texop_samples_identical
:
1254 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1261 /* Returns true if this texture operation queries something about the texture
1262 * rather than actually sampling it.
1265 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1267 switch (instr
->op
) {
1270 case nir_texop_texture_samples
:
1271 case nir_texop_query_levels
:
1272 case nir_texop_txf_ms_mcs
:
1279 case nir_texop_txf_ms
:
1283 unreachable("Invalid texture opcode");
1287 static inline nir_alu_type
1288 nir_tex_instr_src_type(nir_tex_instr
*instr
, unsigned src
)
1290 switch (instr
->src
[src
].src_type
) {
1291 case nir_tex_src_coord
:
1292 switch (instr
->op
) {
1294 case nir_texop_txf_ms
:
1295 case nir_texop_txf_ms_mcs
:
1296 case nir_texop_samples_identical
:
1297 return nir_type_int
;
1300 return nir_type_float
;
1303 case nir_tex_src_lod
:
1304 switch (instr
->op
) {
1307 return nir_type_int
;
1310 return nir_type_float
;
1313 case nir_tex_src_projector
:
1314 case nir_tex_src_comparator
:
1315 case nir_tex_src_bias
:
1316 case nir_tex_src_ddx
:
1317 case nir_tex_src_ddy
:
1318 return nir_type_float
;
1320 case nir_tex_src_offset
:
1321 case nir_tex_src_ms_index
:
1322 case nir_tex_src_texture_offset
:
1323 case nir_tex_src_sampler_offset
:
1324 return nir_type_int
;
1327 unreachable("Invalid texture source type");
1331 static inline unsigned
1332 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1334 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1335 return instr
->coord_components
;
1337 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1338 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1341 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1342 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1343 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1344 if (instr
->is_array
)
1345 return instr
->coord_components
- 1;
1347 return instr
->coord_components
;
1354 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1356 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1357 if (instr
->src
[i
].src_type
== type
)
1363 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1368 nir_const_value value
;
1371 } nir_load_const_instr
;
1384 /* creates a new SSA variable in an undefined state */
1389 } nir_ssa_undef_instr
;
1392 struct exec_node node
;
1394 /* The predecessor block corresponding to this source */
1395 struct nir_block
*pred
;
1400 #define nir_foreach_phi_src(phi_src, phi) \
1401 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1402 #define nir_foreach_phi_src_safe(phi_src, phi) \
1403 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1408 struct exec_list srcs
; /** < list of nir_phi_src */
1414 struct exec_node node
;
1417 } nir_parallel_copy_entry
;
1419 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1420 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1425 /* A list of nir_parallel_copy_entrys. The sources of all of the
1426 * entries are copied to the corresponding destinations "in parallel".
1427 * In other words, if we have two entries: a -> b and b -> a, the values
1430 struct exec_list entries
;
1431 } nir_parallel_copy_instr
;
1433 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1434 type
, nir_instr_type_alu
)
1435 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1436 type
, nir_instr_type_call
)
1437 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1438 type
, nir_instr_type_jump
)
1439 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1440 type
, nir_instr_type_tex
)
1441 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1442 type
, nir_instr_type_intrinsic
)
1443 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1444 type
, nir_instr_type_load_const
)
1445 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1446 type
, nir_instr_type_ssa_undef
)
1447 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1448 type
, nir_instr_type_phi
)
1449 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1450 nir_parallel_copy_instr
, instr
,
1451 type
, nir_instr_type_parallel_copy
)
1456 * Control flow consists of a tree of control flow nodes, which include
1457 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1458 * instructions that always run start-to-finish. Each basic block also keeps
1459 * track of its successors (blocks which may run immediately after the current
1460 * block) and predecessors (blocks which could have run immediately before the
1461 * current block). Each function also has a start block and an end block which
1462 * all return statements point to (which is always empty). Together, all the
1463 * blocks with their predecessors and successors make up the control flow
1464 * graph (CFG) of the function. There are helpers that modify the tree of
1465 * control flow nodes while modifying the CFG appropriately; these should be
1466 * used instead of modifying the tree directly.
1473 nir_cf_node_function
1476 typedef struct nir_cf_node
{
1477 struct exec_node node
;
1478 nir_cf_node_type type
;
1479 struct nir_cf_node
*parent
;
1482 typedef struct nir_block
{
1483 nir_cf_node cf_node
;
1485 struct exec_list instr_list
; /** < list of nir_instr */
1487 /** generic block index; generated by nir_index_blocks */
1491 * Each block can only have up to 2 successors, so we put them in a simple
1492 * array - no need for anything more complicated.
1494 struct nir_block
*successors
[2];
1496 /* Set of nir_block predecessors in the CFG */
1497 struct set
*predecessors
;
1500 * this node's immediate dominator in the dominance tree - set to NULL for
1503 struct nir_block
*imm_dom
;
1505 /* This node's children in the dominance tree */
1506 unsigned num_dom_children
;
1507 struct nir_block
**dom_children
;
1509 /* Set of nir_blocks on the dominance frontier of this block */
1510 struct set
*dom_frontier
;
1513 * These two indices have the property that dom_{pre,post}_index for each
1514 * child of this block in the dominance tree will always be between
1515 * dom_pre_index and dom_post_index for this block, which makes testing if
1516 * a given block is dominated by another block an O(1) operation.
1518 unsigned dom_pre_index
, dom_post_index
;
1520 /* live in and out for this block; used for liveness analysis */
1521 BITSET_WORD
*live_in
;
1522 BITSET_WORD
*live_out
;
1525 static inline nir_instr
*
1526 nir_block_first_instr(nir_block
*block
)
1528 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1529 return exec_node_data(nir_instr
, head
, node
);
1532 static inline nir_instr
*
1533 nir_block_last_instr(nir_block
*block
)
1535 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1536 return exec_node_data(nir_instr
, tail
, node
);
1539 #define nir_foreach_instr(instr, block) \
1540 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1541 #define nir_foreach_instr_reverse(instr, block) \
1542 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1543 #define nir_foreach_instr_safe(instr, block) \
1544 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1545 #define nir_foreach_instr_reverse_safe(instr, block) \
1546 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1548 typedef struct nir_if
{
1549 nir_cf_node cf_node
;
1552 struct exec_list then_list
; /** < list of nir_cf_node */
1553 struct exec_list else_list
; /** < list of nir_cf_node */
1559 nir_instr
*conditional_instr
;
1561 nir_block
*break_block
;
1562 nir_block
*continue_from_block
;
1564 bool continue_from_then
;
1566 struct list_head loop_terminator_link
;
1567 } nir_loop_terminator
;
1570 /* Number of instructions in the loop */
1571 unsigned num_instructions
;
1573 /* How many times the loop is run (if known) */
1574 unsigned trip_count
;
1575 bool is_trip_count_known
;
1577 /* Unroll the loop regardless of its size */
1580 nir_loop_terminator
*limiting_terminator
;
1582 /* A list of loop_terminators terminating this loop. */
1583 struct list_head loop_terminator_list
;
1587 nir_cf_node cf_node
;
1589 struct exec_list body
; /** < list of nir_cf_node */
1591 nir_loop_info
*info
;
1595 * Various bits of metadata that can may be created or required by
1596 * optimization and analysis passes
1599 nir_metadata_none
= 0x0,
1600 nir_metadata_block_index
= 0x1,
1601 nir_metadata_dominance
= 0x2,
1602 nir_metadata_live_ssa_defs
= 0x4,
1603 nir_metadata_not_properly_reset
= 0x8,
1604 nir_metadata_loop_analysis
= 0x10,
1608 nir_cf_node cf_node
;
1610 /** pointer to the function of which this is an implementation */
1611 struct nir_function
*function
;
1613 struct exec_list body
; /** < list of nir_cf_node */
1615 nir_block
*end_block
;
1617 /** list for all local variables in the function */
1618 struct exec_list locals
;
1620 /** array of variables used as parameters */
1621 unsigned num_params
;
1622 nir_variable
**params
;
1624 /** variable used to hold the result of the function */
1625 nir_variable
*return_var
;
1627 /** list of local registers in the function */
1628 struct exec_list registers
;
1630 /** next available local register index */
1633 /** next available SSA value index */
1636 /* total number of basic blocks, only valid when block_index_dirty = false */
1637 unsigned num_blocks
;
1639 nir_metadata valid_metadata
;
1640 } nir_function_impl
;
1642 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1643 nir_start_block(nir_function_impl
*impl
)
1645 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1648 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1649 nir_impl_last_block(nir_function_impl
*impl
)
1651 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1654 static inline nir_cf_node
*
1655 nir_cf_node_next(nir_cf_node
*node
)
1657 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1658 if (exec_node_is_tail_sentinel(next
))
1661 return exec_node_data(nir_cf_node
, next
, node
);
1664 static inline nir_cf_node
*
1665 nir_cf_node_prev(nir_cf_node
*node
)
1667 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1668 if (exec_node_is_head_sentinel(prev
))
1671 return exec_node_data(nir_cf_node
, prev
, node
);
1675 nir_cf_node_is_first(const nir_cf_node
*node
)
1677 return exec_node_is_head_sentinel(node
->node
.prev
);
1681 nir_cf_node_is_last(const nir_cf_node
*node
)
1683 return exec_node_is_tail_sentinel(node
->node
.next
);
1686 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1687 type
, nir_cf_node_block
)
1688 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1689 type
, nir_cf_node_if
)
1690 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1691 type
, nir_cf_node_loop
)
1692 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1693 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1695 static inline nir_block
*
1696 nir_if_first_then_block(nir_if
*if_stmt
)
1698 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1699 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1702 static inline nir_block
*
1703 nir_if_last_then_block(nir_if
*if_stmt
)
1705 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1706 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1709 static inline nir_block
*
1710 nir_if_first_else_block(nir_if
*if_stmt
)
1712 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1713 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1716 static inline nir_block
*
1717 nir_if_last_else_block(nir_if
*if_stmt
)
1719 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1720 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1723 static inline nir_block
*
1724 nir_loop_first_block(nir_loop
*loop
)
1726 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1727 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1730 static inline nir_block
*
1731 nir_loop_last_block(nir_loop
*loop
)
1733 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1734 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1740 nir_parameter_inout
,
1741 } nir_parameter_type
;
1744 nir_parameter_type param_type
;
1745 const struct glsl_type
*type
;
1748 typedef struct nir_function
{
1749 struct exec_node node
;
1752 struct nir_shader
*shader
;
1754 unsigned num_params
;
1755 nir_parameter
*params
;
1756 const struct glsl_type
*return_type
;
1758 /** The implementation of this function.
1760 * If the function is only declared and not implemented, this is NULL.
1762 nir_function_impl
*impl
;
1765 typedef struct nir_shader_compiler_options
{
1770 /** Lowers flrp when it does not support doubles */
1777 bool lower_bitfield_extract
;
1778 bool lower_bitfield_insert
;
1779 bool lower_uadd_carry
;
1780 bool lower_usub_borrow
;
1781 /** lowers fneg and ineg to fsub and isub. */
1783 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1786 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1789 /** enables rules to lower idiv by power-of-two: */
1792 /* Does the native fdot instruction replicate its result for four
1793 * components? If so, then opt_algebraic_late will turn all fdotN
1794 * instructions into fdot_replicatedN instructions.
1796 bool fdot_replicates
;
1798 /** lowers ffract to fsub+ffloor: */
1801 bool lower_pack_half_2x16
;
1802 bool lower_pack_unorm_2x16
;
1803 bool lower_pack_snorm_2x16
;
1804 bool lower_pack_unorm_4x8
;
1805 bool lower_pack_snorm_4x8
;
1806 bool lower_unpack_half_2x16
;
1807 bool lower_unpack_unorm_2x16
;
1808 bool lower_unpack_snorm_2x16
;
1809 bool lower_unpack_unorm_4x8
;
1810 bool lower_unpack_snorm_4x8
;
1812 bool lower_extract_byte
;
1813 bool lower_extract_word
;
1816 * Does the driver support real 32-bit integers? (Otherwise, integers
1817 * are simulated by floats.)
1819 bool native_integers
;
1821 /* Indicates that the driver only has zero-based vertex id */
1822 bool vertex_id_zero_based
;
1824 bool lower_cs_local_index_from_id
;
1827 * Should nir_lower_io() create load_interpolated_input intrinsics?
1829 * If not, it generates regular load_input intrinsics and interpolation
1830 * information must be inferred from the list of input nir_variables.
1832 bool use_interpolated_input_intrinsics
;
1834 unsigned max_unroll_iterations
;
1835 } nir_shader_compiler_options
;
1837 typedef struct nir_shader
{
1838 /** list of uniforms (nir_variable) */
1839 struct exec_list uniforms
;
1841 /** list of inputs (nir_variable) */
1842 struct exec_list inputs
;
1844 /** list of outputs (nir_variable) */
1845 struct exec_list outputs
;
1847 /** list of shared compute variables (nir_variable) */
1848 struct exec_list shared
;
1850 /** Set of driver-specific options for the shader.
1852 * The memory for the options is expected to be kept in a single static
1853 * copy by the driver.
1855 const struct nir_shader_compiler_options
*options
;
1857 /** Various bits of compile-time information about a given shader */
1858 struct shader_info
*info
;
1860 /** list of global variables in the shader (nir_variable) */
1861 struct exec_list globals
;
1863 /** list of system value variables in the shader (nir_variable) */
1864 struct exec_list system_values
;
1866 struct exec_list functions
; /** < list of nir_function */
1868 /** list of global register in the shader */
1869 struct exec_list registers
;
1871 /** next available global register index */
1875 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1878 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1880 /** The shader stage, such as MESA_SHADER_VERTEX. */
1881 gl_shader_stage stage
;
1884 static inline nir_function_impl
*
1885 nir_shader_get_entrypoint(nir_shader
*shader
)
1887 assert(exec_list_length(&shader
->functions
) == 1);
1888 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1889 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1890 assert(func
->return_type
== glsl_void_type());
1891 assert(func
->num_params
== 0);
1896 #define nir_foreach_function(func, shader) \
1897 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1899 nir_shader
*nir_shader_create(void *mem_ctx
,
1900 gl_shader_stage stage
,
1901 const nir_shader_compiler_options
*options
,
1904 /** creates a register, including assigning it an index and adding it to the list */
1905 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1907 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1909 void nir_reg_remove(nir_register
*reg
);
1911 /** Adds a variable to the appropriate list in nir_shader */
1912 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1915 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1917 assert(var
->data
.mode
== nir_var_local
);
1918 exec_list_push_tail(&impl
->locals
, &var
->node
);
1921 /** creates a variable, sets a few defaults, and adds it to the list */
1922 nir_variable
*nir_variable_create(nir_shader
*shader
,
1923 nir_variable_mode mode
,
1924 const struct glsl_type
*type
,
1926 /** creates a local variable and adds it to the list */
1927 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1928 const struct glsl_type
*type
,
1931 /** creates a function and adds it to the shader's list of functions */
1932 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1934 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1935 /** creates a function_impl that isn't tied to any particular function */
1936 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1938 nir_block
*nir_block_create(nir_shader
*shader
);
1939 nir_if
*nir_if_create(nir_shader
*shader
);
1940 nir_loop
*nir_loop_create(nir_shader
*shader
);
1942 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1944 /** requests that the given pieces of metadata be generated */
1945 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
1946 /** dirties all but the preserved metadata */
1947 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1949 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1950 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1952 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1954 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1955 unsigned num_components
,
1958 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1959 nir_intrinsic_op op
);
1961 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1962 nir_function
*callee
);
1964 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1966 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1968 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1970 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1971 unsigned num_components
,
1974 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1975 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1976 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1978 typedef bool (*nir_deref_foreach_leaf_cb
)(nir_deref_var
*deref
, void *state
);
1979 bool nir_deref_foreach_leaf(nir_deref_var
*deref
,
1980 nir_deref_foreach_leaf_cb cb
, void *state
);
1982 nir_load_const_instr
*
1983 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1986 * NIR Cursors and Instruction Insertion API
1989 * A tiny struct representing a point to insert/extract instructions or
1990 * control flow nodes. Helps reduce the combinatorial explosion of possible
1991 * points to insert/extract.
1993 * \sa nir_control_flow.h
1996 nir_cursor_before_block
,
1997 nir_cursor_after_block
,
1998 nir_cursor_before_instr
,
1999 nir_cursor_after_instr
,
2000 } nir_cursor_option
;
2003 nir_cursor_option option
;
2010 static inline nir_block
*
2011 nir_cursor_current_block(nir_cursor cursor
)
2013 if (cursor
.option
== nir_cursor_before_instr
||
2014 cursor
.option
== nir_cursor_after_instr
) {
2015 return cursor
.instr
->block
;
2017 return cursor
.block
;
2021 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2023 static inline nir_cursor
2024 nir_before_block(nir_block
*block
)
2027 cursor
.option
= nir_cursor_before_block
;
2028 cursor
.block
= block
;
2032 static inline nir_cursor
2033 nir_after_block(nir_block
*block
)
2036 cursor
.option
= nir_cursor_after_block
;
2037 cursor
.block
= block
;
2041 static inline nir_cursor
2042 nir_before_instr(nir_instr
*instr
)
2045 cursor
.option
= nir_cursor_before_instr
;
2046 cursor
.instr
= instr
;
2050 static inline nir_cursor
2051 nir_after_instr(nir_instr
*instr
)
2054 cursor
.option
= nir_cursor_after_instr
;
2055 cursor
.instr
= instr
;
2059 static inline nir_cursor
2060 nir_after_block_before_jump(nir_block
*block
)
2062 nir_instr
*last_instr
= nir_block_last_instr(block
);
2063 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2064 return nir_before_instr(last_instr
);
2066 return nir_after_block(block
);
2070 static inline nir_cursor
2071 nir_before_cf_node(nir_cf_node
*node
)
2073 if (node
->type
== nir_cf_node_block
)
2074 return nir_before_block(nir_cf_node_as_block(node
));
2076 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2079 static inline nir_cursor
2080 nir_after_cf_node(nir_cf_node
*node
)
2082 if (node
->type
== nir_cf_node_block
)
2083 return nir_after_block(nir_cf_node_as_block(node
));
2085 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2088 static inline nir_cursor
2089 nir_after_phis(nir_block
*block
)
2091 nir_foreach_instr(instr
, block
) {
2092 if (instr
->type
!= nir_instr_type_phi
)
2093 return nir_before_instr(instr
);
2095 return nir_after_block(block
);
2098 static inline nir_cursor
2099 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2101 if (node
->type
== nir_cf_node_block
)
2102 return nir_after_block(nir_cf_node_as_block(node
));
2104 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2106 return nir_after_phis(block
);
2109 static inline nir_cursor
2110 nir_before_cf_list(struct exec_list
*cf_list
)
2112 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2113 exec_list_get_head(cf_list
), node
);
2114 return nir_before_cf_node(first_node
);
2117 static inline nir_cursor
2118 nir_after_cf_list(struct exec_list
*cf_list
)
2120 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2121 exec_list_get_tail(cf_list
), node
);
2122 return nir_after_cf_node(last_node
);
2126 * Insert a NIR instruction at the given cursor.
2128 * Note: This does not update the cursor.
2130 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2133 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2135 nir_instr_insert(nir_before_instr(instr
), before
);
2139 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2141 nir_instr_insert(nir_after_instr(instr
), after
);
2145 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2147 nir_instr_insert(nir_before_block(block
), before
);
2151 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2153 nir_instr_insert(nir_after_block(block
), after
);
2157 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2159 nir_instr_insert(nir_before_cf_node(node
), before
);
2163 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2165 nir_instr_insert(nir_after_cf_node(node
), after
);
2169 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2171 nir_instr_insert(nir_before_cf_list(list
), before
);
2175 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2177 nir_instr_insert(nir_after_cf_list(list
), after
);
2180 void nir_instr_remove(nir_instr
*instr
);
2184 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2185 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2186 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2187 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2189 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2190 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2192 nir_const_value
*nir_src_as_const_value(nir_src src
);
2193 bool nir_src_is_dynamically_uniform(nir_src src
);
2194 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2195 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2196 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2197 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2198 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2201 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2202 unsigned num_components
, unsigned bit_size
,
2204 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2205 unsigned num_components
, unsigned bit_size
,
2207 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2208 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2209 nir_instr
*after_me
);
2211 uint8_t nir_ssa_def_components_read(nir_ssa_def
*def
);
2214 * finds the next basic block in source-code order, returns NULL if there is
2218 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2220 /* Performs the opposite of nir_block_cf_tree_next() */
2222 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2224 /* Gets the first block in a CF node in source-code order */
2226 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2228 /* Gets the last block in a CF node in source-code order */
2230 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2232 /* Gets the next block after a CF node in source-code order */
2234 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2236 /* Macros for loops that visit blocks in source-code order */
2238 #define nir_foreach_block(block, impl) \
2239 for (nir_block *block = nir_start_block(impl); block != NULL; \
2240 block = nir_block_cf_tree_next(block))
2242 #define nir_foreach_block_safe(block, impl) \
2243 for (nir_block *block = nir_start_block(impl), \
2244 *next = nir_block_cf_tree_next(block); \
2246 block = next, next = nir_block_cf_tree_next(block))
2248 #define nir_foreach_block_reverse(block, impl) \
2249 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2250 block = nir_block_cf_tree_prev(block))
2252 #define nir_foreach_block_reverse_safe(block, impl) \
2253 for (nir_block *block = nir_impl_last_block(impl), \
2254 *prev = nir_block_cf_tree_prev(block); \
2256 block = prev, prev = nir_block_cf_tree_prev(block))
2258 #define nir_foreach_block_in_cf_node(block, node) \
2259 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2260 block != nir_cf_node_cf_tree_next(node); \
2261 block = nir_block_cf_tree_next(block))
2263 /* If the following CF node is an if, this function returns that if.
2264 * Otherwise, it returns NULL.
2266 nir_if
*nir_block_get_following_if(nir_block
*block
);
2268 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2270 void nir_index_local_regs(nir_function_impl
*impl
);
2271 void nir_index_global_regs(nir_shader
*shader
);
2272 void nir_index_ssa_defs(nir_function_impl
*impl
);
2273 unsigned nir_index_instrs(nir_function_impl
*impl
);
2275 void nir_index_blocks(nir_function_impl
*impl
);
2277 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2278 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2279 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2281 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2282 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2283 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2284 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2285 nir_deref
*nir_deref_clone(const nir_deref
*deref
, void *mem_ctx
);
2286 nir_deref_var
*nir_deref_var_clone(const nir_deref_var
*deref
, void *mem_ctx
);
2289 void nir_validate_shader(nir_shader
*shader
);
2290 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2291 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2294 should_clone_nir(void)
2296 static int should_clone
= -1;
2297 if (should_clone
< 0)
2298 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2300 return should_clone
;
2303 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2304 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2305 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2306 static inline bool should_clone_nir(void) { return false; }
2309 #define _PASS(nir, do_pass) do { \
2311 nir_validate_shader(nir); \
2312 if (should_clone_nir()) { \
2313 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2319 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2320 nir_metadata_set_validation_flag(nir); \
2321 if (pass(nir, ##__VA_ARGS__)) { \
2323 nir_metadata_check_validation_flag(nir); \
2327 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2328 pass(nir, ##__VA_ARGS__); \
2331 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2332 void nir_calc_dominance(nir_shader
*shader
);
2334 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2335 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2337 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2338 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2340 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2341 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2343 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2344 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2346 int nir_gs_count_vertices(const nir_shader
*shader
);
2348 bool nir_split_var_copies(nir_shader
*shader
);
2350 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2351 bool nir_lower_returns(nir_shader
*shader
);
2353 bool nir_inline_functions(nir_shader
*shader
);
2355 bool nir_propagate_invariant(nir_shader
*shader
);
2357 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2358 bool nir_lower_var_copies(nir_shader
*shader
);
2360 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2362 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2364 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2366 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2367 nir_function_impl
*entrypoint
,
2368 bool outputs
, bool inputs
);
2370 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2372 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2373 int (*type_size
)(const struct glsl_type
*));
2376 /* If set, this forces all non-flat fragment shader inputs to be
2377 * interpolated as if with the "sample" qualifier. This requires
2378 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2380 nir_lower_io_force_sample_interpolation
= (1 << 1),
2381 } nir_lower_io_options
;
2382 void nir_lower_io(nir_shader
*shader
,
2383 nir_variable_mode modes
,
2384 int (*type_size
)(const struct glsl_type
*),
2385 nir_lower_io_options
);
2386 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2387 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2389 bool nir_is_per_vertex_io(nir_variable
*var
, gl_shader_stage stage
);
2391 void nir_lower_io_types(nir_shader
*shader
);
2392 void nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2393 void nir_lower_regs_to_ssa(nir_shader
*shader
);
2394 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2396 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2397 bool nir_lower_constant_initializers(nir_shader
*shader
,
2398 nir_variable_mode modes
);
2400 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2401 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2402 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2403 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2405 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2406 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2408 bool nir_lower_samplers(nir_shader
*shader
,
2409 const struct gl_shader_program
*shader_program
);
2411 bool nir_lower_system_values(nir_shader
*shader
);
2413 typedef struct nir_lower_tex_options
{
2415 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2416 * sampler types a texture projector is lowered.
2421 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2423 bool lower_txf_offset
;
2426 * If true, lower away nir_tex_src_offset for all rect textures.
2428 bool lower_rect_offset
;
2431 * If true, lower rect textures to 2D, using txs to fetch the
2432 * texture dimensions and dividing the texture coords by the
2433 * texture dims to normalize.
2438 * If true, convert yuv to rgb.
2440 unsigned lower_y_uv_external
;
2441 unsigned lower_y_u_v_external
;
2442 unsigned lower_yx_xuxv_external
;
2445 * To emulate certain texture wrap modes, this can be used
2446 * to saturate the specified tex coord to [0.0, 1.0]. The
2447 * bits are according to sampler #, ie. if, for example:
2449 * (conf->saturate_s & (1 << n))
2451 * is true, then the s coord for sampler n is saturated.
2453 * Note that clamping must happen *after* projector lowering
2454 * so any projected texture sample instruction with a clamped
2455 * coordinate gets automatically lowered, regardless of the
2456 * 'lower_txp' setting.
2458 unsigned saturate_s
;
2459 unsigned saturate_t
;
2460 unsigned saturate_r
;
2462 /* Bitmask of textures that need swizzling.
2464 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2465 * swizzles[texture_index] is applied to the result of the texturing
2468 unsigned swizzle_result
;
2470 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2471 * while 4 and 5 represent 0 and 1 respectively.
2473 uint8_t swizzles
[32][4];
2476 * Bitmap of textures that need srgb to linear conversion. If
2477 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2478 * of the texture are lowered to linear.
2480 unsigned lower_srgb
;
2483 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2485 bool lower_txd_cube_map
;
2488 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
2489 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
2490 * with lower_txd_cube_map.
2492 bool lower_txd_shadow
;
2493 } nir_lower_tex_options
;
2495 bool nir_lower_tex(nir_shader
*shader
,
2496 const nir_lower_tex_options
*options
);
2498 bool nir_lower_idiv(nir_shader
*shader
);
2500 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2501 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2502 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2504 void nir_lower_two_sided_color(nir_shader
*shader
);
2506 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
2508 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2509 void nir_lower_tes_patch_vertices(nir_shader
*tes
, unsigned patch_vertices
);
2511 typedef struct nir_lower_wpos_ytransform_options
{
2512 int state_tokens
[5];
2513 bool fs_coord_origin_upper_left
:1;
2514 bool fs_coord_origin_lower_left
:1;
2515 bool fs_coord_pixel_center_integer
:1;
2516 bool fs_coord_pixel_center_half_integer
:1;
2517 } nir_lower_wpos_ytransform_options
;
2519 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2520 const nir_lower_wpos_ytransform_options
*options
);
2521 bool nir_lower_wpos_center(nir_shader
*shader
);
2523 typedef struct nir_lower_drawpixels_options
{
2524 int texcoord_state_tokens
[5];
2525 int scale_state_tokens
[5];
2526 int bias_state_tokens
[5];
2527 unsigned drawpix_sampler
;
2528 unsigned pixelmap_sampler
;
2530 bool scale_and_bias
:1;
2531 } nir_lower_drawpixels_options
;
2533 void nir_lower_drawpixels(nir_shader
*shader
,
2534 const nir_lower_drawpixels_options
*options
);
2536 typedef struct nir_lower_bitmap_options
{
2539 } nir_lower_bitmap_options
;
2541 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2543 bool nir_lower_atomics(nir_shader
*shader
,
2544 const struct gl_shader_program
*shader_program
);
2545 bool nir_lower_to_source_mods(nir_shader
*shader
);
2547 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2550 nir_lower_imul64
= (1 << 0),
2551 nir_lower_isign64
= (1 << 1),
2552 /** Lower all int64 modulus and division opcodes */
2553 nir_lower_divmod64
= (1 << 2),
2554 } nir_lower_int64_options
;
2556 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
2559 nir_lower_drcp
= (1 << 0),
2560 nir_lower_dsqrt
= (1 << 1),
2561 nir_lower_drsq
= (1 << 2),
2562 nir_lower_dtrunc
= (1 << 3),
2563 nir_lower_dfloor
= (1 << 4),
2564 nir_lower_dceil
= (1 << 5),
2565 nir_lower_dfract
= (1 << 6),
2566 nir_lower_dround_even
= (1 << 7),
2567 nir_lower_dmod
= (1 << 8)
2568 } nir_lower_doubles_options
;
2570 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2571 bool nir_lower_64bit_pack(nir_shader
*shader
);
2573 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2575 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2577 void nir_loop_analyze_impl(nir_function_impl
*impl
,
2578 nir_variable_mode indirect_mask
);
2580 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2582 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2583 bool nir_repair_ssa(nir_shader
*shader
);
2585 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
2587 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2588 * registers. If false, convert all values (even those not involved in a phi
2589 * node) to registers.
2591 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2593 bool nir_lower_phis_to_regs_block(nir_block
*block
);
2594 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
2596 bool nir_opt_algebraic(nir_shader
*shader
);
2597 bool nir_opt_algebraic_late(nir_shader
*shader
);
2598 bool nir_opt_constant_folding(nir_shader
*shader
);
2600 bool nir_opt_global_to_local(nir_shader
*shader
);
2602 bool nir_copy_prop(nir_shader
*shader
);
2604 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
2606 bool nir_opt_cse(nir_shader
*shader
);
2608 bool nir_opt_dce(nir_shader
*shader
);
2610 bool nir_opt_dead_cf(nir_shader
*shader
);
2612 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2614 bool nir_opt_if(nir_shader
*shader
);
2616 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
2618 bool nir_opt_move_comparisons(nir_shader
*shader
);
2620 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2622 bool nir_opt_remove_phis(nir_shader
*shader
);
2624 bool nir_opt_trivial_continues(nir_shader
*shader
);
2626 bool nir_opt_undef(nir_shader
*shader
);
2628 bool nir_opt_conditional_discard(nir_shader
*shader
);
2630 void nir_sweep(nir_shader
*shader
);
2632 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2633 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);