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
92 gl_state_index16 tokens
[STATE_LENGTH
];
97 nir_var_shader_in
= (1 << 0),
98 nir_var_shader_out
= (1 << 1),
99 nir_var_global
= (1 << 2),
100 nir_var_local
= (1 << 3),
101 nir_var_uniform
= (1 << 4),
102 nir_var_shader_storage
= (1 << 5),
103 nir_var_system_value
= (1 << 6),
104 nir_var_param
= (1 << 7),
105 nir_var_shared
= (1 << 8),
113 nir_rounding_mode_undef
= 0,
114 nir_rounding_mode_rtne
= 1, /* round to nearest even */
115 nir_rounding_mode_ru
= 2, /* round up */
116 nir_rounding_mode_rd
= 3, /* round down */
117 nir_rounding_mode_rtz
= 4, /* round towards zero */
133 typedef struct nir_constant
{
135 * Value of the constant.
137 * The field used to back the values supplied by the constant is determined
138 * by the type associated with the \c nir_variable. Constants may be
139 * scalars, vectors, or matrices.
141 nir_const_value values
[4];
143 /* we could get this from the var->type but makes clone *much* easier to
144 * not have to care about the type.
146 unsigned num_elements
;
148 /* Array elements / Structure Fields */
149 struct nir_constant
**elements
;
153 * \brief Layout qualifiers for gl_FragDepth.
155 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
156 * with a layout qualifier.
159 nir_depth_layout_none
, /**< No depth layout is specified. */
160 nir_depth_layout_any
,
161 nir_depth_layout_greater
,
162 nir_depth_layout_less
,
163 nir_depth_layout_unchanged
167 * Either a uniform, global variable, shader input, or shader output. Based on
168 * ir_variable - it should be easy to translate between the two.
171 typedef struct nir_variable
{
172 struct exec_node node
;
175 * Declared type of the variable
177 const struct glsl_type
*type
;
180 * Declared name of the variable
184 struct nir_variable_data
{
186 * Storage class of the variable.
188 * \sa nir_variable_mode
190 nir_variable_mode mode
;
193 * Is the variable read-only?
195 * This is set for variables declared as \c const, shader inputs,
198 unsigned read_only
:1;
202 unsigned invariant
:1;
205 * When separate shader programs are enabled, only input/outputs between
206 * the stages of a multi-stage separate program can be safely removed
207 * from the shader interface. Other input/outputs must remains active.
209 * This is also used to make sure xfb varyings that are unused by the
210 * fragment shader are not removed.
212 unsigned always_active_io
:1;
215 * Interpolation mode for shader inputs / outputs
217 * \sa glsl_interp_mode
219 unsigned interpolation
:2;
222 * \name ARB_fragment_coord_conventions
225 unsigned origin_upper_left
:1;
226 unsigned pixel_center_integer
:1;
230 * If non-zero, then this variable may be packed along with other variables
231 * into a single varying slot, so this offset should be applied when
232 * accessing components. For example, an offset of 1 means that the x
233 * component of this variable is actually stored in component y of the
234 * location specified by \c location.
236 unsigned location_frac
:2;
239 * If true, this variable represents an array of scalars that should
240 * be tightly packed. In other words, consecutive array elements
241 * should be stored one component apart, rather than one slot apart.
246 * Whether this is a fragment shader output implicitly initialized with
247 * the previous contents of the specified render target at the
248 * framebuffer location corresponding to this shader invocation.
250 unsigned fb_fetch_output
:1;
253 * \brief Layout qualifier for gl_FragDepth.
255 * This is not equal to \c ir_depth_layout_none if and only if this
256 * variable is \c gl_FragDepth and a layout qualifier is specified.
258 nir_depth_layout depth_layout
;
261 * Storage location of the base of this variable
263 * The precise meaning of this field depends on the nature of the variable.
265 * - Vertex shader input: one of the values from \c gl_vert_attrib.
266 * - Vertex shader output: one of the values from \c gl_varying_slot.
267 * - Geometry shader input: one of the values from \c gl_varying_slot.
268 * - Geometry shader output: one of the values from \c gl_varying_slot.
269 * - Fragment shader input: one of the values from \c gl_varying_slot.
270 * - Fragment shader output: one of the values from \c gl_frag_result.
271 * - Uniforms: Per-stage uniform slot number for default uniform block.
272 * - Uniforms: Index within the uniform block definition for UBO members.
273 * - Non-UBO Uniforms: uniform slot number.
274 * - Other: This field is not currently used.
276 * If the variable is a uniform, shader input, or shader output, and the
277 * slot has not been assigned, the value will be -1.
282 * The actual location of the variable in the IR. Only valid for inputs
285 unsigned int driver_location
;
288 * Vertex stream output identifier.
290 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
291 * stream of the i-th component.
296 * output index for dual source blending.
301 * Descriptor set binding for sampler or UBO.
306 * Initial binding point for a sampler or UBO.
308 * For array types, this represents the binding point for the first element.
313 * Location an atomic counter is stored at.
318 * ARB_shader_image_load_store qualifiers.
321 bool read_only
; /**< "readonly" qualifier. */
322 bool write_only
; /**< "writeonly" qualifier. */
327 /** Image internal format if specified explicitly, otherwise GL_NONE. */
333 * Built-in state that backs this uniform
335 * Once set at variable creation, \c state_slots must remain invariant.
336 * This is because, ideally, this array would be shared by all clones of
337 * this variable in the IR tree. In other words, we'd really like for it
338 * to be a fly-weight.
340 * If the variable is not a uniform, \c num_state_slots will be zero and
341 * \c state_slots will be \c NULL.
344 unsigned num_state_slots
; /**< Number of state slots used */
345 nir_state_slot
*state_slots
; /**< State descriptors. */
349 * Constant expression assigned in the initializer of the variable
351 * This field should only be used temporarily by creators of NIR shaders
352 * and then lower_constant_initializers can be used to get rid of them.
353 * Most of the rest of NIR ignores this field or asserts that it's NULL.
355 nir_constant
*constant_initializer
;
358 * For variables that are in an interface block or are an instance of an
359 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
361 * \sa ir_variable::location
363 const struct glsl_type
*interface_type
;
366 #define nir_foreach_variable(var, var_list) \
367 foreach_list_typed(nir_variable, var, node, var_list)
369 #define nir_foreach_variable_safe(var, var_list) \
370 foreach_list_typed_safe(nir_variable, var, node, var_list)
373 nir_variable_is_global(const nir_variable
*var
)
375 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
378 typedef struct nir_register
{
379 struct exec_node node
;
381 unsigned num_components
; /** < number of vector components */
382 unsigned num_array_elems
; /** < size of array (0 for no array) */
384 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
387 /** generic register index. */
390 /** only for debug purposes, can be NULL */
393 /** whether this register is local (per-function) or global (per-shader) */
397 * If this flag is set to true, then accessing channels >= num_components
398 * is well-defined, and simply spills over to the next array element. This
399 * is useful for backends that can do per-component accessing, in
400 * particular scalar backends. By setting this flag and making
401 * num_components equal to 1, structures can be packed tightly into
402 * registers and then registers can be accessed per-component to get to
403 * each structure member, even if it crosses vec4 boundaries.
407 /** set of nir_srcs where this register is used (read from) */
408 struct list_head uses
;
410 /** set of nir_dests where this register is defined (written to) */
411 struct list_head defs
;
413 /** set of nir_ifs where this register is used as a condition */
414 struct list_head if_uses
;
417 #define nir_foreach_register(reg, reg_list) \
418 foreach_list_typed(nir_register, reg, node, reg_list)
419 #define nir_foreach_register_safe(reg, reg_list) \
420 foreach_list_typed_safe(nir_register, reg, node, reg_list)
426 nir_instr_type_intrinsic
,
427 nir_instr_type_load_const
,
429 nir_instr_type_ssa_undef
,
431 nir_instr_type_parallel_copy
,
434 typedef struct nir_instr
{
435 struct exec_node node
;
437 struct nir_block
*block
;
439 /** generic instruction index. */
442 /* A temporary for optimization and analysis passes to use for storing
443 * flags. For instance, DCE uses this to store the "dead/live" info.
448 static inline nir_instr
*
449 nir_instr_next(nir_instr
*instr
)
451 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
452 if (exec_node_is_tail_sentinel(next
))
455 return exec_node_data(nir_instr
, next
, node
);
458 static inline nir_instr
*
459 nir_instr_prev(nir_instr
*instr
)
461 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
462 if (exec_node_is_head_sentinel(prev
))
465 return exec_node_data(nir_instr
, prev
, node
);
469 nir_instr_is_first(const nir_instr
*instr
)
471 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
475 nir_instr_is_last(const nir_instr
*instr
)
477 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
480 typedef struct nir_ssa_def
{
481 /** for debugging only, can be NULL */
484 /** generic SSA definition index. */
487 /** Index into the live_in and live_out bitfields */
490 nir_instr
*parent_instr
;
492 /** set of nir_instrs where this register is used (read from) */
493 struct list_head uses
;
495 /** set of nir_ifs where this register is used as a condition */
496 struct list_head if_uses
;
498 uint8_t num_components
;
500 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
508 struct nir_src
*indirect
; /** < NULL for no indirect offset */
509 unsigned base_offset
;
511 /* TODO use-def chain goes here */
515 nir_instr
*parent_instr
;
516 struct list_head def_link
;
519 struct nir_src
*indirect
; /** < NULL for no indirect offset */
520 unsigned base_offset
;
522 /* TODO def-use chain goes here */
527 typedef struct nir_src
{
529 nir_instr
*parent_instr
;
530 struct nir_if
*parent_if
;
533 struct list_head use_link
;
543 static inline nir_src
546 nir_src src
= { { NULL
} };
550 #define NIR_SRC_INIT nir_src_init()
552 #define nir_foreach_use(src, reg_or_ssa_def) \
553 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
555 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
556 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
558 #define nir_foreach_if_use(src, reg_or_ssa_def) \
559 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
561 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
562 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
573 static inline nir_dest
576 nir_dest dest
= { { { NULL
} } };
580 #define NIR_DEST_INIT nir_dest_init()
582 #define nir_foreach_def(dest, reg) \
583 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
585 #define nir_foreach_def_safe(dest, reg) \
586 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
588 static inline nir_src
589 nir_src_for_ssa(nir_ssa_def
*def
)
591 nir_src src
= NIR_SRC_INIT
;
599 static inline nir_src
600 nir_src_for_reg(nir_register
*reg
)
602 nir_src src
= NIR_SRC_INIT
;
606 src
.reg
.indirect
= NULL
;
607 src
.reg
.base_offset
= 0;
612 static inline nir_dest
613 nir_dest_for_reg(nir_register
*reg
)
615 nir_dest dest
= NIR_DEST_INIT
;
622 static inline unsigned
623 nir_src_bit_size(nir_src src
)
625 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
628 static inline unsigned
629 nir_dest_bit_size(nir_dest dest
)
631 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
634 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
635 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
641 * \name input modifiers
645 * For inputs interpreted as floating point, flips the sign bit. For
646 * inputs interpreted as integers, performs the two's complement negation.
651 * Clears the sign bit for floating point values, and computes the integer
652 * absolute value for integers. Note that the negate modifier acts after
653 * the absolute value modifier, therefore if both are set then all inputs
654 * will become negative.
660 * For each input component, says which component of the register it is
661 * chosen from. Note that which elements of the swizzle are used and which
662 * are ignored are based on the write mask for most opcodes - for example,
663 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
664 * a swizzle of {2, x, 1, 0} where x means "don't care."
673 * \name saturate output modifier
675 * Only valid for opcodes that output floating-point numbers. Clamps the
676 * output to between 0.0 and 1.0 inclusive.
681 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
685 nir_type_invalid
= 0, /* Not a valid type */
690 nir_type_bool32
= 32 | nir_type_bool
,
691 nir_type_int8
= 8 | nir_type_int
,
692 nir_type_int16
= 16 | nir_type_int
,
693 nir_type_int32
= 32 | nir_type_int
,
694 nir_type_int64
= 64 | nir_type_int
,
695 nir_type_uint8
= 8 | nir_type_uint
,
696 nir_type_uint16
= 16 | nir_type_uint
,
697 nir_type_uint32
= 32 | nir_type_uint
,
698 nir_type_uint64
= 64 | nir_type_uint
,
699 nir_type_float16
= 16 | nir_type_float
,
700 nir_type_float32
= 32 | nir_type_float
,
701 nir_type_float64
= 64 | nir_type_float
,
704 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
705 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
707 static inline unsigned
708 nir_alu_type_get_type_size(nir_alu_type type
)
710 return type
& NIR_ALU_TYPE_SIZE_MASK
;
713 static inline unsigned
714 nir_alu_type_get_base_type(nir_alu_type type
)
716 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
719 static inline nir_alu_type
720 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
724 return nir_type_bool32
;
727 return nir_type_uint32
;
730 return nir_type_int32
;
732 case GLSL_TYPE_UINT16
:
733 return nir_type_uint16
;
735 case GLSL_TYPE_INT16
:
736 return nir_type_int16
;
738 case GLSL_TYPE_UINT64
:
739 return nir_type_uint64
;
741 case GLSL_TYPE_INT64
:
742 return nir_type_int64
;
744 case GLSL_TYPE_FLOAT
:
745 return nir_type_float32
;
747 case GLSL_TYPE_FLOAT16
:
748 return nir_type_float16
;
750 case GLSL_TYPE_DOUBLE
:
751 return nir_type_float64
;
754 unreachable("unknown type");
758 static inline nir_alu_type
759 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
761 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
764 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
,
765 nir_rounding_mode rnd
);
768 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
769 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
770 } nir_op_algebraic_property
;
778 * The number of components in the output
780 * If non-zero, this is the size of the output and input sizes are
781 * explicitly given; swizzle and writemask are still in effect, but if
782 * the output component is masked out, then the input component may
785 * If zero, the opcode acts in the standard, per-component manner; the
786 * operation is performed on each component (except the ones that are
787 * masked out) with the input being taken from the input swizzle for
790 * The size of some of the inputs may be given (i.e. non-zero) even
791 * though output_size is zero; in that case, the inputs with a zero
792 * size act per-component, while the inputs with non-zero size don't.
794 unsigned output_size
;
797 * The type of vector that the instruction outputs. Note that the
798 * staurate modifier is only allowed on outputs with the float type.
801 nir_alu_type output_type
;
804 * The number of components in each input
806 unsigned input_sizes
[4];
809 * The type of vector that each input takes. Note that negate and
810 * absolute value are only allowed on inputs with int or float type and
811 * behave differently on the two.
813 nir_alu_type input_types
[4];
815 nir_op_algebraic_property algebraic_properties
;
818 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
820 typedef struct nir_alu_instr
{
824 /** Indicates that this ALU instruction generates an exact value
826 * This is kind of a mixture of GLSL "precise" and "invariant" and not
827 * really equivalent to either. This indicates that the value generated by
828 * this operation is high-precision and any code transformations that touch
829 * it must ensure that the resulting value is bit-for-bit identical to the
838 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
839 nir_alu_instr
*instr
);
840 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
841 nir_alu_instr
*instr
);
843 /* is this source channel used? */
845 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
848 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
849 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
851 return (instr
->dest
.write_mask
>> channel
) & 1;
855 * For instructions whose destinations are SSA, get the number of channels
858 static inline unsigned
859 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
861 assert(instr
->dest
.dest
.is_ssa
);
863 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
864 return nir_op_infos
[instr
->op
].input_sizes
[src
];
866 return instr
->dest
.dest
.ssa
.num_components
;
869 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
870 unsigned src1
, unsigned src2
);
874 nir_deref_type_array
,
875 nir_deref_type_struct
878 typedef struct nir_deref
{
879 nir_deref_type deref_type
;
880 struct nir_deref
*child
;
881 const struct glsl_type
*type
;
890 /* This enum describes how the array is referenced. If the deref is
891 * direct then the base_offset is used. If the deref is indirect then
892 * offset is given by base_offset + indirect. If the deref is a wildcard
893 * then the deref refers to all of the elements of the array at the same
894 * time. Wildcard dereferences are only ever allowed in copy_var
895 * intrinsics and the source and destination derefs must have matching
899 nir_deref_array_type_direct
,
900 nir_deref_array_type_indirect
,
901 nir_deref_array_type_wildcard
,
902 } nir_deref_array_type
;
907 nir_deref_array_type deref_array_type
;
908 unsigned base_offset
;
918 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
,
919 deref_type
, nir_deref_type_var
)
920 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
,
921 deref_type
, nir_deref_type_array
)
922 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
,
923 deref_type
, nir_deref_type_struct
)
925 /* Returns the last deref in the chain. */
926 static inline nir_deref
*
927 nir_deref_tail(nir_deref
*deref
)
930 deref
= deref
->child
;
938 nir_deref_var
**params
;
939 nir_deref_var
*return_deref
;
941 struct nir_function
*callee
;
944 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
945 num_variables, num_indices, idx0, idx1, idx2, flags) \
946 nir_intrinsic_##name,
948 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
951 #include "nir_intrinsics.h"
952 nir_num_intrinsics
= nir_last_intrinsic
+ 1
955 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
957 /** Represents an intrinsic
959 * An intrinsic is an instruction type for handling things that are
960 * more-or-less regular operations but don't just consume and produce SSA
961 * values like ALU operations do. Intrinsics are not for things that have
962 * special semantic meaning such as phi nodes and parallel copies.
963 * Examples of intrinsics include variable load/store operations, system
964 * value loads, and the like. Even though texturing more-or-less falls
965 * under this category, texturing is its own instruction type because
966 * trying to represent texturing with intrinsics would lead to a
967 * combinatorial explosion of intrinsic opcodes.
969 * By having a single instruction type for handling a lot of different
970 * cases, optimization passes can look for intrinsics and, for the most
971 * part, completely ignore them. Each intrinsic type also has a few
972 * possible flags that govern whether or not they can be reordered or
973 * eliminated. That way passes like dead code elimination can still work
974 * on intrisics without understanding the meaning of each.
976 * Each intrinsic has some number of constant indices, some number of
977 * variables, and some number of sources. What these sources, variables,
978 * and indices mean depends on the intrinsic and is documented with the
979 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
980 * instructions are the only types of instruction that can operate on
986 nir_intrinsic_op intrinsic
;
990 /** number of components if this is a vectorized intrinsic
992 * Similarly to ALU operations, some intrinsics are vectorized.
993 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
994 * For vectorized intrinsics, the num_components field specifies the
995 * number of destination components and the number of source components
996 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
998 uint8_t num_components
;
1000 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
1002 nir_deref_var
*variables
[2];
1005 } nir_intrinsic_instr
;
1008 * \name NIR intrinsics semantic flags
1010 * information about what the compiler can do with the intrinsics.
1012 * \sa nir_intrinsic_info::flags
1016 * whether the intrinsic can be safely eliminated if none of its output
1017 * value is not being used.
1019 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
1022 * Whether the intrinsic can be reordered with respect to any other
1023 * intrinsic, i.e. whether the only reordering dependencies of the
1024 * intrinsic are due to the register reads/writes.
1026 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
1027 } nir_intrinsic_semantic_flag
;
1030 * \name NIR intrinsics const-index flag
1032 * Indicates the usage of a const_index slot.
1034 * \sa nir_intrinsic_info::index_map
1038 * Generally instructions that take a offset src argument, can encode
1039 * a constant 'base' value which is added to the offset.
1041 NIR_INTRINSIC_BASE
= 1,
1044 * For store instructions, a writemask for the store.
1046 NIR_INTRINSIC_WRMASK
= 2,
1049 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1051 NIR_INTRINSIC_STREAM_ID
= 3,
1054 * The clip-plane id for load_user_clip_plane intrinsic.
1056 NIR_INTRINSIC_UCP_ID
= 4,
1059 * The amount of data, starting from BASE, that this instruction may
1060 * access. This is used to provide bounds if the offset is not constant.
1062 NIR_INTRINSIC_RANGE
= 5,
1065 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1067 NIR_INTRINSIC_DESC_SET
= 6,
1070 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1072 NIR_INTRINSIC_BINDING
= 7,
1077 NIR_INTRINSIC_COMPONENT
= 8,
1080 * Interpolation mode (only meaningful for FS inputs).
1082 NIR_INTRINSIC_INTERP_MODE
= 9,
1085 * A binary nir_op to use when performing a reduction or scan operation
1087 NIR_INTRINSIC_REDUCTION_OP
= 10,
1090 * Cluster size for reduction operations
1092 NIR_INTRINSIC_CLUSTER_SIZE
= 11,
1094 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1096 } nir_intrinsic_index_flag
;
1098 #define NIR_INTRINSIC_MAX_INPUTS 4
1103 unsigned num_srcs
; /** < number of register/SSA inputs */
1105 /** number of components of each input register
1107 * If this value is 0, the number of components is given by the
1108 * num_components field of nir_intrinsic_instr.
1110 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1114 /** number of components of the output register
1116 * If this value is 0, the number of components is given by the
1117 * num_components field of nir_intrinsic_instr.
1119 unsigned dest_components
;
1121 /** the number of inputs/outputs that are variables */
1122 unsigned num_variables
;
1124 /** the number of constant indices used by the intrinsic */
1125 unsigned num_indices
;
1127 /** indicates the usage of intr->const_index[n] */
1128 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1130 /** semantic flags for calls to this intrinsic */
1131 nir_intrinsic_semantic_flag flags
;
1132 } nir_intrinsic_info
;
1134 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1137 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1138 static inline type \
1139 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1141 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1142 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1143 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1145 static inline void \
1146 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1148 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1149 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1150 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1153 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1154 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1155 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1156 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1157 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1158 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1159 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1160 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1161 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1162 INTRINSIC_IDX_ACCESSORS(reduction_op
, REDUCTION_OP
, unsigned)
1163 INTRINSIC_IDX_ACCESSORS(cluster_size
, CLUSTER_SIZE
, unsigned)
1166 * \group texture information
1168 * This gives semantic information about textures which is useful to the
1169 * frontend, the backend, and lowering passes, but not the optimizer.
1174 nir_tex_src_projector
,
1175 nir_tex_src_comparator
, /* shadow comparator */
1179 nir_tex_src_ms_index
, /* MSAA sample index */
1180 nir_tex_src_ms_mcs
, /* MSAA compression value */
1183 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1184 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1185 nir_tex_src_plane
, /* < selects plane for planar textures */
1186 nir_num_tex_src_types
1191 nir_tex_src_type src_type
;
1195 nir_texop_tex
, /**< Regular texture look-up */
1196 nir_texop_txb
, /**< Texture look-up with LOD bias */
1197 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1198 nir_texop_txd
, /**< Texture look-up with partial derivatives */
1199 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1200 nir_texop_txf_ms
, /**< Multisample texture fetch */
1201 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1202 nir_texop_txs
, /**< Texture size */
1203 nir_texop_lod
, /**< Texture lod query */
1204 nir_texop_tg4
, /**< Texture gather */
1205 nir_texop_query_levels
, /**< Texture levels query */
1206 nir_texop_texture_samples
, /**< Texture samples query */
1207 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1215 enum glsl_sampler_dim sampler_dim
;
1216 nir_alu_type dest_type
;
1221 unsigned num_srcs
, coord_components
;
1222 bool is_array
, is_shadow
;
1225 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1226 * components or the new-style shadow that outputs 1 component.
1228 bool is_new_style_shadow
;
1230 /* gather component selector */
1231 unsigned component
: 2;
1233 /** The texture index
1235 * If this texture instruction has a nir_tex_src_texture_offset source,
1236 * then the texture index is given by texture_index + texture_offset.
1238 unsigned texture_index
;
1240 /** The size of the texture array or 0 if it's not an array */
1241 unsigned texture_array_size
;
1243 /** The texture deref
1245 * If this is null, use texture_index instead.
1247 nir_deref_var
*texture
;
1249 /** The sampler index
1251 * The following operations do not require a sampler and, as such, this
1252 * field should be ignored:
1254 * - nir_texop_txf_ms
1257 * - nir_texop_query_levels
1258 * - nir_texop_texture_samples
1259 * - nir_texop_samples_identical
1261 * If this texture instruction has a nir_tex_src_sampler_offset source,
1262 * then the sampler index is given by sampler_index + sampler_offset.
1264 unsigned sampler_index
;
1266 /** The sampler deref
1268 * If this is null, use sampler_index instead.
1270 nir_deref_var
*sampler
;
1273 static inline unsigned
1274 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1276 switch (instr
->op
) {
1277 case nir_texop_txs
: {
1279 switch (instr
->sampler_dim
) {
1280 case GLSL_SAMPLER_DIM_1D
:
1281 case GLSL_SAMPLER_DIM_BUF
:
1284 case GLSL_SAMPLER_DIM_2D
:
1285 case GLSL_SAMPLER_DIM_CUBE
:
1286 case GLSL_SAMPLER_DIM_MS
:
1287 case GLSL_SAMPLER_DIM_RECT
:
1288 case GLSL_SAMPLER_DIM_EXTERNAL
:
1289 case GLSL_SAMPLER_DIM_SUBPASS
:
1292 case GLSL_SAMPLER_DIM_3D
:
1296 unreachable("not reached");
1298 if (instr
->is_array
)
1306 case nir_texop_texture_samples
:
1307 case nir_texop_query_levels
:
1308 case nir_texop_samples_identical
:
1312 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1319 /* Returns true if this texture operation queries something about the texture
1320 * rather than actually sampling it.
1323 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1325 switch (instr
->op
) {
1328 case nir_texop_texture_samples
:
1329 case nir_texop_query_levels
:
1330 case nir_texop_txf_ms_mcs
:
1337 case nir_texop_txf_ms
:
1341 unreachable("Invalid texture opcode");
1345 static inline nir_alu_type
1346 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1348 switch (instr
->src
[src
].src_type
) {
1349 case nir_tex_src_coord
:
1350 switch (instr
->op
) {
1352 case nir_texop_txf_ms
:
1353 case nir_texop_txf_ms_mcs
:
1354 case nir_texop_samples_identical
:
1355 return nir_type_int
;
1358 return nir_type_float
;
1361 case nir_tex_src_lod
:
1362 switch (instr
->op
) {
1365 return nir_type_int
;
1368 return nir_type_float
;
1371 case nir_tex_src_projector
:
1372 case nir_tex_src_comparator
:
1373 case nir_tex_src_bias
:
1374 case nir_tex_src_ddx
:
1375 case nir_tex_src_ddy
:
1376 return nir_type_float
;
1378 case nir_tex_src_offset
:
1379 case nir_tex_src_ms_index
:
1380 case nir_tex_src_texture_offset
:
1381 case nir_tex_src_sampler_offset
:
1382 return nir_type_int
;
1385 unreachable("Invalid texture source type");
1389 static inline unsigned
1390 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1392 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1393 return instr
->coord_components
;
1395 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1396 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1399 if (instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1400 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1401 if (instr
->is_array
)
1402 return instr
->coord_components
- 1;
1404 return instr
->coord_components
;
1407 /* Usual APIs don't allow cube + offset, but we allow it, with 2 coords for
1408 * the offset, since a cube maps to a single face.
1410 if (instr
->src
[src
].src_type
== nir_tex_src_offset
) {
1411 if (instr
->sampler_dim
== GLSL_SAMPLER_DIM_CUBE
)
1413 else if (instr
->is_array
)
1414 return instr
->coord_components
- 1;
1416 return instr
->coord_components
;
1423 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1425 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1426 if (instr
->src
[i
].src_type
== type
)
1432 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1433 nir_tex_src_type src_type
,
1436 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1441 nir_const_value value
;
1444 } nir_load_const_instr
;
1457 /* creates a new SSA variable in an undefined state */
1462 } nir_ssa_undef_instr
;
1465 struct exec_node node
;
1467 /* The predecessor block corresponding to this source */
1468 struct nir_block
*pred
;
1473 #define nir_foreach_phi_src(phi_src, phi) \
1474 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1475 #define nir_foreach_phi_src_safe(phi_src, phi) \
1476 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1481 struct exec_list srcs
; /** < list of nir_phi_src */
1487 struct exec_node node
;
1490 } nir_parallel_copy_entry
;
1492 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1493 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1498 /* A list of nir_parallel_copy_entrys. The sources of all of the
1499 * entries are copied to the corresponding destinations "in parallel".
1500 * In other words, if we have two entries: a -> b and b -> a, the values
1503 struct exec_list entries
;
1504 } nir_parallel_copy_instr
;
1506 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1507 type
, nir_instr_type_alu
)
1508 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1509 type
, nir_instr_type_call
)
1510 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1511 type
, nir_instr_type_jump
)
1512 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1513 type
, nir_instr_type_tex
)
1514 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1515 type
, nir_instr_type_intrinsic
)
1516 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1517 type
, nir_instr_type_load_const
)
1518 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1519 type
, nir_instr_type_ssa_undef
)
1520 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1521 type
, nir_instr_type_phi
)
1522 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1523 nir_parallel_copy_instr
, instr
,
1524 type
, nir_instr_type_parallel_copy
)
1529 * Control flow consists of a tree of control flow nodes, which include
1530 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1531 * instructions that always run start-to-finish. Each basic block also keeps
1532 * track of its successors (blocks which may run immediately after the current
1533 * block) and predecessors (blocks which could have run immediately before the
1534 * current block). Each function also has a start block and an end block which
1535 * all return statements point to (which is always empty). Together, all the
1536 * blocks with their predecessors and successors make up the control flow
1537 * graph (CFG) of the function. There are helpers that modify the tree of
1538 * control flow nodes while modifying the CFG appropriately; these should be
1539 * used instead of modifying the tree directly.
1546 nir_cf_node_function
1549 typedef struct nir_cf_node
{
1550 struct exec_node node
;
1551 nir_cf_node_type type
;
1552 struct nir_cf_node
*parent
;
1555 typedef struct nir_block
{
1556 nir_cf_node cf_node
;
1558 struct exec_list instr_list
; /** < list of nir_instr */
1560 /** generic block index; generated by nir_index_blocks */
1564 * Each block can only have up to 2 successors, so we put them in a simple
1565 * array - no need for anything more complicated.
1567 struct nir_block
*successors
[2];
1569 /* Set of nir_block predecessors in the CFG */
1570 struct set
*predecessors
;
1573 * this node's immediate dominator in the dominance tree - set to NULL for
1576 struct nir_block
*imm_dom
;
1578 /* This node's children in the dominance tree */
1579 unsigned num_dom_children
;
1580 struct nir_block
**dom_children
;
1582 /* Set of nir_blocks on the dominance frontier of this block */
1583 struct set
*dom_frontier
;
1586 * These two indices have the property that dom_{pre,post}_index for each
1587 * child of this block in the dominance tree will always be between
1588 * dom_pre_index and dom_post_index for this block, which makes testing if
1589 * a given block is dominated by another block an O(1) operation.
1591 unsigned dom_pre_index
, dom_post_index
;
1593 /* live in and out for this block; used for liveness analysis */
1594 BITSET_WORD
*live_in
;
1595 BITSET_WORD
*live_out
;
1598 static inline nir_instr
*
1599 nir_block_first_instr(nir_block
*block
)
1601 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1602 return exec_node_data(nir_instr
, head
, node
);
1605 static inline nir_instr
*
1606 nir_block_last_instr(nir_block
*block
)
1608 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1609 return exec_node_data(nir_instr
, tail
, node
);
1612 #define nir_foreach_instr(instr, block) \
1613 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1614 #define nir_foreach_instr_reverse(instr, block) \
1615 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1616 #define nir_foreach_instr_safe(instr, block) \
1617 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1618 #define nir_foreach_instr_reverse_safe(instr, block) \
1619 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1621 typedef struct nir_if
{
1622 nir_cf_node cf_node
;
1625 struct exec_list then_list
; /** < list of nir_cf_node */
1626 struct exec_list else_list
; /** < list of nir_cf_node */
1632 nir_instr
*conditional_instr
;
1634 nir_block
*break_block
;
1635 nir_block
*continue_from_block
;
1637 bool continue_from_then
;
1639 struct list_head loop_terminator_link
;
1640 } nir_loop_terminator
;
1643 /* Number of instructions in the loop */
1644 unsigned num_instructions
;
1646 /* How many times the loop is run (if known) */
1647 unsigned trip_count
;
1648 bool is_trip_count_known
;
1650 /* Unroll the loop regardless of its size */
1653 nir_loop_terminator
*limiting_terminator
;
1655 /* A list of loop_terminators terminating this loop. */
1656 struct list_head loop_terminator_list
;
1660 nir_cf_node cf_node
;
1662 struct exec_list body
; /** < list of nir_cf_node */
1664 nir_loop_info
*info
;
1668 * Various bits of metadata that can may be created or required by
1669 * optimization and analysis passes
1672 nir_metadata_none
= 0x0,
1673 nir_metadata_block_index
= 0x1,
1674 nir_metadata_dominance
= 0x2,
1675 nir_metadata_live_ssa_defs
= 0x4,
1676 nir_metadata_not_properly_reset
= 0x8,
1677 nir_metadata_loop_analysis
= 0x10,
1681 nir_cf_node cf_node
;
1683 /** pointer to the function of which this is an implementation */
1684 struct nir_function
*function
;
1686 struct exec_list body
; /** < list of nir_cf_node */
1688 nir_block
*end_block
;
1690 /** list for all local variables in the function */
1691 struct exec_list locals
;
1693 /** array of variables used as parameters */
1694 unsigned num_params
;
1695 nir_variable
**params
;
1697 /** variable used to hold the result of the function */
1698 nir_variable
*return_var
;
1700 /** list of local registers in the function */
1701 struct exec_list registers
;
1703 /** next available local register index */
1706 /** next available SSA value index */
1709 /* total number of basic blocks, only valid when block_index_dirty = false */
1710 unsigned num_blocks
;
1712 nir_metadata valid_metadata
;
1713 } nir_function_impl
;
1715 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1716 nir_start_block(nir_function_impl
*impl
)
1718 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1721 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1722 nir_impl_last_block(nir_function_impl
*impl
)
1724 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1727 static inline nir_cf_node
*
1728 nir_cf_node_next(nir_cf_node
*node
)
1730 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1731 if (exec_node_is_tail_sentinel(next
))
1734 return exec_node_data(nir_cf_node
, next
, node
);
1737 static inline nir_cf_node
*
1738 nir_cf_node_prev(nir_cf_node
*node
)
1740 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1741 if (exec_node_is_head_sentinel(prev
))
1744 return exec_node_data(nir_cf_node
, prev
, node
);
1748 nir_cf_node_is_first(const nir_cf_node
*node
)
1750 return exec_node_is_head_sentinel(node
->node
.prev
);
1754 nir_cf_node_is_last(const nir_cf_node
*node
)
1756 return exec_node_is_tail_sentinel(node
->node
.next
);
1759 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1760 type
, nir_cf_node_block
)
1761 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1762 type
, nir_cf_node_if
)
1763 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1764 type
, nir_cf_node_loop
)
1765 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1766 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1768 static inline nir_block
*
1769 nir_if_first_then_block(nir_if
*if_stmt
)
1771 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1772 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1775 static inline nir_block
*
1776 nir_if_last_then_block(nir_if
*if_stmt
)
1778 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1779 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1782 static inline nir_block
*
1783 nir_if_first_else_block(nir_if
*if_stmt
)
1785 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1786 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1789 static inline nir_block
*
1790 nir_if_last_else_block(nir_if
*if_stmt
)
1792 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1793 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1796 static inline nir_block
*
1797 nir_loop_first_block(nir_loop
*loop
)
1799 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1800 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1803 static inline nir_block
*
1804 nir_loop_last_block(nir_loop
*loop
)
1806 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1807 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1813 nir_parameter_inout
,
1814 } nir_parameter_type
;
1817 nir_parameter_type param_type
;
1818 const struct glsl_type
*type
;
1821 typedef struct nir_function
{
1822 struct exec_node node
;
1825 struct nir_shader
*shader
;
1827 unsigned num_params
;
1828 nir_parameter
*params
;
1829 const struct glsl_type
*return_type
;
1831 /** The implementation of this function.
1833 * If the function is only declared and not implemented, this is NULL.
1835 nir_function_impl
*impl
;
1838 typedef struct nir_shader_compiler_options
{
1843 /** Lowers flrp when it does not support doubles */
1850 bool lower_bitfield_extract
;
1851 bool lower_bitfield_insert
;
1852 bool lower_uadd_carry
;
1853 bool lower_usub_borrow
;
1854 /** lowers fneg and ineg to fsub and isub. */
1856 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1859 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1862 /** enables rules to lower idiv by power-of-two: */
1865 /* Does the native fdot instruction replicate its result for four
1866 * components? If so, then opt_algebraic_late will turn all fdotN
1867 * instructions into fdot_replicatedN instructions.
1869 bool fdot_replicates
;
1871 /** lowers ffract to fsub+ffloor: */
1876 bool lower_pack_half_2x16
;
1877 bool lower_pack_unorm_2x16
;
1878 bool lower_pack_snorm_2x16
;
1879 bool lower_pack_unorm_4x8
;
1880 bool lower_pack_snorm_4x8
;
1881 bool lower_unpack_half_2x16
;
1882 bool lower_unpack_unorm_2x16
;
1883 bool lower_unpack_snorm_2x16
;
1884 bool lower_unpack_unorm_4x8
;
1885 bool lower_unpack_snorm_4x8
;
1887 bool lower_extract_byte
;
1888 bool lower_extract_word
;
1890 bool lower_all_io_to_temps
;
1893 * Does the driver support real 32-bit integers? (Otherwise, integers
1894 * are simulated by floats.)
1896 bool native_integers
;
1898 /* Indicates that the driver only has zero-based vertex id */
1899 bool vertex_id_zero_based
;
1901 bool lower_cs_local_index_from_id
;
1903 bool lower_device_index_to_zero
;
1906 * Should nir_lower_io() create load_interpolated_input intrinsics?
1908 * If not, it generates regular load_input intrinsics and interpolation
1909 * information must be inferred from the list of input nir_variables.
1911 bool use_interpolated_input_intrinsics
;
1914 * Do vertex shader double inputs use two locations? The Vulkan spec
1915 * requires two locations to be used, OpenGL allows a single location.
1917 bool vs_inputs_dual_locations
;
1919 unsigned max_unroll_iterations
;
1920 } nir_shader_compiler_options
;
1922 typedef struct nir_shader
{
1923 /** list of uniforms (nir_variable) */
1924 struct exec_list uniforms
;
1926 /** list of inputs (nir_variable) */
1927 struct exec_list inputs
;
1929 /** list of outputs (nir_variable) */
1930 struct exec_list outputs
;
1932 /** list of shared compute variables (nir_variable) */
1933 struct exec_list shared
;
1935 /** Set of driver-specific options for the shader.
1937 * The memory for the options is expected to be kept in a single static
1938 * copy by the driver.
1940 const struct nir_shader_compiler_options
*options
;
1942 /** Various bits of compile-time information about a given shader */
1943 struct shader_info info
;
1945 /** list of global variables in the shader (nir_variable) */
1946 struct exec_list globals
;
1948 /** list of system value variables in the shader (nir_variable) */
1949 struct exec_list system_values
;
1951 struct exec_list functions
; /** < list of nir_function */
1953 /** list of global register in the shader */
1954 struct exec_list registers
;
1956 /** next available global register index */
1960 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1963 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1966 static inline nir_function_impl
*
1967 nir_shader_get_entrypoint(nir_shader
*shader
)
1969 assert(exec_list_length(&shader
->functions
) == 1);
1970 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1971 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1972 assert(func
->return_type
== glsl_void_type());
1973 assert(func
->num_params
== 0);
1978 #define nir_foreach_function(func, shader) \
1979 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1981 nir_shader
*nir_shader_create(void *mem_ctx
,
1982 gl_shader_stage stage
,
1983 const nir_shader_compiler_options
*options
,
1986 /** creates a register, including assigning it an index and adding it to the list */
1987 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1989 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1991 void nir_reg_remove(nir_register
*reg
);
1993 /** Adds a variable to the appropriate list in nir_shader */
1994 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1997 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1999 assert(var
->data
.mode
== nir_var_local
);
2000 exec_list_push_tail(&impl
->locals
, &var
->node
);
2003 /** creates a variable, sets a few defaults, and adds it to the list */
2004 nir_variable
*nir_variable_create(nir_shader
*shader
,
2005 nir_variable_mode mode
,
2006 const struct glsl_type
*type
,
2008 /** creates a local variable and adds it to the list */
2009 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
2010 const struct glsl_type
*type
,
2013 /** creates a function and adds it to the shader's list of functions */
2014 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
2016 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
2017 /** creates a function_impl that isn't tied to any particular function */
2018 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
2020 nir_block
*nir_block_create(nir_shader
*shader
);
2021 nir_if
*nir_if_create(nir_shader
*shader
);
2022 nir_loop
*nir_loop_create(nir_shader
*shader
);
2024 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
2026 /** requests that the given pieces of metadata be generated */
2027 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
2028 /** dirties all but the preserved metadata */
2029 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
2031 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
2032 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
2034 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
2036 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
2037 unsigned num_components
,
2040 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
2041 nir_intrinsic_op op
);
2043 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
2044 nir_function
*callee
);
2046 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
2048 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
2050 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
2052 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
2053 unsigned num_components
,
2056 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
2057 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
2058 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
2060 typedef bool (*nir_deref_foreach_leaf_cb
)(nir_deref_var
*deref
, void *state
);
2061 bool nir_deref_foreach_leaf(nir_deref_var
*deref
,
2062 nir_deref_foreach_leaf_cb cb
, void *state
);
2064 nir_load_const_instr
*
2065 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
2067 nir_const_value
nir_alu_binop_identity(nir_op binop
, unsigned bit_size
);
2070 * NIR Cursors and Instruction Insertion API
2073 * A tiny struct representing a point to insert/extract instructions or
2074 * control flow nodes. Helps reduce the combinatorial explosion of possible
2075 * points to insert/extract.
2077 * \sa nir_control_flow.h
2080 nir_cursor_before_block
,
2081 nir_cursor_after_block
,
2082 nir_cursor_before_instr
,
2083 nir_cursor_after_instr
,
2084 } nir_cursor_option
;
2087 nir_cursor_option option
;
2094 static inline nir_block
*
2095 nir_cursor_current_block(nir_cursor cursor
)
2097 if (cursor
.option
== nir_cursor_before_instr
||
2098 cursor
.option
== nir_cursor_after_instr
) {
2099 return cursor
.instr
->block
;
2101 return cursor
.block
;
2105 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2107 static inline nir_cursor
2108 nir_before_block(nir_block
*block
)
2111 cursor
.option
= nir_cursor_before_block
;
2112 cursor
.block
= block
;
2116 static inline nir_cursor
2117 nir_after_block(nir_block
*block
)
2120 cursor
.option
= nir_cursor_after_block
;
2121 cursor
.block
= block
;
2125 static inline nir_cursor
2126 nir_before_instr(nir_instr
*instr
)
2129 cursor
.option
= nir_cursor_before_instr
;
2130 cursor
.instr
= instr
;
2134 static inline nir_cursor
2135 nir_after_instr(nir_instr
*instr
)
2138 cursor
.option
= nir_cursor_after_instr
;
2139 cursor
.instr
= instr
;
2143 static inline nir_cursor
2144 nir_after_block_before_jump(nir_block
*block
)
2146 nir_instr
*last_instr
= nir_block_last_instr(block
);
2147 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2148 return nir_before_instr(last_instr
);
2150 return nir_after_block(block
);
2154 static inline nir_cursor
2155 nir_before_cf_node(nir_cf_node
*node
)
2157 if (node
->type
== nir_cf_node_block
)
2158 return nir_before_block(nir_cf_node_as_block(node
));
2160 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2163 static inline nir_cursor
2164 nir_after_cf_node(nir_cf_node
*node
)
2166 if (node
->type
== nir_cf_node_block
)
2167 return nir_after_block(nir_cf_node_as_block(node
));
2169 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2172 static inline nir_cursor
2173 nir_after_phis(nir_block
*block
)
2175 nir_foreach_instr(instr
, block
) {
2176 if (instr
->type
!= nir_instr_type_phi
)
2177 return nir_before_instr(instr
);
2179 return nir_after_block(block
);
2182 static inline nir_cursor
2183 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2185 if (node
->type
== nir_cf_node_block
)
2186 return nir_after_block(nir_cf_node_as_block(node
));
2188 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2190 return nir_after_phis(block
);
2193 static inline nir_cursor
2194 nir_before_cf_list(struct exec_list
*cf_list
)
2196 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2197 exec_list_get_head(cf_list
), node
);
2198 return nir_before_cf_node(first_node
);
2201 static inline nir_cursor
2202 nir_after_cf_list(struct exec_list
*cf_list
)
2204 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2205 exec_list_get_tail(cf_list
), node
);
2206 return nir_after_cf_node(last_node
);
2210 * Insert a NIR instruction at the given cursor.
2212 * Note: This does not update the cursor.
2214 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2217 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2219 nir_instr_insert(nir_before_instr(instr
), before
);
2223 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2225 nir_instr_insert(nir_after_instr(instr
), after
);
2229 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2231 nir_instr_insert(nir_before_block(block
), before
);
2235 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2237 nir_instr_insert(nir_after_block(block
), after
);
2241 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2243 nir_instr_insert(nir_before_cf_node(node
), before
);
2247 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2249 nir_instr_insert(nir_after_cf_node(node
), after
);
2253 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2255 nir_instr_insert(nir_before_cf_list(list
), before
);
2259 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2261 nir_instr_insert(nir_after_cf_list(list
), after
);
2264 void nir_instr_remove(nir_instr
*instr
);
2268 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2269 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2270 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2271 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2273 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2274 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2276 nir_const_value
*nir_src_as_const_value(nir_src src
);
2277 bool nir_src_is_dynamically_uniform(nir_src src
);
2278 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2279 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2280 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2281 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2282 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2284 void nir_instr_rewrite_deref(nir_instr
*instr
, nir_deref_var
**deref
,
2285 nir_deref_var
*new_deref
);
2287 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2288 unsigned num_components
, unsigned bit_size
,
2290 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2291 unsigned num_components
, unsigned bit_size
,
2294 nir_ssa_dest_init_for_type(nir_instr
*instr
, nir_dest
*dest
,
2295 const struct glsl_type
*type
,
2298 assert(glsl_type_is_vector_or_scalar(type
));
2299 nir_ssa_dest_init(instr
, dest
, glsl_get_components(type
),
2300 glsl_get_bit_size(type
), name
);
2302 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2303 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2304 nir_instr
*after_me
);
2306 uint8_t nir_ssa_def_components_read(const nir_ssa_def
*def
);
2309 * finds the next basic block in source-code order, returns NULL if there is
2313 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2315 /* Performs the opposite of nir_block_cf_tree_next() */
2317 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2319 /* Gets the first block in a CF node in source-code order */
2321 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2323 /* Gets the last block in a CF node in source-code order */
2325 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2327 /* Gets the next block after a CF node in source-code order */
2329 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2331 /* Macros for loops that visit blocks in source-code order */
2333 #define nir_foreach_block(block, impl) \
2334 for (nir_block *block = nir_start_block(impl); block != NULL; \
2335 block = nir_block_cf_tree_next(block))
2337 #define nir_foreach_block_safe(block, impl) \
2338 for (nir_block *block = nir_start_block(impl), \
2339 *next = nir_block_cf_tree_next(block); \
2341 block = next, next = nir_block_cf_tree_next(block))
2343 #define nir_foreach_block_reverse(block, impl) \
2344 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2345 block = nir_block_cf_tree_prev(block))
2347 #define nir_foreach_block_reverse_safe(block, impl) \
2348 for (nir_block *block = nir_impl_last_block(impl), \
2349 *prev = nir_block_cf_tree_prev(block); \
2351 block = prev, prev = nir_block_cf_tree_prev(block))
2353 #define nir_foreach_block_in_cf_node(block, node) \
2354 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2355 block != nir_cf_node_cf_tree_next(node); \
2356 block = nir_block_cf_tree_next(block))
2358 /* If the following CF node is an if, this function returns that if.
2359 * Otherwise, it returns NULL.
2361 nir_if
*nir_block_get_following_if(nir_block
*block
);
2363 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2365 void nir_index_local_regs(nir_function_impl
*impl
);
2366 void nir_index_global_regs(nir_shader
*shader
);
2367 void nir_index_ssa_defs(nir_function_impl
*impl
);
2368 unsigned nir_index_instrs(nir_function_impl
*impl
);
2370 void nir_index_blocks(nir_function_impl
*impl
);
2372 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2373 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2374 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2376 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2377 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2378 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2379 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2380 nir_deref
*nir_deref_clone(const nir_deref
*deref
, void *mem_ctx
);
2381 nir_deref_var
*nir_deref_var_clone(const nir_deref_var
*deref
, void *mem_ctx
);
2383 nir_shader
*nir_shader_serialize_deserialize(void *mem_ctx
, nir_shader
*s
);
2386 void nir_validate_shader(nir_shader
*shader
);
2387 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2388 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2391 should_clone_nir(void)
2393 static int should_clone
= -1;
2394 if (should_clone
< 0)
2395 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2397 return should_clone
;
2401 should_serialize_deserialize_nir(void)
2403 static int test_serialize
= -1;
2404 if (test_serialize
< 0)
2405 test_serialize
= env_var_as_boolean("NIR_TEST_SERIALIZE", false);
2407 return test_serialize
;
2411 should_print_nir(void)
2413 static int should_print
= -1;
2414 if (should_print
< 0)
2415 should_print
= env_var_as_boolean("NIR_PRINT", false);
2417 return should_print
;
2420 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2421 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2422 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2423 static inline bool should_clone_nir(void) { return false; }
2424 static inline bool should_serialize_deserialize_nir(void) { return false; }
2425 static inline bool should_print_nir(void) { return false; }
2428 #define _PASS(nir, do_pass) do { \
2430 nir_validate_shader(nir); \
2431 if (should_clone_nir()) { \
2432 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2436 if (should_serialize_deserialize_nir()) { \
2437 void *mem_ctx = ralloc_parent(nir); \
2438 nir = nir_shader_serialize_deserialize(mem_ctx, nir); \
2442 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2443 nir_metadata_set_validation_flag(nir); \
2444 if (should_print_nir()) \
2445 printf("%s\n", #pass); \
2446 if (pass(nir, ##__VA_ARGS__)) { \
2448 if (should_print_nir()) \
2449 nir_print_shader(nir, stdout); \
2450 nir_metadata_check_validation_flag(nir); \
2454 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2455 if (should_print_nir()) \
2456 printf("%s\n", #pass); \
2457 pass(nir, ##__VA_ARGS__); \
2458 if (should_print_nir()) \
2459 nir_print_shader(nir, stdout); \
2462 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2463 void nir_calc_dominance(nir_shader
*shader
);
2465 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2466 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2468 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2469 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2471 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2472 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2474 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2475 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2477 int nir_gs_count_vertices(const nir_shader
*shader
);
2479 bool nir_split_var_copies(nir_shader
*shader
);
2481 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2482 bool nir_lower_returns(nir_shader
*shader
);
2484 bool nir_inline_functions(nir_shader
*shader
);
2486 bool nir_propagate_invariant(nir_shader
*shader
);
2488 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2489 bool nir_lower_var_copies(nir_shader
*shader
);
2491 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2493 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2495 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2497 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2498 nir_function_impl
*entrypoint
,
2499 bool outputs
, bool inputs
);
2501 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2503 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2504 int (*type_size
)(const struct glsl_type
*));
2506 /* Some helpers to do very simple linking */
2507 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2508 void nir_compact_varyings(nir_shader
*producer
, nir_shader
*consumer
,
2509 bool default_to_smooth_interp
);
2512 /* If set, this forces all non-flat fragment shader inputs to be
2513 * interpolated as if with the "sample" qualifier. This requires
2514 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2516 nir_lower_io_force_sample_interpolation
= (1 << 1),
2517 } nir_lower_io_options
;
2518 bool nir_lower_io(nir_shader
*shader
,
2519 nir_variable_mode modes
,
2520 int (*type_size
)(const struct glsl_type
*),
2521 nir_lower_io_options
);
2522 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2523 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2525 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2527 void nir_lower_io_types(nir_shader
*shader
);
2528 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2529 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2530 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2532 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2533 bool nir_lower_constant_initializers(nir_shader
*shader
,
2534 nir_variable_mode modes
);
2536 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2537 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2538 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
2540 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2541 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2542 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
2543 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2544 void nir_lower_io_arrays_to_elements(nir_shader
*producer
, nir_shader
*consumer
);
2545 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader
*shader
,
2547 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2548 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
2550 bool nir_lower_samplers(nir_shader
*shader
,
2551 const struct gl_shader_program
*shader_program
);
2552 bool nir_lower_samplers_as_deref(nir_shader
*shader
,
2553 const struct gl_shader_program
*shader_program
);
2555 typedef struct nir_lower_subgroups_options
{
2556 uint8_t subgroup_size
;
2557 uint8_t ballot_bit_size
;
2558 bool lower_to_scalar
:1;
2559 bool lower_vote_trivial
:1;
2560 bool lower_subgroup_masks
:1;
2561 bool lower_shuffle
:1;
2563 } nir_lower_subgroups_options
;
2565 bool nir_lower_subgroups(nir_shader
*shader
,
2566 const nir_lower_subgroups_options
*options
);
2568 bool nir_lower_system_values(nir_shader
*shader
);
2570 typedef struct nir_lower_tex_options
{
2572 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2573 * sampler types a texture projector is lowered.
2578 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2580 bool lower_txf_offset
;
2583 * If true, lower away nir_tex_src_offset for all rect textures.
2585 bool lower_rect_offset
;
2588 * If true, lower rect textures to 2D, using txs to fetch the
2589 * texture dimensions and dividing the texture coords by the
2590 * texture dims to normalize.
2595 * If true, convert yuv to rgb.
2597 unsigned lower_y_uv_external
;
2598 unsigned lower_y_u_v_external
;
2599 unsigned lower_yx_xuxv_external
;
2600 unsigned lower_xy_uxvx_external
;
2603 * To emulate certain texture wrap modes, this can be used
2604 * to saturate the specified tex coord to [0.0, 1.0]. The
2605 * bits are according to sampler #, ie. if, for example:
2607 * (conf->saturate_s & (1 << n))
2609 * is true, then the s coord for sampler n is saturated.
2611 * Note that clamping must happen *after* projector lowering
2612 * so any projected texture sample instruction with a clamped
2613 * coordinate gets automatically lowered, regardless of the
2614 * 'lower_txp' setting.
2616 unsigned saturate_s
;
2617 unsigned saturate_t
;
2618 unsigned saturate_r
;
2620 /* Bitmask of textures that need swizzling.
2622 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2623 * swizzles[texture_index] is applied to the result of the texturing
2626 unsigned swizzle_result
;
2628 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2629 * while 4 and 5 represent 0 and 1 respectively.
2631 uint8_t swizzles
[32][4];
2634 * Bitmap of textures that need srgb to linear conversion. If
2635 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2636 * of the texture are lowered to linear.
2638 unsigned lower_srgb
;
2641 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2643 bool lower_txd_cube_map
;
2646 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
2647 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
2648 * with lower_txd_cube_map.
2650 bool lower_txd_shadow
;
2653 * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
2654 * Implies lower_txd_cube_map and lower_txd_shadow.
2657 } nir_lower_tex_options
;
2659 bool nir_lower_tex(nir_shader
*shader
,
2660 const nir_lower_tex_options
*options
);
2662 bool nir_lower_idiv(nir_shader
*shader
);
2664 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2665 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2666 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2668 void nir_lower_two_sided_color(nir_shader
*shader
);
2670 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
2672 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2673 void nir_lower_tes_patch_vertices(nir_shader
*tes
, unsigned patch_vertices
);
2675 typedef struct nir_lower_wpos_ytransform_options
{
2676 gl_state_index16 state_tokens
[STATE_LENGTH
];
2677 bool fs_coord_origin_upper_left
:1;
2678 bool fs_coord_origin_lower_left
:1;
2679 bool fs_coord_pixel_center_integer
:1;
2680 bool fs_coord_pixel_center_half_integer
:1;
2681 } nir_lower_wpos_ytransform_options
;
2683 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2684 const nir_lower_wpos_ytransform_options
*options
);
2685 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
2687 typedef struct nir_lower_drawpixels_options
{
2688 gl_state_index16 texcoord_state_tokens
[STATE_LENGTH
];
2689 gl_state_index16 scale_state_tokens
[STATE_LENGTH
];
2690 gl_state_index16 bias_state_tokens
[STATE_LENGTH
];
2691 unsigned drawpix_sampler
;
2692 unsigned pixelmap_sampler
;
2694 bool scale_and_bias
:1;
2695 } nir_lower_drawpixels_options
;
2697 void nir_lower_drawpixels(nir_shader
*shader
,
2698 const nir_lower_drawpixels_options
*options
);
2700 typedef struct nir_lower_bitmap_options
{
2703 } nir_lower_bitmap_options
;
2705 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2707 bool nir_lower_atomics(nir_shader
*shader
,
2708 const struct gl_shader_program
*shader_program
);
2709 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
2710 bool nir_lower_uniforms_to_ubo(nir_shader
*shader
);
2711 bool nir_lower_to_source_mods(nir_shader
*shader
);
2713 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2716 nir_lower_imul64
= (1 << 0),
2717 nir_lower_isign64
= (1 << 1),
2718 /** Lower all int64 modulus and division opcodes */
2719 nir_lower_divmod64
= (1 << 2),
2720 } nir_lower_int64_options
;
2722 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
2725 nir_lower_drcp
= (1 << 0),
2726 nir_lower_dsqrt
= (1 << 1),
2727 nir_lower_drsq
= (1 << 2),
2728 nir_lower_dtrunc
= (1 << 3),
2729 nir_lower_dfloor
= (1 << 4),
2730 nir_lower_dceil
= (1 << 5),
2731 nir_lower_dfract
= (1 << 6),
2732 nir_lower_dround_even
= (1 << 7),
2733 nir_lower_dmod
= (1 << 8)
2734 } nir_lower_doubles_options
;
2736 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2737 bool nir_lower_64bit_pack(nir_shader
*shader
);
2739 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2741 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2743 void nir_loop_analyze_impl(nir_function_impl
*impl
,
2744 nir_variable_mode indirect_mask
);
2746 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2748 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2749 bool nir_repair_ssa(nir_shader
*shader
);
2751 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
2753 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2754 * registers. If false, convert all values (even those not involved in a phi
2755 * node) to registers.
2757 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2759 bool nir_lower_phis_to_regs_block(nir_block
*block
);
2760 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
2762 bool nir_opt_algebraic(nir_shader
*shader
);
2763 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
2764 bool nir_opt_algebraic_late(nir_shader
*shader
);
2765 bool nir_opt_constant_folding(nir_shader
*shader
);
2767 bool nir_opt_global_to_local(nir_shader
*shader
);
2769 bool nir_copy_prop(nir_shader
*shader
);
2771 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
2773 bool nir_opt_cse(nir_shader
*shader
);
2775 bool nir_opt_dce(nir_shader
*shader
);
2777 bool nir_opt_dead_cf(nir_shader
*shader
);
2779 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2781 bool nir_opt_if(nir_shader
*shader
);
2783 bool nir_opt_intrinsics(nir_shader
*shader
);
2785 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
2787 bool nir_opt_move_comparisons(nir_shader
*shader
);
2789 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2791 bool nir_opt_remove_phis(nir_shader
*shader
);
2793 bool nir_opt_shrink_load(nir_shader
*shader
);
2795 bool nir_opt_trivial_continues(nir_shader
*shader
);
2797 bool nir_opt_undef(nir_shader
*shader
);
2799 bool nir_opt_conditional_discard(nir_shader
*shader
);
2801 void nir_sweep(nir_shader
*shader
);
2803 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2804 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);