2 * Copyright © 2014 Connor Abbott
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Connor Abbott (cwabbott0@gmail.com)
30 #include "util/hash_table.h"
31 #include "compiler/glsl/list.h"
32 #include "GL/gl.h" /* GLenum */
33 #include "util/list.h"
34 #include "util/ralloc.h"
36 #include "util/bitset.h"
37 #include "compiler/nir_types.h"
38 #include "compiler/shader_enums.h"
41 #include "nir_opcodes.h"
48 struct gl_shader_program
;
51 #define NIR_TRUE (~0u)
53 /** Defines a cast function
55 * This macro defines a cast function from in_type to out_type where
56 * out_type is some structure type that contains a field of type out_type.
58 * Note that you have to be a bit careful as the generated cast function
61 #define NIR_DEFINE_CAST(name, in_type, out_type, field) \
62 static inline out_type * \
63 name(const in_type *parent) \
65 return exec_node_data(out_type, parent, field); \
74 * Description of built-in state associated with a uniform
76 * \sa nir_variable::state_slots
84 nir_var_shader_in
= (1 << 0),
85 nir_var_shader_out
= (1 << 1),
86 nir_var_global
= (1 << 2),
87 nir_var_local
= (1 << 3),
88 nir_var_uniform
= (1 << 4),
89 nir_var_shader_storage
= (1 << 5),
90 nir_var_system_value
= (1 << 6),
91 nir_var_param
= (1 << 7),
92 nir_var_shared
= (1 << 8),
97 * Data stored in an nir_constant
99 union nir_constant_data
{
107 typedef struct nir_constant
{
109 * Value of the constant.
111 * The field used to back the values supplied by the constant is determined
112 * by the type associated with the \c nir_variable. Constants may be
113 * scalars, vectors, or matrices.
115 union nir_constant_data value
;
117 /* we could get this from the var->type but makes clone *much* easier to
118 * not have to care about the type.
120 unsigned num_elements
;
122 /* Array elements / Structure Fields */
123 struct nir_constant
**elements
;
127 * \brief Layout qualifiers for gl_FragDepth.
129 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
130 * with a layout qualifier.
133 nir_depth_layout_none
, /**< No depth layout is specified. */
134 nir_depth_layout_any
,
135 nir_depth_layout_greater
,
136 nir_depth_layout_less
,
137 nir_depth_layout_unchanged
141 * Either a uniform, global variable, shader input, or shader output. Based on
142 * ir_variable - it should be easy to translate between the two.
145 typedef struct nir_variable
{
146 struct exec_node node
;
149 * Declared type of the variable
151 const struct glsl_type
*type
;
154 * Declared name of the variable
158 struct nir_variable_data
{
160 * Storage class of the variable.
162 * \sa nir_variable_mode
164 nir_variable_mode mode
;
167 * Is the variable read-only?
169 * This is set for variables declared as \c const, shader inputs,
172 unsigned read_only
:1;
176 unsigned invariant
:1;
179 * Interpolation mode for shader inputs / outputs
181 * \sa glsl_interp_qualifier
183 unsigned interpolation
:2;
186 * \name ARB_fragment_coord_conventions
189 unsigned origin_upper_left
:1;
190 unsigned pixel_center_integer
:1;
194 * Was the location explicitly set in the shader?
196 * If the location is explicitly set in the shader, it \b cannot be changed
197 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
200 unsigned explicit_location
:1;
201 unsigned explicit_index
:1;
204 * Was an initial binding explicitly set in the shader?
206 * If so, constant_initializer contains an integer nir_constant
207 * representing the initial binding point.
209 unsigned explicit_binding
:1;
212 * Does this variable have an initializer?
214 * This is used by the linker to cross-validiate initializers of global
217 unsigned has_initializer
:1;
220 * If non-zero, then this variable may be packed along with other variables
221 * into a single varying slot, so this offset should be applied when
222 * accessing components. For example, an offset of 1 means that the x
223 * component of this variable is actually stored in component y of the
224 * location specified by \c location.
226 unsigned location_frac
:2;
229 * \brief Layout qualifier for gl_FragDepth.
231 * This is not equal to \c ir_depth_layout_none if and only if this
232 * variable is \c gl_FragDepth and a layout qualifier is specified.
234 nir_depth_layout depth_layout
;
237 * Storage location of the base of this variable
239 * The precise meaning of this field depends on the nature of the variable.
241 * - Vertex shader input: one of the values from \c gl_vert_attrib.
242 * - Vertex shader output: one of the values from \c gl_varying_slot.
243 * - Geometry shader input: one of the values from \c gl_varying_slot.
244 * - Geometry shader output: one of the values from \c gl_varying_slot.
245 * - Fragment shader input: one of the values from \c gl_varying_slot.
246 * - Fragment shader output: one of the values from \c gl_frag_result.
247 * - Uniforms: Per-stage uniform slot number for default uniform block.
248 * - Uniforms: Index within the uniform block definition for UBO members.
249 * - Non-UBO Uniforms: uniform slot number.
250 * - Other: This field is not currently used.
252 * If the variable is a uniform, shader input, or shader output, and the
253 * slot has not been assigned, the value will be -1.
258 * The actual location of the variable in the IR. Only valid for inputs
261 unsigned int driver_location
;
264 * output index for dual source blending.
269 * Descriptor set binding for sampler or UBO.
274 * Initial binding point for a sampler or UBO.
276 * For array types, this represents the binding point for the first element.
281 * Location an atomic counter is stored at.
286 * ARB_shader_image_load_store qualifiers.
289 bool read_only
; /**< "readonly" qualifier. */
290 bool write_only
; /**< "writeonly" qualifier. */
295 /** Image internal format if specified explicitly, otherwise GL_NONE. */
300 * Highest element accessed with a constant expression array index
302 * Not used for non-array variables.
304 unsigned max_array_access
;
309 * Built-in state that backs this uniform
311 * Once set at variable creation, \c state_slots must remain invariant.
312 * This is because, ideally, this array would be shared by all clones of
313 * this variable in the IR tree. In other words, we'd really like for it
314 * to be a fly-weight.
316 * If the variable is not a uniform, \c num_state_slots will be zero and
317 * \c state_slots will be \c NULL.
320 unsigned num_state_slots
; /**< Number of state slots used */
321 nir_state_slot
*state_slots
; /**< State descriptors. */
325 * Constant expression assigned in the initializer of the variable
327 nir_constant
*constant_initializer
;
330 * For variables that are in an interface block or are an instance of an
331 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
333 * \sa ir_variable::location
335 const struct glsl_type
*interface_type
;
338 #define nir_foreach_variable(var, var_list) \
339 foreach_list_typed(nir_variable, var, node, var_list)
341 #define nir_foreach_variable_safe(var, var_list) \
342 foreach_list_typed_safe(nir_variable, var, node, var_list)
345 nir_variable_is_global(const nir_variable
*var
)
347 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
350 typedef struct nir_register
{
351 struct exec_node node
;
353 unsigned num_components
; /** < number of vector components */
354 unsigned num_array_elems
; /** < size of array (0 for no array) */
356 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
359 /** generic register index. */
362 /** only for debug purposes, can be NULL */
365 /** whether this register is local (per-function) or global (per-shader) */
369 * If this flag is set to true, then accessing channels >= num_components
370 * is well-defined, and simply spills over to the next array element. This
371 * is useful for backends that can do per-component accessing, in
372 * particular scalar backends. By setting this flag and making
373 * num_components equal to 1, structures can be packed tightly into
374 * registers and then registers can be accessed per-component to get to
375 * each structure member, even if it crosses vec4 boundaries.
379 /** set of nir_src's where this register is used (read from) */
380 struct list_head uses
;
382 /** set of nir_dest's where this register is defined (written to) */
383 struct list_head defs
;
385 /** set of nir_if's where this register is used as a condition */
386 struct list_head if_uses
;
393 nir_instr_type_intrinsic
,
394 nir_instr_type_load_const
,
396 nir_instr_type_ssa_undef
,
398 nir_instr_type_parallel_copy
,
401 typedef struct nir_instr
{
402 struct exec_node node
;
404 struct nir_block
*block
;
406 /** generic instruction index. */
409 /* A temporary for optimization and analysis passes to use for storing
410 * flags. For instance, DCE uses this to store the "dead/live" info.
415 static inline nir_instr
*
416 nir_instr_next(nir_instr
*instr
)
418 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
419 if (exec_node_is_tail_sentinel(next
))
422 return exec_node_data(nir_instr
, next
, node
);
425 static inline nir_instr
*
426 nir_instr_prev(nir_instr
*instr
)
428 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
429 if (exec_node_is_head_sentinel(prev
))
432 return exec_node_data(nir_instr
, prev
, node
);
436 nir_instr_is_first(nir_instr
*instr
)
438 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
442 nir_instr_is_last(nir_instr
*instr
)
444 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
447 typedef struct nir_ssa_def
{
448 /** for debugging only, can be NULL */
451 /** generic SSA definition index. */
454 /** Index into the live_in and live_out bitfields */
457 nir_instr
*parent_instr
;
459 /** set of nir_instr's where this register is used (read from) */
460 struct list_head uses
;
462 /** set of nir_if's where this register is used as a condition */
463 struct list_head if_uses
;
465 uint8_t num_components
;
467 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
475 struct nir_src
*indirect
; /** < NULL for no indirect offset */
476 unsigned base_offset
;
478 /* TODO use-def chain goes here */
482 nir_instr
*parent_instr
;
483 struct list_head def_link
;
486 struct nir_src
*indirect
; /** < NULL for no indirect offset */
487 unsigned base_offset
;
489 /* TODO def-use chain goes here */
494 typedef struct nir_src
{
496 nir_instr
*parent_instr
;
497 struct nir_if
*parent_if
;
500 struct list_head use_link
;
510 static inline nir_src
513 nir_src src
= { { NULL
} };
517 #define NIR_SRC_INIT nir_src_init()
519 #define nir_foreach_use(src, reg_or_ssa_def) \
520 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
522 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
523 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
525 #define nir_foreach_if_use(src, reg_or_ssa_def) \
526 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
528 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
529 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
540 static inline nir_dest
543 nir_dest dest
= { { { NULL
} } };
547 #define NIR_DEST_INIT nir_dest_init()
549 #define nir_foreach_def(dest, reg) \
550 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
552 #define nir_foreach_def_safe(dest, reg) \
553 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
555 static inline nir_src
556 nir_src_for_ssa(nir_ssa_def
*def
)
558 nir_src src
= NIR_SRC_INIT
;
566 static inline nir_src
567 nir_src_for_reg(nir_register
*reg
)
569 nir_src src
= NIR_SRC_INIT
;
573 src
.reg
.indirect
= NULL
;
574 src
.reg
.base_offset
= 0;
579 static inline nir_dest
580 nir_dest_for_reg(nir_register
*reg
)
582 nir_dest dest
= NIR_DEST_INIT
;
589 static inline unsigned
590 nir_src_bit_size(nir_src src
)
592 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
595 static inline unsigned
596 nir_dest_bit_size(nir_dest dest
)
598 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
601 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
602 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
608 * \name input modifiers
612 * For inputs interpreted as floating point, flips the sign bit. For
613 * inputs interpreted as integers, performs the two's complement negation.
618 * Clears the sign bit for floating point values, and computes the integer
619 * absolute value for integers. Note that the negate modifier acts after
620 * the absolute value modifier, therefore if both are set then all inputs
621 * will become negative.
627 * For each input component, says which component of the register it is
628 * chosen from. Note that which elements of the swizzle are used and which
629 * are ignored are based on the write mask for most opcodes - for example,
630 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
631 * a swizzle of {2, x, 1, 0} where x means "don't care."
640 * \name saturate output modifier
642 * Only valid for opcodes that output floating-point numbers. Clamps the
643 * output to between 0.0 and 1.0 inclusive.
648 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
652 nir_type_invalid
= 0, /* Not a valid type */
657 nir_type_bool32
= 32 | nir_type_bool
,
658 nir_type_int8
= 8 | nir_type_int
,
659 nir_type_int16
= 16 | nir_type_int
,
660 nir_type_int32
= 32 | nir_type_int
,
661 nir_type_int64
= 64 | nir_type_int
,
662 nir_type_uint8
= 8 | nir_type_uint
,
663 nir_type_uint16
= 16 | nir_type_uint
,
664 nir_type_uint32
= 32 | nir_type_uint
,
665 nir_type_uint64
= 64 | nir_type_uint
,
666 nir_type_float16
= 16 | nir_type_float
,
667 nir_type_float32
= 32 | nir_type_float
,
668 nir_type_float64
= 64 | nir_type_float
,
671 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
672 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
674 static inline unsigned
675 nir_alu_type_get_type_size(nir_alu_type type
)
677 return type
& NIR_ALU_TYPE_SIZE_MASK
;
680 static inline unsigned
681 nir_alu_type_get_base_type(nir_alu_type type
)
683 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
687 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
688 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
689 } nir_op_algebraic_property
;
697 * The number of components in the output
699 * If non-zero, this is the size of the output and input sizes are
700 * explicitly given; swizzle and writemask are still in effect, but if
701 * the output component is masked out, then the input component may
704 * If zero, the opcode acts in the standard, per-component manner; the
705 * operation is performed on each component (except the ones that are
706 * masked out) with the input being taken from the input swizzle for
709 * The size of some of the inputs may be given (i.e. non-zero) even
710 * though output_size is zero; in that case, the inputs with a zero
711 * size act per-component, while the inputs with non-zero size don't.
713 unsigned output_size
;
716 * The type of vector that the instruction outputs. Note that the
717 * staurate modifier is only allowed on outputs with the float type.
720 nir_alu_type output_type
;
723 * The number of components in each input
725 unsigned input_sizes
[4];
728 * The type of vector that each input takes. Note that negate and
729 * absolute value are only allowed on inputs with int or float type and
730 * behave differently on the two.
732 nir_alu_type input_types
[4];
734 nir_op_algebraic_property algebraic_properties
;
737 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
739 typedef struct nir_alu_instr
{
743 /** Indicates that this ALU instruction generates an exact value
745 * This is kind of a mixture of GLSL "precise" and "invariant" and not
746 * really equivalent to either. This indicates that the value generated by
747 * this operation is high-precision and any code transformations that touch
748 * it must ensure that the resulting value is bit-for-bit identical to the
757 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
758 nir_alu_instr
*instr
);
759 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
760 nir_alu_instr
*instr
);
762 /* is this source channel used? */
764 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
766 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
767 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
769 return (instr
->dest
.write_mask
>> channel
) & 1;
773 * For instructions whose destinations are SSA, get the number of channels
776 static inline unsigned
777 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
779 assert(instr
->dest
.dest
.is_ssa
);
781 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
782 return nir_op_infos
[instr
->op
].input_sizes
[src
];
784 return instr
->dest
.dest
.ssa
.num_components
;
789 nir_deref_type_array
,
790 nir_deref_type_struct
793 typedef struct nir_deref
{
794 nir_deref_type deref_type
;
795 struct nir_deref
*child
;
796 const struct glsl_type
*type
;
805 /* This enum describes how the array is referenced. If the deref is
806 * direct then the base_offset is used. If the deref is indirect then then
807 * offset is given by base_offset + indirect. If the deref is a wildcard
808 * then the deref refers to all of the elements of the array at the same
809 * time. Wildcard dereferences are only ever allowed in copy_var
810 * intrinsics and the source and destination derefs must have matching
814 nir_deref_array_type_direct
,
815 nir_deref_array_type_indirect
,
816 nir_deref_array_type_wildcard
,
817 } nir_deref_array_type
;
822 nir_deref_array_type deref_array_type
;
823 unsigned base_offset
;
833 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
834 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
835 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
837 /* Returns the last deref in the chain. */
838 static inline nir_deref
*
839 nir_deref_tail(nir_deref
*deref
)
842 deref
= deref
->child
;
850 nir_deref_var
**params
;
851 nir_deref_var
*return_deref
;
853 struct nir_function
*callee
;
856 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
857 num_variables, num_indices, idx0, idx1, idx2, flags) \
858 nir_intrinsic_##name,
860 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
863 #include "nir_intrinsics.h"
864 nir_num_intrinsics
= nir_last_intrinsic
+ 1
868 #undef LAST_INTRINSIC
870 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
872 /** Represents an intrinsic
874 * An intrinsic is an instruction type for handling things that are
875 * more-or-less regular operations but don't just consume and produce SSA
876 * values like ALU operations do. Intrinsics are not for things that have
877 * special semantic meaning such as phi nodes and parallel copies.
878 * Examples of intrinsics include variable load/store operations, system
879 * value loads, and the like. Even though texturing more-or-less falls
880 * under this category, texturing is its own instruction type because
881 * trying to represent texturing with intrinsics would lead to a
882 * combinatorial explosion of intrinsic opcodes.
884 * By having a single instruction type for handling a lot of different
885 * cases, optimization passes can look for intrinsics and, for the most
886 * part, completely ignore them. Each intrinsic type also has a few
887 * possible flags that govern whether or not they can be reordered or
888 * eliminated. That way passes like dead code elimination can still work
889 * on intrisics without understanding the meaning of each.
891 * Each intrinsic has some number of constant indices, some number of
892 * variables, and some number of sources. What these sources, variables,
893 * and indices mean depends on the intrinsic and is documented with the
894 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
895 * instructions are the only types of instruction that can operate on
901 nir_intrinsic_op intrinsic
;
905 /** number of components if this is a vectorized intrinsic
907 * Similarly to ALU operations, some intrinsics are vectorized.
908 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
909 * For vectorized intrinsics, the num_components field specifies the
910 * number of destination components and the number of source components
911 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
913 uint8_t num_components
;
915 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
917 nir_deref_var
*variables
[2];
920 } nir_intrinsic_instr
;
923 * \name NIR intrinsics semantic flags
925 * information about what the compiler can do with the intrinsics.
927 * \sa nir_intrinsic_info::flags
931 * whether the intrinsic can be safely eliminated if none of its output
932 * value is not being used.
934 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
937 * Whether the intrinsic can be reordered with respect to any other
938 * intrinsic, i.e. whether the only reordering dependencies of the
939 * intrinsic are due to the register reads/writes.
941 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
942 } nir_intrinsic_semantic_flag
;
945 * \name NIR intrinsics const-index flag
947 * Indicates the usage of a const_index slot.
949 * \sa nir_intrinsic_info::index_map
953 * Generally instructions that take a offset src argument, can encode
954 * a constant 'base' value which is added to the offset.
956 NIR_INTRINSIC_BASE
= 1,
959 * For store instructions, a writemask for the store.
961 NIR_INTRINSIC_WRMASK
= 2,
964 * The stream-id for GS emit_vertex/end_primitive intrinsics.
966 NIR_INTRINSIC_STREAM_ID
= 3,
969 * The clip-plane id for load_user_clip_plane intrinsic.
971 NIR_INTRINSIC_UCP_ID
= 4,
974 * The amount of data, starting from BASE, that this instruction may
975 * access. This is used to provide bounds if the offset is not constant.
977 NIR_INTRINSIC_RANGE
= 5,
980 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
982 NIR_INTRINSIC_DESC_SET
= 6,
985 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
987 NIR_INTRINSIC_BINDING
= 7,
989 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
991 } nir_intrinsic_index_flag
;
993 #define NIR_INTRINSIC_MAX_INPUTS 4
998 unsigned num_srcs
; /** < number of register/SSA inputs */
1000 /** number of components of each input register
1002 * If this value is 0, the number of components is given by the
1003 * num_components field of nir_intrinsic_instr.
1005 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1009 /** number of components of the output register
1011 * If this value is 0, the number of components is given by the
1012 * num_components field of nir_intrinsic_instr.
1014 unsigned dest_components
;
1016 /** the number of inputs/outputs that are variables */
1017 unsigned num_variables
;
1019 /** the number of constant indices used by the intrinsic */
1020 unsigned num_indices
;
1022 /** indicates the usage of intr->const_index[n] */
1023 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1025 /** semantic flags for calls to this intrinsic */
1026 nir_intrinsic_semantic_flag flags
;
1027 } nir_intrinsic_info
;
1029 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1032 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1033 static inline type \
1034 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
1036 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1037 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1038 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1040 static inline void \
1041 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1043 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1044 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1045 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1048 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1049 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1050 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1051 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1052 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1053 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1054 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1057 * \group texture information
1059 * This gives semantic information about textures which is useful to the
1060 * frontend, the backend, and lowering passes, but not the optimizer.
1065 nir_tex_src_projector
,
1066 nir_tex_src_comparitor
, /* shadow comparitor */
1070 nir_tex_src_ms_index
, /* MSAA sample index */
1071 nir_tex_src_ms_mcs
, /* MSAA compression value */
1074 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1075 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1076 nir_num_tex_src_types
1081 nir_tex_src_type src_type
;
1085 nir_texop_tex
, /**< Regular texture look-up */
1086 nir_texop_txb
, /**< Texture look-up with LOD bias */
1087 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1088 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1089 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1090 nir_texop_txf_ms
, /**< Multisample texture fetch */
1091 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1092 nir_texop_txs
, /**< Texture size */
1093 nir_texop_lod
, /**< Texture lod query */
1094 nir_texop_tg4
, /**< Texture gather */
1095 nir_texop_query_levels
, /**< Texture levels query */
1096 nir_texop_texture_samples
, /**< Texture samples query */
1097 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1105 enum glsl_sampler_dim sampler_dim
;
1106 nir_alu_type dest_type
;
1111 unsigned num_srcs
, coord_components
;
1112 bool is_array
, is_shadow
;
1115 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1116 * components or the new-style shadow that outputs 1 component.
1118 bool is_new_style_shadow
;
1120 /* gather component selector */
1121 unsigned component
: 2;
1123 /** The texture index
1125 * If this texture instruction has a nir_tex_src_texture_offset source,
1126 * then the texture index is given by texture_index + texture_offset.
1128 unsigned texture_index
;
1130 /** The size of the texture array or 0 if it's not an array */
1131 unsigned texture_array_size
;
1133 /** The texture deref
1135 * If this is null, use texture_index instead.
1137 nir_deref_var
*texture
;
1139 /** The sampler index
1141 * The following operations do not require a sampler and, as such, this
1142 * field should be ignored:
1144 * - nir_texop_txf_ms
1148 * - nir_texop_query_levels
1149 * - nir_texop_texture_samples
1150 * - nir_texop_samples_identical
1152 * If this texture instruction has a nir_tex_src_sampler_offset source,
1153 * then the sampler index is given by sampler_index + sampler_offset.
1155 unsigned sampler_index
;
1157 /** The sampler deref
1159 * If this is null, use sampler_index instead.
1161 nir_deref_var
*sampler
;
1164 static inline unsigned
1165 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1167 switch (instr
->op
) {
1168 case nir_texop_txs
: {
1170 switch (instr
->sampler_dim
) {
1171 case GLSL_SAMPLER_DIM_1D
:
1172 case GLSL_SAMPLER_DIM_BUF
:
1175 case GLSL_SAMPLER_DIM_2D
:
1176 case GLSL_SAMPLER_DIM_CUBE
:
1177 case GLSL_SAMPLER_DIM_MS
:
1178 case GLSL_SAMPLER_DIM_RECT
:
1179 case GLSL_SAMPLER_DIM_EXTERNAL
:
1182 case GLSL_SAMPLER_DIM_3D
:
1186 unreachable("not reached");
1188 if (instr
->is_array
)
1196 case nir_texop_texture_samples
:
1197 case nir_texop_query_levels
:
1198 case nir_texop_samples_identical
:
1202 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1209 /* Returns true if this texture operation queries something about the texture
1210 * rather than actually sampling it.
1213 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1215 switch (instr
->op
) {
1218 case nir_texop_texture_samples
:
1219 case nir_texop_query_levels
:
1220 case nir_texop_txf_ms_mcs
:
1227 case nir_texop_txf_ms
:
1231 unreachable("Invalid texture opcode");
1235 static inline unsigned
1236 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1238 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1239 return instr
->coord_components
;
1241 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1242 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1245 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1246 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1247 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1248 if (instr
->is_array
)
1249 return instr
->coord_components
- 1;
1251 return instr
->coord_components
;
1258 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1260 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1261 if (instr
->src
[i
].src_type
== type
)
1281 nir_const_value value
;
1284 } nir_load_const_instr
;
1297 /* creates a new SSA variable in an undefined state */
1302 } nir_ssa_undef_instr
;
1305 struct exec_node node
;
1307 /* The predecessor block corresponding to this source */
1308 struct nir_block
*pred
;
1313 #define nir_foreach_phi_src(phi_src, phi) \
1314 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1315 #define nir_foreach_phi_src_safe(phi_src, phi) \
1316 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1321 struct exec_list srcs
; /** < list of nir_phi_src */
1327 struct exec_node node
;
1330 } nir_parallel_copy_entry
;
1332 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1333 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1338 /* A list of nir_parallel_copy_entry's. The sources of all of the
1339 * entries are copied to the corresponding destinations "in parallel".
1340 * In other words, if we have two entries: a -> b and b -> a, the values
1343 struct exec_list entries
;
1344 } nir_parallel_copy_instr
;
1346 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1347 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1348 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1349 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1350 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1351 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1352 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1353 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1354 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1355 nir_parallel_copy_instr
, instr
)
1360 * Control flow consists of a tree of control flow nodes, which include
1361 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1362 * instructions that always run start-to-finish. Each basic block also keeps
1363 * track of its successors (blocks which may run immediately after the current
1364 * block) and predecessors (blocks which could have run immediately before the
1365 * current block). Each function also has a start block and an end block which
1366 * all return statements point to (which is always empty). Together, all the
1367 * blocks with their predecessors and successors make up the control flow
1368 * graph (CFG) of the function. There are helpers that modify the tree of
1369 * control flow nodes while modifying the CFG appropriately; these should be
1370 * used instead of modifying the tree directly.
1377 nir_cf_node_function
1380 typedef struct nir_cf_node
{
1381 struct exec_node node
;
1382 nir_cf_node_type type
;
1383 struct nir_cf_node
*parent
;
1386 typedef struct nir_block
{
1387 nir_cf_node cf_node
;
1389 struct exec_list instr_list
; /** < list of nir_instr */
1391 /** generic block index; generated by nir_index_blocks */
1395 * Each block can only have up to 2 successors, so we put them in a simple
1396 * array - no need for anything more complicated.
1398 struct nir_block
*successors
[2];
1400 /* Set of nir_block predecessors in the CFG */
1401 struct set
*predecessors
;
1404 * this node's immediate dominator in the dominance tree - set to NULL for
1407 struct nir_block
*imm_dom
;
1409 /* This node's children in the dominance tree */
1410 unsigned num_dom_children
;
1411 struct nir_block
**dom_children
;
1413 /* Set of nir_block's on the dominance frontier of this block */
1414 struct set
*dom_frontier
;
1417 * These two indices have the property that dom_{pre,post}_index for each
1418 * child of this block in the dominance tree will always be between
1419 * dom_pre_index and dom_post_index for this block, which makes testing if
1420 * a given block is dominated by another block an O(1) operation.
1422 unsigned dom_pre_index
, dom_post_index
;
1424 /* live in and out for this block; used for liveness analysis */
1425 BITSET_WORD
*live_in
;
1426 BITSET_WORD
*live_out
;
1429 static inline nir_instr
*
1430 nir_block_first_instr(nir_block
*block
)
1432 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1433 return exec_node_data(nir_instr
, head
, node
);
1436 static inline nir_instr
*
1437 nir_block_last_instr(nir_block
*block
)
1439 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1440 return exec_node_data(nir_instr
, tail
, node
);
1443 #define nir_foreach_instr(instr, block) \
1444 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1445 #define nir_foreach_instr_reverse(instr, block) \
1446 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1447 #define nir_foreach_instr_safe(instr, block) \
1448 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1449 #define nir_foreach_instr_reverse_safe(instr, block) \
1450 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1452 typedef struct nir_if
{
1453 nir_cf_node cf_node
;
1456 struct exec_list then_list
; /** < list of nir_cf_node */
1457 struct exec_list else_list
; /** < list of nir_cf_node */
1460 static inline nir_cf_node
*
1461 nir_if_first_then_node(nir_if
*if_stmt
)
1463 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1464 return exec_node_data(nir_cf_node
, head
, node
);
1467 static inline nir_cf_node
*
1468 nir_if_last_then_node(nir_if
*if_stmt
)
1470 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1471 return exec_node_data(nir_cf_node
, tail
, node
);
1474 static inline nir_cf_node
*
1475 nir_if_first_else_node(nir_if
*if_stmt
)
1477 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1478 return exec_node_data(nir_cf_node
, head
, node
);
1481 static inline nir_cf_node
*
1482 nir_if_last_else_node(nir_if
*if_stmt
)
1484 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1485 return exec_node_data(nir_cf_node
, tail
, node
);
1489 nir_cf_node cf_node
;
1491 struct exec_list body
; /** < list of nir_cf_node */
1494 static inline nir_cf_node
*
1495 nir_loop_first_cf_node(nir_loop
*loop
)
1497 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1500 static inline nir_cf_node
*
1501 nir_loop_last_cf_node(nir_loop
*loop
)
1503 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1507 * Various bits of metadata that can may be created or required by
1508 * optimization and analysis passes
1511 nir_metadata_none
= 0x0,
1512 nir_metadata_block_index
= 0x1,
1513 nir_metadata_dominance
= 0x2,
1514 nir_metadata_live_ssa_defs
= 0x4,
1515 nir_metadata_not_properly_reset
= 0x8,
1519 nir_cf_node cf_node
;
1521 /** pointer to the function of which this is an implementation */
1522 struct nir_function
*function
;
1524 struct exec_list body
; /** < list of nir_cf_node */
1526 nir_block
*end_block
;
1528 /** list for all local variables in the function */
1529 struct exec_list locals
;
1531 /** array of variables used as parameters */
1532 unsigned num_params
;
1533 nir_variable
**params
;
1535 /** variable used to hold the result of the function */
1536 nir_variable
*return_var
;
1538 /** list of local registers in the function */
1539 struct exec_list registers
;
1541 /** next available local register index */
1544 /** next available SSA value index */
1547 /* total number of basic blocks, only valid when block_index_dirty = false */
1548 unsigned num_blocks
;
1550 nir_metadata valid_metadata
;
1551 } nir_function_impl
;
1553 static inline nir_block
*
1554 nir_start_block(nir_function_impl
*impl
)
1556 return (nir_block
*) exec_list_get_head(&impl
->body
);
1559 static inline nir_block
*
1560 nir_impl_last_block(nir_function_impl
*impl
)
1562 return (nir_block
*) exec_list_get_tail(&impl
->body
);
1565 static inline nir_cf_node
*
1566 nir_cf_node_next(nir_cf_node
*node
)
1568 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1569 if (exec_node_is_tail_sentinel(next
))
1572 return exec_node_data(nir_cf_node
, next
, node
);
1575 static inline nir_cf_node
*
1576 nir_cf_node_prev(nir_cf_node
*node
)
1578 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1579 if (exec_node_is_head_sentinel(prev
))
1582 return exec_node_data(nir_cf_node
, prev
, node
);
1586 nir_cf_node_is_first(const nir_cf_node
*node
)
1588 return exec_node_is_head_sentinel(node
->node
.prev
);
1592 nir_cf_node_is_last(const nir_cf_node
*node
)
1594 return exec_node_is_tail_sentinel(node
->node
.next
);
1597 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1598 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1599 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1600 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1605 nir_parameter_inout
,
1606 } nir_parameter_type
;
1609 nir_parameter_type param_type
;
1610 const struct glsl_type
*type
;
1613 typedef struct nir_function
{
1614 struct exec_node node
;
1617 struct nir_shader
*shader
;
1619 unsigned num_params
;
1620 nir_parameter
*params
;
1621 const struct glsl_type
*return_type
;
1623 /** The implementation of this function.
1625 * If the function is only declared and not implemented, this is NULL.
1627 nir_function_impl
*impl
;
1630 typedef struct nir_shader_compiler_options
{
1635 /** Lowers flrp when it does not support doubles */
1642 bool lower_bitfield_extract
;
1643 bool lower_bitfield_insert
;
1644 bool lower_uadd_carry
;
1645 bool lower_usub_borrow
;
1646 /** lowers fneg and ineg to fsub and isub. */
1648 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1651 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1654 /* Does the native fdot instruction replicate its result for four
1655 * components? If so, then opt_algebraic_late will turn all fdotN
1656 * instructions into fdot_replicatedN instructions.
1658 bool fdot_replicates
;
1660 /** lowers ffract to fsub+ffloor: */
1663 bool lower_pack_half_2x16
;
1664 bool lower_pack_unorm_2x16
;
1665 bool lower_pack_snorm_2x16
;
1666 bool lower_pack_unorm_4x8
;
1667 bool lower_pack_snorm_4x8
;
1668 bool lower_unpack_half_2x16
;
1669 bool lower_unpack_unorm_2x16
;
1670 bool lower_unpack_snorm_2x16
;
1671 bool lower_unpack_unorm_4x8
;
1672 bool lower_unpack_snorm_4x8
;
1674 bool lower_extract_byte
;
1675 bool lower_extract_word
;
1678 * Does the driver support real 32-bit integers? (Otherwise, integers
1679 * are simulated by floats.)
1681 bool native_integers
;
1683 /* Indicates that the driver only has zero-based vertex id */
1684 bool vertex_id_zero_based
;
1685 } nir_shader_compiler_options
;
1687 typedef struct nir_shader_info
{
1690 /* Descriptive name provided by the client; may be NULL */
1693 /* Number of textures used by this shader */
1694 unsigned num_textures
;
1695 /* Number of uniform buffers used by this shader */
1697 /* Number of atomic buffers used by this shader */
1699 /* Number of shader storage buffers used by this shader */
1701 /* Number of images used by this shader */
1702 unsigned num_images
;
1704 /* Which inputs are actually read */
1705 uint64_t inputs_read
;
1706 /* Which outputs are actually written */
1707 uint64_t outputs_written
;
1708 /* Which system values are actually read */
1709 uint64_t system_values_read
;
1711 /* Which patch inputs are actually read */
1712 uint32_t patch_inputs_read
;
1713 /* Which patch outputs are actually written */
1714 uint32_t patch_outputs_written
;
1716 /* Whether or not this shader ever uses textureGather() */
1717 bool uses_texture_gather
;
1719 /* Whether or not this shader uses the gl_ClipDistance output */
1720 bool uses_clip_distance_out
;
1722 /* Whether or not separate shader objects were used */
1723 bool separate_shader
;
1725 /** Was this shader linked with any transform feedback varyings? */
1726 bool has_transform_feedback_varyings
;
1730 /** The number of vertices recieves per input primitive */
1731 unsigned vertices_in
;
1733 /** The output primitive type (GL enum value) */
1734 unsigned output_primitive
;
1736 /** The maximum number of vertices the geometry shader might write. */
1737 unsigned vertices_out
;
1739 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1740 unsigned invocations
;
1742 /** Whether or not this shader uses EndPrimitive */
1743 bool uses_end_primitive
;
1745 /** Whether or not this shader uses non-zero streams */
1753 * Whether any inputs are declared with the "sample" qualifier.
1755 bool uses_sample_qualifier
;
1758 * Whether early fragment tests are enabled as defined by
1759 * ARB_shader_image_load_store.
1761 bool early_fragment_tests
;
1763 /** gl_FragDepth layout for ARB_conservative_depth. */
1764 enum gl_frag_depth_layout depth_layout
;
1768 unsigned local_size
[3];
1772 /** The number of vertices in the TCS output patch. */
1773 unsigned vertices_out
;
1778 typedef struct nir_shader
{
1779 /** list of uniforms (nir_variable) */
1780 struct exec_list uniforms
;
1782 /** list of inputs (nir_variable) */
1783 struct exec_list inputs
;
1785 /** list of outputs (nir_variable) */
1786 struct exec_list outputs
;
1788 /** list of shared compute variables (nir_variable) */
1789 struct exec_list shared
;
1791 /** Set of driver-specific options for the shader.
1793 * The memory for the options is expected to be kept in a single static
1794 * copy by the driver.
1796 const struct nir_shader_compiler_options
*options
;
1798 /** Various bits of compile-time information about a given shader */
1799 struct nir_shader_info info
;
1801 /** list of global variables in the shader (nir_variable) */
1802 struct exec_list globals
;
1804 /** list of system value variables in the shader (nir_variable) */
1805 struct exec_list system_values
;
1807 struct exec_list functions
; /** < list of nir_function */
1809 /** list of global register in the shader */
1810 struct exec_list registers
;
1812 /** next available global register index */
1816 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1819 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1821 /** The shader stage, such as MESA_SHADER_VERTEX. */
1822 gl_shader_stage stage
;
1825 static inline nir_function
*
1826 nir_shader_get_entrypoint(nir_shader
*shader
)
1828 assert(exec_list_length(&shader
->functions
) == 1);
1829 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1830 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1831 assert(func
->return_type
== glsl_void_type());
1832 assert(func
->num_params
== 0);
1836 #define nir_foreach_function(func, shader) \
1837 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1839 nir_shader
*nir_shader_create(void *mem_ctx
,
1840 gl_shader_stage stage
,
1841 const nir_shader_compiler_options
*options
);
1843 /** creates a register, including assigning it an index and adding it to the list */
1844 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1846 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1848 void nir_reg_remove(nir_register
*reg
);
1850 /** Adds a variable to the appropreate list in nir_shader */
1851 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1854 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1856 assert(var
->data
.mode
== nir_var_local
);
1857 exec_list_push_tail(&impl
->locals
, &var
->node
);
1860 /** creates a variable, sets a few defaults, and adds it to the list */
1861 nir_variable
*nir_variable_create(nir_shader
*shader
,
1862 nir_variable_mode mode
,
1863 const struct glsl_type
*type
,
1865 /** creates a local variable and adds it to the list */
1866 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1867 const struct glsl_type
*type
,
1870 /** creates a function and adds it to the shader's list of functions */
1871 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1873 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1874 /** creates a function_impl that isn't tied to any particular function */
1875 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1877 nir_block
*nir_block_create(nir_shader
*shader
);
1878 nir_if
*nir_if_create(nir_shader
*shader
);
1879 nir_loop
*nir_loop_create(nir_shader
*shader
);
1881 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1883 /** requests that the given pieces of metadata be generated */
1884 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1885 /** dirties all but the preserved metadata */
1886 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1888 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1889 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1891 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1893 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1894 unsigned num_components
,
1897 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1898 nir_intrinsic_op op
);
1900 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1901 nir_function
*callee
);
1903 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1905 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1907 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1909 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1910 unsigned num_components
,
1913 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1914 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1915 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1917 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1919 nir_load_const_instr
*
1920 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1923 * NIR Cursors and Instruction Insertion API
1926 * A tiny struct representing a point to insert/extract instructions or
1927 * control flow nodes. Helps reduce the combinatorial explosion of possible
1928 * points to insert/extract.
1930 * \sa nir_control_flow.h
1933 nir_cursor_before_block
,
1934 nir_cursor_after_block
,
1935 nir_cursor_before_instr
,
1936 nir_cursor_after_instr
,
1937 } nir_cursor_option
;
1940 nir_cursor_option option
;
1947 static inline nir_block
*
1948 nir_cursor_current_block(nir_cursor cursor
)
1950 if (cursor
.option
== nir_cursor_before_instr
||
1951 cursor
.option
== nir_cursor_after_instr
) {
1952 return cursor
.instr
->block
;
1954 return cursor
.block
;
1958 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
1960 static inline nir_cursor
1961 nir_before_block(nir_block
*block
)
1964 cursor
.option
= nir_cursor_before_block
;
1965 cursor
.block
= block
;
1969 static inline nir_cursor
1970 nir_after_block(nir_block
*block
)
1973 cursor
.option
= nir_cursor_after_block
;
1974 cursor
.block
= block
;
1978 static inline nir_cursor
1979 nir_before_instr(nir_instr
*instr
)
1982 cursor
.option
= nir_cursor_before_instr
;
1983 cursor
.instr
= instr
;
1987 static inline nir_cursor
1988 nir_after_instr(nir_instr
*instr
)
1991 cursor
.option
= nir_cursor_after_instr
;
1992 cursor
.instr
= instr
;
1996 static inline nir_cursor
1997 nir_after_block_before_jump(nir_block
*block
)
1999 nir_instr
*last_instr
= nir_block_last_instr(block
);
2000 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2001 return nir_before_instr(last_instr
);
2003 return nir_after_block(block
);
2007 static inline nir_cursor
2008 nir_before_cf_node(nir_cf_node
*node
)
2010 if (node
->type
== nir_cf_node_block
)
2011 return nir_before_block(nir_cf_node_as_block(node
));
2013 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2016 static inline nir_cursor
2017 nir_after_cf_node(nir_cf_node
*node
)
2019 if (node
->type
== nir_cf_node_block
)
2020 return nir_after_block(nir_cf_node_as_block(node
));
2022 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2025 static inline nir_cursor
2026 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2028 if (node
->type
== nir_cf_node_block
)
2029 return nir_after_block(nir_cf_node_as_block(node
));
2031 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2032 assert(block
->cf_node
.type
== nir_cf_node_block
);
2034 nir_foreach_instr(instr
, block
) {
2035 if (instr
->type
!= nir_instr_type_phi
)
2036 return nir_before_instr(instr
);
2038 return nir_after_block(block
);
2041 static inline nir_cursor
2042 nir_before_cf_list(struct exec_list
*cf_list
)
2044 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2045 exec_list_get_head(cf_list
), node
);
2046 return nir_before_cf_node(first_node
);
2049 static inline nir_cursor
2050 nir_after_cf_list(struct exec_list
*cf_list
)
2052 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2053 exec_list_get_tail(cf_list
), node
);
2054 return nir_after_cf_node(last_node
);
2058 * Insert a NIR instruction at the given cursor.
2060 * Note: This does not update the cursor.
2062 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2065 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2067 nir_instr_insert(nir_before_instr(instr
), before
);
2071 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2073 nir_instr_insert(nir_after_instr(instr
), after
);
2077 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2079 nir_instr_insert(nir_before_block(block
), before
);
2083 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2085 nir_instr_insert(nir_after_block(block
), after
);
2089 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2091 nir_instr_insert(nir_before_cf_node(node
), before
);
2095 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2097 nir_instr_insert(nir_after_cf_node(node
), after
);
2101 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2103 nir_instr_insert(nir_before_cf_list(list
), before
);
2107 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2109 nir_instr_insert(nir_after_cf_list(list
), after
);
2112 void nir_instr_remove(nir_instr
*instr
);
2116 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2117 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2118 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2119 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2121 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2122 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2124 nir_const_value
*nir_src_as_const_value(nir_src src
);
2125 bool nir_src_is_dynamically_uniform(nir_src src
);
2126 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2127 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2128 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2129 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2130 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2133 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2134 unsigned num_components
, unsigned bit_size
,
2136 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2137 unsigned num_components
, unsigned bit_size
,
2139 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2140 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2141 nir_instr
*after_me
);
2143 uint8_t nir_ssa_def_components_read(nir_ssa_def
*def
);
2146 * finds the next basic block in source-code order, returns NULL if there is
2150 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2152 /* Performs the opposite of nir_block_cf_tree_next() */
2154 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2156 /* Gets the first block in a CF node in source-code order */
2158 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2160 /* Gets the last block in a CF node in source-code order */
2162 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2164 /* Gets the next block after a CF node in source-code order */
2166 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2168 /* Macros for loops that visit blocks in source-code order */
2170 #define nir_foreach_block(block, impl) \
2171 for (nir_block *block = nir_start_block(impl); block != NULL; \
2172 block = nir_block_cf_tree_next(block))
2174 #define nir_foreach_block_safe(block, impl) \
2175 for (nir_block *block = nir_start_block(impl), \
2176 *next = nir_block_cf_tree_next(block); \
2178 block = next, next = nir_block_cf_tree_next(block))
2180 #define nir_foreach_block_reverse(block, impl) \
2181 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2182 block = nir_block_cf_tree_prev(block))
2184 #define nir_foreach_block_reverse_safe(block, impl) \
2185 for (nir_block *block = nir_impl_last_block(impl), \
2186 *prev = nir_block_cf_tree_prev(block); \
2188 block = prev, prev = nir_block_cf_tree_prev(block))
2190 #define nir_foreach_block_in_cf_node(block, node) \
2191 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2192 block != nir_cf_node_cf_tree_next(node); \
2193 block = nir_block_cf_tree_next(block))
2195 /* If the following CF node is an if, this function returns that if.
2196 * Otherwise, it returns NULL.
2198 nir_if
*nir_block_get_following_if(nir_block
*block
);
2200 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2202 void nir_index_local_regs(nir_function_impl
*impl
);
2203 void nir_index_global_regs(nir_shader
*shader
);
2204 void nir_index_ssa_defs(nir_function_impl
*impl
);
2205 unsigned nir_index_instrs(nir_function_impl
*impl
);
2207 void nir_index_blocks(nir_function_impl
*impl
);
2209 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2210 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2212 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2213 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2214 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2215 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2218 void nir_validate_shader(nir_shader
*shader
);
2219 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2220 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2222 #include "util/debug.h"
2224 should_clone_nir(void)
2226 static int should_clone
= -1;
2227 if (should_clone
< 0)
2228 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2230 return should_clone
;
2233 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2234 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2235 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2236 static inline bool should_clone_nir(void) { return false; }
2239 #define _PASS(nir, do_pass) do { \
2241 nir_validate_shader(nir); \
2242 if (should_clone_nir()) { \
2243 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2249 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2250 nir_metadata_set_validation_flag(nir); \
2251 if (pass(nir, ##__VA_ARGS__)) { \
2253 nir_metadata_check_validation_flag(nir); \
2257 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2258 pass(nir, ##__VA_ARGS__); \
2261 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2262 void nir_calc_dominance(nir_shader
*shader
);
2264 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2265 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2267 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2268 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2270 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2271 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2273 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2274 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2276 int nir_gs_count_vertices(const nir_shader
*shader
);
2278 bool nir_split_var_copies(nir_shader
*shader
);
2280 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2281 bool nir_lower_returns(nir_shader
*shader
);
2283 bool nir_inline_functions(nir_shader
*shader
);
2285 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2286 void nir_lower_var_copies(nir_shader
*shader
);
2288 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2290 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2292 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2294 void nir_lower_io_to_temporaries(nir_shader
*shader
, nir_function
*entrypoint
,
2295 bool outputs
, bool inputs
);
2297 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2299 void nir_assign_var_locations(struct exec_list
*var_list
,
2301 int (*type_size
)(const struct glsl_type
*));
2303 void nir_lower_io(nir_shader
*shader
,
2304 nir_variable_mode modes
,
2305 int (*type_size
)(const struct glsl_type
*));
2306 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2307 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2309 void nir_lower_io_types(nir_shader
*shader
);
2310 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2312 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2314 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2315 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2316 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2317 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2319 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2321 void nir_lower_samplers(nir_shader
*shader
,
2322 const struct gl_shader_program
*shader_program
);
2324 bool nir_lower_system_values(nir_shader
*shader
);
2326 typedef struct nir_lower_tex_options
{
2328 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2329 * sampler types a texture projector is lowered.
2334 * If true, lower rect textures to 2D, using txs to fetch the
2335 * texture dimensions and dividing the texture coords by the
2336 * texture dims to normalize.
2341 * To emulate certain texture wrap modes, this can be used
2342 * to saturate the specified tex coord to [0.0, 1.0]. The
2343 * bits are according to sampler #, ie. if, for example:
2345 * (conf->saturate_s & (1 << n))
2347 * is true, then the s coord for sampler n is saturated.
2349 * Note that clamping must happen *after* projector lowering
2350 * so any projected texture sample instruction with a clamped
2351 * coordinate gets automatically lowered, regardless of the
2352 * 'lower_txp' setting.
2354 unsigned saturate_s
;
2355 unsigned saturate_t
;
2356 unsigned saturate_r
;
2358 /* Bitmask of textures that need swizzling.
2360 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2361 * swizzles[texture_index] is applied to the result of the texturing
2364 unsigned swizzle_result
;
2366 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2367 * while 4 and 5 represent 0 and 1 respectively.
2369 uint8_t swizzles
[32][4];
2372 * Bitmap of textures that need srgb to linear conversion. If
2373 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2374 * of the texture are lowered to linear.
2376 unsigned lower_srgb
;
2377 } nir_lower_tex_options
;
2379 bool nir_lower_tex(nir_shader
*shader
,
2380 const nir_lower_tex_options
*options
);
2382 bool nir_lower_idiv(nir_shader
*shader
);
2384 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2385 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2387 void nir_lower_two_sided_color(nir_shader
*shader
);
2389 void nir_lower_clamp_color_outputs(nir_shader
*shader
);
2391 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2393 typedef struct nir_lower_wpos_ytransform_options
{
2394 int state_tokens
[5];
2395 bool fs_coord_origin_upper_left
:1;
2396 bool fs_coord_origin_lower_left
:1;
2397 bool fs_coord_pixel_center_integer
:1;
2398 bool fs_coord_pixel_center_half_integer
:1;
2399 } nir_lower_wpos_ytransform_options
;
2401 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2402 const nir_lower_wpos_ytransform_options
*options
);
2404 typedef struct nir_lower_drawpixels_options
{
2405 int texcoord_state_tokens
[5];
2406 int scale_state_tokens
[5];
2407 int bias_state_tokens
[5];
2408 unsigned drawpix_sampler
;
2409 unsigned pixelmap_sampler
;
2411 bool scale_and_bias
:1;
2412 } nir_lower_drawpixels_options
;
2414 void nir_lower_drawpixels(nir_shader
*shader
,
2415 const nir_lower_drawpixels_options
*options
);
2417 typedef struct nir_lower_bitmap_options
{
2420 } nir_lower_bitmap_options
;
2422 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2424 void nir_lower_atomics(nir_shader
*shader
,
2425 const struct gl_shader_program
*shader_program
);
2426 void nir_lower_to_source_mods(nir_shader
*shader
);
2428 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2431 nir_lower_drcp
= (1 << 0),
2432 nir_lower_dsqrt
= (1 << 1),
2433 nir_lower_drsq
= (1 << 2),
2434 nir_lower_dtrunc
= (1 << 3),
2435 nir_lower_dfloor
= (1 << 4),
2436 nir_lower_dceil
= (1 << 5),
2437 nir_lower_dfract
= (1 << 6),
2438 nir_lower_dround_even
= (1 << 7),
2439 nir_lower_dmod
= (1 << 8)
2440 } nir_lower_doubles_options
;
2442 void nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2443 void nir_lower_double_pack(nir_shader
*shader
);
2445 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2447 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2448 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2450 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2451 void nir_convert_to_ssa(nir_shader
*shader
);
2453 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2454 bool nir_repair_ssa(nir_shader
*shader
);
2456 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2457 * registers. If false, convert all values (even those not involved in a phi
2458 * node) to registers.
2460 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2462 bool nir_opt_algebraic(nir_shader
*shader
);
2463 bool nir_opt_algebraic_late(nir_shader
*shader
);
2464 bool nir_opt_constant_folding(nir_shader
*shader
);
2466 bool nir_opt_global_to_local(nir_shader
*shader
);
2468 bool nir_copy_prop(nir_shader
*shader
);
2470 bool nir_opt_cse(nir_shader
*shader
);
2472 bool nir_opt_dce(nir_shader
*shader
);
2474 bool nir_opt_dead_cf(nir_shader
*shader
);
2476 void nir_opt_gcm(nir_shader
*shader
);
2478 bool nir_opt_peephole_select(nir_shader
*shader
);
2480 bool nir_opt_remove_phis(nir_shader
*shader
);
2482 bool nir_opt_undef(nir_shader
*shader
);
2484 void nir_sweep(nir_shader
*shader
);
2486 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2487 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);