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
90 nir_var_shader_storage
,
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
{
161 * Is the variable read-only?
163 * This is set for variables declared as \c const, shader inputs,
166 unsigned read_only
:1;
170 unsigned invariant
:1;
173 * Storage class of the variable.
175 * \sa nir_variable_mode
177 nir_variable_mode mode
:5;
180 * Interpolation mode for shader inputs / outputs
182 * \sa glsl_interp_qualifier
184 unsigned interpolation
:2;
187 * \name ARB_fragment_coord_conventions
190 unsigned origin_upper_left
:1;
191 unsigned pixel_center_integer
:1;
195 * Was the location explicitly set in the shader?
197 * If the location is explicitly set in the shader, it \b cannot be changed
198 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
201 unsigned explicit_location
:1;
202 unsigned explicit_index
:1;
205 * Was an initial binding explicitly set in the shader?
207 * If so, constant_initializer contains an integer nir_constant
208 * representing the initial binding point.
210 unsigned explicit_binding
:1;
213 * Does this variable have an initializer?
215 * This is used by the linker to cross-validiate initializers of global
218 unsigned has_initializer
:1;
221 * If non-zero, then this variable may be packed along with other variables
222 * into a single varying slot, so this offset should be applied when
223 * accessing components. For example, an offset of 1 means that the x
224 * component of this variable is actually stored in component y of the
225 * location specified by \c location.
227 unsigned location_frac
:2;
230 * \brief Layout qualifier for gl_FragDepth.
232 * This is not equal to \c ir_depth_layout_none if and only if this
233 * variable is \c gl_FragDepth and a layout qualifier is specified.
235 nir_depth_layout depth_layout
;
238 * Storage location of the base of this variable
240 * The precise meaning of this field depends on the nature of the variable.
242 * - Vertex shader input: one of the values from \c gl_vert_attrib.
243 * - Vertex shader output: one of the values from \c gl_varying_slot.
244 * - Geometry shader input: one of the values from \c gl_varying_slot.
245 * - Geometry shader output: one of the values from \c gl_varying_slot.
246 * - Fragment shader input: one of the values from \c gl_varying_slot.
247 * - Fragment shader output: one of the values from \c gl_frag_result.
248 * - Uniforms: Per-stage uniform slot number for default uniform block.
249 * - Uniforms: Index within the uniform block definition for UBO members.
250 * - Non-UBO Uniforms: uniform slot number.
251 * - Other: This field is not currently used.
253 * If the variable is a uniform, shader input, or shader output, and the
254 * slot has not been assigned, the value will be -1.
259 * The actual location of the variable in the IR. Only valid for inputs
262 unsigned int driver_location
;
265 * output index for dual source blending.
270 * Initial binding point for a sampler or UBO.
272 * For array types, this represents the binding point for the first element.
277 * Location an atomic counter is stored at.
282 * ARB_shader_image_load_store qualifiers.
285 bool read_only
; /**< "readonly" qualifier. */
286 bool write_only
; /**< "writeonly" qualifier. */
291 /** Image internal format if specified explicitly, otherwise GL_NONE. */
296 * Highest element accessed with a constant expression array index
298 * Not used for non-array variables.
300 unsigned max_array_access
;
305 * Built-in state that backs this uniform
307 * Once set at variable creation, \c state_slots must remain invariant.
308 * This is because, ideally, this array would be shared by all clones of
309 * this variable in the IR tree. In other words, we'd really like for it
310 * to be a fly-weight.
312 * If the variable is not a uniform, \c num_state_slots will be zero and
313 * \c state_slots will be \c NULL.
316 unsigned num_state_slots
; /**< Number of state slots used */
317 nir_state_slot
*state_slots
; /**< State descriptors. */
321 * Constant expression assigned in the initializer of the variable
323 nir_constant
*constant_initializer
;
326 * For variables that are in an interface block or are an instance of an
327 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
329 * \sa ir_variable::location
331 const struct glsl_type
*interface_type
;
334 #define nir_foreach_variable(var, var_list) \
335 foreach_list_typed(nir_variable, var, node, var_list)
337 #define nir_foreach_variable_safe(var, var_list) \
338 foreach_list_typed_safe(nir_variable, var, node, var_list)
341 nir_variable_is_global(const nir_variable
*var
)
343 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
346 typedef struct nir_register
{
347 struct exec_node node
;
349 unsigned num_components
; /** < number of vector components */
350 unsigned num_array_elems
; /** < size of array (0 for no array) */
352 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
355 /** generic register index. */
358 /** only for debug purposes, can be NULL */
361 /** whether this register is local (per-function) or global (per-shader) */
365 * If this flag is set to true, then accessing channels >= num_components
366 * is well-defined, and simply spills over to the next array element. This
367 * is useful for backends that can do per-component accessing, in
368 * particular scalar backends. By setting this flag and making
369 * num_components equal to 1, structures can be packed tightly into
370 * registers and then registers can be accessed per-component to get to
371 * each structure member, even if it crosses vec4 boundaries.
375 /** set of nir_src's where this register is used (read from) */
376 struct list_head uses
;
378 /** set of nir_dest's where this register is defined (written to) */
379 struct list_head defs
;
381 /** set of nir_if's where this register is used as a condition */
382 struct list_head if_uses
;
389 nir_instr_type_intrinsic
,
390 nir_instr_type_load_const
,
392 nir_instr_type_ssa_undef
,
394 nir_instr_type_parallel_copy
,
397 typedef struct nir_instr
{
398 struct exec_node node
;
400 struct nir_block
*block
;
402 /** generic instruction index. */
405 /* A temporary for optimization and analysis passes to use for storing
406 * flags. For instance, DCE uses this to store the "dead/live" info.
411 static inline nir_instr
*
412 nir_instr_next(nir_instr
*instr
)
414 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
415 if (exec_node_is_tail_sentinel(next
))
418 return exec_node_data(nir_instr
, next
, node
);
421 static inline nir_instr
*
422 nir_instr_prev(nir_instr
*instr
)
424 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
425 if (exec_node_is_head_sentinel(prev
))
428 return exec_node_data(nir_instr
, prev
, node
);
432 nir_instr_is_first(nir_instr
*instr
)
434 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
438 nir_instr_is_last(nir_instr
*instr
)
440 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
443 typedef struct nir_ssa_def
{
444 /** for debugging only, can be NULL */
447 /** generic SSA definition index. */
450 /** Index into the live_in and live_out bitfields */
453 nir_instr
*parent_instr
;
455 /** set of nir_instr's where this register is used (read from) */
456 struct list_head uses
;
458 /** set of nir_if's where this register is used as a condition */
459 struct list_head if_uses
;
461 uint8_t num_components
;
463 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
471 struct nir_src
*indirect
; /** < NULL for no indirect offset */
472 unsigned base_offset
;
474 /* TODO use-def chain goes here */
478 nir_instr
*parent_instr
;
479 struct list_head def_link
;
482 struct nir_src
*indirect
; /** < NULL for no indirect offset */
483 unsigned base_offset
;
485 /* TODO def-use chain goes here */
490 typedef struct nir_src
{
492 nir_instr
*parent_instr
;
493 struct nir_if
*parent_if
;
496 struct list_head use_link
;
506 #define NIR_SRC_INIT (nir_src) { { NULL } }
508 #define nir_foreach_use(reg_or_ssa_def, src) \
509 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
511 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
512 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
514 #define nir_foreach_if_use(reg_or_ssa_def, src) \
515 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
517 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
518 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
529 #define NIR_DEST_INIT (nir_dest) { { { NULL } } }
531 #define nir_foreach_def(reg, dest) \
532 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
534 #define nir_foreach_def_safe(reg, dest) \
535 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
537 static inline nir_src
538 nir_src_for_ssa(nir_ssa_def
*def
)
540 nir_src src
= NIR_SRC_INIT
;
548 static inline nir_src
549 nir_src_for_reg(nir_register
*reg
)
551 nir_src src
= NIR_SRC_INIT
;
555 src
.reg
.indirect
= NULL
;
556 src
.reg
.base_offset
= 0;
561 static inline nir_dest
562 nir_dest_for_reg(nir_register
*reg
)
564 nir_dest dest
= NIR_DEST_INIT
;
571 static inline unsigned
572 nir_src_bit_size(nir_src src
)
574 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
577 static inline unsigned
578 nir_dest_bit_size(nir_dest dest
)
580 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
583 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
584 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
590 * \name input modifiers
594 * For inputs interpreted as floating point, flips the sign bit. For
595 * inputs interpreted as integers, performs the two's complement negation.
600 * Clears the sign bit for floating point values, and computes the integer
601 * absolute value for integers. Note that the negate modifier acts after
602 * the absolute value modifier, therefore if both are set then all inputs
603 * will become negative.
609 * For each input component, says which component of the register it is
610 * chosen from. Note that which elements of the swizzle are used and which
611 * are ignored are based on the write mask for most opcodes - for example,
612 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
613 * a swizzle of {2, x, 1, 0} where x means "don't care."
622 * \name saturate output modifier
624 * Only valid for opcodes that output floating-point numbers. Clamps the
625 * output to between 0.0 and 1.0 inclusive.
630 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
634 nir_type_invalid
= 0, /* Not a valid type */
639 nir_type_bool32
= 32 | nir_type_bool
,
640 nir_type_int8
= 8 | nir_type_int
,
641 nir_type_int16
= 16 | nir_type_int
,
642 nir_type_int32
= 32 | nir_type_int
,
643 nir_type_int64
= 64 | nir_type_int
,
644 nir_type_uint8
= 8 | nir_type_uint
,
645 nir_type_uint16
= 16 | nir_type_uint
,
646 nir_type_uint32
= 32 | nir_type_uint
,
647 nir_type_uint64
= 64 | nir_type_uint
,
648 nir_type_float16
= 16 | nir_type_float
,
649 nir_type_float32
= 32 | nir_type_float
,
650 nir_type_float64
= 64 | nir_type_float
,
653 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
654 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
656 static inline unsigned
657 nir_alu_type_get_type_size(nir_alu_type type
)
659 return type
& NIR_ALU_TYPE_SIZE_MASK
;
662 static inline unsigned
663 nir_alu_type_get_base_type(nir_alu_type type
)
665 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
669 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
670 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
671 } nir_op_algebraic_property
;
679 * The number of components in the output
681 * If non-zero, this is the size of the output and input sizes are
682 * explicitly given; swizzle and writemask are still in effect, but if
683 * the output component is masked out, then the input component may
686 * If zero, the opcode acts in the standard, per-component manner; the
687 * operation is performed on each component (except the ones that are
688 * masked out) with the input being taken from the input swizzle for
691 * The size of some of the inputs may be given (i.e. non-zero) even
692 * though output_size is zero; in that case, the inputs with a zero
693 * size act per-component, while the inputs with non-zero size don't.
695 unsigned output_size
;
698 * The type of vector that the instruction outputs. Note that the
699 * staurate modifier is only allowed on outputs with the float type.
702 nir_alu_type output_type
;
705 * The number of components in each input
707 unsigned input_sizes
[4];
710 * The type of vector that each input takes. Note that negate and
711 * absolute value are only allowed on inputs with int or float type and
712 * behave differently on the two.
714 nir_alu_type input_types
[4];
716 nir_op_algebraic_property algebraic_properties
;
719 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
721 typedef struct nir_alu_instr
{
725 /** Indicates that this ALU instruction generates an exact value
727 * This is kind of a mixture of GLSL "precise" and "invariant" and not
728 * really equivalent to either. This indicates that the value generated by
729 * this operation is high-precision and any code transformations that touch
730 * it must ensure that the resulting value is bit-for-bit identical to the
739 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
740 nir_alu_instr
*instr
);
741 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
742 nir_alu_instr
*instr
);
744 /* is this source channel used? */
746 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
748 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
749 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
751 return (instr
->dest
.write_mask
>> channel
) & 1;
755 * For instructions whose destinations are SSA, get the number of channels
758 static inline unsigned
759 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
761 assert(instr
->dest
.dest
.is_ssa
);
763 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
764 return nir_op_infos
[instr
->op
].input_sizes
[src
];
766 return instr
->dest
.dest
.ssa
.num_components
;
771 nir_deref_type_array
,
772 nir_deref_type_struct
775 typedef struct nir_deref
{
776 nir_deref_type deref_type
;
777 struct nir_deref
*child
;
778 const struct glsl_type
*type
;
787 /* This enum describes how the array is referenced. If the deref is
788 * direct then the base_offset is used. If the deref is indirect then then
789 * offset is given by base_offset + indirect. If the deref is a wildcard
790 * then the deref refers to all of the elements of the array at the same
791 * time. Wildcard dereferences are only ever allowed in copy_var
792 * intrinsics and the source and destination derefs must have matching
796 nir_deref_array_type_direct
,
797 nir_deref_array_type_indirect
,
798 nir_deref_array_type_wildcard
,
799 } nir_deref_array_type
;
804 nir_deref_array_type deref_array_type
;
805 unsigned base_offset
;
815 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
816 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
817 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
819 /* Returns the last deref in the chain. */
820 static inline nir_deref
*
821 nir_deref_tail(nir_deref
*deref
)
824 deref
= deref
->child
;
832 nir_deref_var
**params
;
833 nir_deref_var
*return_deref
;
835 struct nir_function
*callee
;
838 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
839 num_variables, num_indices, idx0, idx1, idx2, flags) \
840 nir_intrinsic_##name,
842 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
845 #include "nir_intrinsics.h"
846 nir_num_intrinsics
= nir_last_intrinsic
+ 1
850 #undef LAST_INTRINSIC
852 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
854 /** Represents an intrinsic
856 * An intrinsic is an instruction type for handling things that are
857 * more-or-less regular operations but don't just consume and produce SSA
858 * values like ALU operations do. Intrinsics are not for things that have
859 * special semantic meaning such as phi nodes and parallel copies.
860 * Examples of intrinsics include variable load/store operations, system
861 * value loads, and the like. Even though texturing more-or-less falls
862 * under this category, texturing is its own instruction type because
863 * trying to represent texturing with intrinsics would lead to a
864 * combinatorial explosion of intrinsic opcodes.
866 * By having a single instruction type for handling a lot of different
867 * cases, optimization passes can look for intrinsics and, for the most
868 * part, completely ignore them. Each intrinsic type also has a few
869 * possible flags that govern whether or not they can be reordered or
870 * eliminated. That way passes like dead code elimination can still work
871 * on intrisics without understanding the meaning of each.
873 * Each intrinsic has some number of constant indices, some number of
874 * variables, and some number of sources. What these sources, variables,
875 * and indices mean depends on the intrinsic and is documented with the
876 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
877 * instructions are the only types of instruction that can operate on
883 nir_intrinsic_op intrinsic
;
887 /** number of components if this is a vectorized intrinsic
889 * Similarly to ALU operations, some intrinsics are vectorized.
890 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
891 * For vectorized intrinsics, the num_components field specifies the
892 * number of destination components and the number of source components
893 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
895 uint8_t num_components
;
897 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
899 nir_deref_var
*variables
[2];
902 } nir_intrinsic_instr
;
905 * \name NIR intrinsics semantic flags
907 * information about what the compiler can do with the intrinsics.
909 * \sa nir_intrinsic_info::flags
913 * whether the intrinsic can be safely eliminated if none of its output
914 * value is not being used.
916 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
919 * Whether the intrinsic can be reordered with respect to any other
920 * intrinsic, i.e. whether the only reordering dependencies of the
921 * intrinsic are due to the register reads/writes.
923 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
924 } nir_intrinsic_semantic_flag
;
927 * \name NIR intrinsics const-index flag
929 * Indicates the usage of a const_index slot.
931 * \sa nir_intrinsic_info::index_map
935 * Generally instructions that take a offset src argument, can encode
936 * a constant 'base' value which is added to the offset.
938 NIR_INTRINSIC_BASE
= 1,
941 * For store instructions, a writemask for the store.
943 NIR_INTRINSIC_WRMASK
= 2,
946 * The stream-id for GS emit_vertex/end_primitive intrinsics.
948 NIR_INTRINSIC_STREAM_ID
= 3,
951 * The clip-plane id for load_user_clip_plane intrinsic.
953 NIR_INTRINSIC_UCP_ID
= 4,
955 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
957 } nir_intrinsic_index_flag
;
959 #define NIR_INTRINSIC_MAX_INPUTS 4
964 unsigned num_srcs
; /** < number of register/SSA inputs */
966 /** number of components of each input register
968 * If this value is 0, the number of components is given by the
969 * num_components field of nir_intrinsic_instr.
971 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
975 /** number of components of the output register
977 * If this value is 0, the number of components is given by the
978 * num_components field of nir_intrinsic_instr.
980 unsigned dest_components
;
982 /** the number of inputs/outputs that are variables */
983 unsigned num_variables
;
985 /** the number of constant indices used by the intrinsic */
986 unsigned num_indices
;
988 /** indicates the usage of intr->const_index[n] */
989 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
991 /** semantic flags for calls to this intrinsic */
992 nir_intrinsic_semantic_flag flags
;
993 } nir_intrinsic_info
;
995 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
998 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1000 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
1002 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1003 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1004 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1006 static inline void \
1007 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1009 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1010 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1011 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1014 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1015 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1016 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1017 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1020 * \group texture information
1022 * This gives semantic information about textures which is useful to the
1023 * frontend, the backend, and lowering passes, but not the optimizer.
1028 nir_tex_src_projector
,
1029 nir_tex_src_comparitor
, /* shadow comparitor */
1033 nir_tex_src_ms_index
, /* MSAA sample index */
1036 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1037 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1038 nir_num_tex_src_types
1043 nir_tex_src_type src_type
;
1047 nir_texop_tex
, /**< Regular texture look-up */
1048 nir_texop_txb
, /**< Texture look-up with LOD bias */
1049 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1050 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1051 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1052 nir_texop_txf_ms
, /**< Multisample texture fetch */
1053 nir_texop_txs
, /**< Texture size */
1054 nir_texop_lod
, /**< Texture lod query */
1055 nir_texop_tg4
, /**< Texture gather */
1056 nir_texop_query_levels
, /**< Texture levels query */
1057 nir_texop_texture_samples
, /**< Texture samples query */
1058 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1066 enum glsl_sampler_dim sampler_dim
;
1067 nir_alu_type dest_type
;
1072 unsigned num_srcs
, coord_components
;
1073 bool is_array
, is_shadow
;
1076 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1077 * components or the new-style shadow that outputs 1 component.
1079 bool is_new_style_shadow
;
1081 /* gather component selector */
1082 unsigned component
: 2;
1084 /** The texture index
1086 * If this texture instruction has a nir_tex_src_texture_offset source,
1087 * then the texture index is given by texture_index + texture_offset.
1089 unsigned texture_index
;
1091 /** The size of the texture array or 0 if it's not an array */
1092 unsigned texture_array_size
;
1094 /** The texture deref
1096 * If this is null, use texture_index instead.
1098 nir_deref_var
*texture
;
1100 /** The sampler index
1102 * The following operations do not require a sampler and, as such, this
1103 * field should be ignored:
1105 * - nir_texop_txf_ms
1109 * - nir_texop_query_levels
1110 * - nir_texop_texture_samples
1111 * - nir_texop_samples_identical
1113 * If this texture instruction has a nir_tex_src_sampler_offset source,
1114 * then the sampler index is given by sampler_index + sampler_offset.
1116 unsigned sampler_index
;
1118 /** The sampler deref
1120 * If this is null, use sampler_index instead.
1122 nir_deref_var
*sampler
;
1125 static inline unsigned
1126 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1128 switch (instr
->op
) {
1129 case nir_texop_txs
: {
1131 switch (instr
->sampler_dim
) {
1132 case GLSL_SAMPLER_DIM_1D
:
1133 case GLSL_SAMPLER_DIM_BUF
:
1136 case GLSL_SAMPLER_DIM_2D
:
1137 case GLSL_SAMPLER_DIM_CUBE
:
1138 case GLSL_SAMPLER_DIM_MS
:
1139 case GLSL_SAMPLER_DIM_RECT
:
1140 case GLSL_SAMPLER_DIM_EXTERNAL
:
1143 case GLSL_SAMPLER_DIM_3D
:
1147 unreachable("not reached");
1149 if (instr
->is_array
)
1157 case nir_texop_texture_samples
:
1158 case nir_texop_query_levels
:
1159 case nir_texop_samples_identical
:
1163 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1170 /* Returns true if this texture operation queries something about the texture
1171 * rather than actually sampling it.
1174 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1176 switch (instr
->op
) {
1179 case nir_texop_texture_samples
:
1180 case nir_texop_query_levels
:
1187 case nir_texop_txf_ms
:
1191 unreachable("Invalid texture opcode");
1195 static inline unsigned
1196 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1198 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1199 return instr
->coord_components
;
1202 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1203 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1204 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1205 if (instr
->is_array
)
1206 return instr
->coord_components
- 1;
1208 return instr
->coord_components
;
1215 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1217 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1218 if (instr
->src
[i
].src_type
== type
)
1238 nir_const_value value
;
1241 } nir_load_const_instr
;
1254 /* creates a new SSA variable in an undefined state */
1259 } nir_ssa_undef_instr
;
1262 struct exec_node node
;
1264 /* The predecessor block corresponding to this source */
1265 struct nir_block
*pred
;
1270 #define nir_foreach_phi_src(phi, entry) \
1271 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1272 #define nir_foreach_phi_src_safe(phi, entry) \
1273 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1278 struct exec_list srcs
; /** < list of nir_phi_src */
1284 struct exec_node node
;
1287 } nir_parallel_copy_entry
;
1289 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1290 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1295 /* A list of nir_parallel_copy_entry's. The sources of all of the
1296 * entries are copied to the corresponding destinations "in parallel".
1297 * In other words, if we have two entries: a -> b and b -> a, the values
1300 struct exec_list entries
;
1301 } nir_parallel_copy_instr
;
1303 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1304 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1305 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1306 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1307 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1308 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1309 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1310 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1311 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1312 nir_parallel_copy_instr
, instr
)
1317 * Control flow consists of a tree of control flow nodes, which include
1318 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1319 * instructions that always run start-to-finish. Each basic block also keeps
1320 * track of its successors (blocks which may run immediately after the current
1321 * block) and predecessors (blocks which could have run immediately before the
1322 * current block). Each function also has a start block and an end block which
1323 * all return statements point to (which is always empty). Together, all the
1324 * blocks with their predecessors and successors make up the control flow
1325 * graph (CFG) of the function. There are helpers that modify the tree of
1326 * control flow nodes while modifying the CFG appropriately; these should be
1327 * used instead of modifying the tree directly.
1334 nir_cf_node_function
1337 typedef struct nir_cf_node
{
1338 struct exec_node node
;
1339 nir_cf_node_type type
;
1340 struct nir_cf_node
*parent
;
1343 typedef struct nir_block
{
1344 nir_cf_node cf_node
;
1346 struct exec_list instr_list
; /** < list of nir_instr */
1348 /** generic block index; generated by nir_index_blocks */
1352 * Each block can only have up to 2 successors, so we put them in a simple
1353 * array - no need for anything more complicated.
1355 struct nir_block
*successors
[2];
1357 /* Set of nir_block predecessors in the CFG */
1358 struct set
*predecessors
;
1361 * this node's immediate dominator in the dominance tree - set to NULL for
1364 struct nir_block
*imm_dom
;
1366 /* This node's children in the dominance tree */
1367 unsigned num_dom_children
;
1368 struct nir_block
**dom_children
;
1370 /* Set of nir_block's on the dominance frontier of this block */
1371 struct set
*dom_frontier
;
1374 * These two indices have the property that dom_{pre,post}_index for each
1375 * child of this block in the dominance tree will always be between
1376 * dom_pre_index and dom_post_index for this block, which makes testing if
1377 * a given block is dominated by another block an O(1) operation.
1379 unsigned dom_pre_index
, dom_post_index
;
1381 /* live in and out for this block; used for liveness analysis */
1382 BITSET_WORD
*live_in
;
1383 BITSET_WORD
*live_out
;
1386 static inline nir_instr
*
1387 nir_block_first_instr(nir_block
*block
)
1389 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1390 return exec_node_data(nir_instr
, head
, node
);
1393 static inline nir_instr
*
1394 nir_block_last_instr(nir_block
*block
)
1396 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1397 return exec_node_data(nir_instr
, tail
, node
);
1400 #define nir_foreach_instr(block, instr) \
1401 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1402 #define nir_foreach_instr_reverse(block, instr) \
1403 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1404 #define nir_foreach_instr_safe(block, instr) \
1405 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1406 #define nir_foreach_instr_reverse_safe(block, instr) \
1407 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1409 typedef struct nir_if
{
1410 nir_cf_node cf_node
;
1413 struct exec_list then_list
; /** < list of nir_cf_node */
1414 struct exec_list else_list
; /** < list of nir_cf_node */
1417 static inline nir_cf_node
*
1418 nir_if_first_then_node(nir_if
*if_stmt
)
1420 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1421 return exec_node_data(nir_cf_node
, head
, node
);
1424 static inline nir_cf_node
*
1425 nir_if_last_then_node(nir_if
*if_stmt
)
1427 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1428 return exec_node_data(nir_cf_node
, tail
, node
);
1431 static inline nir_cf_node
*
1432 nir_if_first_else_node(nir_if
*if_stmt
)
1434 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1435 return exec_node_data(nir_cf_node
, head
, node
);
1438 static inline nir_cf_node
*
1439 nir_if_last_else_node(nir_if
*if_stmt
)
1441 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1442 return exec_node_data(nir_cf_node
, tail
, node
);
1446 nir_cf_node cf_node
;
1448 struct exec_list body
; /** < list of nir_cf_node */
1451 static inline nir_cf_node
*
1452 nir_loop_first_cf_node(nir_loop
*loop
)
1454 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1457 static inline nir_cf_node
*
1458 nir_loop_last_cf_node(nir_loop
*loop
)
1460 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1464 * Various bits of metadata that can may be created or required by
1465 * optimization and analysis passes
1468 nir_metadata_none
= 0x0,
1469 nir_metadata_block_index
= 0x1,
1470 nir_metadata_dominance
= 0x2,
1471 nir_metadata_live_ssa_defs
= 0x4,
1472 nir_metadata_not_properly_reset
= 0x8,
1476 nir_cf_node cf_node
;
1478 /** pointer to the function of which this is an implementation */
1479 struct nir_function
*function
;
1481 struct exec_list body
; /** < list of nir_cf_node */
1483 nir_block
*end_block
;
1485 /** list for all local variables in the function */
1486 struct exec_list locals
;
1488 /** array of variables used as parameters */
1489 unsigned num_params
;
1490 nir_variable
**params
;
1492 /** variable used to hold the result of the function */
1493 nir_variable
*return_var
;
1495 /** list of local registers in the function */
1496 struct exec_list registers
;
1498 /** next available local register index */
1501 /** next available SSA value index */
1504 /* total number of basic blocks, only valid when block_index_dirty = false */
1505 unsigned num_blocks
;
1507 nir_metadata valid_metadata
;
1508 } nir_function_impl
;
1510 static inline nir_block
*
1511 nir_start_block(nir_function_impl
*impl
)
1513 return (nir_block
*) exec_list_get_head(&impl
->body
);
1516 static inline nir_cf_node
*
1517 nir_cf_node_next(nir_cf_node
*node
)
1519 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1520 if (exec_node_is_tail_sentinel(next
))
1523 return exec_node_data(nir_cf_node
, next
, node
);
1526 static inline nir_cf_node
*
1527 nir_cf_node_prev(nir_cf_node
*node
)
1529 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1530 if (exec_node_is_head_sentinel(prev
))
1533 return exec_node_data(nir_cf_node
, prev
, node
);
1537 nir_cf_node_is_first(const nir_cf_node
*node
)
1539 return exec_node_is_head_sentinel(node
->node
.prev
);
1543 nir_cf_node_is_last(const nir_cf_node
*node
)
1545 return exec_node_is_tail_sentinel(node
->node
.next
);
1548 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1549 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1550 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1551 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1556 nir_parameter_inout
,
1557 } nir_parameter_type
;
1560 nir_parameter_type param_type
;
1561 const struct glsl_type
*type
;
1564 typedef struct nir_function
{
1565 struct exec_node node
;
1568 struct nir_shader
*shader
;
1570 unsigned num_params
;
1571 nir_parameter
*params
;
1572 const struct glsl_type
*return_type
;
1574 /** The implementation of this function.
1576 * If the function is only declared and not implemented, this is NULL.
1578 nir_function_impl
*impl
;
1581 typedef struct nir_shader_compiler_options
{
1589 bool lower_bitfield_extract
;
1590 bool lower_bitfield_insert
;
1591 bool lower_uadd_carry
;
1592 bool lower_usub_borrow
;
1593 /** lowers fneg and ineg to fsub and isub. */
1595 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1598 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1601 /* Does the native fdot instruction replicate its result for four
1602 * components? If so, then opt_algebraic_late will turn all fdotN
1603 * instructions into fdot_replicatedN instructions.
1605 bool fdot_replicates
;
1607 /** lowers ffract to fsub+ffloor: */
1610 bool lower_pack_half_2x16
;
1611 bool lower_pack_unorm_2x16
;
1612 bool lower_pack_snorm_2x16
;
1613 bool lower_pack_unorm_4x8
;
1614 bool lower_pack_snorm_4x8
;
1615 bool lower_unpack_half_2x16
;
1616 bool lower_unpack_unorm_2x16
;
1617 bool lower_unpack_snorm_2x16
;
1618 bool lower_unpack_unorm_4x8
;
1619 bool lower_unpack_snorm_4x8
;
1621 bool lower_extract_byte
;
1622 bool lower_extract_word
;
1625 * Does the driver support real 32-bit integers? (Otherwise, integers
1626 * are simulated by floats.)
1628 bool native_integers
;
1629 } nir_shader_compiler_options
;
1631 typedef struct nir_shader_info
{
1634 /* Descriptive name provided by the client; may be NULL */
1637 /* Number of textures used by this shader */
1638 unsigned num_textures
;
1639 /* Number of uniform buffers used by this shader */
1641 /* Number of atomic buffers used by this shader */
1643 /* Number of shader storage buffers used by this shader */
1645 /* Number of images used by this shader */
1646 unsigned num_images
;
1648 /* Which inputs are actually read */
1649 uint64_t inputs_read
;
1650 /* Which outputs are actually written */
1651 uint64_t outputs_written
;
1652 /* Which system values are actually read */
1653 uint64_t system_values_read
;
1655 /* Which patch inputs are actually read */
1656 uint32_t patch_inputs_read
;
1657 /* Which patch outputs are actually written */
1658 uint32_t patch_outputs_written
;
1660 /* Whether or not this shader ever uses textureGather() */
1661 bool uses_texture_gather
;
1663 /* Whether or not this shader uses the gl_ClipDistance output */
1664 bool uses_clip_distance_out
;
1666 /* Whether or not separate shader objects were used */
1667 bool separate_shader
;
1669 /** Was this shader linked with any transform feedback varyings? */
1670 bool has_transform_feedback_varyings
;
1674 /** The number of vertices recieves per input primitive */
1675 unsigned vertices_in
;
1677 /** The output primitive type (GL enum value) */
1678 unsigned output_primitive
;
1680 /** The maximum number of vertices the geometry shader might write. */
1681 unsigned vertices_out
;
1683 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1684 unsigned invocations
;
1686 /** Whether or not this shader uses EndPrimitive */
1687 bool uses_end_primitive
;
1689 /** Whether or not this shader uses non-zero streams */
1697 * Whether early fragment tests are enabled as defined by
1698 * ARB_shader_image_load_store.
1700 bool early_fragment_tests
;
1702 /** gl_FragDepth layout for ARB_conservative_depth. */
1703 enum gl_frag_depth_layout depth_layout
;
1707 unsigned local_size
[3];
1711 /** The number of vertices in the TCS output patch. */
1712 unsigned vertices_out
;
1717 typedef struct nir_shader
{
1718 /** list of uniforms (nir_variable) */
1719 struct exec_list uniforms
;
1721 /** list of inputs (nir_variable) */
1722 struct exec_list inputs
;
1724 /** list of outputs (nir_variable) */
1725 struct exec_list outputs
;
1727 /** list of shared compute variables (nir_variable) */
1728 struct exec_list shared
;
1730 /** Set of driver-specific options for the shader.
1732 * The memory for the options is expected to be kept in a single static
1733 * copy by the driver.
1735 const struct nir_shader_compiler_options
*options
;
1737 /** Various bits of compile-time information about a given shader */
1738 struct nir_shader_info info
;
1740 /** list of global variables in the shader (nir_variable) */
1741 struct exec_list globals
;
1743 /** list of system value variables in the shader (nir_variable) */
1744 struct exec_list system_values
;
1746 struct exec_list functions
; /** < list of nir_function */
1748 /** list of global register in the shader */
1749 struct exec_list registers
;
1751 /** next available global register index */
1755 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1758 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1760 /** The shader stage, such as MESA_SHADER_VERTEX. */
1761 gl_shader_stage stage
;
1764 static inline nir_function
*
1765 nir_shader_get_entrypoint(nir_shader
*shader
)
1767 assert(exec_list_length(&shader
->functions
) == 1);
1768 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1769 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1770 assert(func
->return_type
== glsl_void_type());
1771 assert(func
->num_params
== 0);
1775 #define nir_foreach_function(shader, func) \
1776 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1778 nir_shader
*nir_shader_create(void *mem_ctx
,
1779 gl_shader_stage stage
,
1780 const nir_shader_compiler_options
*options
);
1782 /** creates a register, including assigning it an index and adding it to the list */
1783 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1785 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1787 void nir_reg_remove(nir_register
*reg
);
1789 /** Adds a variable to the appropreate list in nir_shader */
1790 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1793 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1795 assert(var
->data
.mode
== nir_var_local
);
1796 exec_list_push_tail(&impl
->locals
, &var
->node
);
1799 /** creates a variable, sets a few defaults, and adds it to the list */
1800 nir_variable
*nir_variable_create(nir_shader
*shader
,
1801 nir_variable_mode mode
,
1802 const struct glsl_type
*type
,
1804 /** creates a local variable and adds it to the list */
1805 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1806 const struct glsl_type
*type
,
1809 /** creates a function and adds it to the shader's list of functions */
1810 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1812 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1813 /** creates a function_impl that isn't tied to any particular function */
1814 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1816 nir_block
*nir_block_create(nir_shader
*shader
);
1817 nir_if
*nir_if_create(nir_shader
*shader
);
1818 nir_loop
*nir_loop_create(nir_shader
*shader
);
1820 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1822 /** requests that the given pieces of metadata be generated */
1823 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1824 /** dirties all but the preserved metadata */
1825 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1827 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1828 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1830 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1832 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1833 unsigned num_components
);
1835 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1836 nir_intrinsic_op op
);
1838 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1839 nir_function
*callee
);
1841 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1843 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1845 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1847 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1848 unsigned num_components
);
1850 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1851 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1852 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1854 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1856 nir_load_const_instr
*
1857 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1860 * NIR Cursors and Instruction Insertion API
1863 * A tiny struct representing a point to insert/extract instructions or
1864 * control flow nodes. Helps reduce the combinatorial explosion of possible
1865 * points to insert/extract.
1867 * \sa nir_control_flow.h
1870 nir_cursor_before_block
,
1871 nir_cursor_after_block
,
1872 nir_cursor_before_instr
,
1873 nir_cursor_after_instr
,
1874 } nir_cursor_option
;
1877 nir_cursor_option option
;
1884 static inline nir_block
*
1885 nir_cursor_current_block(nir_cursor cursor
)
1887 if (cursor
.option
== nir_cursor_before_instr
||
1888 cursor
.option
== nir_cursor_after_instr
) {
1889 return cursor
.instr
->block
;
1891 return cursor
.block
;
1895 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
1897 static inline nir_cursor
1898 nir_before_block(nir_block
*block
)
1901 cursor
.option
= nir_cursor_before_block
;
1902 cursor
.block
= block
;
1906 static inline nir_cursor
1907 nir_after_block(nir_block
*block
)
1910 cursor
.option
= nir_cursor_after_block
;
1911 cursor
.block
= block
;
1915 static inline nir_cursor
1916 nir_before_instr(nir_instr
*instr
)
1919 cursor
.option
= nir_cursor_before_instr
;
1920 cursor
.instr
= instr
;
1924 static inline nir_cursor
1925 nir_after_instr(nir_instr
*instr
)
1928 cursor
.option
= nir_cursor_after_instr
;
1929 cursor
.instr
= instr
;
1933 static inline nir_cursor
1934 nir_after_block_before_jump(nir_block
*block
)
1936 nir_instr
*last_instr
= nir_block_last_instr(block
);
1937 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1938 return nir_before_instr(last_instr
);
1940 return nir_after_block(block
);
1944 static inline nir_cursor
1945 nir_before_cf_node(nir_cf_node
*node
)
1947 if (node
->type
== nir_cf_node_block
)
1948 return nir_before_block(nir_cf_node_as_block(node
));
1950 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1953 static inline nir_cursor
1954 nir_after_cf_node(nir_cf_node
*node
)
1956 if (node
->type
== nir_cf_node_block
)
1957 return nir_after_block(nir_cf_node_as_block(node
));
1959 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1962 static inline nir_cursor
1963 nir_after_cf_node_and_phis(nir_cf_node
*node
)
1965 if (node
->type
== nir_cf_node_block
)
1966 return nir_after_block(nir_cf_node_as_block(node
));
1968 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
1969 assert(block
->cf_node
.type
== nir_cf_node_block
);
1971 nir_foreach_instr(block
, instr
) {
1972 if (instr
->type
!= nir_instr_type_phi
)
1973 return nir_before_instr(instr
);
1975 return nir_after_block(block
);
1978 static inline nir_cursor
1979 nir_before_cf_list(struct exec_list
*cf_list
)
1981 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1982 exec_list_get_head(cf_list
), node
);
1983 return nir_before_cf_node(first_node
);
1986 static inline nir_cursor
1987 nir_after_cf_list(struct exec_list
*cf_list
)
1989 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1990 exec_list_get_tail(cf_list
), node
);
1991 return nir_after_cf_node(last_node
);
1995 * Insert a NIR instruction at the given cursor.
1997 * Note: This does not update the cursor.
1999 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2002 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2004 nir_instr_insert(nir_before_instr(instr
), before
);
2008 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2010 nir_instr_insert(nir_after_instr(instr
), after
);
2014 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2016 nir_instr_insert(nir_before_block(block
), before
);
2020 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2022 nir_instr_insert(nir_after_block(block
), after
);
2026 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2028 nir_instr_insert(nir_before_cf_node(node
), before
);
2032 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2034 nir_instr_insert(nir_after_cf_node(node
), after
);
2038 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2040 nir_instr_insert(nir_before_cf_list(list
), before
);
2044 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2046 nir_instr_insert(nir_after_cf_list(list
), after
);
2049 void nir_instr_remove(nir_instr
*instr
);
2053 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2054 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2055 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2056 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2058 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2059 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2061 nir_const_value
*nir_src_as_const_value(nir_src src
);
2062 bool nir_src_is_dynamically_uniform(nir_src src
);
2063 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2064 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2065 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2066 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2067 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2070 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2071 unsigned num_components
, unsigned bit_size
,
2073 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2074 unsigned num_components
, unsigned bit_size
,
2076 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2077 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2078 nir_instr
*after_me
);
2080 /* visits basic blocks in source-code order */
2081 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
2082 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
2084 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
2086 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
2089 /* If the following CF node is an if, this function returns that if.
2090 * Otherwise, it returns NULL.
2092 nir_if
*nir_block_get_following_if(nir_block
*block
);
2094 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2096 void nir_index_local_regs(nir_function_impl
*impl
);
2097 void nir_index_global_regs(nir_shader
*shader
);
2098 void nir_index_ssa_defs(nir_function_impl
*impl
);
2099 unsigned nir_index_instrs(nir_function_impl
*impl
);
2101 void nir_index_blocks(nir_function_impl
*impl
);
2103 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2104 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2106 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2107 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2108 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2109 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2112 void nir_validate_shader(nir_shader
*shader
);
2113 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2114 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2116 #include "util/debug.h"
2118 should_clone_nir(void)
2120 static int should_clone
= -1;
2121 if (should_clone
< 0)
2122 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2124 return should_clone
;
2127 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2128 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2129 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2130 static inline bool should_clone_nir(void) { return false; }
2133 #define _PASS(nir, do_pass) do { \
2135 nir_validate_shader(nir); \
2136 if (should_clone_nir()) { \
2137 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2143 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2144 nir_metadata_set_validation_flag(nir); \
2145 if (pass(nir, ##__VA_ARGS__)) { \
2147 nir_metadata_check_validation_flag(nir); \
2151 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2152 pass(nir, ##__VA_ARGS__); \
2155 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2156 void nir_calc_dominance(nir_shader
*shader
);
2158 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2159 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2161 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2162 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2164 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2165 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2167 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2168 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2170 int nir_gs_count_vertices(const nir_shader
*shader
);
2172 bool nir_split_var_copies(nir_shader
*shader
);
2174 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2175 bool nir_lower_returns(nir_shader
*shader
);
2177 bool nir_inline_functions(nir_shader
*shader
);
2179 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2180 void nir_lower_var_copies(nir_shader
*shader
);
2182 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2184 bool nir_lower_indirect_derefs(nir_shader
*shader
, uint32_t mode_mask
);
2186 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2188 void nir_lower_outputs_to_temporaries(nir_shader
*shader
,
2189 nir_function
*entrypoint
);
2190 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2192 void nir_assign_var_locations(struct exec_list
*var_list
,
2194 int (*type_size
)(const struct glsl_type
*));
2196 void nir_lower_io(nir_shader
*shader
,
2197 nir_variable_mode mode
,
2198 int (*type_size
)(const struct glsl_type
*));
2199 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2200 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2202 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2204 bool nir_remove_dead_variables(nir_shader
*shader
);
2206 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2207 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2208 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2209 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2211 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2213 void nir_lower_samplers(nir_shader
*shader
,
2214 const struct gl_shader_program
*shader_program
);
2216 bool nir_lower_system_values(nir_shader
*shader
);
2218 typedef struct nir_lower_tex_options
{
2220 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2221 * sampler types a texture projector is lowered.
2226 * If true, lower rect textures to 2D, using txs to fetch the
2227 * texture dimensions and dividing the texture coords by the
2228 * texture dims to normalize.
2233 * To emulate certain texture wrap modes, this can be used
2234 * to saturate the specified tex coord to [0.0, 1.0]. The
2235 * bits are according to sampler #, ie. if, for example:
2237 * (conf->saturate_s & (1 << n))
2239 * is true, then the s coord for sampler n is saturated.
2241 * Note that clamping must happen *after* projector lowering
2242 * so any projected texture sample instruction with a clamped
2243 * coordinate gets automatically lowered, regardless of the
2244 * 'lower_txp' setting.
2246 unsigned saturate_s
;
2247 unsigned saturate_t
;
2248 unsigned saturate_r
;
2250 /* Bitmask of textures that need swizzling.
2252 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2253 * swizzles[texture_index] is applied to the result of the texturing
2256 unsigned swizzle_result
;
2258 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2259 * while 4 and 5 represent 0 and 1 respectively.
2261 uint8_t swizzles
[32][4];
2262 } nir_lower_tex_options
;
2264 bool nir_lower_tex(nir_shader
*shader
,
2265 const nir_lower_tex_options
*options
);
2267 void nir_lower_idiv(nir_shader
*shader
);
2269 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2270 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2272 void nir_lower_two_sided_color(nir_shader
*shader
);
2274 void nir_lower_atomics(nir_shader
*shader
,
2275 const struct gl_shader_program
*shader_program
);
2276 void nir_lower_to_source_mods(nir_shader
*shader
);
2278 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2280 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2282 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2283 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2285 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2286 void nir_convert_to_ssa(nir_shader
*shader
);
2288 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2289 bool nir_repair_ssa(nir_shader
*shader
);
2291 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2292 * registers. If false, convert all values (even those not involved in a phi
2293 * node) to registers.
2295 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2297 bool nir_opt_algebraic(nir_shader
*shader
);
2298 bool nir_opt_algebraic_late(nir_shader
*shader
);
2299 bool nir_opt_constant_folding(nir_shader
*shader
);
2301 bool nir_opt_global_to_local(nir_shader
*shader
);
2303 bool nir_copy_prop(nir_shader
*shader
);
2305 bool nir_opt_cse(nir_shader
*shader
);
2307 bool nir_opt_dce(nir_shader
*shader
);
2309 bool nir_opt_dead_cf(nir_shader
*shader
);
2311 void nir_opt_gcm(nir_shader
*shader
);
2313 bool nir_opt_peephole_select(nir_shader
*shader
);
2315 bool nir_opt_remove_phis(nir_shader
*shader
);
2317 bool nir_opt_undef(nir_shader
*shader
);
2319 void nir_sweep(nir_shader
*shader
);
2321 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2322 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);