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 * Initial binding point for a sampler or UBO.
271 * For array types, this represents the binding point for the first element.
276 * Location an atomic counter is stored at.
281 * ARB_shader_image_load_store qualifiers.
284 bool read_only
; /**< "readonly" qualifier. */
285 bool write_only
; /**< "writeonly" qualifier. */
290 /** Image internal format if specified explicitly, otherwise GL_NONE. */
295 * Highest element accessed with a constant expression array index
297 * Not used for non-array variables.
299 unsigned max_array_access
;
304 * Built-in state that backs this uniform
306 * Once set at variable creation, \c state_slots must remain invariant.
307 * This is because, ideally, this array would be shared by all clones of
308 * this variable in the IR tree. In other words, we'd really like for it
309 * to be a fly-weight.
311 * If the variable is not a uniform, \c num_state_slots will be zero and
312 * \c state_slots will be \c NULL.
315 unsigned num_state_slots
; /**< Number of state slots used */
316 nir_state_slot
*state_slots
; /**< State descriptors. */
320 * Constant expression assigned in the initializer of the variable
322 nir_constant
*constant_initializer
;
325 * For variables that are in an interface block or are an instance of an
326 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
328 * \sa ir_variable::location
330 const struct glsl_type
*interface_type
;
333 #define nir_foreach_variable(var, var_list) \
334 foreach_list_typed(nir_variable, var, node, var_list)
336 #define nir_foreach_variable_safe(var, var_list) \
337 foreach_list_typed_safe(nir_variable, var, node, var_list)
340 nir_variable_is_global(const nir_variable
*var
)
342 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
345 typedef struct nir_register
{
346 struct exec_node node
;
348 unsigned num_components
; /** < number of vector components */
349 unsigned num_array_elems
; /** < size of array (0 for no array) */
351 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
354 /** generic register index. */
357 /** only for debug purposes, can be NULL */
360 /** whether this register is local (per-function) or global (per-shader) */
364 * If this flag is set to true, then accessing channels >= num_components
365 * is well-defined, and simply spills over to the next array element. This
366 * is useful for backends that can do per-component accessing, in
367 * particular scalar backends. By setting this flag and making
368 * num_components equal to 1, structures can be packed tightly into
369 * registers and then registers can be accessed per-component to get to
370 * each structure member, even if it crosses vec4 boundaries.
374 /** set of nir_src's where this register is used (read from) */
375 struct list_head uses
;
377 /** set of nir_dest's where this register is defined (written to) */
378 struct list_head defs
;
380 /** set of nir_if's where this register is used as a condition */
381 struct list_head if_uses
;
388 nir_instr_type_intrinsic
,
389 nir_instr_type_load_const
,
391 nir_instr_type_ssa_undef
,
393 nir_instr_type_parallel_copy
,
396 typedef struct nir_instr
{
397 struct exec_node node
;
399 struct nir_block
*block
;
401 /** generic instruction index. */
404 /* A temporary for optimization and analysis passes to use for storing
405 * flags. For instance, DCE uses this to store the "dead/live" info.
410 static inline nir_instr
*
411 nir_instr_next(nir_instr
*instr
)
413 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
414 if (exec_node_is_tail_sentinel(next
))
417 return exec_node_data(nir_instr
, next
, node
);
420 static inline nir_instr
*
421 nir_instr_prev(nir_instr
*instr
)
423 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
424 if (exec_node_is_head_sentinel(prev
))
427 return exec_node_data(nir_instr
, prev
, node
);
431 nir_instr_is_first(nir_instr
*instr
)
433 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
437 nir_instr_is_last(nir_instr
*instr
)
439 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
442 typedef struct nir_ssa_def
{
443 /** for debugging only, can be NULL */
446 /** generic SSA definition index. */
449 /** Index into the live_in and live_out bitfields */
452 nir_instr
*parent_instr
;
454 /** set of nir_instr's where this register is used (read from) */
455 struct list_head uses
;
457 /** set of nir_if's where this register is used as a condition */
458 struct list_head if_uses
;
460 uint8_t num_components
;
462 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
470 struct nir_src
*indirect
; /** < NULL for no indirect offset */
471 unsigned base_offset
;
473 /* TODO use-def chain goes here */
477 nir_instr
*parent_instr
;
478 struct list_head def_link
;
481 struct nir_src
*indirect
; /** < NULL for no indirect offset */
482 unsigned base_offset
;
484 /* TODO def-use chain goes here */
489 typedef struct nir_src
{
491 nir_instr
*parent_instr
;
492 struct nir_if
*parent_if
;
495 struct list_head use_link
;
505 #define NIR_SRC_INIT (nir_src) { { NULL } }
507 #define nir_foreach_use(reg_or_ssa_def, src) \
508 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
510 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
511 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
513 #define nir_foreach_if_use(reg_or_ssa_def, src) \
514 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
516 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
517 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
528 #define NIR_DEST_INIT (nir_dest) { { { NULL } } }
530 #define nir_foreach_def(reg, dest) \
531 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
533 #define nir_foreach_def_safe(reg, dest) \
534 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
536 static inline nir_src
537 nir_src_for_ssa(nir_ssa_def
*def
)
539 nir_src src
= NIR_SRC_INIT
;
547 static inline nir_src
548 nir_src_for_reg(nir_register
*reg
)
550 nir_src src
= NIR_SRC_INIT
;
554 src
.reg
.indirect
= NULL
;
555 src
.reg
.base_offset
= 0;
560 static inline nir_dest
561 nir_dest_for_reg(nir_register
*reg
)
563 nir_dest dest
= NIR_DEST_INIT
;
570 static inline unsigned
571 nir_src_bit_size(nir_src src
)
573 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
576 static inline unsigned
577 nir_dest_bit_size(nir_dest dest
)
579 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
582 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
583 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
589 * \name input modifiers
593 * For inputs interpreted as floating point, flips the sign bit. For
594 * inputs interpreted as integers, performs the two's complement negation.
599 * Clears the sign bit for floating point values, and computes the integer
600 * absolute value for integers. Note that the negate modifier acts after
601 * the absolute value modifier, therefore if both are set then all inputs
602 * will become negative.
608 * For each input component, says which component of the register it is
609 * chosen from. Note that which elements of the swizzle are used and which
610 * are ignored are based on the write mask for most opcodes - for example,
611 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
612 * a swizzle of {2, x, 1, 0} where x means "don't care."
621 * \name saturate output modifier
623 * Only valid for opcodes that output floating-point numbers. Clamps the
624 * output to between 0.0 and 1.0 inclusive.
629 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
633 nir_type_invalid
= 0, /* Not a valid type */
638 nir_type_bool32
= 32 | nir_type_bool
,
639 nir_type_int8
= 8 | nir_type_int
,
640 nir_type_int16
= 16 | nir_type_int
,
641 nir_type_int32
= 32 | nir_type_int
,
642 nir_type_int64
= 64 | nir_type_int
,
643 nir_type_uint8
= 8 | nir_type_uint
,
644 nir_type_uint16
= 16 | nir_type_uint
,
645 nir_type_uint32
= 32 | nir_type_uint
,
646 nir_type_uint64
= 64 | nir_type_uint
,
647 nir_type_float16
= 16 | nir_type_float
,
648 nir_type_float32
= 32 | nir_type_float
,
649 nir_type_float64
= 64 | nir_type_float
,
652 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
653 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
655 static inline unsigned
656 nir_alu_type_get_type_size(nir_alu_type type
)
658 return type
& NIR_ALU_TYPE_SIZE_MASK
;
661 static inline unsigned
662 nir_alu_type_get_base_type(nir_alu_type type
)
664 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
668 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
669 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
670 } nir_op_algebraic_property
;
678 * The number of components in the output
680 * If non-zero, this is the size of the output and input sizes are
681 * explicitly given; swizzle and writemask are still in effect, but if
682 * the output component is masked out, then the input component may
685 * If zero, the opcode acts in the standard, per-component manner; the
686 * operation is performed on each component (except the ones that are
687 * masked out) with the input being taken from the input swizzle for
690 * The size of some of the inputs may be given (i.e. non-zero) even
691 * though output_size is zero; in that case, the inputs with a zero
692 * size act per-component, while the inputs with non-zero size don't.
694 unsigned output_size
;
697 * The type of vector that the instruction outputs. Note that the
698 * staurate modifier is only allowed on outputs with the float type.
701 nir_alu_type output_type
;
704 * The number of components in each input
706 unsigned input_sizes
[4];
709 * The type of vector that each input takes. Note that negate and
710 * absolute value are only allowed on inputs with int or float type and
711 * behave differently on the two.
713 nir_alu_type input_types
[4];
715 nir_op_algebraic_property algebraic_properties
;
718 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
720 typedef struct nir_alu_instr
{
724 /** Indicates that this ALU instruction generates an exact value
726 * This is kind of a mixture of GLSL "precise" and "invariant" and not
727 * really equivalent to either. This indicates that the value generated by
728 * this operation is high-precision and any code transformations that touch
729 * it must ensure that the resulting value is bit-for-bit identical to the
738 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
739 nir_alu_instr
*instr
);
740 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
741 nir_alu_instr
*instr
);
743 /* is this source channel used? */
745 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
747 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
748 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
750 return (instr
->dest
.write_mask
>> channel
) & 1;
754 * For instructions whose destinations are SSA, get the number of channels
757 static inline unsigned
758 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
760 assert(instr
->dest
.dest
.is_ssa
);
762 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
763 return nir_op_infos
[instr
->op
].input_sizes
[src
];
765 return instr
->dest
.dest
.ssa
.num_components
;
770 nir_deref_type_array
,
771 nir_deref_type_struct
774 typedef struct nir_deref
{
775 nir_deref_type deref_type
;
776 struct nir_deref
*child
;
777 const struct glsl_type
*type
;
786 /* This enum describes how the array is referenced. If the deref is
787 * direct then the base_offset is used. If the deref is indirect then then
788 * offset is given by base_offset + indirect. If the deref is a wildcard
789 * then the deref refers to all of the elements of the array at the same
790 * time. Wildcard dereferences are only ever allowed in copy_var
791 * intrinsics and the source and destination derefs must have matching
795 nir_deref_array_type_direct
,
796 nir_deref_array_type_indirect
,
797 nir_deref_array_type_wildcard
,
798 } nir_deref_array_type
;
803 nir_deref_array_type deref_array_type
;
804 unsigned base_offset
;
814 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
815 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
816 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
818 /* Returns the last deref in the chain. */
819 static inline nir_deref
*
820 nir_deref_tail(nir_deref
*deref
)
823 deref
= deref
->child
;
831 nir_deref_var
**params
;
832 nir_deref_var
*return_deref
;
834 struct nir_function
*callee
;
837 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
838 num_variables, num_indices, idx0, idx1, idx2, flags) \
839 nir_intrinsic_##name,
841 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
844 #include "nir_intrinsics.h"
845 nir_num_intrinsics
= nir_last_intrinsic
+ 1
849 #undef LAST_INTRINSIC
851 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
853 /** Represents an intrinsic
855 * An intrinsic is an instruction type for handling things that are
856 * more-or-less regular operations but don't just consume and produce SSA
857 * values like ALU operations do. Intrinsics are not for things that have
858 * special semantic meaning such as phi nodes and parallel copies.
859 * Examples of intrinsics include variable load/store operations, system
860 * value loads, and the like. Even though texturing more-or-less falls
861 * under this category, texturing is its own instruction type because
862 * trying to represent texturing with intrinsics would lead to a
863 * combinatorial explosion of intrinsic opcodes.
865 * By having a single instruction type for handling a lot of different
866 * cases, optimization passes can look for intrinsics and, for the most
867 * part, completely ignore them. Each intrinsic type also has a few
868 * possible flags that govern whether or not they can be reordered or
869 * eliminated. That way passes like dead code elimination can still work
870 * on intrisics without understanding the meaning of each.
872 * Each intrinsic has some number of constant indices, some number of
873 * variables, and some number of sources. What these sources, variables,
874 * and indices mean depends on the intrinsic and is documented with the
875 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
876 * instructions are the only types of instruction that can operate on
882 nir_intrinsic_op intrinsic
;
886 /** number of components if this is a vectorized intrinsic
888 * Similarly to ALU operations, some intrinsics are vectorized.
889 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
890 * For vectorized intrinsics, the num_components field specifies the
891 * number of destination components and the number of source components
892 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
894 uint8_t num_components
;
896 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
898 nir_deref_var
*variables
[2];
901 } nir_intrinsic_instr
;
904 * \name NIR intrinsics semantic flags
906 * information about what the compiler can do with the intrinsics.
908 * \sa nir_intrinsic_info::flags
912 * whether the intrinsic can be safely eliminated if none of its output
913 * value is not being used.
915 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
918 * Whether the intrinsic can be reordered with respect to any other
919 * intrinsic, i.e. whether the only reordering dependencies of the
920 * intrinsic are due to the register reads/writes.
922 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
923 } nir_intrinsic_semantic_flag
;
926 * \name NIR intrinsics const-index flag
928 * Indicates the usage of a const_index slot.
930 * \sa nir_intrinsic_info::index_map
934 * Generally instructions that take a offset src argument, can encode
935 * a constant 'base' value which is added to the offset.
937 NIR_INTRINSIC_BASE
= 1,
940 * For store instructions, a writemask for the store.
942 NIR_INTRINSIC_WRMASK
= 2,
945 * The stream-id for GS emit_vertex/end_primitive intrinsics.
947 NIR_INTRINSIC_STREAM_ID
= 3,
950 * The clip-plane id for load_user_clip_plane intrinsic.
952 NIR_INTRINSIC_UCP_ID
= 4,
954 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
956 } nir_intrinsic_index_flag
;
958 #define NIR_INTRINSIC_MAX_INPUTS 4
963 unsigned num_srcs
; /** < number of register/SSA inputs */
965 /** number of components of each input register
967 * If this value is 0, the number of components is given by the
968 * num_components field of nir_intrinsic_instr.
970 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
974 /** number of components of the output register
976 * If this value is 0, the number of components is given by the
977 * num_components field of nir_intrinsic_instr.
979 unsigned dest_components
;
981 /** the number of inputs/outputs that are variables */
982 unsigned num_variables
;
984 /** the number of constant indices used by the intrinsic */
985 unsigned num_indices
;
987 /** indicates the usage of intr->const_index[n] */
988 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
990 /** semantic flags for calls to this intrinsic */
991 nir_intrinsic_semantic_flag flags
;
992 } nir_intrinsic_info
;
994 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
997 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
999 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
1001 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1002 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1003 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1005 static inline void \
1006 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1008 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1009 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1010 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1013 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1014 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1015 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1016 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1019 * \group texture information
1021 * This gives semantic information about textures which is useful to the
1022 * frontend, the backend, and lowering passes, but not the optimizer.
1027 nir_tex_src_projector
,
1028 nir_tex_src_comparitor
, /* shadow comparitor */
1032 nir_tex_src_ms_index
, /* MSAA sample index */
1035 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1036 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1037 nir_num_tex_src_types
1042 nir_tex_src_type src_type
;
1046 nir_texop_tex
, /**< Regular texture look-up */
1047 nir_texop_txb
, /**< Texture look-up with LOD bias */
1048 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1049 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1050 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1051 nir_texop_txf_ms
, /**< Multisample texture fetch */
1052 nir_texop_txs
, /**< Texture size */
1053 nir_texop_lod
, /**< Texture lod query */
1054 nir_texop_tg4
, /**< Texture gather */
1055 nir_texop_query_levels
, /**< Texture levels query */
1056 nir_texop_texture_samples
, /**< Texture samples query */
1057 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1065 enum glsl_sampler_dim sampler_dim
;
1066 nir_alu_type dest_type
;
1071 unsigned num_srcs
, coord_components
;
1072 bool is_array
, is_shadow
;
1075 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1076 * components or the new-style shadow that outputs 1 component.
1078 bool is_new_style_shadow
;
1080 /* gather component selector */
1081 unsigned component
: 2;
1083 /** The texture index
1085 * If this texture instruction has a nir_tex_src_texture_offset source,
1086 * then the texture index is given by texture_index + texture_offset.
1088 unsigned texture_index
;
1090 /** The size of the texture array or 0 if it's not an array */
1091 unsigned texture_array_size
;
1093 /** The texture deref
1095 * If this is null, use texture_index instead.
1097 nir_deref_var
*texture
;
1099 /** The sampler index
1101 * The following operations do not require a sampler and, as such, this
1102 * field should be ignored:
1104 * - nir_texop_txf_ms
1108 * - nir_texop_query_levels
1109 * - nir_texop_texture_samples
1110 * - nir_texop_samples_identical
1112 * If this texture instruction has a nir_tex_src_sampler_offset source,
1113 * then the sampler index is given by sampler_index + sampler_offset.
1115 unsigned sampler_index
;
1117 /** The sampler deref
1119 * If this is null, use sampler_index instead.
1121 nir_deref_var
*sampler
;
1124 static inline unsigned
1125 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1127 switch (instr
->op
) {
1128 case nir_texop_txs
: {
1130 switch (instr
->sampler_dim
) {
1131 case GLSL_SAMPLER_DIM_1D
:
1132 case GLSL_SAMPLER_DIM_BUF
:
1135 case GLSL_SAMPLER_DIM_2D
:
1136 case GLSL_SAMPLER_DIM_CUBE
:
1137 case GLSL_SAMPLER_DIM_MS
:
1138 case GLSL_SAMPLER_DIM_RECT
:
1139 case GLSL_SAMPLER_DIM_EXTERNAL
:
1142 case GLSL_SAMPLER_DIM_3D
:
1146 unreachable("not reached");
1148 if (instr
->is_array
)
1156 case nir_texop_texture_samples
:
1157 case nir_texop_query_levels
:
1158 case nir_texop_samples_identical
:
1162 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1169 /* Returns true if this texture operation queries something about the texture
1170 * rather than actually sampling it.
1173 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1175 switch (instr
->op
) {
1178 case nir_texop_texture_samples
:
1179 case nir_texop_query_levels
:
1186 case nir_texop_txf_ms
:
1190 unreachable("Invalid texture opcode");
1194 static inline unsigned
1195 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1197 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1198 return instr
->coord_components
;
1201 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1202 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1203 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1204 if (instr
->is_array
)
1205 return instr
->coord_components
- 1;
1207 return instr
->coord_components
;
1214 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1216 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1217 if (instr
->src
[i
].src_type
== type
)
1237 nir_const_value value
;
1240 } nir_load_const_instr
;
1253 /* creates a new SSA variable in an undefined state */
1258 } nir_ssa_undef_instr
;
1261 struct exec_node node
;
1263 /* The predecessor block corresponding to this source */
1264 struct nir_block
*pred
;
1269 #define nir_foreach_phi_src(phi, entry) \
1270 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1271 #define nir_foreach_phi_src_safe(phi, entry) \
1272 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1277 struct exec_list srcs
; /** < list of nir_phi_src */
1283 struct exec_node node
;
1286 } nir_parallel_copy_entry
;
1288 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1289 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1294 /* A list of nir_parallel_copy_entry's. The sources of all of the
1295 * entries are copied to the corresponding destinations "in parallel".
1296 * In other words, if we have two entries: a -> b and b -> a, the values
1299 struct exec_list entries
;
1300 } nir_parallel_copy_instr
;
1302 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1303 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1304 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1305 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1306 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1307 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1308 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1309 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1310 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1311 nir_parallel_copy_instr
, instr
)
1316 * Control flow consists of a tree of control flow nodes, which include
1317 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1318 * instructions that always run start-to-finish. Each basic block also keeps
1319 * track of its successors (blocks which may run immediately after the current
1320 * block) and predecessors (blocks which could have run immediately before the
1321 * current block). Each function also has a start block and an end block which
1322 * all return statements point to (which is always empty). Together, all the
1323 * blocks with their predecessors and successors make up the control flow
1324 * graph (CFG) of the function. There are helpers that modify the tree of
1325 * control flow nodes while modifying the CFG appropriately; these should be
1326 * used instead of modifying the tree directly.
1333 nir_cf_node_function
1336 typedef struct nir_cf_node
{
1337 struct exec_node node
;
1338 nir_cf_node_type type
;
1339 struct nir_cf_node
*parent
;
1342 typedef struct nir_block
{
1343 nir_cf_node cf_node
;
1345 struct exec_list instr_list
; /** < list of nir_instr */
1347 /** generic block index; generated by nir_index_blocks */
1351 * Each block can only have up to 2 successors, so we put them in a simple
1352 * array - no need for anything more complicated.
1354 struct nir_block
*successors
[2];
1356 /* Set of nir_block predecessors in the CFG */
1357 struct set
*predecessors
;
1360 * this node's immediate dominator in the dominance tree - set to NULL for
1363 struct nir_block
*imm_dom
;
1365 /* This node's children in the dominance tree */
1366 unsigned num_dom_children
;
1367 struct nir_block
**dom_children
;
1369 /* Set of nir_block's on the dominance frontier of this block */
1370 struct set
*dom_frontier
;
1373 * These two indices have the property that dom_{pre,post}_index for each
1374 * child of this block in the dominance tree will always be between
1375 * dom_pre_index and dom_post_index for this block, which makes testing if
1376 * a given block is dominated by another block an O(1) operation.
1378 unsigned dom_pre_index
, dom_post_index
;
1380 /* live in and out for this block; used for liveness analysis */
1381 BITSET_WORD
*live_in
;
1382 BITSET_WORD
*live_out
;
1385 static inline nir_instr
*
1386 nir_block_first_instr(nir_block
*block
)
1388 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1389 return exec_node_data(nir_instr
, head
, node
);
1392 static inline nir_instr
*
1393 nir_block_last_instr(nir_block
*block
)
1395 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1396 return exec_node_data(nir_instr
, tail
, node
);
1399 #define nir_foreach_instr(block, instr) \
1400 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1401 #define nir_foreach_instr_reverse(block, instr) \
1402 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1403 #define nir_foreach_instr_safe(block, instr) \
1404 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1405 #define nir_foreach_instr_reverse_safe(block, instr) \
1406 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1408 typedef struct nir_if
{
1409 nir_cf_node cf_node
;
1412 struct exec_list then_list
; /** < list of nir_cf_node */
1413 struct exec_list else_list
; /** < list of nir_cf_node */
1416 static inline nir_cf_node
*
1417 nir_if_first_then_node(nir_if
*if_stmt
)
1419 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1420 return exec_node_data(nir_cf_node
, head
, node
);
1423 static inline nir_cf_node
*
1424 nir_if_last_then_node(nir_if
*if_stmt
)
1426 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1427 return exec_node_data(nir_cf_node
, tail
, node
);
1430 static inline nir_cf_node
*
1431 nir_if_first_else_node(nir_if
*if_stmt
)
1433 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1434 return exec_node_data(nir_cf_node
, head
, node
);
1437 static inline nir_cf_node
*
1438 nir_if_last_else_node(nir_if
*if_stmt
)
1440 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1441 return exec_node_data(nir_cf_node
, tail
, node
);
1445 nir_cf_node cf_node
;
1447 struct exec_list body
; /** < list of nir_cf_node */
1450 static inline nir_cf_node
*
1451 nir_loop_first_cf_node(nir_loop
*loop
)
1453 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1456 static inline nir_cf_node
*
1457 nir_loop_last_cf_node(nir_loop
*loop
)
1459 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1463 * Various bits of metadata that can may be created or required by
1464 * optimization and analysis passes
1467 nir_metadata_none
= 0x0,
1468 nir_metadata_block_index
= 0x1,
1469 nir_metadata_dominance
= 0x2,
1470 nir_metadata_live_ssa_defs
= 0x4,
1471 nir_metadata_not_properly_reset
= 0x8,
1475 nir_cf_node cf_node
;
1477 /** pointer to the function of which this is an implementation */
1478 struct nir_function
*function
;
1480 struct exec_list body
; /** < list of nir_cf_node */
1482 nir_block
*end_block
;
1484 /** list for all local variables in the function */
1485 struct exec_list locals
;
1487 /** array of variables used as parameters */
1488 unsigned num_params
;
1489 nir_variable
**params
;
1491 /** variable used to hold the result of the function */
1492 nir_variable
*return_var
;
1494 /** list of local registers in the function */
1495 struct exec_list registers
;
1497 /** next available local register index */
1500 /** next available SSA value index */
1503 /* total number of basic blocks, only valid when block_index_dirty = false */
1504 unsigned num_blocks
;
1506 nir_metadata valid_metadata
;
1507 } nir_function_impl
;
1509 static inline nir_block
*
1510 nir_start_block(nir_function_impl
*impl
)
1512 return (nir_block
*) exec_list_get_head(&impl
->body
);
1515 static inline nir_cf_node
*
1516 nir_cf_node_next(nir_cf_node
*node
)
1518 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1519 if (exec_node_is_tail_sentinel(next
))
1522 return exec_node_data(nir_cf_node
, next
, node
);
1525 static inline nir_cf_node
*
1526 nir_cf_node_prev(nir_cf_node
*node
)
1528 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1529 if (exec_node_is_head_sentinel(prev
))
1532 return exec_node_data(nir_cf_node
, prev
, node
);
1536 nir_cf_node_is_first(const nir_cf_node
*node
)
1538 return exec_node_is_head_sentinel(node
->node
.prev
);
1542 nir_cf_node_is_last(const nir_cf_node
*node
)
1544 return exec_node_is_tail_sentinel(node
->node
.next
);
1547 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1548 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1549 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1550 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1555 nir_parameter_inout
,
1556 } nir_parameter_type
;
1559 nir_parameter_type param_type
;
1560 const struct glsl_type
*type
;
1563 typedef struct nir_function
{
1564 struct exec_node node
;
1567 struct nir_shader
*shader
;
1569 unsigned num_params
;
1570 nir_parameter
*params
;
1571 const struct glsl_type
*return_type
;
1573 /** The implementation of this function.
1575 * If the function is only declared and not implemented, this is NULL.
1577 nir_function_impl
*impl
;
1580 typedef struct nir_shader_compiler_options
{
1588 bool lower_bitfield_extract
;
1589 bool lower_bitfield_insert
;
1590 bool lower_uadd_carry
;
1591 bool lower_usub_borrow
;
1592 /** lowers fneg and ineg to fsub and isub. */
1594 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1597 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1600 /* Does the native fdot instruction replicate its result for four
1601 * components? If so, then opt_algebraic_late will turn all fdotN
1602 * instructions into fdot_replicatedN instructions.
1604 bool fdot_replicates
;
1606 /** lowers ffract to fsub+ffloor: */
1609 bool lower_pack_half_2x16
;
1610 bool lower_pack_unorm_2x16
;
1611 bool lower_pack_snorm_2x16
;
1612 bool lower_pack_unorm_4x8
;
1613 bool lower_pack_snorm_4x8
;
1614 bool lower_unpack_half_2x16
;
1615 bool lower_unpack_unorm_2x16
;
1616 bool lower_unpack_snorm_2x16
;
1617 bool lower_unpack_unorm_4x8
;
1618 bool lower_unpack_snorm_4x8
;
1620 bool lower_extract_byte
;
1621 bool lower_extract_word
;
1624 * Does the driver support real 32-bit integers? (Otherwise, integers
1625 * are simulated by floats.)
1627 bool native_integers
;
1629 /* Indicates that the driver only has zero-based vertex id */
1630 bool vertex_id_zero_based
;
1631 } nir_shader_compiler_options
;
1633 typedef struct nir_shader_info
{
1636 /* Descriptive name provided by the client; may be NULL */
1639 /* Number of textures used by this shader */
1640 unsigned num_textures
;
1641 /* Number of uniform buffers used by this shader */
1643 /* Number of atomic buffers used by this shader */
1645 /* Number of shader storage buffers used by this shader */
1647 /* Number of images used by this shader */
1648 unsigned num_images
;
1650 /* Which inputs are actually read */
1651 uint64_t inputs_read
;
1652 /* Which outputs are actually written */
1653 uint64_t outputs_written
;
1654 /* Which system values are actually read */
1655 uint64_t system_values_read
;
1657 /* Which patch inputs are actually read */
1658 uint32_t patch_inputs_read
;
1659 /* Which patch outputs are actually written */
1660 uint32_t patch_outputs_written
;
1662 /* Whether or not this shader ever uses textureGather() */
1663 bool uses_texture_gather
;
1665 /* Whether or not this shader uses the gl_ClipDistance output */
1666 bool uses_clip_distance_out
;
1668 /* Whether or not separate shader objects were used */
1669 bool separate_shader
;
1671 /** Was this shader linked with any transform feedback varyings? */
1672 bool has_transform_feedback_varyings
;
1676 /** The number of vertices recieves per input primitive */
1677 unsigned vertices_in
;
1679 /** The output primitive type (GL enum value) */
1680 unsigned output_primitive
;
1682 /** The maximum number of vertices the geometry shader might write. */
1683 unsigned vertices_out
;
1685 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1686 unsigned invocations
;
1688 /** Whether or not this shader uses EndPrimitive */
1689 bool uses_end_primitive
;
1691 /** Whether or not this shader uses non-zero streams */
1699 * Whether early fragment tests are enabled as defined by
1700 * ARB_shader_image_load_store.
1702 bool early_fragment_tests
;
1704 /** gl_FragDepth layout for ARB_conservative_depth. */
1705 enum gl_frag_depth_layout depth_layout
;
1709 unsigned local_size
[3];
1713 /** The number of vertices in the TCS output patch. */
1714 unsigned vertices_out
;
1719 typedef struct nir_shader
{
1720 /** list of uniforms (nir_variable) */
1721 struct exec_list uniforms
;
1723 /** list of inputs (nir_variable) */
1724 struct exec_list inputs
;
1726 /** list of outputs (nir_variable) */
1727 struct exec_list outputs
;
1729 /** list of shared compute variables (nir_variable) */
1730 struct exec_list shared
;
1732 /** Set of driver-specific options for the shader.
1734 * The memory for the options is expected to be kept in a single static
1735 * copy by the driver.
1737 const struct nir_shader_compiler_options
*options
;
1739 /** Various bits of compile-time information about a given shader */
1740 struct nir_shader_info info
;
1742 /** list of global variables in the shader (nir_variable) */
1743 struct exec_list globals
;
1745 /** list of system value variables in the shader (nir_variable) */
1746 struct exec_list system_values
;
1748 struct exec_list functions
; /** < list of nir_function */
1750 /** list of global register in the shader */
1751 struct exec_list registers
;
1753 /** next available global register index */
1757 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1760 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1762 /** The shader stage, such as MESA_SHADER_VERTEX. */
1763 gl_shader_stage stage
;
1766 static inline nir_function
*
1767 nir_shader_get_entrypoint(nir_shader
*shader
)
1769 assert(exec_list_length(&shader
->functions
) == 1);
1770 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1771 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1772 assert(func
->return_type
== glsl_void_type());
1773 assert(func
->num_params
== 0);
1777 #define nir_foreach_function(shader, func) \
1778 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1780 nir_shader
*nir_shader_create(void *mem_ctx
,
1781 gl_shader_stage stage
,
1782 const nir_shader_compiler_options
*options
);
1784 /** creates a register, including assigning it an index and adding it to the list */
1785 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1787 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1789 void nir_reg_remove(nir_register
*reg
);
1791 /** Adds a variable to the appropreate list in nir_shader */
1792 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1795 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1797 assert(var
->data
.mode
== nir_var_local
);
1798 exec_list_push_tail(&impl
->locals
, &var
->node
);
1801 /** creates a variable, sets a few defaults, and adds it to the list */
1802 nir_variable
*nir_variable_create(nir_shader
*shader
,
1803 nir_variable_mode mode
,
1804 const struct glsl_type
*type
,
1806 /** creates a local variable and adds it to the list */
1807 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1808 const struct glsl_type
*type
,
1811 /** creates a function and adds it to the shader's list of functions */
1812 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1814 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1815 /** creates a function_impl that isn't tied to any particular function */
1816 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1818 nir_block
*nir_block_create(nir_shader
*shader
);
1819 nir_if
*nir_if_create(nir_shader
*shader
);
1820 nir_loop
*nir_loop_create(nir_shader
*shader
);
1822 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1824 /** requests that the given pieces of metadata be generated */
1825 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1826 /** dirties all but the preserved metadata */
1827 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1829 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1830 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1832 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1834 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1835 unsigned num_components
,
1838 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1839 nir_intrinsic_op op
);
1841 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1842 nir_function
*callee
);
1844 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1846 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1848 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1850 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1851 unsigned num_components
,
1854 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1855 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1856 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1858 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1860 nir_load_const_instr
*
1861 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1864 * NIR Cursors and Instruction Insertion API
1867 * A tiny struct representing a point to insert/extract instructions or
1868 * control flow nodes. Helps reduce the combinatorial explosion of possible
1869 * points to insert/extract.
1871 * \sa nir_control_flow.h
1874 nir_cursor_before_block
,
1875 nir_cursor_after_block
,
1876 nir_cursor_before_instr
,
1877 nir_cursor_after_instr
,
1878 } nir_cursor_option
;
1881 nir_cursor_option option
;
1888 static inline nir_block
*
1889 nir_cursor_current_block(nir_cursor cursor
)
1891 if (cursor
.option
== nir_cursor_before_instr
||
1892 cursor
.option
== nir_cursor_after_instr
) {
1893 return cursor
.instr
->block
;
1895 return cursor
.block
;
1899 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
1901 static inline nir_cursor
1902 nir_before_block(nir_block
*block
)
1905 cursor
.option
= nir_cursor_before_block
;
1906 cursor
.block
= block
;
1910 static inline nir_cursor
1911 nir_after_block(nir_block
*block
)
1914 cursor
.option
= nir_cursor_after_block
;
1915 cursor
.block
= block
;
1919 static inline nir_cursor
1920 nir_before_instr(nir_instr
*instr
)
1923 cursor
.option
= nir_cursor_before_instr
;
1924 cursor
.instr
= instr
;
1928 static inline nir_cursor
1929 nir_after_instr(nir_instr
*instr
)
1932 cursor
.option
= nir_cursor_after_instr
;
1933 cursor
.instr
= instr
;
1937 static inline nir_cursor
1938 nir_after_block_before_jump(nir_block
*block
)
1940 nir_instr
*last_instr
= nir_block_last_instr(block
);
1941 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1942 return nir_before_instr(last_instr
);
1944 return nir_after_block(block
);
1948 static inline nir_cursor
1949 nir_before_cf_node(nir_cf_node
*node
)
1951 if (node
->type
== nir_cf_node_block
)
1952 return nir_before_block(nir_cf_node_as_block(node
));
1954 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1957 static inline nir_cursor
1958 nir_after_cf_node(nir_cf_node
*node
)
1960 if (node
->type
== nir_cf_node_block
)
1961 return nir_after_block(nir_cf_node_as_block(node
));
1963 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1966 static inline nir_cursor
1967 nir_after_cf_node_and_phis(nir_cf_node
*node
)
1969 if (node
->type
== nir_cf_node_block
)
1970 return nir_after_block(nir_cf_node_as_block(node
));
1972 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
1973 assert(block
->cf_node
.type
== nir_cf_node_block
);
1975 nir_foreach_instr(block
, instr
) {
1976 if (instr
->type
!= nir_instr_type_phi
)
1977 return nir_before_instr(instr
);
1979 return nir_after_block(block
);
1982 static inline nir_cursor
1983 nir_before_cf_list(struct exec_list
*cf_list
)
1985 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1986 exec_list_get_head(cf_list
), node
);
1987 return nir_before_cf_node(first_node
);
1990 static inline nir_cursor
1991 nir_after_cf_list(struct exec_list
*cf_list
)
1993 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1994 exec_list_get_tail(cf_list
), node
);
1995 return nir_after_cf_node(last_node
);
1999 * Insert a NIR instruction at the given cursor.
2001 * Note: This does not update the cursor.
2003 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2006 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2008 nir_instr_insert(nir_before_instr(instr
), before
);
2012 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2014 nir_instr_insert(nir_after_instr(instr
), after
);
2018 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2020 nir_instr_insert(nir_before_block(block
), before
);
2024 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2026 nir_instr_insert(nir_after_block(block
), after
);
2030 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2032 nir_instr_insert(nir_before_cf_node(node
), before
);
2036 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2038 nir_instr_insert(nir_after_cf_node(node
), after
);
2042 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2044 nir_instr_insert(nir_before_cf_list(list
), before
);
2048 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2050 nir_instr_insert(nir_after_cf_list(list
), after
);
2053 void nir_instr_remove(nir_instr
*instr
);
2057 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2058 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2059 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2060 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2062 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2063 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2065 nir_const_value
*nir_src_as_const_value(nir_src src
);
2066 bool nir_src_is_dynamically_uniform(nir_src src
);
2067 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2068 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2069 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2070 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2071 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2074 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2075 unsigned num_components
, unsigned bit_size
,
2077 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2078 unsigned num_components
, unsigned bit_size
,
2080 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2081 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2082 nir_instr
*after_me
);
2084 /* visits basic blocks in source-code order */
2085 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
2086 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
2088 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
2090 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
2093 /* If the following CF node is an if, this function returns that if.
2094 * Otherwise, it returns NULL.
2096 nir_if
*nir_block_get_following_if(nir_block
*block
);
2098 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2100 void nir_index_local_regs(nir_function_impl
*impl
);
2101 void nir_index_global_regs(nir_shader
*shader
);
2102 void nir_index_ssa_defs(nir_function_impl
*impl
);
2103 unsigned nir_index_instrs(nir_function_impl
*impl
);
2105 void nir_index_blocks(nir_function_impl
*impl
);
2107 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2108 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2110 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2111 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2112 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2113 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2116 void nir_validate_shader(nir_shader
*shader
);
2117 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2118 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2120 #include "util/debug.h"
2122 should_clone_nir(void)
2124 static int should_clone
= -1;
2125 if (should_clone
< 0)
2126 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2128 return should_clone
;
2131 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2132 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2133 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2134 static inline bool should_clone_nir(void) { return false; }
2137 #define _PASS(nir, do_pass) do { \
2139 nir_validate_shader(nir); \
2140 if (should_clone_nir()) { \
2141 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2147 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2148 nir_metadata_set_validation_flag(nir); \
2149 if (pass(nir, ##__VA_ARGS__)) { \
2151 nir_metadata_check_validation_flag(nir); \
2155 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2156 pass(nir, ##__VA_ARGS__); \
2159 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2160 void nir_calc_dominance(nir_shader
*shader
);
2162 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2163 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2165 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2166 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2168 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2169 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2171 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2172 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2174 int nir_gs_count_vertices(const nir_shader
*shader
);
2176 bool nir_split_var_copies(nir_shader
*shader
);
2178 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2179 bool nir_lower_returns(nir_shader
*shader
);
2181 bool nir_inline_functions(nir_shader
*shader
);
2183 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2184 void nir_lower_var_copies(nir_shader
*shader
);
2186 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2188 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2190 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2192 void nir_lower_outputs_to_temporaries(nir_shader
*shader
,
2193 nir_function
*entrypoint
);
2194 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2196 void nir_assign_var_locations(struct exec_list
*var_list
,
2198 int (*type_size
)(const struct glsl_type
*));
2200 void nir_lower_io(nir_shader
*shader
,
2201 nir_variable_mode modes
,
2202 int (*type_size
)(const struct glsl_type
*));
2203 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2204 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2206 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2208 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2210 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2211 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2212 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2213 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2215 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2217 void nir_lower_samplers(nir_shader
*shader
,
2218 const struct gl_shader_program
*shader_program
);
2220 bool nir_lower_system_values(nir_shader
*shader
);
2222 typedef struct nir_lower_tex_options
{
2224 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2225 * sampler types a texture projector is lowered.
2230 * If true, lower rect textures to 2D, using txs to fetch the
2231 * texture dimensions and dividing the texture coords by the
2232 * texture dims to normalize.
2237 * To emulate certain texture wrap modes, this can be used
2238 * to saturate the specified tex coord to [0.0, 1.0]. The
2239 * bits are according to sampler #, ie. if, for example:
2241 * (conf->saturate_s & (1 << n))
2243 * is true, then the s coord for sampler n is saturated.
2245 * Note that clamping must happen *after* projector lowering
2246 * so any projected texture sample instruction with a clamped
2247 * coordinate gets automatically lowered, regardless of the
2248 * 'lower_txp' setting.
2250 unsigned saturate_s
;
2251 unsigned saturate_t
;
2252 unsigned saturate_r
;
2254 /* Bitmask of textures that need swizzling.
2256 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2257 * swizzles[texture_index] is applied to the result of the texturing
2260 unsigned swizzle_result
;
2262 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2263 * while 4 and 5 represent 0 and 1 respectively.
2265 uint8_t swizzles
[32][4];
2266 } nir_lower_tex_options
;
2268 bool nir_lower_tex(nir_shader
*shader
,
2269 const nir_lower_tex_options
*options
);
2271 void nir_lower_idiv(nir_shader
*shader
);
2273 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2274 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2276 void nir_lower_two_sided_color(nir_shader
*shader
);
2278 void nir_lower_atomics(nir_shader
*shader
,
2279 const struct gl_shader_program
*shader_program
);
2280 void nir_lower_to_source_mods(nir_shader
*shader
);
2282 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2284 void nir_lower_double_pack(nir_shader
*shader
);
2286 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2288 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2289 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2291 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2292 void nir_convert_to_ssa(nir_shader
*shader
);
2294 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2295 bool nir_repair_ssa(nir_shader
*shader
);
2297 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2298 * registers. If false, convert all values (even those not involved in a phi
2299 * node) to registers.
2301 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2303 bool nir_opt_algebraic(nir_shader
*shader
);
2304 bool nir_opt_algebraic_late(nir_shader
*shader
);
2305 bool nir_opt_constant_folding(nir_shader
*shader
);
2307 bool nir_opt_global_to_local(nir_shader
*shader
);
2309 bool nir_copy_prop(nir_shader
*shader
);
2311 bool nir_opt_cse(nir_shader
*shader
);
2313 bool nir_opt_dce(nir_shader
*shader
);
2315 bool nir_opt_dead_cf(nir_shader
*shader
);
2317 void nir_opt_gcm(nir_shader
*shader
);
2319 bool nir_opt_peephole_select(nir_shader
*shader
);
2321 bool nir_opt_remove_phis(nir_shader
*shader
);
2323 bool nir_opt_undef(nir_shader
*shader
);
2325 void nir_sweep(nir_shader
*shader
);
2327 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2328 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);