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"
32 #include "GL/gl.h" /* GLenum */
33 #include "util/list.h"
34 #include "util/ralloc.h"
36 #include "util/bitset.h"
37 #include "nir_types.h"
38 #include "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); \
68 struct nir_function_overload
;
75 * Description of built-in state associated with a uniform
77 * \sa nir_variable::state_slots
90 nir_var_shader_storage
,
95 * Data stored in an nir_constant
97 union nir_constant_data
{
104 typedef struct nir_constant
{
106 * Value of the constant.
108 * The field used to back the values supplied by the constant is determined
109 * by the type associated with the \c nir_variable. Constants may be
110 * scalars, vectors, or matrices.
112 union nir_constant_data value
;
114 /* Array elements / Structure Fields */
115 struct nir_constant
**elements
;
119 * \brief Layout qualifiers for gl_FragDepth.
121 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
122 * with a layout qualifier.
125 nir_depth_layout_none
, /**< No depth layout is specified. */
126 nir_depth_layout_any
,
127 nir_depth_layout_greater
,
128 nir_depth_layout_less
,
129 nir_depth_layout_unchanged
133 * Either a uniform, global variable, shader input, or shader output. Based on
134 * ir_variable - it should be easy to translate between the two.
138 struct exec_node node
;
141 * Declared type of the variable
143 const struct glsl_type
*type
;
146 * Declared name of the variable
151 * For variables which satisfy the is_interface_instance() predicate, this
152 * points to an array of integers such that if the ith member of the
153 * interface block is an array, max_ifc_array_access[i] is the maximum
154 * array element of that member that has been accessed. If the ith member
155 * of the interface block is not an array, max_ifc_array_access[i] is
158 * For variables whose type is not an interface block, this pointer is
161 unsigned *max_ifc_array_access
;
163 struct nir_variable_data
{
166 * Is the variable read-only?
168 * This is set for variables declared as \c const, shader inputs,
171 unsigned read_only
:1;
174 unsigned invariant
:1;
177 * Storage class of the variable.
179 * \sa nir_variable_mode
181 nir_variable_mode mode
:4;
184 * Interpolation mode for shader inputs / outputs
186 * \sa glsl_interp_qualifier
188 unsigned interpolation
:2;
191 * \name ARB_fragment_coord_conventions
194 unsigned origin_upper_left
:1;
195 unsigned pixel_center_integer
:1;
199 * Was the location explicitly set in the shader?
201 * If the location is explicitly set in the shader, it \b cannot be changed
202 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
205 unsigned explicit_location
:1;
206 unsigned explicit_index
:1;
209 * Was an initial binding explicitly set in the shader?
211 * If so, constant_initializer contains an integer nir_constant
212 * representing the initial binding point.
214 unsigned explicit_binding
:1;
217 * Does this variable have an initializer?
219 * This is used by the linker to cross-validiate initializers of global
222 unsigned has_initializer
:1;
225 * Is this variable a generic output or input that has not yet been matched
226 * up to a variable in another stage of the pipeline?
228 * This is used by the linker as scratch storage while assigning locations
229 * to generic inputs and outputs.
231 unsigned is_unmatched_generic_inout
:1;
234 * If non-zero, then this variable may be packed along with other variables
235 * into a single varying slot, so this offset should be applied when
236 * accessing components. For example, an offset of 1 means that the x
237 * component of this variable is actually stored in component y of the
238 * location specified by \c location.
240 unsigned location_frac
:2;
243 * Non-zero if this variable was created by lowering a named interface
244 * block which was not an array.
246 * Note that this variable and \c from_named_ifc_block_array will never
249 unsigned from_named_ifc_block_nonarray
:1;
252 * Non-zero if this variable was created by lowering a named interface
253 * block which was an array.
255 * Note that this variable and \c from_named_ifc_block_nonarray will never
258 unsigned from_named_ifc_block_array
:1;
261 * \brief Layout qualifier for gl_FragDepth.
263 * This is not equal to \c ir_depth_layout_none if and only if this
264 * variable is \c gl_FragDepth and a layout qualifier is specified.
266 nir_depth_layout depth_layout
;
269 * Storage location of the base of this variable
271 * The precise meaning of this field depends on the nature of the variable.
273 * - Vertex shader input: one of the values from \c gl_vert_attrib.
274 * - Vertex shader output: one of the values from \c gl_varying_slot.
275 * - Geometry shader input: one of the values from \c gl_varying_slot.
276 * - Geometry shader output: one of the values from \c gl_varying_slot.
277 * - Fragment shader input: one of the values from \c gl_varying_slot.
278 * - Fragment shader output: one of the values from \c gl_frag_result.
279 * - Uniforms: Per-stage uniform slot number for default uniform block.
280 * - Uniforms: Index within the uniform block definition for UBO members.
281 * - Non-UBO Uniforms: uniform slot number.
282 * - Other: This field is not currently used.
284 * If the variable is a uniform, shader input, or shader output, and the
285 * slot has not been assigned, the value will be -1.
290 * The actual location of the variable in the IR. Only valid for inputs
293 unsigned int driver_location
;
296 * output index for dual source blending.
301 * Initial binding point for a sampler or UBO.
303 * For array types, this represents the binding point for the first element.
308 * Location an atomic counter is stored at.
311 unsigned buffer_index
;
316 * ARB_shader_image_load_store qualifiers.
319 bool read_only
; /**< "readonly" qualifier. */
320 bool write_only
; /**< "writeonly" qualifier. */
325 /** Image internal format if specified explicitly, otherwise GL_NONE. */
330 * Highest element accessed with a constant expression array index
332 * Not used for non-array variables.
334 unsigned max_array_access
;
339 * Built-in state that backs this uniform
341 * Once set at variable creation, \c state_slots must remain invariant.
342 * This is because, ideally, this array would be shared by all clones of
343 * this variable in the IR tree. In other words, we'd really like for it
344 * to be a fly-weight.
346 * If the variable is not a uniform, \c num_state_slots will be zero and
347 * \c state_slots will be \c NULL.
350 unsigned num_state_slots
; /**< Number of state slots used */
351 nir_state_slot
*state_slots
; /**< State descriptors. */
355 * Constant expression assigned in the initializer of the variable
357 nir_constant
*constant_initializer
;
360 * For variables that are in an interface block or are an instance of an
361 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
363 * \sa ir_variable::location
365 const struct glsl_type
*interface_type
;
368 #define nir_foreach_variable(var, var_list) \
369 foreach_list_typed(nir_variable, var, node, var_list)
372 struct exec_node node
;
374 unsigned num_components
; /** < number of vector components */
375 unsigned num_array_elems
; /** < size of array (0 for no array) */
377 /** generic register index. */
380 /** only for debug purposes, can be NULL */
383 /** whether this register is local (per-function) or global (per-shader) */
387 * If this flag is set to true, then accessing channels >= num_components
388 * is well-defined, and simply spills over to the next array element. This
389 * is useful for backends that can do per-component accessing, in
390 * particular scalar backends. By setting this flag and making
391 * num_components equal to 1, structures can be packed tightly into
392 * registers and then registers can be accessed per-component to get to
393 * each structure member, even if it crosses vec4 boundaries.
397 /** set of nir_instr's where this register is used (read from) */
398 struct list_head uses
;
400 /** set of nir_instr's where this register is defined (written to) */
401 struct list_head defs
;
403 /** set of nir_if's where this register is used as a condition */
404 struct list_head if_uses
;
411 nir_instr_type_intrinsic
,
412 nir_instr_type_load_const
,
414 nir_instr_type_ssa_undef
,
416 nir_instr_type_parallel_copy
,
419 typedef struct nir_instr
{
420 struct exec_node node
;
422 struct nir_block
*block
;
424 /** generic instruction index. */
427 /* A temporary for optimization and analysis passes to use for storing
428 * flags. For instance, DCE uses this to store the "dead/live" info.
433 static inline nir_instr
*
434 nir_instr_next(nir_instr
*instr
)
436 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
437 if (exec_node_is_tail_sentinel(next
))
440 return exec_node_data(nir_instr
, next
, node
);
443 static inline nir_instr
*
444 nir_instr_prev(nir_instr
*instr
)
446 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
447 if (exec_node_is_head_sentinel(prev
))
450 return exec_node_data(nir_instr
, prev
, node
);
454 nir_instr_is_first(nir_instr
*instr
)
456 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
460 nir_instr_is_last(nir_instr
*instr
)
462 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
466 /** for debugging only, can be NULL */
469 /** generic SSA definition index. */
472 /** Index into the live_in and live_out bitfields */
475 nir_instr
*parent_instr
;
477 /** set of nir_instr's where this register is used (read from) */
478 struct list_head uses
;
480 /** set of nir_if's where this register is used as a condition */
481 struct list_head if_uses
;
483 uint8_t num_components
;
490 struct nir_src
*indirect
; /** < NULL for no indirect offset */
491 unsigned base_offset
;
493 /* TODO use-def chain goes here */
497 nir_instr
*parent_instr
;
498 struct list_head def_link
;
501 struct nir_src
*indirect
; /** < NULL for no indirect offset */
502 unsigned base_offset
;
504 /* TODO def-use chain goes here */
509 typedef struct nir_src
{
511 nir_instr
*parent_instr
;
512 struct nir_if
*parent_if
;
515 struct list_head use_link
;
525 #define NIR_SRC_INIT (nir_src) { { NULL } }
527 #define nir_foreach_use(reg_or_ssa_def, src) \
528 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
530 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
531 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
533 #define nir_foreach_if_use(reg_or_ssa_def, src) \
534 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
536 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
537 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
548 #define NIR_DEST_INIT (nir_dest) { { { NULL } } }
550 #define nir_foreach_def(reg, dest) \
551 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
553 #define nir_foreach_def_safe(reg, dest) \
554 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
556 static inline nir_src
557 nir_src_for_ssa(nir_ssa_def
*def
)
559 nir_src src
= NIR_SRC_INIT
;
567 static inline nir_src
568 nir_src_for_reg(nir_register
*reg
)
570 nir_src src
= NIR_SRC_INIT
;
574 src
.reg
.indirect
= NULL
;
575 src
.reg
.base_offset
= 0;
580 static inline nir_dest
581 nir_dest_for_reg(nir_register
*reg
)
583 nir_dest dest
= NIR_DEST_INIT
;
590 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
591 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
597 * \name input modifiers
601 * For inputs interpreted as floating point, flips the sign bit. For
602 * inputs interpreted as integers, performs the two's complement negation.
607 * Clears the sign bit for floating point values, and computes the integer
608 * absolute value for integers. Note that the negate modifier acts after
609 * the absolute value modifier, therefore if both are set then all inputs
610 * will become negative.
616 * For each input component, says which component of the register it is
617 * chosen from. Note that which elements of the swizzle are used and which
618 * are ignored are based on the write mask for most opcodes - for example,
619 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
620 * a swizzle of {2, x, 1, 0} where x means "don't care."
629 * \name saturate output modifier
631 * Only valid for opcodes that output floating-point numbers. Clamps the
632 * output to between 0.0 and 1.0 inclusive.
637 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
641 nir_type_invalid
= 0, /* Not a valid type */
649 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
650 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
651 } nir_op_algebraic_property
;
659 * The number of components in the output
661 * If non-zero, this is the size of the output and input sizes are
662 * explicitly given; swizzle and writemask are still in effect, but if
663 * the output component is masked out, then the input component may
666 * If zero, the opcode acts in the standard, per-component manner; the
667 * operation is performed on each component (except the ones that are
668 * masked out) with the input being taken from the input swizzle for
671 * The size of some of the inputs may be given (i.e. non-zero) even
672 * though output_size is zero; in that case, the inputs with a zero
673 * size act per-component, while the inputs with non-zero size don't.
675 unsigned output_size
;
678 * The type of vector that the instruction outputs. Note that the
679 * staurate modifier is only allowed on outputs with the float type.
682 nir_alu_type output_type
;
685 * The number of components in each input
687 unsigned input_sizes
[4];
690 * The type of vector that each input takes. Note that negate and
691 * absolute value are only allowed on inputs with int or float type and
692 * behave differently on the two.
694 nir_alu_type input_types
[4];
696 nir_op_algebraic_property algebraic_properties
;
699 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
701 typedef struct nir_alu_instr
{
708 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
709 nir_alu_instr
*instr
);
710 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
711 nir_alu_instr
*instr
);
713 /* is this source channel used? */
715 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
717 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
718 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
720 return (instr
->dest
.write_mask
>> channel
) & 1;
724 * For instructions whose destinations are SSA, get the number of channels
727 static inline unsigned
728 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
730 assert(instr
->dest
.dest
.is_ssa
);
732 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
733 return nir_op_infos
[instr
->op
].input_sizes
[src
];
735 return instr
->dest
.dest
.ssa
.num_components
;
740 nir_deref_type_array
,
741 nir_deref_type_struct
744 typedef struct nir_deref
{
745 nir_deref_type deref_type
;
746 struct nir_deref
*child
;
747 const struct glsl_type
*type
;
756 /* This enum describes how the array is referenced. If the deref is
757 * direct then the base_offset is used. If the deref is indirect then then
758 * offset is given by base_offset + indirect. If the deref is a wildcard
759 * then the deref refers to all of the elements of the array at the same
760 * time. Wildcard dereferences are only ever allowed in copy_var
761 * intrinsics and the source and destination derefs must have matching
765 nir_deref_array_type_direct
,
766 nir_deref_array_type_indirect
,
767 nir_deref_array_type_wildcard
,
768 } nir_deref_array_type
;
773 nir_deref_array_type deref_array_type
;
774 unsigned base_offset
;
784 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
785 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
786 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
792 nir_deref_var
**params
;
793 nir_deref_var
*return_deref
;
795 struct nir_function_overload
*callee
;
798 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
799 num_variables, num_indices, flags) \
800 nir_intrinsic_##name,
802 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
805 #include "nir_intrinsics.h"
806 nir_num_intrinsics
= nir_last_intrinsic
+ 1
810 #undef LAST_INTRINSIC
812 /** Represents an intrinsic
814 * An intrinsic is an instruction type for handling things that are
815 * more-or-less regular operations but don't just consume and produce SSA
816 * values like ALU operations do. Intrinsics are not for things that have
817 * special semantic meaning such as phi nodes and parallel copies.
818 * Examples of intrinsics include variable load/store operations, system
819 * value loads, and the like. Even though texturing more-or-less falls
820 * under this category, texturing is its own instruction type because
821 * trying to represent texturing with intrinsics would lead to a
822 * combinatorial explosion of intrinsic opcodes.
824 * By having a single instruction type for handling a lot of different
825 * cases, optimization passes can look for intrinsics and, for the most
826 * part, completely ignore them. Each intrinsic type also has a few
827 * possible flags that govern whether or not they can be reordered or
828 * eliminated. That way passes like dead code elimination can still work
829 * on intrisics without understanding the meaning of each.
831 * Each intrinsic has some number of constant indices, some number of
832 * variables, and some number of sources. What these sources, variables,
833 * and indices mean depends on the intrinsic and is documented with the
834 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
835 * instructions are the only types of instruction that can operate on
841 nir_intrinsic_op intrinsic
;
845 /** number of components if this is a vectorized intrinsic
847 * Similarly to ALU operations, some intrinsics are vectorized.
848 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
849 * For vectorized intrinsics, the num_components field specifies the
850 * number of destination components and the number of source components
851 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
853 uint8_t num_components
;
857 nir_deref_var
*variables
[2];
860 } nir_intrinsic_instr
;
863 * \name NIR intrinsics semantic flags
865 * information about what the compiler can do with the intrinsics.
867 * \sa nir_intrinsic_info::flags
871 * whether the intrinsic can be safely eliminated if none of its output
872 * value is not being used.
874 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
877 * Whether the intrinsic can be reordered with respect to any other
878 * intrinsic, i.e. whether the only reordering dependencies of the
879 * intrinsic are due to the register reads/writes.
881 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
882 } nir_intrinsic_semantic_flag
;
884 #define NIR_INTRINSIC_MAX_INPUTS 4
889 unsigned num_srcs
; /** < number of register/SSA inputs */
891 /** number of components of each input register
893 * If this value is 0, the number of components is given by the
894 * num_components field of nir_intrinsic_instr.
896 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
900 /** number of components of the output register
902 * If this value is 0, the number of components is given by the
903 * num_components field of nir_intrinsic_instr.
905 unsigned dest_components
;
907 /** the number of inputs/outputs that are variables */
908 unsigned num_variables
;
910 /** the number of constant indices used by the intrinsic */
911 unsigned num_indices
;
913 /** semantic flags for calls to this intrinsic */
914 nir_intrinsic_semantic_flag flags
;
915 } nir_intrinsic_info
;
917 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
920 * \group texture information
922 * This gives semantic information about textures which is useful to the
923 * frontend, the backend, and lowering passes, but not the optimizer.
928 nir_tex_src_projector
,
929 nir_tex_src_comparitor
, /* shadow comparitor */
933 nir_tex_src_ms_index
, /* MSAA sample index */
936 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
937 nir_num_tex_src_types
942 nir_tex_src_type src_type
;
946 nir_texop_tex
, /**< Regular texture look-up */
947 nir_texop_txb
, /**< Texture look-up with LOD bias */
948 nir_texop_txl
, /**< Texture look-up with explicit LOD */
949 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
950 nir_texop_txf
, /**< Texel fetch with explicit LOD */
951 nir_texop_txf_ms
, /**< Multisample texture fetch */
952 nir_texop_txs
, /**< Texture size */
953 nir_texop_lod
, /**< Texture lod query */
954 nir_texop_tg4
, /**< Texture gather */
955 nir_texop_query_levels
, /**< Texture levels query */
956 nir_texop_texture_samples
, /**< Texture samples query */
962 enum glsl_sampler_dim sampler_dim
;
963 nir_alu_type dest_type
;
968 unsigned num_srcs
, coord_components
;
969 bool is_array
, is_shadow
;
972 * If is_shadow is true, whether this is the old-style shadow that outputs 4
973 * components or the new-style shadow that outputs 1 component.
975 bool is_new_style_shadow
;
977 /* constant offset - must be 0 if the offset source is used */
980 /* gather component selector */
981 unsigned component
: 2;
983 /** The sampler index
985 * If this texture instruction has a nir_tex_src_sampler_offset source,
986 * then the sampler index is given by sampler_index + sampler_offset.
988 unsigned sampler_index
;
990 /** The size of the sampler array or 0 if it's not an array */
991 unsigned sampler_array_size
;
993 nir_deref_var
*sampler
; /* if this is NULL, use sampler_index instead */
996 static inline unsigned
997 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1000 case nir_texop_txs
: {
1002 switch (instr
->sampler_dim
) {
1003 case GLSL_SAMPLER_DIM_1D
:
1004 case GLSL_SAMPLER_DIM_BUF
:
1007 case GLSL_SAMPLER_DIM_2D
:
1008 case GLSL_SAMPLER_DIM_CUBE
:
1009 case GLSL_SAMPLER_DIM_MS
:
1010 case GLSL_SAMPLER_DIM_RECT
:
1011 case GLSL_SAMPLER_DIM_EXTERNAL
:
1014 case GLSL_SAMPLER_DIM_3D
:
1018 unreachable("not reached");
1020 if (instr
->is_array
)
1028 case nir_texop_texture_samples
:
1029 case nir_texop_query_levels
:
1033 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1040 static inline unsigned
1041 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1043 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1044 return instr
->coord_components
;
1047 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1048 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1049 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1050 if (instr
->is_array
)
1051 return instr
->coord_components
- 1;
1053 return instr
->coord_components
;
1060 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1062 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1063 if (instr
->src
[i
].src_type
== type
)
1080 nir_const_value value
;
1083 } nir_load_const_instr
;
1096 /* creates a new SSA variable in an undefined state */
1101 } nir_ssa_undef_instr
;
1104 struct exec_node node
;
1106 /* The predecessor block corresponding to this source */
1107 struct nir_block
*pred
;
1112 #define nir_foreach_phi_src(phi, entry) \
1113 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1114 #define nir_foreach_phi_src_safe(phi, entry) \
1115 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1120 struct exec_list srcs
; /** < list of nir_phi_src */
1126 struct exec_node node
;
1129 } nir_parallel_copy_entry
;
1131 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1132 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1137 /* A list of nir_parallel_copy_entry's. The sources of all of the
1138 * entries are copied to the corresponding destinations "in parallel".
1139 * In other words, if we have two entries: a -> b and b -> a, the values
1142 struct exec_list entries
;
1143 } nir_parallel_copy_instr
;
1145 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1146 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1147 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1148 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1149 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1150 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1151 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1152 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1153 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1154 nir_parallel_copy_instr
, instr
)
1159 * Control flow consists of a tree of control flow nodes, which include
1160 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1161 * instructions that always run start-to-finish. Each basic block also keeps
1162 * track of its successors (blocks which may run immediately after the current
1163 * block) and predecessors (blocks which could have run immediately before the
1164 * current block). Each function also has a start block and an end block which
1165 * all return statements point to (which is always empty). Together, all the
1166 * blocks with their predecessors and successors make up the control flow
1167 * graph (CFG) of the function. There are helpers that modify the tree of
1168 * control flow nodes while modifying the CFG appropriately; these should be
1169 * used instead of modifying the tree directly.
1176 nir_cf_node_function
1179 typedef struct nir_cf_node
{
1180 struct exec_node node
;
1181 nir_cf_node_type type
;
1182 struct nir_cf_node
*parent
;
1185 typedef struct nir_block
{
1186 nir_cf_node cf_node
;
1188 struct exec_list instr_list
; /** < list of nir_instr */
1190 /** generic block index; generated by nir_index_blocks */
1194 * Each block can only have up to 2 successors, so we put them in a simple
1195 * array - no need for anything more complicated.
1197 struct nir_block
*successors
[2];
1199 /* Set of nir_block predecessors in the CFG */
1200 struct set
*predecessors
;
1203 * this node's immediate dominator in the dominance tree - set to NULL for
1206 struct nir_block
*imm_dom
;
1208 /* This node's children in the dominance tree */
1209 unsigned num_dom_children
;
1210 struct nir_block
**dom_children
;
1212 /* Set of nir_block's on the dominance frontier of this block */
1213 struct set
*dom_frontier
;
1216 * These two indices have the property that dom_{pre,post}_index for each
1217 * child of this block in the dominance tree will always be between
1218 * dom_pre_index and dom_post_index for this block, which makes testing if
1219 * a given block is dominated by another block an O(1) operation.
1221 unsigned dom_pre_index
, dom_post_index
;
1223 /* live in and out for this block; used for liveness analysis */
1224 BITSET_WORD
*live_in
;
1225 BITSET_WORD
*live_out
;
1228 static inline nir_instr
*
1229 nir_block_first_instr(nir_block
*block
)
1231 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1232 return exec_node_data(nir_instr
, head
, node
);
1235 static inline nir_instr
*
1236 nir_block_last_instr(nir_block
*block
)
1238 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1239 return exec_node_data(nir_instr
, tail
, node
);
1242 #define nir_foreach_instr(block, instr) \
1243 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1244 #define nir_foreach_instr_reverse(block, instr) \
1245 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1246 #define nir_foreach_instr_safe(block, instr) \
1247 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1248 #define nir_foreach_instr_safe_reverse(block, instr) \
1249 foreach_list_typed_safe_reverse(nir_instr, instr, node, &(block)->instr_list)
1251 typedef struct nir_if
{
1252 nir_cf_node cf_node
;
1255 struct exec_list then_list
; /** < list of nir_cf_node */
1256 struct exec_list else_list
; /** < list of nir_cf_node */
1259 static inline nir_cf_node
*
1260 nir_if_first_then_node(nir_if
*if_stmt
)
1262 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1263 return exec_node_data(nir_cf_node
, head
, node
);
1266 static inline nir_cf_node
*
1267 nir_if_last_then_node(nir_if
*if_stmt
)
1269 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1270 return exec_node_data(nir_cf_node
, tail
, node
);
1273 static inline nir_cf_node
*
1274 nir_if_first_else_node(nir_if
*if_stmt
)
1276 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1277 return exec_node_data(nir_cf_node
, head
, node
);
1280 static inline nir_cf_node
*
1281 nir_if_last_else_node(nir_if
*if_stmt
)
1283 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1284 return exec_node_data(nir_cf_node
, tail
, node
);
1288 nir_cf_node cf_node
;
1290 struct exec_list body
; /** < list of nir_cf_node */
1293 static inline nir_cf_node
*
1294 nir_loop_first_cf_node(nir_loop
*loop
)
1296 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1299 static inline nir_cf_node
*
1300 nir_loop_last_cf_node(nir_loop
*loop
)
1302 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1306 * Various bits of metadata that can may be created or required by
1307 * optimization and analysis passes
1310 nir_metadata_none
= 0x0,
1311 nir_metadata_block_index
= 0x1,
1312 nir_metadata_dominance
= 0x2,
1313 nir_metadata_live_variables
= 0x4,
1317 nir_cf_node cf_node
;
1319 /** pointer to the overload of which this is an implementation */
1320 struct nir_function_overload
*overload
;
1322 struct exec_list body
; /** < list of nir_cf_node */
1324 nir_block
*end_block
;
1326 /** list for all local variables in the function */
1327 struct exec_list locals
;
1329 /** array of variables used as parameters */
1330 unsigned num_params
;
1331 nir_variable
**params
;
1333 /** variable used to hold the result of the function */
1334 nir_variable
*return_var
;
1336 /** list of local registers in the function */
1337 struct exec_list registers
;
1339 /** next available local register index */
1342 /** next available SSA value index */
1345 /* total number of basic blocks, only valid when block_index_dirty = false */
1346 unsigned num_blocks
;
1348 nir_metadata valid_metadata
;
1349 } nir_function_impl
;
1351 static inline nir_block
*
1352 nir_start_block(nir_function_impl
*impl
)
1354 return (nir_block
*) exec_list_get_head(&impl
->body
);
1357 static inline nir_cf_node
*
1358 nir_cf_node_next(nir_cf_node
*node
)
1360 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1361 if (exec_node_is_tail_sentinel(next
))
1364 return exec_node_data(nir_cf_node
, next
, node
);
1367 static inline nir_cf_node
*
1368 nir_cf_node_prev(nir_cf_node
*node
)
1370 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1371 if (exec_node_is_head_sentinel(prev
))
1374 return exec_node_data(nir_cf_node
, prev
, node
);
1378 nir_cf_node_is_first(const nir_cf_node
*node
)
1380 return exec_node_is_head_sentinel(node
->node
.prev
);
1384 nir_cf_node_is_last(const nir_cf_node
*node
)
1386 return exec_node_is_tail_sentinel(node
->node
.next
);
1389 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1390 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1391 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1392 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1397 nir_parameter_inout
,
1398 } nir_parameter_type
;
1401 nir_parameter_type param_type
;
1402 const struct glsl_type
*type
;
1405 typedef struct nir_function_overload
{
1406 struct exec_node node
;
1408 unsigned num_params
;
1409 nir_parameter
*params
;
1410 const struct glsl_type
*return_type
;
1412 nir_function_impl
*impl
; /** < NULL if the overload is only declared yet */
1414 /** pointer to the function of which this is an overload */
1415 struct nir_function
*function
;
1416 } nir_function_overload
;
1418 typedef struct nir_function
{
1419 struct exec_node node
;
1421 struct exec_list overload_list
; /** < list of nir_function_overload */
1423 struct nir_shader
*shader
;
1426 #define nir_function_first_overload(func) \
1427 exec_node_data(nir_function_overload, \
1428 exec_list_get_head(&(func)->overload_list), node)
1430 typedef struct nir_shader_compiler_options
{
1436 /** lowers fneg and ineg to fsub and isub. */
1438 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1441 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1444 /* Does the native fdot instruction replicate its result for four
1445 * components? If so, then opt_algebraic_late will turn all fdotN
1446 * instructions into fdot_replicatedN instructions.
1448 bool fdot_replicates
;
1450 /** lowers ffract to fsub+ffloor: */
1454 * Does the driver support real 32-bit integers? (Otherwise, integers
1455 * are simulated by floats.)
1457 bool native_integers
;
1458 } nir_shader_compiler_options
;
1460 typedef struct nir_shader_info
{
1463 /* Descriptive name provided by the client; may be NULL */
1466 /* Number of textures used by this shader */
1467 unsigned num_textures
;
1468 /* Number of uniform buffers used by this shader */
1470 /* Number of atomic buffers used by this shader */
1472 /* Number of shader storage buffers used by this shader */
1474 /* Number of images used by this shader */
1475 unsigned num_images
;
1477 /* Which inputs are actually read */
1478 uint64_t inputs_read
;
1479 /* Which outputs are actually written */
1480 uint64_t outputs_written
;
1481 /* Which system values are actually read */
1482 uint64_t system_values_read
;
1484 /* Whether or not this shader ever uses textureGather() */
1485 bool uses_texture_gather
;
1487 /* Whether or not this shader uses the gl_ClipDistance output */
1488 bool uses_clip_distance_out
;
1490 /* Whether or not separate shader objects were used */
1491 bool separate_shader
;
1493 /** Was this shader linked with any transform feedback varyings? */
1494 bool has_transform_feedback_varyings
;
1498 /** The maximum number of vertices the geometry shader might write. */
1499 unsigned vertices_out
;
1501 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1502 unsigned invocations
;
1506 unsigned local_size
[3];
1511 typedef struct nir_shader
{
1512 /** list of uniforms (nir_variable) */
1513 struct exec_list uniforms
;
1515 /** list of inputs (nir_variable) */
1516 struct exec_list inputs
;
1518 /** list of outputs (nir_variable) */
1519 struct exec_list outputs
;
1521 /** Set of driver-specific options for the shader.
1523 * The memory for the options is expected to be kept in a single static
1524 * copy by the driver.
1526 const struct nir_shader_compiler_options
*options
;
1528 /** Various bits of compile-time information about a given shader */
1529 struct nir_shader_info info
;
1531 /** list of global variables in the shader (nir_variable) */
1532 struct exec_list globals
;
1534 /** list of system value variables in the shader (nir_variable) */
1535 struct exec_list system_values
;
1537 struct exec_list functions
; /** < list of nir_function */
1539 /** list of global register in the shader */
1540 struct exec_list registers
;
1542 /** next available global register index */
1546 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1549 unsigned num_inputs
, num_uniforms
, num_outputs
;
1551 /** The shader stage, such as MESA_SHADER_VERTEX. */
1552 gl_shader_stage stage
;
1555 #define nir_foreach_overload(shader, overload) \
1556 foreach_list_typed(nir_function, func, node, &(shader)->functions) \
1557 foreach_list_typed(nir_function_overload, overload, node, \
1558 &(func)->overload_list)
1560 nir_shader
*nir_shader_create(void *mem_ctx
,
1561 gl_shader_stage stage
,
1562 const nir_shader_compiler_options
*options
);
1564 /** creates a register, including assigning it an index and adding it to the list */
1565 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1567 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1569 void nir_reg_remove(nir_register
*reg
);
1571 /** Adds a variable to the appropreate list in nir_shader */
1572 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1575 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1577 assert(var
->data
.mode
== nir_var_local
);
1578 exec_list_push_tail(&impl
->locals
, &var
->node
);
1581 /** creates a variable, sets a few defaults, and adds it to the list */
1582 nir_variable
*nir_variable_create(nir_shader
*shader
,
1583 nir_variable_mode mode
,
1584 const struct glsl_type
*type
,
1586 /** creates a local variable and adds it to the list */
1587 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1588 const struct glsl_type
*type
,
1591 /** creates a function and adds it to the shader's list of functions */
1592 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1594 /** creates a null function returning null */
1595 nir_function_overload
*nir_function_overload_create(nir_function
*func
);
1597 nir_function_impl
*nir_function_impl_create(nir_function_overload
*func
);
1599 nir_block
*nir_block_create(void *mem_ctx
);
1600 nir_if
*nir_if_create(void *mem_ctx
);
1601 nir_loop
*nir_loop_create(void *mem_ctx
);
1603 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1605 /** requests that the given pieces of metadata be generated */
1606 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1607 /** dirties all but the preserved metadata */
1608 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1610 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1611 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1613 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1615 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1616 unsigned num_components
);
1618 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1619 nir_intrinsic_op op
);
1621 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1622 nir_function_overload
*callee
);
1624 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1626 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1628 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1630 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1631 unsigned num_components
);
1633 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1634 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1635 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1637 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1639 nir_load_const_instr
*
1640 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1643 * NIR Cursors and Instruction Insertion API
1646 * A tiny struct representing a point to insert/extract instructions or
1647 * control flow nodes. Helps reduce the combinatorial explosion of possible
1648 * points to insert/extract.
1650 * \sa nir_control_flow.h
1653 nir_cursor_before_block
,
1654 nir_cursor_after_block
,
1655 nir_cursor_before_instr
,
1656 nir_cursor_after_instr
,
1657 } nir_cursor_option
;
1660 nir_cursor_option option
;
1667 static inline nir_cursor
1668 nir_before_block(nir_block
*block
)
1671 cursor
.option
= nir_cursor_before_block
;
1672 cursor
.block
= block
;
1676 static inline nir_cursor
1677 nir_after_block(nir_block
*block
)
1680 cursor
.option
= nir_cursor_after_block
;
1681 cursor
.block
= block
;
1685 static inline nir_cursor
1686 nir_before_instr(nir_instr
*instr
)
1689 cursor
.option
= nir_cursor_before_instr
;
1690 cursor
.instr
= instr
;
1694 static inline nir_cursor
1695 nir_after_instr(nir_instr
*instr
)
1698 cursor
.option
= nir_cursor_after_instr
;
1699 cursor
.instr
= instr
;
1703 static inline nir_cursor
1704 nir_after_block_before_jump(nir_block
*block
)
1706 nir_instr
*last_instr
= nir_block_last_instr(block
);
1707 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1708 return nir_before_instr(last_instr
);
1710 return nir_after_block(block
);
1714 static inline nir_cursor
1715 nir_before_cf_node(nir_cf_node
*node
)
1717 if (node
->type
== nir_cf_node_block
)
1718 return nir_before_block(nir_cf_node_as_block(node
));
1720 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1723 static inline nir_cursor
1724 nir_after_cf_node(nir_cf_node
*node
)
1726 if (node
->type
== nir_cf_node_block
)
1727 return nir_after_block(nir_cf_node_as_block(node
));
1729 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1732 static inline nir_cursor
1733 nir_before_cf_list(struct exec_list
*cf_list
)
1735 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1736 exec_list_get_head(cf_list
), node
);
1737 return nir_before_cf_node(first_node
);
1740 static inline nir_cursor
1741 nir_after_cf_list(struct exec_list
*cf_list
)
1743 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1744 exec_list_get_tail(cf_list
), node
);
1745 return nir_after_cf_node(last_node
);
1749 * Insert a NIR instruction at the given cursor.
1751 * Note: This does not update the cursor.
1753 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
1756 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1758 nir_instr_insert(nir_before_instr(instr
), before
);
1762 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
1764 nir_instr_insert(nir_after_instr(instr
), after
);
1768 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
1770 nir_instr_insert(nir_before_block(block
), before
);
1774 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
1776 nir_instr_insert(nir_after_block(block
), after
);
1780 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
1782 nir_instr_insert(nir_before_cf_node(node
), before
);
1786 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
1788 nir_instr_insert(nir_after_cf_node(node
), after
);
1792 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
1794 nir_instr_insert(nir_before_cf_list(list
), before
);
1798 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
1800 nir_instr_insert(nir_after_cf_list(list
), after
);
1803 void nir_instr_remove(nir_instr
*instr
);
1807 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
1808 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
1809 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
1810 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
1812 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
1813 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
1815 nir_const_value
*nir_src_as_const_value(nir_src src
);
1816 bool nir_src_is_dynamically_uniform(nir_src src
);
1817 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
1818 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
1819 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
1820 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
1821 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
1824 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1825 unsigned num_components
, const char *name
);
1826 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1827 unsigned num_components
, const char *name
);
1828 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
1830 /* visits basic blocks in source-code order */
1831 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
1832 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1834 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1836 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1839 /* If the following CF node is an if, this function returns that if.
1840 * Otherwise, it returns NULL.
1842 nir_if
*nir_block_get_following_if(nir_block
*block
);
1844 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
1846 void nir_index_local_regs(nir_function_impl
*impl
);
1847 void nir_index_global_regs(nir_shader
*shader
);
1848 void nir_index_ssa_defs(nir_function_impl
*impl
);
1849 unsigned nir_index_instrs(nir_function_impl
*impl
);
1851 void nir_index_blocks(nir_function_impl
*impl
);
1853 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
1854 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
1857 void nir_validate_shader(nir_shader
*shader
);
1859 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
1862 void nir_calc_dominance_impl(nir_function_impl
*impl
);
1863 void nir_calc_dominance(nir_shader
*shader
);
1865 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
1866 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
1868 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
1869 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
1871 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
1872 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
1874 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
1875 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
1877 int nir_gs_count_vertices(nir_shader
*shader
);
1879 bool nir_split_var_copies(nir_shader
*shader
);
1881 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
1882 void nir_lower_var_copies(nir_shader
*shader
);
1884 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
1886 bool nir_lower_locals_to_regs(nir_shader
*shader
);
1888 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
1890 void nir_assign_var_locations(struct exec_list
*var_list
,
1892 int (*type_size
)(const struct glsl_type
*));
1894 void nir_lower_io(nir_shader
*shader
,
1895 nir_variable_mode mode
,
1896 int (*type_size
)(const struct glsl_type
*));
1897 void nir_lower_vars_to_ssa(nir_shader
*shader
);
1899 bool nir_remove_dead_variables(nir_shader
*shader
);
1901 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
1902 bool nir_lower_vec_to_movs(nir_shader
*shader
);
1903 void nir_lower_alu_to_scalar(nir_shader
*shader
);
1904 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
1906 void nir_lower_phis_to_scalar(nir_shader
*shader
);
1908 void nir_lower_samplers(nir_shader
*shader
,
1909 const struct gl_shader_program
*shader_program
);
1911 bool nir_lower_system_values(nir_shader
*shader
);
1913 typedef struct nir_lower_tex_options
{
1915 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
1916 * sampler types a texture projector is lowered.
1921 * If true, lower rect textures to 2D, using txs to fetch the
1922 * texture dimensions and dividing the texture coords by the
1923 * texture dims to normalize.
1928 * To emulate certain texture wrap modes, this can be used
1929 * to saturate the specified tex coord to [0.0, 1.0]. The
1930 * bits are according to sampler #, ie. if, for example:
1932 * (conf->saturate_s & (1 << n))
1934 * is true, then the s coord for sampler n is saturated.
1936 * Note that clamping must happen *after* projector lowering
1937 * so any projected texture sample instruction with a clamped
1938 * coordinate gets automatically lowered, regardless of the
1939 * 'lower_txp' setting.
1941 unsigned saturate_s
;
1942 unsigned saturate_t
;
1943 unsigned saturate_r
;
1944 } nir_lower_tex_options
;
1946 void nir_lower_tex(nir_shader
*shader
,
1947 const nir_lower_tex_options
*options
);
1949 void nir_lower_idiv(nir_shader
*shader
);
1951 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
1952 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
1954 void nir_lower_two_sided_color(nir_shader
*shader
);
1956 void nir_lower_atomics(nir_shader
*shader
);
1957 void nir_lower_to_source_mods(nir_shader
*shader
);
1959 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
1961 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
1963 void nir_live_variables_impl(nir_function_impl
*impl
);
1964 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
1966 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
1967 void nir_convert_to_ssa(nir_shader
*shader
);
1969 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
1970 * registers. If false, convert all values (even those not involved in a phi
1971 * node) to registers.
1973 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
1975 bool nir_opt_algebraic(nir_shader
*shader
);
1976 bool nir_opt_algebraic_late(nir_shader
*shader
);
1977 bool nir_opt_constant_folding(nir_shader
*shader
);
1979 bool nir_opt_global_to_local(nir_shader
*shader
);
1981 bool nir_copy_prop_impl(nir_function_impl
*impl
);
1982 bool nir_copy_prop(nir_shader
*shader
);
1984 bool nir_opt_cse(nir_shader
*shader
);
1986 bool nir_opt_dce_impl(nir_function_impl
*impl
);
1987 bool nir_opt_dce(nir_shader
*shader
);
1989 bool nir_opt_dead_cf(nir_shader
*shader
);
1991 void nir_opt_gcm(nir_shader
*shader
);
1993 bool nir_opt_peephole_select(nir_shader
*shader
);
1994 bool nir_opt_peephole_ffma(nir_shader
*shader
);
1996 bool nir_opt_remove_phis(nir_shader
*shader
);
1998 bool nir_opt_undef(nir_shader
*shader
);
2000 void nir_sweep(nir_shader
*shader
);
2002 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2003 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);