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 "glsl/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
;
369 struct exec_node node
;
371 unsigned num_components
; /** < number of vector components */
372 unsigned num_array_elems
; /** < size of array (0 for no array) */
374 /** generic register index. */
377 /** only for debug purposes, can be NULL */
380 /** whether this register is local (per-function) or global (per-shader) */
384 * If this flag is set to true, then accessing channels >= num_components
385 * is well-defined, and simply spills over to the next array element. This
386 * is useful for backends that can do per-component accessing, in
387 * particular scalar backends. By setting this flag and making
388 * num_components equal to 1, structures can be packed tightly into
389 * registers and then registers can be accessed per-component to get to
390 * each structure member, even if it crosses vec4 boundaries.
394 /** set of nir_instr's where this register is used (read from) */
395 struct list_head uses
;
397 /** set of nir_instr's where this register is defined (written to) */
398 struct list_head defs
;
400 /** set of nir_if's where this register is used as a condition */
401 struct list_head if_uses
;
408 nir_instr_type_intrinsic
,
409 nir_instr_type_load_const
,
411 nir_instr_type_ssa_undef
,
413 nir_instr_type_parallel_copy
,
416 typedef struct nir_instr
{
417 struct exec_node node
;
419 struct nir_block
*block
;
421 /** generic instruction index. */
424 /* A temporary for optimization and analysis passes to use for storing
425 * flags. For instance, DCE uses this to store the "dead/live" info.
430 static inline nir_instr
*
431 nir_instr_next(nir_instr
*instr
)
433 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
434 if (exec_node_is_tail_sentinel(next
))
437 return exec_node_data(nir_instr
, next
, node
);
440 static inline nir_instr
*
441 nir_instr_prev(nir_instr
*instr
)
443 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
444 if (exec_node_is_head_sentinel(prev
))
447 return exec_node_data(nir_instr
, prev
, node
);
451 nir_instr_is_first(nir_instr
*instr
)
453 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
457 nir_instr_is_last(nir_instr
*instr
)
459 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
463 /** for debugging only, can be NULL */
466 /** generic SSA definition index. */
469 /** Index into the live_in and live_out bitfields */
472 nir_instr
*parent_instr
;
474 /** set of nir_instr's where this register is used (read from) */
475 struct list_head uses
;
477 /** set of nir_if's where this register is used as a condition */
478 struct list_head if_uses
;
480 uint8_t num_components
;
487 struct nir_src
*indirect
; /** < NULL for no indirect offset */
488 unsigned base_offset
;
490 /* TODO use-def chain goes here */
494 nir_instr
*parent_instr
;
495 struct list_head def_link
;
498 struct nir_src
*indirect
; /** < NULL for no indirect offset */
499 unsigned base_offset
;
501 /* TODO def-use chain goes here */
506 typedef struct nir_src
{
508 nir_instr
*parent_instr
;
509 struct nir_if
*parent_if
;
512 struct list_head use_link
;
522 #define NIR_SRC_INIT (nir_src) { { NULL } }
524 #define nir_foreach_use(reg_or_ssa_def, src) \
525 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
527 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
528 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
530 #define nir_foreach_if_use(reg_or_ssa_def, src) \
531 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
533 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
534 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
545 #define NIR_DEST_INIT (nir_dest) { { { NULL } } }
547 #define nir_foreach_def(reg, dest) \
548 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
550 #define nir_foreach_def_safe(reg, dest) \
551 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
553 static inline nir_src
554 nir_src_for_ssa(nir_ssa_def
*def
)
556 nir_src src
= NIR_SRC_INIT
;
564 static inline nir_src
565 nir_src_for_reg(nir_register
*reg
)
567 nir_src src
= NIR_SRC_INIT
;
571 src
.reg
.indirect
= NULL
;
572 src
.reg
.base_offset
= 0;
577 static inline nir_dest
578 nir_dest_for_reg(nir_register
*reg
)
580 nir_dest dest
= NIR_DEST_INIT
;
587 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
588 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
594 * \name input modifiers
598 * For inputs interpreted as floating point, flips the sign bit. For
599 * inputs interpreted as integers, performs the two's complement negation.
604 * Clears the sign bit for floating point values, and computes the integer
605 * absolute value for integers. Note that the negate modifier acts after
606 * the absolute value modifier, therefore if both are set then all inputs
607 * will become negative.
613 * For each input component, says which component of the register it is
614 * chosen from. Note that which elements of the swizzle are used and which
615 * are ignored are based on the write mask for most opcodes - for example,
616 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
617 * a swizzle of {2, x, 1, 0} where x means "don't care."
626 * \name saturate output modifier
628 * Only valid for opcodes that output floating-point numbers. Clamps the
629 * output to between 0.0 and 1.0 inclusive.
634 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
638 nir_type_invalid
= 0, /* Not a valid type */
646 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
647 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
648 } nir_op_algebraic_property
;
656 * The number of components in the output
658 * If non-zero, this is the size of the output and input sizes are
659 * explicitly given; swizzle and writemask are still in effect, but if
660 * the output component is masked out, then the input component may
663 * If zero, the opcode acts in the standard, per-component manner; the
664 * operation is performed on each component (except the ones that are
665 * masked out) with the input being taken from the input swizzle for
668 * The size of some of the inputs may be given (i.e. non-zero) even
669 * though output_size is zero; in that case, the inputs with a zero
670 * size act per-component, while the inputs with non-zero size don't.
672 unsigned output_size
;
675 * The type of vector that the instruction outputs. Note that the
676 * staurate modifier is only allowed on outputs with the float type.
679 nir_alu_type output_type
;
682 * The number of components in each input
684 unsigned input_sizes
[4];
687 * The type of vector that each input takes. Note that negate and
688 * absolute value are only allowed on inputs with int or float type and
689 * behave differently on the two.
691 nir_alu_type input_types
[4];
693 nir_op_algebraic_property algebraic_properties
;
696 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
698 typedef struct nir_alu_instr
{
705 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
706 nir_alu_instr
*instr
);
707 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
708 nir_alu_instr
*instr
);
710 /* is this source channel used? */
712 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
714 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
715 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
717 return (instr
->dest
.write_mask
>> channel
) & 1;
721 * For instructions whose destinations are SSA, get the number of channels
724 static inline unsigned
725 nir_ssa_alu_instr_src_components(nir_alu_instr
*instr
, unsigned src
)
727 assert(instr
->dest
.dest
.is_ssa
);
729 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
730 return nir_op_infos
[instr
->op
].input_sizes
[src
];
732 return instr
->dest
.dest
.ssa
.num_components
;
737 nir_deref_type_array
,
738 nir_deref_type_struct
741 typedef struct nir_deref
{
742 nir_deref_type deref_type
;
743 struct nir_deref
*child
;
744 const struct glsl_type
*type
;
753 /* This enum describes how the array is referenced. If the deref is
754 * direct then the base_offset is used. If the deref is indirect then then
755 * offset is given by base_offset + indirect. If the deref is a wildcard
756 * then the deref refers to all of the elements of the array at the same
757 * time. Wildcard dereferences are only ever allowed in copy_var
758 * intrinsics and the source and destination derefs must have matching
762 nir_deref_array_type_direct
,
763 nir_deref_array_type_indirect
,
764 nir_deref_array_type_wildcard
,
765 } nir_deref_array_type
;
770 nir_deref_array_type deref_array_type
;
771 unsigned base_offset
;
781 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
782 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
783 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
789 nir_deref_var
**params
;
790 nir_deref_var
*return_deref
;
792 struct nir_function_overload
*callee
;
795 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
796 num_variables, num_indices, flags) \
797 nir_intrinsic_##name,
799 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
802 #include "nir_intrinsics.h"
803 nir_num_intrinsics
= nir_last_intrinsic
+ 1
807 #undef LAST_INTRINSIC
809 /** Represents an intrinsic
811 * An intrinsic is an instruction type for handling things that are
812 * more-or-less regular operations but don't just consume and produce SSA
813 * values like ALU operations do. Intrinsics are not for things that have
814 * special semantic meaning such as phi nodes and parallel copies.
815 * Examples of intrinsics include variable load/store operations, system
816 * value loads, and the like. Even though texturing more-or-less falls
817 * under this category, texturing is its own instruction type because
818 * trying to represent texturing with intrinsics would lead to a
819 * combinatorial explosion of intrinsic opcodes.
821 * By having a single instruction type for handling a lot of different
822 * cases, optimization passes can look for intrinsics and, for the most
823 * part, completely ignore them. Each intrinsic type also has a few
824 * possible flags that govern whether or not they can be reordered or
825 * eliminated. That way passes like dead code elimination can still work
826 * on intrisics without understanding the meaning of each.
828 * Each intrinsic has some number of constant indices, some number of
829 * variables, and some number of sources. What these sources, variables,
830 * and indices mean depends on the intrinsic and is documented with the
831 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
832 * instructions are the only types of instruction that can operate on
838 nir_intrinsic_op intrinsic
;
842 /** number of components if this is a vectorized intrinsic
844 * Similarly to ALU operations, some intrinsics are vectorized.
845 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
846 * For vectorized intrinsics, the num_components field specifies the
847 * number of destination components and the number of source components
848 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
850 uint8_t num_components
;
854 nir_deref_var
*variables
[2];
857 } nir_intrinsic_instr
;
860 * \name NIR intrinsics semantic flags
862 * information about what the compiler can do with the intrinsics.
864 * \sa nir_intrinsic_info::flags
868 * whether the intrinsic can be safely eliminated if none of its output
869 * value is not being used.
871 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
874 * Whether the intrinsic can be reordered with respect to any other
875 * intrinsic, i.e. whether the only reordering dependencies of the
876 * intrinsic are due to the register reads/writes.
878 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
879 } nir_intrinsic_semantic_flag
;
881 #define NIR_INTRINSIC_MAX_INPUTS 4
886 unsigned num_srcs
; /** < number of register/SSA inputs */
888 /** number of components of each input register
890 * If this value is 0, the number of components is given by the
891 * num_components field of nir_intrinsic_instr.
893 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
897 /** number of components of the output register
899 * If this value is 0, the number of components is given by the
900 * num_components field of nir_intrinsic_instr.
902 unsigned dest_components
;
904 /** the number of inputs/outputs that are variables */
905 unsigned num_variables
;
907 /** the number of constant indices used by the intrinsic */
908 unsigned num_indices
;
910 /** semantic flags for calls to this intrinsic */
911 nir_intrinsic_semantic_flag flags
;
912 } nir_intrinsic_info
;
914 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
917 * \group texture information
919 * This gives semantic information about textures which is useful to the
920 * frontend, the backend, and lowering passes, but not the optimizer.
925 nir_tex_src_projector
,
926 nir_tex_src_comparitor
, /* shadow comparitor */
930 nir_tex_src_ms_index
, /* MSAA sample index */
933 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
934 nir_num_tex_src_types
939 nir_tex_src_type src_type
;
943 nir_texop_tex
, /**< Regular texture look-up */
944 nir_texop_txb
, /**< Texture look-up with LOD bias */
945 nir_texop_txl
, /**< Texture look-up with explicit LOD */
946 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
947 nir_texop_txf
, /**< Texel fetch with explicit LOD */
948 nir_texop_txf_ms
, /**< Multisample texture fetch */
949 nir_texop_txs
, /**< Texture size */
950 nir_texop_lod
, /**< Texture lod query */
951 nir_texop_tg4
, /**< Texture gather */
952 nir_texop_query_levels
, /**< Texture levels query */
953 nir_texop_texture_samples
, /**< Texture samples query */
959 enum glsl_sampler_dim sampler_dim
;
960 nir_alu_type dest_type
;
965 unsigned num_srcs
, coord_components
;
966 bool is_array
, is_shadow
;
969 * If is_shadow is true, whether this is the old-style shadow that outputs 4
970 * components or the new-style shadow that outputs 1 component.
972 bool is_new_style_shadow
;
974 /* constant offset - must be 0 if the offset source is used */
977 /* gather component selector */
978 unsigned component
: 2;
980 /** The sampler index
982 * If this texture instruction has a nir_tex_src_sampler_offset source,
983 * then the sampler index is given by sampler_index + sampler_offset.
985 unsigned sampler_index
;
987 /** The size of the sampler array or 0 if it's not an array */
988 unsigned sampler_array_size
;
990 nir_deref_var
*sampler
; /* if this is NULL, use sampler_index instead */
993 static inline unsigned
994 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
997 case nir_texop_txs
: {
999 switch (instr
->sampler_dim
) {
1000 case GLSL_SAMPLER_DIM_1D
:
1001 case GLSL_SAMPLER_DIM_BUF
:
1004 case GLSL_SAMPLER_DIM_2D
:
1005 case GLSL_SAMPLER_DIM_CUBE
:
1006 case GLSL_SAMPLER_DIM_MS
:
1007 case GLSL_SAMPLER_DIM_RECT
:
1008 case GLSL_SAMPLER_DIM_EXTERNAL
:
1011 case GLSL_SAMPLER_DIM_3D
:
1015 unreachable("not reached");
1017 if (instr
->is_array
)
1025 case nir_texop_texture_samples
:
1026 case nir_texop_query_levels
:
1030 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1037 static inline unsigned
1038 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1040 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1041 return instr
->coord_components
;
1044 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1045 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1046 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1047 if (instr
->is_array
)
1048 return instr
->coord_components
- 1;
1050 return instr
->coord_components
;
1057 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1059 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1060 if (instr
->src
[i
].src_type
== type
)
1077 nir_const_value value
;
1080 } nir_load_const_instr
;
1093 /* creates a new SSA variable in an undefined state */
1098 } nir_ssa_undef_instr
;
1101 struct exec_node node
;
1103 /* The predecessor block corresponding to this source */
1104 struct nir_block
*pred
;
1109 #define nir_foreach_phi_src(phi, entry) \
1110 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1111 #define nir_foreach_phi_src_safe(phi, entry) \
1112 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1117 struct exec_list srcs
; /** < list of nir_phi_src */
1123 struct exec_node node
;
1126 } nir_parallel_copy_entry
;
1128 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1129 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1134 /* A list of nir_parallel_copy_entry's. The sources of all of the
1135 * entries are copied to the corresponding destinations "in parallel".
1136 * In other words, if we have two entries: a -> b and b -> a, the values
1139 struct exec_list entries
;
1140 } nir_parallel_copy_instr
;
1142 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1143 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1144 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1145 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1146 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1147 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1148 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1149 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1150 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1151 nir_parallel_copy_instr
, instr
)
1156 * Control flow consists of a tree of control flow nodes, which include
1157 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1158 * instructions that always run start-to-finish. Each basic block also keeps
1159 * track of its successors (blocks which may run immediately after the current
1160 * block) and predecessors (blocks which could have run immediately before the
1161 * current block). Each function also has a start block and an end block which
1162 * all return statements point to (which is always empty). Together, all the
1163 * blocks with their predecessors and successors make up the control flow
1164 * graph (CFG) of the function. There are helpers that modify the tree of
1165 * control flow nodes while modifying the CFG appropriately; these should be
1166 * used instead of modifying the tree directly.
1173 nir_cf_node_function
1176 typedef struct nir_cf_node
{
1177 struct exec_node node
;
1178 nir_cf_node_type type
;
1179 struct nir_cf_node
*parent
;
1182 typedef struct nir_block
{
1183 nir_cf_node cf_node
;
1185 struct exec_list instr_list
; /** < list of nir_instr */
1187 /** generic block index; generated by nir_index_blocks */
1191 * Each block can only have up to 2 successors, so we put them in a simple
1192 * array - no need for anything more complicated.
1194 struct nir_block
*successors
[2];
1196 /* Set of nir_block predecessors in the CFG */
1197 struct set
*predecessors
;
1200 * this node's immediate dominator in the dominance tree - set to NULL for
1203 struct nir_block
*imm_dom
;
1205 /* This node's children in the dominance tree */
1206 unsigned num_dom_children
;
1207 struct nir_block
**dom_children
;
1209 /* Set of nir_block's on the dominance frontier of this block */
1210 struct set
*dom_frontier
;
1213 * These two indices have the property that dom_{pre,post}_index for each
1214 * child of this block in the dominance tree will always be between
1215 * dom_pre_index and dom_post_index for this block, which makes testing if
1216 * a given block is dominated by another block an O(1) operation.
1218 unsigned dom_pre_index
, dom_post_index
;
1220 /* live in and out for this block; used for liveness analysis */
1221 BITSET_WORD
*live_in
;
1222 BITSET_WORD
*live_out
;
1225 static inline nir_instr
*
1226 nir_block_first_instr(nir_block
*block
)
1228 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1229 return exec_node_data(nir_instr
, head
, node
);
1232 static inline nir_instr
*
1233 nir_block_last_instr(nir_block
*block
)
1235 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1236 return exec_node_data(nir_instr
, tail
, node
);
1239 #define nir_foreach_instr(block, instr) \
1240 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1241 #define nir_foreach_instr_reverse(block, instr) \
1242 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1243 #define nir_foreach_instr_safe(block, instr) \
1244 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1245 #define nir_foreach_instr_safe_reverse(block, instr) \
1246 foreach_list_typed_safe_reverse(nir_instr, instr, node, &(block)->instr_list)
1248 typedef struct nir_if
{
1249 nir_cf_node cf_node
;
1252 struct exec_list then_list
; /** < list of nir_cf_node */
1253 struct exec_list else_list
; /** < list of nir_cf_node */
1256 static inline nir_cf_node
*
1257 nir_if_first_then_node(nir_if
*if_stmt
)
1259 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1260 return exec_node_data(nir_cf_node
, head
, node
);
1263 static inline nir_cf_node
*
1264 nir_if_last_then_node(nir_if
*if_stmt
)
1266 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1267 return exec_node_data(nir_cf_node
, tail
, node
);
1270 static inline nir_cf_node
*
1271 nir_if_first_else_node(nir_if
*if_stmt
)
1273 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1274 return exec_node_data(nir_cf_node
, head
, node
);
1277 static inline nir_cf_node
*
1278 nir_if_last_else_node(nir_if
*if_stmt
)
1280 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1281 return exec_node_data(nir_cf_node
, tail
, node
);
1285 nir_cf_node cf_node
;
1287 struct exec_list body
; /** < list of nir_cf_node */
1290 static inline nir_cf_node
*
1291 nir_loop_first_cf_node(nir_loop
*loop
)
1293 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1296 static inline nir_cf_node
*
1297 nir_loop_last_cf_node(nir_loop
*loop
)
1299 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1303 * Various bits of metadata that can may be created or required by
1304 * optimization and analysis passes
1307 nir_metadata_none
= 0x0,
1308 nir_metadata_block_index
= 0x1,
1309 nir_metadata_dominance
= 0x2,
1310 nir_metadata_live_variables
= 0x4,
1314 nir_cf_node cf_node
;
1316 /** pointer to the overload of which this is an implementation */
1317 struct nir_function_overload
*overload
;
1319 struct exec_list body
; /** < list of nir_cf_node */
1321 nir_block
*end_block
;
1323 /** list for all local variables in the function */
1324 struct exec_list locals
;
1326 /** array of variables used as parameters */
1327 unsigned num_params
;
1328 nir_variable
**params
;
1330 /** variable used to hold the result of the function */
1331 nir_variable
*return_var
;
1333 /** list of local registers in the function */
1334 struct exec_list registers
;
1336 /** next available local register index */
1339 /** next available SSA value index */
1342 /* total number of basic blocks, only valid when block_index_dirty = false */
1343 unsigned num_blocks
;
1345 nir_metadata valid_metadata
;
1346 } nir_function_impl
;
1348 static inline nir_block
*
1349 nir_start_block(nir_function_impl
*impl
)
1351 return (nir_block
*) exec_list_get_head(&impl
->body
);
1354 static inline nir_cf_node
*
1355 nir_cf_node_next(nir_cf_node
*node
)
1357 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1358 if (exec_node_is_tail_sentinel(next
))
1361 return exec_node_data(nir_cf_node
, next
, node
);
1364 static inline nir_cf_node
*
1365 nir_cf_node_prev(nir_cf_node
*node
)
1367 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1368 if (exec_node_is_head_sentinel(prev
))
1371 return exec_node_data(nir_cf_node
, prev
, node
);
1375 nir_cf_node_is_first(const nir_cf_node
*node
)
1377 return exec_node_is_head_sentinel(node
->node
.prev
);
1381 nir_cf_node_is_last(const nir_cf_node
*node
)
1383 return exec_node_is_tail_sentinel(node
->node
.next
);
1386 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1387 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1388 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1389 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1394 nir_parameter_inout
,
1395 } nir_parameter_type
;
1398 nir_parameter_type param_type
;
1399 const struct glsl_type
*type
;
1402 typedef struct nir_function_overload
{
1403 struct exec_node node
;
1405 unsigned num_params
;
1406 nir_parameter
*params
;
1407 const struct glsl_type
*return_type
;
1409 nir_function_impl
*impl
; /** < NULL if the overload is only declared yet */
1411 /** pointer to the function of which this is an overload */
1412 struct nir_function
*function
;
1413 } nir_function_overload
;
1415 typedef struct nir_function
{
1416 struct exec_node node
;
1418 struct exec_list overload_list
; /** < list of nir_function_overload */
1420 struct nir_shader
*shader
;
1423 #define nir_function_first_overload(func) \
1424 exec_node_data(nir_function_overload, \
1425 exec_list_get_head(&(func)->overload_list), node)
1427 typedef struct nir_shader_compiler_options
{
1433 /** lowers fneg and ineg to fsub and isub. */
1435 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1438 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1441 /* Does the native fdot instruction replicate its result for four
1442 * components? If so, then opt_algebraic_late will turn all fdotN
1443 * instructions into fdot_replicatedN instructions.
1445 bool fdot_replicates
;
1447 /** lowers ffract to fsub+ffloor: */
1451 * Does the driver support real 32-bit integers? (Otherwise, integers
1452 * are simulated by floats.)
1454 bool native_integers
;
1455 } nir_shader_compiler_options
;
1457 typedef struct nir_shader
{
1458 /** hash table of name -> uniform nir_variable */
1459 struct exec_list uniforms
;
1461 /** hash table of name -> input nir_variable */
1462 struct exec_list inputs
;
1464 /** hash table of name -> output nir_variable */
1465 struct exec_list outputs
;
1467 /** Set of driver-specific options for the shader.
1469 * The memory for the options is expected to be kept in a single static
1470 * copy by the driver.
1472 const struct nir_shader_compiler_options
*options
;
1474 /** list of global variables in the shader */
1475 struct exec_list globals
;
1477 /** list of system value variables in the shader */
1478 struct exec_list system_values
;
1480 struct exec_list functions
; /** < list of nir_function */
1482 /** list of global register in the shader */
1483 struct exec_list registers
;
1485 /** next available global register index */
1489 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1492 unsigned num_inputs
, num_uniforms
, num_outputs
;
1494 /** The shader stage, such as MESA_SHADER_VERTEX. */
1495 gl_shader_stage stage
;
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 #define nir_foreach_overload(shader, overload) \
1507 foreach_list_typed(nir_function, func, node, &(shader)->functions) \
1508 foreach_list_typed(nir_function_overload, overload, node, \
1509 &(func)->overload_list)
1511 nir_shader
*nir_shader_create(void *mem_ctx
,
1512 gl_shader_stage stage
,
1513 const nir_shader_compiler_options
*options
);
1515 /** creates a register, including assigning it an index and adding it to the list */
1516 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1518 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1520 void nir_reg_remove(nir_register
*reg
);
1522 /** creates a function and adds it to the shader's list of functions */
1523 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1525 /** creates a null function returning null */
1526 nir_function_overload
*nir_function_overload_create(nir_function
*func
);
1528 nir_function_impl
*nir_function_impl_create(nir_function_overload
*func
);
1530 nir_block
*nir_block_create(void *mem_ctx
);
1531 nir_if
*nir_if_create(void *mem_ctx
);
1532 nir_loop
*nir_loop_create(void *mem_ctx
);
1534 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1536 /** requests that the given pieces of metadata be generated */
1537 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1538 /** dirties all but the preserved metadata */
1539 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1541 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1542 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1544 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1546 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1547 unsigned num_components
);
1549 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1550 nir_intrinsic_op op
);
1552 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1553 nir_function_overload
*callee
);
1555 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1557 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1559 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1561 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1562 unsigned num_components
);
1564 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1565 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1566 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1568 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1570 nir_load_const_instr
*
1571 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1574 * NIR Cursors and Instruction Insertion API
1577 * A tiny struct representing a point to insert/extract instructions or
1578 * control flow nodes. Helps reduce the combinatorial explosion of possible
1579 * points to insert/extract.
1581 * \sa nir_control_flow.h
1584 nir_cursor_before_block
,
1585 nir_cursor_after_block
,
1586 nir_cursor_before_instr
,
1587 nir_cursor_after_instr
,
1588 } nir_cursor_option
;
1591 nir_cursor_option option
;
1598 static inline nir_cursor
1599 nir_before_block(nir_block
*block
)
1602 cursor
.option
= nir_cursor_before_block
;
1603 cursor
.block
= block
;
1607 static inline nir_cursor
1608 nir_after_block(nir_block
*block
)
1611 cursor
.option
= nir_cursor_after_block
;
1612 cursor
.block
= block
;
1616 static inline nir_cursor
1617 nir_before_instr(nir_instr
*instr
)
1620 cursor
.option
= nir_cursor_before_instr
;
1621 cursor
.instr
= instr
;
1625 static inline nir_cursor
1626 nir_after_instr(nir_instr
*instr
)
1629 cursor
.option
= nir_cursor_after_instr
;
1630 cursor
.instr
= instr
;
1634 static inline nir_cursor
1635 nir_after_block_before_jump(nir_block
*block
)
1637 nir_instr
*last_instr
= nir_block_last_instr(block
);
1638 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1639 return nir_before_instr(last_instr
);
1641 return nir_after_block(block
);
1645 static inline nir_cursor
1646 nir_before_cf_node(nir_cf_node
*node
)
1648 if (node
->type
== nir_cf_node_block
)
1649 return nir_before_block(nir_cf_node_as_block(node
));
1651 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1654 static inline nir_cursor
1655 nir_after_cf_node(nir_cf_node
*node
)
1657 if (node
->type
== nir_cf_node_block
)
1658 return nir_after_block(nir_cf_node_as_block(node
));
1660 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1663 static inline nir_cursor
1664 nir_before_cf_list(struct exec_list
*cf_list
)
1666 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1667 exec_list_get_head(cf_list
), node
);
1668 return nir_before_cf_node(first_node
);
1671 static inline nir_cursor
1672 nir_after_cf_list(struct exec_list
*cf_list
)
1674 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1675 exec_list_get_tail(cf_list
), node
);
1676 return nir_after_cf_node(last_node
);
1680 * Insert a NIR instruction at the given cursor.
1682 * Note: This does not update the cursor.
1684 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
1687 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1689 nir_instr_insert(nir_before_instr(instr
), before
);
1693 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
1695 nir_instr_insert(nir_after_instr(instr
), after
);
1699 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
1701 nir_instr_insert(nir_before_block(block
), before
);
1705 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
1707 nir_instr_insert(nir_after_block(block
), after
);
1711 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
1713 nir_instr_insert(nir_before_cf_node(node
), before
);
1717 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
1719 nir_instr_insert(nir_after_cf_node(node
), after
);
1723 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
1725 nir_instr_insert(nir_before_cf_list(list
), before
);
1729 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
1731 nir_instr_insert(nir_after_cf_list(list
), after
);
1734 void nir_instr_remove(nir_instr
*instr
);
1738 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
1739 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
1740 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
1741 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
1743 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
1744 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
1746 nir_const_value
*nir_src_as_const_value(nir_src src
);
1747 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
1748 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
1749 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
1750 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
1751 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
1754 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1755 unsigned num_components
, const char *name
);
1756 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1757 unsigned num_components
, const char *name
);
1758 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
1760 /* visits basic blocks in source-code order */
1761 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
1762 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1764 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1766 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1769 /* If the following CF node is an if, this function returns that if.
1770 * Otherwise, it returns NULL.
1772 nir_if
*nir_block_get_following_if(nir_block
*block
);
1774 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
1776 void nir_index_local_regs(nir_function_impl
*impl
);
1777 void nir_index_global_regs(nir_shader
*shader
);
1778 void nir_index_ssa_defs(nir_function_impl
*impl
);
1779 unsigned nir_index_instrs(nir_function_impl
*impl
);
1781 void nir_index_blocks(nir_function_impl
*impl
);
1783 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
1784 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
1787 void nir_validate_shader(nir_shader
*shader
);
1789 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
1792 void nir_calc_dominance_impl(nir_function_impl
*impl
);
1793 void nir_calc_dominance(nir_shader
*shader
);
1795 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
1796 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
1798 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
1799 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
1801 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
1802 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
1804 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
1805 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
1807 void nir_split_var_copies(nir_shader
*shader
);
1809 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
1810 void nir_lower_var_copies(nir_shader
*shader
);
1812 void nir_lower_global_vars_to_local(nir_shader
*shader
);
1814 void nir_lower_locals_to_regs(nir_shader
*shader
);
1816 void nir_lower_outputs_to_temporaries(nir_shader
*shader
);
1818 void nir_assign_var_locations(struct exec_list
*var_list
,
1820 int (*type_size
)(const struct glsl_type
*));
1822 void nir_lower_io(nir_shader
*shader
,
1823 int (*type_size
)(const struct glsl_type
*));
1824 void nir_lower_vars_to_ssa(nir_shader
*shader
);
1826 void nir_remove_dead_variables(nir_shader
*shader
);
1828 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
1829 void nir_lower_vec_to_movs(nir_shader
*shader
);
1830 void nir_lower_alu_to_scalar(nir_shader
*shader
);
1831 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
1833 void nir_lower_phis_to_scalar(nir_shader
*shader
);
1835 void nir_lower_samplers(nir_shader
*shader
,
1836 const struct gl_shader_program
*shader_program
);
1838 void nir_lower_system_values(nir_shader
*shader
);
1839 void nir_lower_tex_projector(nir_shader
*shader
);
1840 void nir_lower_idiv(nir_shader
*shader
);
1842 void nir_lower_atomics(nir_shader
*shader
);
1843 void nir_lower_to_source_mods(nir_shader
*shader
);
1845 void nir_normalize_cubemap_coords(nir_shader
*shader
);
1847 void nir_live_variables_impl(nir_function_impl
*impl
);
1848 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
1850 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
1851 void nir_convert_to_ssa(nir_shader
*shader
);
1853 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
1854 * registers. If false, convert all values (even those not involved in a phi
1855 * node) to registers.
1857 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
1859 bool nir_opt_algebraic(nir_shader
*shader
);
1860 bool nir_opt_algebraic_late(nir_shader
*shader
);
1861 bool nir_opt_constant_folding(nir_shader
*shader
);
1863 bool nir_opt_global_to_local(nir_shader
*shader
);
1865 bool nir_copy_prop_impl(nir_function_impl
*impl
);
1866 bool nir_copy_prop(nir_shader
*shader
);
1868 bool nir_opt_cse(nir_shader
*shader
);
1870 bool nir_opt_dce_impl(nir_function_impl
*impl
);
1871 bool nir_opt_dce(nir_shader
*shader
);
1873 bool nir_opt_dead_cf(nir_shader
*shader
);
1875 void nir_opt_gcm(nir_shader
*shader
);
1877 bool nir_opt_peephole_select(nir_shader
*shader
);
1878 bool nir_opt_peephole_ffma(nir_shader
*shader
);
1880 bool nir_opt_remove_phis(nir_shader
*shader
);
1882 bool nir_opt_undef(nir_shader
*shader
);
1884 void nir_sweep(nir_shader
*shader
);
1886 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);