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
94 * Data stored in an nir_constant
96 union nir_constant_data
{
103 typedef struct nir_constant
{
105 * Value of the constant.
107 * The field used to back the values supplied by the constant is determined
108 * by the type associated with the \c nir_variable. Constants may be
109 * scalars, vectors, or matrices.
111 union nir_constant_data value
;
113 /* Array elements / Structure Fields */
114 struct nir_constant
**elements
;
118 * \brief Layout qualifiers for gl_FragDepth.
120 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
121 * with a layout qualifier.
124 nir_depth_layout_none
, /**< No depth layout is specified. */
125 nir_depth_layout_any
,
126 nir_depth_layout_greater
,
127 nir_depth_layout_less
,
128 nir_depth_layout_unchanged
132 * Either a uniform, global variable, shader input, or shader output. Based on
133 * ir_variable - it should be easy to translate between the two.
137 struct exec_node node
;
140 * Declared type of the variable
142 const struct glsl_type
*type
;
145 * Declared name of the variable
150 * For variables which satisfy the is_interface_instance() predicate, this
151 * points to an array of integers such that if the ith member of the
152 * interface block is an array, max_ifc_array_access[i] is the maximum
153 * array element of that member that has been accessed. If the ith member
154 * of the interface block is not an array, max_ifc_array_access[i] is
157 * For variables whose type is not an interface block, this pointer is
160 unsigned *max_ifc_array_access
;
162 struct nir_variable_data
{
165 * Is the variable read-only?
167 * This is set for variables declared as \c const, shader inputs,
170 unsigned read_only
:1;
173 unsigned invariant
:1;
176 * Storage class of the variable.
178 * \sa nir_variable_mode
180 nir_variable_mode mode
:4;
183 * Interpolation mode for shader inputs / outputs
185 * \sa glsl_interp_qualifier
187 unsigned interpolation
:2;
190 * \name ARB_fragment_coord_conventions
193 unsigned origin_upper_left
:1;
194 unsigned pixel_center_integer
:1;
198 * Was the location explicitly set in the shader?
200 * If the location is explicitly set in the shader, it \b cannot be changed
201 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
204 unsigned explicit_location
:1;
205 unsigned explicit_index
:1;
208 * Was an initial binding explicitly set in the shader?
210 * If so, constant_initializer contains an integer nir_constant
211 * representing the initial binding point.
213 unsigned explicit_binding
:1;
216 * Does this variable have an initializer?
218 * This is used by the linker to cross-validiate initializers of global
221 unsigned has_initializer
:1;
224 * Is this variable a generic output or input that has not yet been matched
225 * up to a variable in another stage of the pipeline?
227 * This is used by the linker as scratch storage while assigning locations
228 * to generic inputs and outputs.
230 unsigned is_unmatched_generic_inout
:1;
233 * If non-zero, then this variable may be packed along with other variables
234 * into a single varying slot, so this offset should be applied when
235 * accessing components. For example, an offset of 1 means that the x
236 * component of this variable is actually stored in component y of the
237 * location specified by \c location.
239 unsigned location_frac
:2;
242 * Non-zero if this variable was created by lowering a named interface
243 * block which was not an array.
245 * Note that this variable and \c from_named_ifc_block_array will never
248 unsigned from_named_ifc_block_nonarray
:1;
251 * Non-zero if this variable was created by lowering a named interface
252 * block which was an array.
254 * Note that this variable and \c from_named_ifc_block_nonarray will never
257 unsigned from_named_ifc_block_array
:1;
260 * \brief Layout qualifier for gl_FragDepth.
262 * This is not equal to \c ir_depth_layout_none if and only if this
263 * variable is \c gl_FragDepth and a layout qualifier is specified.
265 nir_depth_layout depth_layout
;
268 * Storage location of the base of this variable
270 * The precise meaning of this field depends on the nature of the variable.
272 * - Vertex shader input: one of the values from \c gl_vert_attrib.
273 * - Vertex shader output: one of the values from \c gl_varying_slot.
274 * - Geometry shader input: one of the values from \c gl_varying_slot.
275 * - Geometry shader output: one of the values from \c gl_varying_slot.
276 * - Fragment shader input: one of the values from \c gl_varying_slot.
277 * - Fragment shader output: one of the values from \c gl_frag_result.
278 * - Uniforms: Per-stage uniform slot number for default uniform block.
279 * - Uniforms: Index within the uniform block definition for UBO members.
280 * - Other: This field is not currently used.
282 * If the variable is a uniform, shader input, or shader output, and the
283 * slot has not been assigned, the value will be -1.
288 * The actual location of the variable in the IR. Only valid for inputs
291 unsigned int driver_location
;
294 * output index for dual source blending.
299 * Initial binding point for a sampler or UBO.
301 * For array types, this represents the binding point for the first element.
306 * Location an atomic counter is stored at.
309 unsigned buffer_index
;
314 * ARB_shader_image_load_store qualifiers.
317 bool read_only
; /**< "readonly" qualifier. */
318 bool write_only
; /**< "writeonly" qualifier. */
323 /** Image internal format if specified explicitly, otherwise GL_NONE. */
328 * Highest element accessed with a constant expression array index
330 * Not used for non-array variables.
332 unsigned max_array_access
;
337 * Built-in state that backs this uniform
339 * Once set at variable creation, \c state_slots must remain invariant.
340 * This is because, ideally, this array would be shared by all clones of
341 * this variable in the IR tree. In other words, we'd really like for it
342 * to be a fly-weight.
344 * If the variable is not a uniform, \c num_state_slots will be zero and
345 * \c state_slots will be \c NULL.
348 unsigned num_state_slots
; /**< Number of state slots used */
349 nir_state_slot
*state_slots
; /**< State descriptors. */
353 * Constant expression assigned in the initializer of the variable
355 nir_constant
*constant_initializer
;
358 * For variables that are in an interface block or are an instance of an
359 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
361 * \sa ir_variable::location
363 const struct glsl_type
*interface_type
;
367 struct exec_node node
;
369 unsigned num_components
; /** < number of vector components */
370 unsigned num_array_elems
; /** < size of array (0 for no array) */
372 /** generic register index. */
375 /** only for debug purposes, can be NULL */
378 /** whether this register is local (per-function) or global (per-shader) */
382 * If this flag is set to true, then accessing channels >= num_components
383 * is well-defined, and simply spills over to the next array element. This
384 * is useful for backends that can do per-component accessing, in
385 * particular scalar backends. By setting this flag and making
386 * num_components equal to 1, structures can be packed tightly into
387 * registers and then registers can be accessed per-component to get to
388 * each structure member, even if it crosses vec4 boundaries.
393 * If this pointer is non-NULL then this register has exactly one
394 * definition and that definition dominates all of its uses. This is
395 * set by the out-of-SSA pass so that backends can get SSA-like
396 * information even once they have gone out of SSA.
398 struct nir_instr
*parent_instr
;
400 /** set of nir_instr's where this register is used (read from) */
401 struct list_head uses
;
403 /** set of nir_instr's where this register is defined (written to) */
404 struct list_head defs
;
406 /** set of nir_if's where this register is used as a condition */
407 struct list_head if_uses
;
414 nir_instr_type_intrinsic
,
415 nir_instr_type_load_const
,
417 nir_instr_type_ssa_undef
,
419 nir_instr_type_parallel_copy
,
422 typedef struct nir_instr
{
423 struct exec_node node
;
425 struct nir_block
*block
;
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 /** for debugging only, can be NULL */
457 /** generic SSA definition index. */
460 /** Index into the live_in and live_out bitfields */
463 nir_instr
*parent_instr
;
465 /** set of nir_instr's where this register is used (read from) */
466 struct list_head uses
;
468 /** set of nir_if's where this register is used as a condition */
469 struct list_head if_uses
;
471 uint8_t num_components
;
478 struct nir_src
*indirect
; /** < NULL for no indirect offset */
479 unsigned base_offset
;
481 /* TODO use-def chain goes here */
485 nir_instr
*parent_instr
;
486 struct list_head def_link
;
489 struct nir_src
*indirect
; /** < NULL for no indirect offset */
490 unsigned base_offset
;
492 /* TODO def-use chain goes here */
497 typedef struct nir_src
{
499 nir_instr
*parent_instr
;
500 struct nir_if
*parent_if
;
503 struct list_head use_link
;
513 #define NIR_SRC_INIT (nir_src) { { NULL } }
515 #define nir_foreach_use(reg_or_ssa_def, src) \
516 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
518 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
519 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
521 #define nir_foreach_if_use(reg_or_ssa_def, src) \
522 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
524 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
525 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
536 #define NIR_DEST_INIT (nir_dest) { { { NULL } } }
538 #define nir_foreach_def(reg, dest) \
539 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
541 #define nir_foreach_def_safe(reg, dest) \
542 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
544 static inline nir_src
545 nir_src_for_ssa(nir_ssa_def
*def
)
547 nir_src src
= NIR_SRC_INIT
;
555 static inline nir_src
556 nir_src_for_reg(nir_register
*reg
)
558 nir_src src
= NIR_SRC_INIT
;
562 src
.reg
.indirect
= NULL
;
563 src
.reg
.base_offset
= 0;
568 static inline nir_instr
*
569 nir_src_get_parent_instr(const nir_src
*src
)
572 return src
->ssa
->parent_instr
;
574 return src
->reg
.reg
->parent_instr
;
578 static inline nir_dest
579 nir_dest_for_reg(nir_register
*reg
)
581 nir_dest dest
= NIR_DEST_INIT
;
588 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *mem_ctx
);
589 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, void *mem_ctx
);
595 * \name input modifiers
599 * For inputs interpreted as floating point, flips the sign bit. For
600 * inputs interpreted as integers, performs the two's complement negation.
605 * Clears the sign bit for floating point values, and computes the integer
606 * absolute value for integers. Note that the negate modifier acts after
607 * the absolute value modifier, therefore if both are set then all inputs
608 * will become negative.
614 * For each input component, says which component of the register it is
615 * chosen from. Note that which elements of the swizzle are used and which
616 * are ignored are based on the write mask for most opcodes - for example,
617 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
618 * a swizzle of {2, x, 1, 0} where x means "don't care."
627 * \name saturate output modifier
629 * Only valid for opcodes that output floating-point numbers. Clamps the
630 * output to between 0.0 and 1.0 inclusive.
635 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
638 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
, void *mem_ctx
);
639 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
643 nir_type_invalid
= 0, /* Not a valid type */
651 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
652 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
653 } nir_op_algebraic_property
;
661 * The number of components in the output
663 * If non-zero, this is the size of the output and input sizes are
664 * explicitly given; swizzle and writemask are still in effect, but if
665 * the output component is masked out, then the input component may
668 * If zero, the opcode acts in the standard, per-component manner; the
669 * operation is performed on each component (except the ones that are
670 * masked out) with the input being taken from the input swizzle for
673 * The size of some of the inputs may be given (i.e. non-zero) even
674 * though output_size is zero; in that case, the inputs with a zero
675 * size act per-component, while the inputs with non-zero size don't.
677 unsigned output_size
;
680 * The type of vector that the instruction outputs. Note that the
681 * staurate modifier is only allowed on outputs with the float type.
684 nir_alu_type output_type
;
687 * The number of components in each input
689 unsigned input_sizes
[4];
692 * The type of vector that each input takes. Note that negate and
693 * absolute value are only allowed on inputs with int or float type and
694 * behave differently on the two.
696 nir_alu_type input_types
[4];
698 nir_op_algebraic_property algebraic_properties
;
701 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
703 typedef struct nir_alu_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 */
958 enum glsl_sampler_dim sampler_dim
;
959 nir_alu_type dest_type
;
964 unsigned num_srcs
, coord_components
;
965 bool is_array
, is_shadow
;
968 * If is_shadow is true, whether this is the old-style shadow that outputs 4
969 * components or the new-style shadow that outputs 1 component.
971 bool is_new_style_shadow
;
973 /* constant offset - must be 0 if the offset source is used */
976 /* gather component selector */
977 unsigned component
: 2;
979 /** The sampler index
981 * If this texture instruction has a nir_tex_src_sampler_offset source,
982 * then the sampler index is given by sampler_index + sampler_offset.
984 unsigned sampler_index
;
986 /** The size of the sampler array or 0 if it's not an array */
987 unsigned sampler_array_size
;
989 nir_deref_var
*sampler
; /* if this is NULL, use sampler_index instead */
992 static inline unsigned
993 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
996 case nir_texop_txs
: {
998 switch (instr
->sampler_dim
) {
999 case GLSL_SAMPLER_DIM_1D
:
1000 case GLSL_SAMPLER_DIM_BUF
:
1003 case GLSL_SAMPLER_DIM_2D
:
1004 case GLSL_SAMPLER_DIM_CUBE
:
1005 case GLSL_SAMPLER_DIM_MS
:
1006 case GLSL_SAMPLER_DIM_RECT
:
1007 case GLSL_SAMPLER_DIM_EXTERNAL
:
1010 case GLSL_SAMPLER_DIM_3D
:
1014 unreachable("not reached");
1016 if (instr
->is_array
)
1024 case nir_texop_query_levels
:
1028 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1035 static inline unsigned
1036 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1038 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1039 return instr
->coord_components
;
1042 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1043 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1044 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1045 if (instr
->is_array
)
1046 return instr
->coord_components
- 1;
1048 return instr
->coord_components
;
1055 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1057 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1058 if (instr
->src
[i
].src_type
== type
)
1075 nir_const_value value
;
1078 } nir_load_const_instr
;
1091 /* creates a new SSA variable in an undefined state */
1096 } nir_ssa_undef_instr
;
1099 struct exec_node node
;
1101 /* The predecessor block corresponding to this source */
1102 struct nir_block
*pred
;
1107 #define nir_foreach_phi_src(phi, entry) \
1108 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1113 struct exec_list srcs
; /** < list of nir_phi_src */
1119 struct exec_node node
;
1122 } nir_parallel_copy_entry
;
1124 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1125 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1130 /* A list of nir_parallel_copy_entry's. The sources of all of the
1131 * entries are copied to the corresponding destinations "in parallel".
1132 * In other words, if we have two entries: a -> b and b -> a, the values
1135 struct exec_list entries
;
1136 } nir_parallel_copy_instr
;
1138 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1139 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1140 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1141 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1142 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1143 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1144 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1145 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1146 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1147 nir_parallel_copy_instr
, instr
)
1152 * Control flow consists of a tree of control flow nodes, which include
1153 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1154 * instructions that always run start-to-finish. Each basic block also keeps
1155 * track of its successors (blocks which may run immediately after the current
1156 * block) and predecessors (blocks which could have run immediately before the
1157 * current block). Each function also has a start block and an end block which
1158 * all return statements point to (which is always empty). Together, all the
1159 * blocks with their predecessors and successors make up the control flow
1160 * graph (CFG) of the function. There are helpers that modify the tree of
1161 * control flow nodes while modifying the CFG appropriately; these should be
1162 * used instead of modifying the tree directly.
1169 nir_cf_node_function
1172 typedef struct nir_cf_node
{
1173 struct exec_node node
;
1174 nir_cf_node_type type
;
1175 struct nir_cf_node
*parent
;
1178 typedef struct nir_block
{
1179 nir_cf_node cf_node
;
1181 struct exec_list instr_list
; /** < list of nir_instr */
1183 /** generic block index; generated by nir_index_blocks */
1187 * Each block can only have up to 2 successors, so we put them in a simple
1188 * array - no need for anything more complicated.
1190 struct nir_block
*successors
[2];
1192 /* Set of nir_block predecessors in the CFG */
1193 struct set
*predecessors
;
1196 * this node's immediate dominator in the dominance tree - set to NULL for
1199 struct nir_block
*imm_dom
;
1201 /* This node's children in the dominance tree */
1202 unsigned num_dom_children
;
1203 struct nir_block
**dom_children
;
1205 /* Set of nir_block's on the dominance frontier of this block */
1206 struct set
*dom_frontier
;
1209 * These two indices have the property that dom_{pre,post}_index for each
1210 * child of this block in the dominance tree will always be between
1211 * dom_pre_index and dom_post_index for this block, which makes testing if
1212 * a given block is dominated by another block an O(1) operation.
1214 unsigned dom_pre_index
, dom_post_index
;
1216 /* live in and out for this block; used for liveness analysis */
1217 BITSET_WORD
*live_in
;
1218 BITSET_WORD
*live_out
;
1221 static inline nir_instr
*
1222 nir_block_first_instr(nir_block
*block
)
1224 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1225 return exec_node_data(nir_instr
, head
, node
);
1228 static inline nir_instr
*
1229 nir_block_last_instr(nir_block
*block
)
1231 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1232 return exec_node_data(nir_instr
, tail
, node
);
1235 #define nir_foreach_instr(block, instr) \
1236 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1237 #define nir_foreach_instr_reverse(block, instr) \
1238 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1239 #define nir_foreach_instr_safe(block, instr) \
1240 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1242 typedef struct nir_if
{
1243 nir_cf_node cf_node
;
1246 struct exec_list then_list
; /** < list of nir_cf_node */
1247 struct exec_list else_list
; /** < list of nir_cf_node */
1250 static inline nir_cf_node
*
1251 nir_if_first_then_node(nir_if
*if_stmt
)
1253 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1254 return exec_node_data(nir_cf_node
, head
, node
);
1257 static inline nir_cf_node
*
1258 nir_if_last_then_node(nir_if
*if_stmt
)
1260 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1261 return exec_node_data(nir_cf_node
, tail
, node
);
1264 static inline nir_cf_node
*
1265 nir_if_first_else_node(nir_if
*if_stmt
)
1267 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1268 return exec_node_data(nir_cf_node
, head
, node
);
1271 static inline nir_cf_node
*
1272 nir_if_last_else_node(nir_if
*if_stmt
)
1274 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1275 return exec_node_data(nir_cf_node
, tail
, node
);
1279 nir_cf_node cf_node
;
1281 struct exec_list body
; /** < list of nir_cf_node */
1284 static inline nir_cf_node
*
1285 nir_loop_first_cf_node(nir_loop
*loop
)
1287 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1290 static inline nir_cf_node
*
1291 nir_loop_last_cf_node(nir_loop
*loop
)
1293 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1297 * Various bits of metadata that can may be created or required by
1298 * optimization and analysis passes
1301 nir_metadata_none
= 0x0,
1302 nir_metadata_block_index
= 0x1,
1303 nir_metadata_dominance
= 0x2,
1304 nir_metadata_live_variables
= 0x4,
1308 nir_cf_node cf_node
;
1310 /** pointer to the overload of which this is an implementation */
1311 struct nir_function_overload
*overload
;
1313 struct exec_list body
; /** < list of nir_cf_node */
1315 nir_block
*start_block
, *end_block
;
1317 /** list for all local variables in the function */
1318 struct exec_list locals
;
1320 /** array of variables used as parameters */
1321 unsigned num_params
;
1322 nir_variable
**params
;
1324 /** variable used to hold the result of the function */
1325 nir_variable
*return_var
;
1327 /** list of local registers in the function */
1328 struct exec_list registers
;
1330 /** next available local register index */
1333 /** next available SSA value index */
1336 /* total number of basic blocks, only valid when block_index_dirty = false */
1337 unsigned num_blocks
;
1339 nir_metadata valid_metadata
;
1340 } nir_function_impl
;
1342 static inline nir_cf_node
*
1343 nir_cf_node_next(nir_cf_node
*node
)
1345 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1346 if (exec_node_is_tail_sentinel(next
))
1349 return exec_node_data(nir_cf_node
, next
, node
);
1352 static inline nir_cf_node
*
1353 nir_cf_node_prev(nir_cf_node
*node
)
1355 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1356 if (exec_node_is_head_sentinel(prev
))
1359 return exec_node_data(nir_cf_node
, prev
, node
);
1363 nir_cf_node_is_first(const nir_cf_node
*node
)
1365 return exec_node_is_head_sentinel(node
->node
.prev
);
1369 nir_cf_node_is_last(const nir_cf_node
*node
)
1371 return exec_node_is_tail_sentinel(node
->node
.next
);
1374 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1375 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1376 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1377 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1382 nir_parameter_inout
,
1383 } nir_parameter_type
;
1386 nir_parameter_type param_type
;
1387 const struct glsl_type
*type
;
1390 typedef struct nir_function_overload
{
1391 struct exec_node node
;
1393 unsigned num_params
;
1394 nir_parameter
*params
;
1395 const struct glsl_type
*return_type
;
1397 nir_function_impl
*impl
; /** < NULL if the overload is only declared yet */
1399 /** pointer to the function of which this is an overload */
1400 struct nir_function
*function
;
1401 } nir_function_overload
;
1403 typedef struct nir_function
{
1404 struct exec_node node
;
1406 struct exec_list overload_list
; /** < list of nir_function_overload */
1408 struct nir_shader
*shader
;
1411 #define nir_function_first_overload(func) \
1412 exec_node_data(nir_function_overload, \
1413 exec_list_get_head(&(func)->overload_list), node)
1415 typedef struct nir_shader_compiler_options
{
1421 /** lowers fneg and ineg to fsub and isub. */
1423 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1426 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1430 * Does the driver support real 32-bit integers? (Otherwise, integers
1431 * are simulated by floats.)
1433 bool native_integers
;
1434 } nir_shader_compiler_options
;
1436 typedef struct nir_shader
{
1437 /** hash table of name -> uniform nir_variable */
1438 struct exec_list uniforms
;
1440 /** hash table of name -> input nir_variable */
1441 struct exec_list inputs
;
1443 /** hash table of name -> output nir_variable */
1444 struct exec_list outputs
;
1446 /** Set of driver-specific options for the shader.
1448 * The memory for the options is expected to be kept in a single static
1449 * copy by the driver.
1451 const struct nir_shader_compiler_options
*options
;
1453 /** list of global variables in the shader */
1454 struct exec_list globals
;
1456 /** list of system value variables in the shader */
1457 struct exec_list system_values
;
1459 struct exec_list functions
; /** < list of nir_function */
1461 /** list of global register in the shader */
1462 struct exec_list registers
;
1464 /** next available global register index */
1468 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1471 unsigned num_inputs
, num_uniforms
, num_outputs
;
1473 /** the number of uniforms that are only accessed directly */
1474 unsigned num_direct_uniforms
;
1477 #define nir_foreach_overload(shader, overload) \
1478 foreach_list_typed(nir_function, func, node, &(shader)->functions) \
1479 foreach_list_typed(nir_function_overload, overload, node, \
1480 &(func)->overload_list)
1482 nir_shader
*nir_shader_create(void *mem_ctx
,
1483 const nir_shader_compiler_options
*options
);
1485 /** creates a register, including assigning it an index and adding it to the list */
1486 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1488 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1490 void nir_reg_remove(nir_register
*reg
);
1492 /** creates a function and adds it to the shader's list of functions */
1493 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1495 /** creates a null function returning null */
1496 nir_function_overload
*nir_function_overload_create(nir_function
*func
);
1498 nir_function_impl
*nir_function_impl_create(nir_function_overload
*func
);
1500 nir_block
*nir_block_create(void *mem_ctx
);
1501 nir_if
*nir_if_create(void *mem_ctx
);
1502 nir_loop
*nir_loop_create(void *mem_ctx
);
1504 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1506 /** puts a control flow node immediately after another control flow node */
1507 void nir_cf_node_insert_after(nir_cf_node
*node
, nir_cf_node
*after
);
1509 /** puts a control flow node immediately before another control flow node */
1510 void nir_cf_node_insert_before(nir_cf_node
*node
, nir_cf_node
*before
);
1512 /** puts a control flow node at the beginning of a list from an if, loop, or function */
1513 void nir_cf_node_insert_begin(struct exec_list
*list
, nir_cf_node
*node
);
1515 /** puts a control flow node at the end of a list from an if, loop, or function */
1516 void nir_cf_node_insert_end(struct exec_list
*list
, nir_cf_node
*node
);
1518 /** removes a control flow node, doing any cleanup necessary */
1519 void nir_cf_node_remove(nir_cf_node
*node
);
1521 /** requests that the given pieces of metadata be generated */
1522 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1523 /** dirties all but the preserved metadata */
1524 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1526 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1527 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1529 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1531 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1532 unsigned num_components
);
1534 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1535 nir_intrinsic_op op
);
1537 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1538 nir_function_overload
*callee
);
1540 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1542 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1544 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1546 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1547 unsigned num_components
);
1549 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1550 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1551 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1553 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1555 nir_load_const_instr
*
1556 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1558 void nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
);
1559 void nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
);
1561 void nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
);
1562 void nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
);
1564 void nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
);
1565 void nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
);
1567 void nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
);
1568 void nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
);
1570 void nir_instr_remove(nir_instr
*instr
);
1572 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
1573 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
1574 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
1575 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
1577 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
1578 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
1580 nir_const_value
*nir_src_as_const_value(nir_src src
);
1581 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
1582 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
1583 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
1584 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
1586 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1587 unsigned num_components
, const char *name
);
1588 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1589 unsigned num_components
, const char *name
);
1590 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
, void *mem_ctx
);
1592 /* visits basic blocks in source-code order */
1593 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
1594 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1596 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1599 /* If the following CF node is an if, this function returns that if.
1600 * Otherwise, it returns NULL.
1602 nir_if
*nir_block_get_following_if(nir_block
*block
);
1604 void nir_index_local_regs(nir_function_impl
*impl
);
1605 void nir_index_global_regs(nir_shader
*shader
);
1606 void nir_index_ssa_defs(nir_function_impl
*impl
);
1608 void nir_index_blocks(nir_function_impl
*impl
);
1610 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
1611 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
1614 void nir_validate_shader(nir_shader
*shader
);
1616 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
1619 void nir_calc_dominance_impl(nir_function_impl
*impl
);
1620 void nir_calc_dominance(nir_shader
*shader
);
1622 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
1623 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
1625 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
1626 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
1628 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
1629 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
1631 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
1632 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
1634 void nir_split_var_copies(nir_shader
*shader
);
1636 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
1637 void nir_lower_var_copies(nir_shader
*shader
);
1639 void nir_lower_global_vars_to_local(nir_shader
*shader
);
1641 void nir_lower_locals_to_regs(nir_shader
*shader
);
1643 void nir_assign_var_locations_scalar(struct exec_list
*var_list
,
1645 void nir_assign_var_locations_scalar_direct_first(nir_shader
*shader
,
1646 struct exec_list
*var_list
,
1647 unsigned *direct_size
,
1650 void nir_lower_io(nir_shader
*shader
);
1652 void nir_lower_vars_to_ssa(nir_shader
*shader
);
1654 void nir_remove_dead_variables(nir_shader
*shader
);
1656 void nir_lower_vec_to_movs(nir_shader
*shader
);
1657 void nir_lower_alu_to_scalar(nir_shader
*shader
);
1659 void nir_lower_phis_to_scalar(nir_shader
*shader
);
1661 void nir_lower_samplers(nir_shader
*shader
,
1662 const struct gl_shader_program
*shader_program
,
1663 gl_shader_stage stage
);
1665 void nir_lower_system_values(nir_shader
*shader
);
1666 void nir_lower_tex_projector(nir_shader
*shader
);
1667 void nir_lower_idiv(nir_shader
*shader
);
1669 void nir_lower_atomics(nir_shader
*shader
);
1670 void nir_lower_to_source_mods(nir_shader
*shader
);
1672 void nir_normalize_cubemap_coords(nir_shader
*shader
);
1674 void nir_live_variables_impl(nir_function_impl
*impl
);
1675 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
1677 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
1678 void nir_convert_to_ssa(nir_shader
*shader
);
1679 void nir_convert_from_ssa(nir_shader
*shader
);
1681 bool nir_opt_algebraic(nir_shader
*shader
);
1682 bool nir_opt_algebraic_late(nir_shader
*shader
);
1683 bool nir_opt_constant_folding(nir_shader
*shader
);
1685 bool nir_opt_global_to_local(nir_shader
*shader
);
1687 bool nir_copy_prop_impl(nir_function_impl
*impl
);
1688 bool nir_copy_prop(nir_shader
*shader
);
1690 bool nir_opt_cse(nir_shader
*shader
);
1692 bool nir_opt_dce_impl(nir_function_impl
*impl
);
1693 bool nir_opt_dce(nir_shader
*shader
);
1695 void nir_opt_gcm(nir_shader
*shader
);
1697 bool nir_opt_peephole_select(nir_shader
*shader
);
1698 bool nir_opt_peephole_ffma(nir_shader
*shader
);
1700 bool nir_opt_remove_phis(nir_shader
*shader
);
1702 void nir_sweep(nir_shader
*shader
);