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"
33 #include "GL/gl.h" /* GLenum */
34 #include "util/ralloc.h"
35 #include "main/mtypes.h"
36 #include "main/bitset.h"
37 #include "nir_types.h"
45 #define NIR_TRUE (~0u)
47 /** Defines a cast function
49 * This macro defines a cast function from in_type to out_type where
50 * out_type is some structure type that contains a field of type out_type.
52 * Note that you have to be a bit careful as the generated cast function
55 #define NIR_DEFINE_CAST(name, in_type, out_type, field) \
56 static inline out_type * \
57 name(const in_type *parent) \
59 return exec_node_data(out_type, parent, field); \
62 struct nir_function_overload
;
68 * Description of built-in state associated with a uniform
70 * \sa nir_variable::state_slots
87 * Data stored in an nir_constant
89 union nir_constant_data
{
96 typedef struct nir_constant
{
98 * Value of the constant.
100 * The field used to back the values supplied by the constant is determined
101 * by the type associated with the \c ir_instruction. Constants may be
102 * scalars, vectors, or matrices.
104 union nir_constant_data value
;
106 /* Array elements / Structure Fields */
107 struct nir_constant
**elements
;
111 * \brief Layout qualifiers for gl_FragDepth.
113 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
114 * with a layout qualifier.
117 nir_depth_layout_none
, /**< No depth layout is specified. */
118 nir_depth_layout_any
,
119 nir_depth_layout_greater
,
120 nir_depth_layout_less
,
121 nir_depth_layout_unchanged
125 * Either a uniform, global variable, shader input, or shader output. Based on
126 * ir_variable - it should be easy to translate between the two.
130 struct exec_node node
;
133 * Declared type of the variable
135 const struct glsl_type
*type
;
138 * Declared name of the variable
143 * For variables which satisfy the is_interface_instance() predicate, this
144 * points to an array of integers such that if the ith member of the
145 * interface block is an array, max_ifc_array_access[i] is the maximum
146 * array element of that member that has been accessed. If the ith member
147 * of the interface block is not an array, max_ifc_array_access[i] is
150 * For variables whose type is not an interface block, this pointer is
153 unsigned *max_ifc_array_access
;
155 struct nir_variable_data
{
158 * Is the variable read-only?
160 * This is set for variables declared as \c const, shader inputs,
163 unsigned read_only
:1;
166 unsigned invariant
:1;
169 * Storage class of the variable.
171 * \sa nir_variable_mode
173 nir_variable_mode mode
:4;
176 * Interpolation mode for shader inputs / outputs
178 * \sa ir_variable_interpolation
180 unsigned interpolation
:2;
183 * \name ARB_fragment_coord_conventions
186 unsigned origin_upper_left
:1;
187 unsigned pixel_center_integer
:1;
191 * Was the location explicitly set in the shader?
193 * If the location is explicitly set in the shader, it \b cannot be changed
194 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
197 unsigned explicit_location
:1;
198 unsigned explicit_index
:1;
201 * Was an initial binding explicitly set in the shader?
203 * If so, constant_value contains an integer ir_constant representing the
204 * initial binding point.
206 unsigned explicit_binding
:1;
209 * Does this variable have an initializer?
211 * This is used by the linker to cross-validiate initializers of global
214 unsigned has_initializer
:1;
217 * Is this variable a generic output or input that has not yet been matched
218 * up to a variable in another stage of the pipeline?
220 * This is used by the linker as scratch storage while assigning locations
221 * to generic inputs and outputs.
223 unsigned is_unmatched_generic_inout
:1;
226 * If non-zero, then this variable may be packed along with other variables
227 * into a single varying slot, so this offset should be applied when
228 * accessing components. For example, an offset of 1 means that the x
229 * component of this variable is actually stored in component y of the
230 * location specified by \c location.
232 unsigned location_frac
:2;
235 * Non-zero if this variable was created by lowering a named interface
236 * block which was not an array.
238 * Note that this variable and \c from_named_ifc_block_array will never
241 unsigned from_named_ifc_block_nonarray
:1;
244 * Non-zero if this variable was created by lowering a named interface
245 * block which was an array.
247 * Note that this variable and \c from_named_ifc_block_nonarray will never
250 unsigned from_named_ifc_block_array
:1;
253 * \brief Layout qualifier for gl_FragDepth.
255 * This is not equal to \c ir_depth_layout_none if and only if this
256 * variable is \c gl_FragDepth and a layout qualifier is specified.
258 nir_depth_layout depth_layout
;
261 * Storage location of the base of this variable
263 * The precise meaning of this field depends on the nature of the variable.
265 * - Vertex shader input: one of the values from \c gl_vert_attrib.
266 * - Vertex shader output: one of the values from \c gl_varying_slot.
267 * - Geometry shader input: one of the values from \c gl_varying_slot.
268 * - Geometry shader output: one of the values from \c gl_varying_slot.
269 * - Fragment shader input: one of the values from \c gl_varying_slot.
270 * - Fragment shader output: one of the values from \c gl_frag_result.
271 * - Uniforms: Per-stage uniform slot number for default uniform block.
272 * - Uniforms: Index within the uniform block definition for UBO members.
273 * - Other: This field is not currently used.
275 * If the variable is a uniform, shader input, or shader output, and the
276 * slot has not been assigned, the value will be -1.
281 * The actual location of the variable in the IR. Only valid for inputs
284 unsigned int driver_location
;
287 * output index for dual source blending.
292 * Initial binding point for a sampler or UBO.
294 * For array types, this represents the binding point for the first element.
299 * Location an atomic counter is stored at.
302 unsigned buffer_index
;
307 * ARB_shader_image_load_store qualifiers.
310 bool read_only
; /**< "readonly" qualifier. */
311 bool write_only
; /**< "writeonly" qualifier. */
316 /** Image internal format if specified explicitly, otherwise GL_NONE. */
321 * Highest element accessed with a constant expression array index
323 * Not used for non-array variables.
325 unsigned max_array_access
;
330 * Built-in state that backs this uniform
332 * Once set at variable creation, \c state_slots must remain invariant.
333 * This is because, ideally, this array would be shared by all clones of
334 * this variable in the IR tree. In other words, we'd really like for it
335 * to be a fly-weight.
337 * If the variable is not a uniform, \c num_state_slots will be zero and
338 * \c state_slots will be \c NULL.
341 unsigned num_state_slots
; /**< Number of state slots used */
342 nir_state_slot
*state_slots
; /**< State descriptors. */
346 * Value assigned in the initializer of a variable declared "const"
348 nir_constant
*constant_value
;
351 * Constant expression assigned in the initializer of the variable
354 * This field and \c ::constant_value are distinct. Even if the two fields
355 * refer to constants with the same value, they must point to separate
358 nir_constant
*constant_initializer
;
361 * For variables that are in an interface block or are an instance of an
362 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
364 * \sa ir_variable::location
366 const struct glsl_type
*interface_type
;
370 struct exec_node node
;
372 unsigned num_components
; /** < number of vector components */
373 unsigned num_array_elems
; /** < size of array (0 for no array) */
375 /** generic register index. */
378 /** only for debug purposes, can be NULL */
381 /** whether this register is local (per-function) or global (per-shader) */
385 * If this flag is set to true, then accessing channels >= num_components
386 * is well-defined, and simply spills over to the next array element. This
387 * is useful for backends that can do per-component accessing, in
388 * particular scalar backends. By setting this flag and making
389 * num_components equal to 1, structures can be packed tightly into
390 * registers and then registers can be accessed per-component to get to
391 * each structure member, even if it crosses vec4 boundaries.
395 /** set of nir_instr's where this register is used (read from) */
398 /** set of nir_instr's where this register is defined (written to) */
401 /** set of ifs where this register is used as a condition */
408 nir_instr_type_texture
,
409 nir_instr_type_intrinsic
,
410 nir_instr_type_load_const
,
412 nir_instr_type_ssa_undef
,
414 nir_instr_type_parallel_copy
,
418 struct exec_node node
;
420 struct nir_block
*block
;
422 /* flag for dead code elimination (see nir_opt_dce.c) */
426 #define nir_instr_next(instr) \
427 exec_node_data(nir_instr, (instr)->node.next, node)
429 #define nir_instr_prev(instr) \
430 exec_node_data(nir_instr, (instr)->node.prev, node)
433 /** for debugging only, can be NULL */
436 /** generic SSA definition index. */
439 /** Index into the live_in and live_out bitfields */
442 nir_instr
*parent_instr
;
447 uint8_t num_components
;
454 struct nir_src
*indirect
; /** < NULL for no indirect offset */
455 unsigned base_offset
;
457 /* TODO use-def chain goes here */
462 struct nir_src
*indirect
; /** < NULL for no indirect offset */
463 unsigned base_offset
;
465 /* TODO def-use chain goes here */
468 typedef struct nir_src
{
486 nir_src
nir_src_copy(nir_src src
, void *mem_ctx
);
487 nir_dest
nir_dest_copy(nir_dest dest
, void *mem_ctx
);
493 * \name input modifiers
497 * For inputs interpreted as a floating point, flips the sign bit. For inputs
498 * interpreted as an integer, performs the two's complement negation.
503 * Clears the sign bit for floating point values, and computes the integer
504 * absolute value for integers. Note that the negate modifier acts after
505 * the absolute value modifier, therefore if both are set then all inputs
506 * will become negative.
512 * For each input component, says which component of the register it is
513 * chosen from. Note that which elements of the swizzle are used and which
514 * are ignored are based on the write mask for most opcodes - for example,
515 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
516 * a swizzle of {2, x, 1, 0} where x means "don't care."
525 * \name saturate output modifier
527 * Only valid for opcodes that output floating-point numbers. Clamps the
528 * output to between 0.0 and 1.0 inclusive.
533 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
536 #define OPCODE(name, num_inputs, per_component, output_size, output_type, \
537 input_sizes, input_types) \
540 #define LAST_OPCODE(name) nir_last_opcode = nir_op_##name,
543 #include "nir_opcodes.h"
544 nir_num_opcodes
= nir_last_opcode
+ 1
563 * If true, the opcode acts in the standard, per-component manner; the
564 * operation is performed on each component (except the ones that are masked
565 * out) with the input being taken from the input swizzle for that component.
567 * If false, the size of the output and inputs are explicitly given; swizzle
568 * and writemask are still in effect, but if the output component is masked
569 * out, then the input component may still be in use.
571 * The size of some of the inputs may be given (i.e. non-zero) even though
572 * per_component is false; in that case, each component of the input acts
573 * per-component, while the rest of the inputs and the output are normal.
574 * For example, for conditional select the condition is per-component but
575 * everything else is normal.
580 * If per_component is false, the number of components in the output.
582 unsigned output_size
;
585 * The type of vector that the instruction outputs. Note that this
586 * determines whether the saturate modifier is allowed.
589 nir_alu_type output_type
;
592 * If per_component is false, the number of components in each input.
594 unsigned input_sizes
[4];
597 * The type of vector that each input takes. Note that negate is only
598 * allowed on inputs with int or float type, and behaves differently on the
599 * two, and absolute value is only allowed on float type inputs.
601 nir_alu_type input_types
[4];
604 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
606 typedef struct nir_alu_instr
{
615 /* is this source channel used? */
617 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
619 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
620 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
622 return (instr
->dest
.write_mask
>> channel
) & 1;
627 nir_deref_type_array
,
628 nir_deref_type_struct
631 typedef struct nir_deref
{
632 nir_deref_type deref_type
;
633 struct nir_deref
*child
;
634 const struct glsl_type
*type
;
643 /* This enum describes how the array is referenced. If the deref is
644 * direct then the base_offset is used. If the deref is indirect then then
645 * offset is given by base_offset + indirect. If the deref is a wildcard
646 * then the deref refers to all of the elements of the array at the same
647 * time. Wildcard dereferences are only ever allowed in copy_var
648 * intrinsics and the source and destination derefs must have matching
652 nir_deref_array_type_direct
,
653 nir_deref_array_type_indirect
,
654 nir_deref_array_type_wildcard
,
655 } nir_deref_array_type
;
660 nir_deref_array_type deref_array_type
;
661 unsigned base_offset
;
671 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
672 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
673 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
679 nir_deref_var
**params
;
680 nir_deref_var
*return_deref
;
685 struct nir_function_overload
*callee
;
688 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
689 num_variables, num_indices, flags) \
690 nir_intrinsic_##name,
692 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
695 #include "nir_intrinsics.h"
696 nir_num_intrinsics
= nir_last_intrinsic
+ 1
700 #undef LAST_INTRINSIC
705 nir_intrinsic_op intrinsic
;
709 /** number of components if this is a vectorized intrinsic */
710 uint8_t num_components
;
714 nir_deref_var
*variables
[2];
720 } nir_intrinsic_instr
;
723 * \name NIR intrinsics semantic flags
725 * information about what the compiler can do with the intrinsics.
727 * \sa nir_intrinsic_info::flags
731 * whether the intrinsic can be safely eliminated if none of its register
732 * outputs are being used.
734 #define NIR_INTRINSIC_CAN_ELIMINATE (1 << 0)
737 * Whether the intrinsic can be reordered with respect to any other intrinsic,
738 * i.e. whether the only reodering dependencies of the intrinsic are due to the
739 * register reads/writes.
741 #define NIR_INTRINSIC_CAN_REORDER (1 << 1)
744 #define NIR_INTRINSIC_MAX_INPUTS 4
749 unsigned num_srcs
; /** < number of register/SSA inputs */
751 /** number of components of each input register
753 * If this value is 0, the number of components is given by the
754 * num_components field of nir_intrinsic_instr.
756 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
760 /** number of components of the output register
762 * If this value is 0, the number of components is given by the
763 * num_components field of nir_intrinsic_instr.
765 unsigned dest_components
;
767 /** the number of inputs/outputs that are variables */
768 unsigned num_variables
;
770 /** the number of constant indices used by the intrinsic */
771 unsigned num_indices
;
773 /** semantic flags for calls to this intrinsic */
775 } nir_intrinsic_info
;
777 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
780 * \group texture information
782 * This gives semantic information about textures which is useful to the
783 * frontend, the backend, and lowering passes, but not the optimizer.
788 nir_tex_src_projector
,
789 nir_tex_src_comparitor
, /* shadow comparitor */
793 nir_tex_src_ms_index
, /* MSAA sample index */
796 nir_tex_src_sampler_index
, /* < dynamically uniform indirect index */
797 nir_num_texinput_types
801 nir_texop_tex
, /**< Regular texture look-up */
802 nir_texop_txb
, /**< Texture look-up with LOD bias */
803 nir_texop_txl
, /**< Texture look-up with explicit LOD */
804 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
805 nir_texop_txf
, /**< Texel fetch with explicit LOD */
806 nir_texop_txf_ms
, /**< Multisample texture fetch */
807 nir_texop_txs
, /**< Texture size */
808 nir_texop_lod
, /**< Texture lod query */
809 nir_texop_tg4
, /**< Texture gather */
810 nir_texop_query_levels
/**< Texture levels query */
819 enum glsl_sampler_dim sampler_dim
;
820 nir_alu_type dest_type
;
825 nir_texinput_type src_type
[4];
826 unsigned num_srcs
, coord_components
;
827 bool is_array
, is_shadow
;
830 * If is_shadow is true, whether this is the old-style shadow that outputs 4
831 * components or the new-style shadow that outputs 1 component.
833 bool is_new_style_shadow
;
835 /* constant offset - must be 0 if the offset source is used */
838 /* gather component selector */
839 unsigned component
: 2;
841 unsigned sampler_index
;
842 nir_deref_var
*sampler
; /* if this is NULL, use sampler_index instead */
845 static inline unsigned
846 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
848 if (instr
->op
== nir_texop_txs
) {
850 switch (instr
->sampler_dim
) {
851 case GLSL_SAMPLER_DIM_1D
:
852 case GLSL_SAMPLER_DIM_BUF
:
855 case GLSL_SAMPLER_DIM_2D
:
856 case GLSL_SAMPLER_DIM_CUBE
:
857 case GLSL_SAMPLER_DIM_MS
:
858 case GLSL_SAMPLER_DIM_RECT
:
859 case GLSL_SAMPLER_DIM_EXTERNAL
:
862 case GLSL_SAMPLER_DIM_3D
:
874 if (instr
->op
== nir_texop_query_levels
)
877 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
883 static inline unsigned
884 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
886 if (instr
->src_type
[src
] == nir_tex_src_coord
)
887 return instr
->coord_components
;
890 if (instr
->src_type
[src
] == nir_tex_src_offset
||
891 instr
->src_type
[src
] == nir_tex_src_ddx
||
892 instr
->src_type
[src
] == nir_tex_src_ddy
) {
894 return instr
->coord_components
- 1;
896 return instr
->coord_components
;
903 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_texinput_type type
)
905 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
906 if (instr
->src_type
[i
] == type
)
924 nir_const_value value
;
925 nir_const_value
*array
;
928 unsigned num_components
;
931 * The number of constant array elements to be copied into the variable. If
932 * this != 0, then value.array holds the array of size array_elems;
933 * otherwise, value.value holds the single vector constant (the more common
934 * case, and the only case for SSA destinations).
936 unsigned array_elems
;
942 } nir_load_const_instr
;
955 /* creates a new SSA variable in an undefined state */
960 } nir_ssa_undef_instr
;
963 struct exec_node node
;
964 struct nir_block
*pred
;
971 struct exec_list srcs
;
976 struct exec_node node
;
979 } nir_parallel_copy_copy
;
984 /* Indicates that this is the parallel copy at the end of the block.
985 * When isolating phi nodes, we create 2 parallel copies in most blocks;
986 * this flag helps tell them apart.
990 struct exec_list copies
;
991 } nir_parallel_copy_instr
;
993 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
994 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
995 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
996 NIR_DEFINE_CAST(nir_instr_as_texture
, nir_instr
, nir_tex_instr
, instr
)
997 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
998 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
999 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1000 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1001 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1002 nir_parallel_copy_instr
, instr
)
1007 * Control flow consists of a tree of control flow nodes, which include
1008 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1009 * instructions that always run start-to-finish. Each basic block also keeps
1010 * track of its successors (blocks which may run immediately after the current
1011 * block) and predecessors (blocks which could have run immediately before the
1012 * current block). Each function also has a start block and an end block which
1013 * all return statements point to (which is always empty). Together, all the
1014 * blocks with their predecessors and successors make up the control flow
1015 * graph (CFG) of the function. There are helpers that modify the tree of
1016 * control flow nodes while modifying the CFG appropriately; these should be
1017 * used instead of modifying the tree directly.
1024 nir_cf_node_function
1027 typedef struct nir_cf_node
{
1028 struct exec_node node
;
1029 nir_cf_node_type type
;
1030 struct nir_cf_node
*parent
;
1033 typedef struct nir_block
{
1034 nir_cf_node cf_node
;
1035 struct exec_list instr_list
;
1040 * Each block can only have up to 2 successors, so we put them in a simple
1041 * array - no need for anything more complicated.
1043 struct nir_block
*successors
[2];
1045 struct set
*predecessors
;
1048 * this node's immediate dominator in the dominance tree - set to NULL for
1051 struct nir_block
*imm_dom
;
1053 /* This node's children in the dominance tree */
1054 unsigned num_dom_children
;
1055 struct nir_block
**dom_children
;
1057 struct set
*dom_frontier
;
1059 /* live in and out for this block; used for liveness analysis */
1060 BITSET_WORD
*live_in
;
1061 BITSET_WORD
*live_out
;
1064 #define nir_block_first_instr(block) \
1065 exec_node_data(nir_instr, exec_list_get_head(&(block)->instr_list), node)
1066 #define nir_block_last_instr(block) \
1067 exec_node_data(nir_instr, exec_list_get_tail(&(block)->instr_list), node)
1069 #define nir_foreach_instr(block, instr) \
1070 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1071 #define nir_foreach_instr_reverse(block, instr) \
1072 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1073 #define nir_foreach_instr_safe(block, instr) \
1074 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1077 nir_cf_node cf_node
;
1079 struct exec_list then_list
;
1080 struct exec_list else_list
;
1083 #define nir_if_first_then_node(if) \
1084 exec_node_data(nir_cf_node, exec_list_get_head(&(if)->then_list), node)
1085 #define nir_if_last_then_node(if) \
1086 exec_node_data(nir_cf_node, exec_list_get_tail(&(if)->then_list), node)
1087 #define nir_if_first_else_node(if) \
1088 exec_node_data(nir_cf_node, exec_list_get_head(&(if)->else_list), node)
1089 #define nir_if_last_else_node(if) \
1090 exec_node_data(nir_cf_node, exec_list_get_tail(&(if)->else_list), node)
1093 nir_cf_node cf_node
;
1094 struct exec_list body
;
1097 #define nir_loop_first_cf_node(loop) \
1098 exec_node_data(nir_cf_node, exec_list_get_head(&(loop)->body), node)
1099 #define nir_loop_last_cf_node(loop) \
1100 exec_node_data(nir_cf_node, exec_list_get_tail(&(loop)->body), node)
1103 * Various bits of metadata that can may be created or required by
1104 * optimization and analysis passes
1107 nir_metadata_none
= 0x0,
1108 nir_metadata_block_index
= 0x1,
1109 nir_metadata_dominance
= 0x2,
1110 nir_metadata_live_variables
= 0x4,
1114 nir_cf_node cf_node
;
1116 /** pointer to the overload of which this is an implementation */
1117 struct nir_function_overload
*overload
;
1119 struct exec_list body
; /** < list of nir_cf_node */
1121 nir_block
*start_block
, *end_block
;
1123 /** list for all local variables in the function */
1124 struct exec_list locals
;
1126 /** array of variables used as parameters */
1127 unsigned num_params
;
1128 nir_variable
**params
;
1130 /** variable used to hold the result of the function */
1131 nir_variable
*return_var
;
1133 /** list of local registers in the function */
1134 struct exec_list registers
;
1136 /** next available local register index */
1139 /** next available SSA value index */
1142 /* total number of basic blocks, only valid when block_index_dirty = false */
1143 unsigned num_blocks
;
1145 nir_metadata valid_metadata
;
1146 } nir_function_impl
;
1148 #define nir_cf_node_next(_node) \
1149 exec_node_data(nir_cf_node, exec_node_get_next(&(_node)->node), node)
1151 #define nir_cf_node_prev(_node) \
1152 exec_node_data(nir_cf_node, exec_node_get_prev(&(_node)->node), node)
1154 #define nir_cf_node_is_first(_node) \
1155 exec_node_is_head_sentinel((_node)->node.prev)
1157 #define nir_cf_node_is_last(_node) \
1158 exec_node_is_tail_sentinel((_node)->node.next)
1160 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1161 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1162 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1163 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1168 nir_parameter_inout
,
1169 } nir_parameter_type
;
1172 nir_parameter_type param_type
;
1173 const struct glsl_type
*type
;
1176 typedef struct nir_function_overload
{
1177 struct exec_node node
;
1179 unsigned num_params
;
1180 nir_parameter
*params
;
1181 const struct glsl_type
*return_type
;
1183 nir_function_impl
*impl
; /** < NULL if the overload is only declared yet */
1185 /** pointer to the function of which this is an overload */
1186 struct nir_function
*function
;
1187 } nir_function_overload
;
1189 typedef struct nir_function
{
1190 struct exec_node node
;
1192 struct exec_list overload_list
;
1194 struct nir_shader
*shader
;
1197 #define nir_function_first_overload(func) \
1198 exec_node_data(nir_function_overload, \
1199 exec_list_get_head(&(func)->overload_list), node)
1201 typedef struct nir_shader
{
1202 /** hash table of name -> uniform */
1203 struct hash_table
*uniforms
;
1205 /** hash table of name -> input */
1206 struct hash_table
*inputs
;
1208 /** hash table of name -> output */
1209 struct hash_table
*outputs
;
1211 /** list of global variables in the shader */
1212 struct exec_list globals
;
1214 struct exec_list system_values
;
1216 struct exec_list functions
;
1218 /** list of global registers in the shader */
1219 struct exec_list registers
;
1221 /** structures used in this shader */
1222 unsigned num_user_structures
;
1223 struct glsl_type
**user_structures
;
1225 /** next available global register index */
1229 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1232 unsigned num_inputs
, num_uniforms
, num_outputs
;
1235 #define nir_foreach_overload(shader, overload) \
1236 foreach_list_typed(nir_function, func, node, &(shader)->functions) \
1237 foreach_list_typed(nir_function_overload, overload, node, \
1238 &(func)->overload_list)
1240 nir_shader
*nir_shader_create(void *mem_ctx
);
1242 /** creates a register, including assigning it an index and adding it to the list */
1243 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1245 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1247 void nir_reg_remove(nir_register
*reg
);
1249 /** creates a function and adds it to the shader's list of functions */
1250 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1252 /** creates a null function returning null */
1253 nir_function_overload
*nir_function_overload_create(nir_function
*func
);
1255 nir_function_impl
*nir_function_impl_create(nir_function_overload
*func
);
1257 nir_block
*nir_block_create(void *mem_ctx
);
1258 nir_if
*nir_if_create(void *mem_ctx
);
1259 nir_loop
*nir_loop_create(void *mem_ctx
);
1261 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1263 /** puts a control flow node immediately after another control flow node */
1264 void nir_cf_node_insert_after(nir_cf_node
*node
, nir_cf_node
*after
);
1266 /** puts a control flow node immediately before another control flow node */
1267 void nir_cf_node_insert_before(nir_cf_node
*node
, nir_cf_node
*before
);
1269 /** puts a control flow node at the beginning of a list from an if, loop, or function */
1270 void nir_cf_node_insert_begin(struct exec_list
*list
, nir_cf_node
*node
);
1272 /** puts a control flow node at the end of a list from an if, loop, or function */
1273 void nir_cf_node_insert_end(struct exec_list
*list
, nir_cf_node
*node
);
1275 /** removes a control flow node, doing any cleanup necessary */
1276 void nir_cf_node_remove(nir_cf_node
*node
);
1278 /** requests that the given pieces of metadata be generated */
1279 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1280 /** dirties all but the preserved metadata */
1281 void nir_metadata_dirty(nir_function_impl
*impl
, nir_metadata preserved
);
1283 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1284 nir_alu_instr
*nir_alu_instr_create(void *mem_ctx
, nir_op op
);
1286 nir_jump_instr
*nir_jump_instr_create(void *mem_ctx
, nir_jump_type type
);
1288 nir_load_const_instr
*nir_load_const_instr_create(void *mem_ctx
);
1290 nir_intrinsic_instr
*nir_intrinsic_instr_create(void *mem_ctx
,
1291 nir_intrinsic_op op
);
1293 nir_call_instr
*nir_call_instr_create(void *mem_ctx
,
1294 nir_function_overload
*callee
);
1296 nir_tex_instr
*nir_tex_instr_create(void *mem_ctx
, unsigned num_srcs
);
1298 nir_phi_instr
*nir_phi_instr_create(void *mem_ctx
);
1300 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(void *mem_ctx
);
1302 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(void *mem_ctx
);
1304 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1305 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1306 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1308 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1310 void nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
);
1311 void nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
);
1313 void nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
);
1314 void nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
);
1316 void nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
);
1317 void nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
);
1319 void nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
);
1320 void nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
);
1322 void nir_instr_remove(nir_instr
*instr
);
1324 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
1325 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
1326 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
1327 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
1329 nir_const_value
*nir_src_as_const_value(nir_src src
);
1330 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
1331 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
1333 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1334 unsigned num_components
, const char *name
);
1335 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
, void *mem_ctx
);
1337 /* visits basic blocks in source-code order */
1338 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
1339 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1341 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1344 /* If the following CF node is an if, this function returns that if.
1345 * Otherwise, it returns NULL.
1347 nir_if
*nir_block_following_if(nir_block
*block
);
1349 void nir_index_local_regs(nir_function_impl
*impl
);
1350 void nir_index_global_regs(nir_shader
*shader
);
1351 void nir_index_ssa_defs(nir_function_impl
*impl
);
1353 void nir_index_blocks(nir_function_impl
*impl
);
1355 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
1357 void nir_validate_shader(nir_shader
*shader
);
1359 void nir_calc_dominance_impl(nir_function_impl
*impl
);
1360 void nir_calc_dominance(nir_shader
*shader
);
1362 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
1363 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
1365 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
1366 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
1368 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
1369 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
1371 void nir_split_var_copies(nir_shader
*shader
);
1373 void nir_lower_global_vars_to_local(nir_shader
*shader
);
1375 void nir_lower_locals_to_regs(nir_shader
*shader
);
1377 void nir_lower_io(nir_shader
*shader
);
1379 void nir_lower_variables(nir_shader
*shader
);
1381 void nir_remove_dead_variables(nir_shader
*shader
);
1383 void nir_lower_vec_to_movs(nir_shader
*shader
);
1385 void nir_lower_samplers(nir_shader
*shader
,
1386 struct gl_shader_program
*shader_program
,
1387 struct gl_program
*prog
);
1389 void nir_lower_system_values(nir_shader
*shader
);
1391 void nir_lower_atomics(nir_shader
*shader
);
1392 void nir_lower_to_source_mods(nir_shader
*shader
);
1394 void nir_live_variables_impl(nir_function_impl
*impl
);
1395 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
1397 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
1398 void nir_convert_to_ssa(nir_shader
*shader
);
1399 void nir_convert_from_ssa(nir_shader
*shader
);
1401 bool nir_opt_global_to_local(nir_shader
*shader
);
1403 bool nir_copy_prop_impl(nir_function_impl
*impl
);
1404 bool nir_copy_prop(nir_shader
*shader
);
1406 bool nir_opt_cse(nir_shader
*shader
);
1408 bool nir_opt_dce_impl(nir_function_impl
*impl
);
1409 bool nir_opt_dce(nir_shader
*shader
);
1411 bool nir_opt_peephole_select(nir_shader
*shader
);
1412 bool nir_opt_peephole_ffma(nir_shader
*shader
);