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)
31 #include "util/hash_table.h"
32 #include "compiler/glsl/list.h"
33 #include "GL/gl.h" /* GLenum */
34 #include "util/list.h"
35 #include "util/ralloc.h"
37 #include "util/bitset.h"
38 #include "util/macros.h"
39 #include "compiler/nir_types.h"
40 #include "compiler/shader_enums.h"
41 #include "compiler/shader_info.h"
45 #include "util/debug.h"
48 #include "nir_opcodes.h"
50 #if defined(_WIN32) && !defined(snprintf)
51 #define snprintf _snprintf
59 #define NIR_TRUE (~0u)
61 /** Defines a cast function
63 * This macro defines a cast function from in_type to out_type where
64 * out_type is some structure type that contains a field of type out_type.
66 * Note that you have to be a bit careful as the generated cast function
69 #define NIR_DEFINE_CAST(name, in_type, out_type, field, \
70 type_field, type_value) \
71 static inline out_type * \
72 name(const in_type *parent) \
74 assert(parent && parent->type_field == type_value); \
75 return exec_node_data(out_type, parent, field); \
84 * Description of built-in state associated with a uniform
86 * \sa nir_variable::state_slots
89 gl_state_index16 tokens
[STATE_LENGTH
];
94 nir_var_shader_in
= (1 << 0),
95 nir_var_shader_out
= (1 << 1),
96 nir_var_global
= (1 << 2),
97 nir_var_local
= (1 << 3),
98 nir_var_uniform
= (1 << 4),
99 nir_var_shader_storage
= (1 << 5),
100 nir_var_system_value
= (1 << 6),
101 nir_var_param
= (1 << 7),
102 nir_var_shared
= (1 << 8),
110 nir_rounding_mode_undef
= 0,
111 nir_rounding_mode_rtne
= 1, /* round to nearest even */
112 nir_rounding_mode_ru
= 2, /* round up */
113 nir_rounding_mode_rd
= 3, /* round down */
114 nir_rounding_mode_rtz
= 4, /* round towards zero */
130 typedef struct nir_constant
{
132 * Value of the constant.
134 * The field used to back the values supplied by the constant is determined
135 * by the type associated with the \c nir_variable. Constants may be
136 * scalars, vectors, or matrices.
138 nir_const_value values
[4];
140 /* we could get this from the var->type but makes clone *much* easier to
141 * not have to care about the type.
143 unsigned num_elements
;
145 /* Array elements / Structure Fields */
146 struct nir_constant
**elements
;
150 * \brief Layout qualifiers for gl_FragDepth.
152 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
153 * with a layout qualifier.
156 nir_depth_layout_none
, /**< No depth layout is specified. */
157 nir_depth_layout_any
,
158 nir_depth_layout_greater
,
159 nir_depth_layout_less
,
160 nir_depth_layout_unchanged
164 * Either a uniform, global variable, shader input, or shader output. Based on
165 * ir_variable - it should be easy to translate between the two.
168 typedef struct nir_variable
{
169 struct exec_node node
;
172 * Declared type of the variable
174 const struct glsl_type
*type
;
177 * Declared name of the variable
181 struct nir_variable_data
{
183 * Storage class of the variable.
185 * \sa nir_variable_mode
187 nir_variable_mode mode
;
190 * Is the variable read-only?
192 * This is set for variables declared as \c const, shader inputs,
195 unsigned read_only
:1;
199 unsigned invariant
:1;
202 * When separate shader programs are enabled, only input/outputs between
203 * the stages of a multi-stage separate program can be safely removed
204 * from the shader interface. Other input/outputs must remains active.
206 * This is also used to make sure xfb varyings that are unused by the
207 * fragment shader are not removed.
209 unsigned always_active_io
:1;
212 * Interpolation mode for shader inputs / outputs
214 * \sa glsl_interp_mode
216 unsigned interpolation
:2;
219 * \name ARB_fragment_coord_conventions
222 unsigned origin_upper_left
:1;
223 unsigned pixel_center_integer
:1;
227 * If non-zero, then this variable may be packed along with other variables
228 * into a single varying slot, so this offset should be applied when
229 * accessing components. For example, an offset of 1 means that the x
230 * component of this variable is actually stored in component y of the
231 * location specified by \c location.
233 unsigned location_frac
:2;
236 * If true, this variable represents an array of scalars that should
237 * be tightly packed. In other words, consecutive array elements
238 * should be stored one component apart, rather than one slot apart.
243 * Whether this is a fragment shader output implicitly initialized with
244 * the previous contents of the specified render target at the
245 * framebuffer location corresponding to this shader invocation.
247 unsigned fb_fetch_output
:1;
250 * Non-zero if this variable is considered bindless as defined by
251 * ARB_bindless_texture.
256 * \brief Layout qualifier for gl_FragDepth.
258 * This is not equal to \c ir_depth_layout_none if and only if this
259 * variable is \c gl_FragDepth and a layout qualifier is specified.
261 nir_depth_layout depth_layout
;
264 * Storage location of the base of this variable
266 * The precise meaning of this field depends on the nature of the variable.
268 * - Vertex shader input: one of the values from \c gl_vert_attrib.
269 * - Vertex shader output: one of the values from \c gl_varying_slot.
270 * - Geometry shader input: one of the values from \c gl_varying_slot.
271 * - Geometry shader output: one of the values from \c gl_varying_slot.
272 * - Fragment shader input: one of the values from \c gl_varying_slot.
273 * - Fragment shader output: one of the values from \c gl_frag_result.
274 * - Uniforms: Per-stage uniform slot number for default uniform block.
275 * - Uniforms: Index within the uniform block definition for UBO members.
276 * - Non-UBO Uniforms: uniform slot number.
277 * - Other: This field is not currently used.
279 * If the variable is a uniform, shader input, or shader output, and the
280 * slot has not been assigned, the value will be -1.
285 * The actual location of the variable in the IR. Only valid for inputs
288 unsigned int driver_location
;
291 * Vertex stream output identifier.
293 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
294 * stream of the i-th component.
299 * output index for dual source blending.
304 * Descriptor set binding for sampler or UBO.
309 * Initial binding point for a sampler or UBO.
311 * For array types, this represents the binding point for the first element.
316 * Location an atomic counter is stored at.
321 * ARB_shader_image_load_store qualifiers.
324 bool read_only
; /**< "readonly" qualifier. */
325 bool write_only
; /**< "writeonly" qualifier. */
330 /** Image internal format if specified explicitly, otherwise GL_NONE. */
336 * Built-in state that backs this uniform
338 * Once set at variable creation, \c state_slots must remain invariant.
339 * This is because, ideally, this array would be shared by all clones of
340 * this variable in the IR tree. In other words, we'd really like for it
341 * to be a fly-weight.
343 * If the variable is not a uniform, \c num_state_slots will be zero and
344 * \c state_slots will be \c NULL.
347 unsigned num_state_slots
; /**< Number of state slots used */
348 nir_state_slot
*state_slots
; /**< State descriptors. */
352 * Constant expression assigned in the initializer of the variable
354 * This field should only be used temporarily by creators of NIR shaders
355 * and then lower_constant_initializers can be used to get rid of them.
356 * Most of the rest of NIR ignores this field or asserts that it's NULL.
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
;
369 #define nir_foreach_variable(var, var_list) \
370 foreach_list_typed(nir_variable, var, node, var_list)
372 #define nir_foreach_variable_safe(var, var_list) \
373 foreach_list_typed_safe(nir_variable, var, node, var_list)
376 nir_variable_is_global(const nir_variable
*var
)
378 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
381 typedef struct nir_register
{
382 struct exec_node node
;
384 unsigned num_components
; /** < number of vector components */
385 unsigned num_array_elems
; /** < size of array (0 for no array) */
387 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
390 /** generic register index. */
393 /** only for debug purposes, can be NULL */
396 /** whether this register is local (per-function) or global (per-shader) */
400 * If this flag is set to true, then accessing channels >= num_components
401 * is well-defined, and simply spills over to the next array element. This
402 * is useful for backends that can do per-component accessing, in
403 * particular scalar backends. By setting this flag and making
404 * num_components equal to 1, structures can be packed tightly into
405 * registers and then registers can be accessed per-component to get to
406 * each structure member, even if it crosses vec4 boundaries.
410 /** set of nir_srcs where this register is used (read from) */
411 struct list_head uses
;
413 /** set of nir_dests where this register is defined (written to) */
414 struct list_head defs
;
416 /** set of nir_ifs where this register is used as a condition */
417 struct list_head if_uses
;
420 #define nir_foreach_register(reg, reg_list) \
421 foreach_list_typed(nir_register, reg, node, reg_list)
422 #define nir_foreach_register_safe(reg, reg_list) \
423 foreach_list_typed_safe(nir_register, reg, node, reg_list)
429 nir_instr_type_intrinsic
,
430 nir_instr_type_load_const
,
432 nir_instr_type_ssa_undef
,
434 nir_instr_type_parallel_copy
,
437 typedef struct nir_instr
{
438 struct exec_node node
;
440 struct nir_block
*block
;
442 /** generic instruction index. */
445 /* A temporary for optimization and analysis passes to use for storing
446 * flags. For instance, DCE uses this to store the "dead/live" info.
451 static inline nir_instr
*
452 nir_instr_next(nir_instr
*instr
)
454 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
455 if (exec_node_is_tail_sentinel(next
))
458 return exec_node_data(nir_instr
, next
, node
);
461 static inline nir_instr
*
462 nir_instr_prev(nir_instr
*instr
)
464 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
465 if (exec_node_is_head_sentinel(prev
))
468 return exec_node_data(nir_instr
, prev
, node
);
472 nir_instr_is_first(const nir_instr
*instr
)
474 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
478 nir_instr_is_last(const nir_instr
*instr
)
480 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
483 typedef struct nir_ssa_def
{
484 /** for debugging only, can be NULL */
487 /** generic SSA definition index. */
490 /** Index into the live_in and live_out bitfields */
493 nir_instr
*parent_instr
;
495 /** set of nir_instrs where this register is used (read from) */
496 struct list_head uses
;
498 /** set of nir_ifs where this register is used as a condition */
499 struct list_head if_uses
;
501 uint8_t num_components
;
503 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
511 struct nir_src
*indirect
; /** < NULL for no indirect offset */
512 unsigned base_offset
;
514 /* TODO use-def chain goes here */
518 nir_instr
*parent_instr
;
519 struct list_head def_link
;
522 struct nir_src
*indirect
; /** < NULL for no indirect offset */
523 unsigned base_offset
;
525 /* TODO def-use chain goes here */
530 typedef struct nir_src
{
532 nir_instr
*parent_instr
;
533 struct nir_if
*parent_if
;
536 struct list_head use_link
;
546 static inline nir_src
549 nir_src src
= { { NULL
} };
553 #define NIR_SRC_INIT nir_src_init()
555 #define nir_foreach_use(src, reg_or_ssa_def) \
556 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
558 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
559 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
561 #define nir_foreach_if_use(src, reg_or_ssa_def) \
562 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
564 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
565 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
576 static inline nir_dest
579 nir_dest dest
= { { { NULL
} } };
583 #define NIR_DEST_INIT nir_dest_init()
585 #define nir_foreach_def(dest, reg) \
586 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
588 #define nir_foreach_def_safe(dest, reg) \
589 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
591 static inline nir_src
592 nir_src_for_ssa(nir_ssa_def
*def
)
594 nir_src src
= NIR_SRC_INIT
;
602 static inline nir_src
603 nir_src_for_reg(nir_register
*reg
)
605 nir_src src
= NIR_SRC_INIT
;
609 src
.reg
.indirect
= NULL
;
610 src
.reg
.base_offset
= 0;
615 static inline nir_dest
616 nir_dest_for_reg(nir_register
*reg
)
618 nir_dest dest
= NIR_DEST_INIT
;
625 static inline unsigned
626 nir_src_bit_size(nir_src src
)
628 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
631 static inline unsigned
632 nir_src_num_components(nir_src src
)
634 return src
.is_ssa
? src
.ssa
->num_components
: src
.reg
.reg
->num_components
;
637 static inline unsigned
638 nir_dest_bit_size(nir_dest dest
)
640 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
643 static inline unsigned
644 nir_dest_num_components(nir_dest dest
)
646 return dest
.is_ssa
? dest
.ssa
.num_components
: dest
.reg
.reg
->num_components
;
649 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
650 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
656 * \name input modifiers
660 * For inputs interpreted as floating point, flips the sign bit. For
661 * inputs interpreted as integers, performs the two's complement negation.
666 * Clears the sign bit for floating point values, and computes the integer
667 * absolute value for integers. Note that the negate modifier acts after
668 * the absolute value modifier, therefore if both are set then all inputs
669 * will become negative.
675 * For each input component, says which component of the register it is
676 * chosen from. Note that which elements of the swizzle are used and which
677 * are ignored are based on the write mask for most opcodes - for example,
678 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
679 * a swizzle of {2, x, 1, 0} where x means "don't care."
688 * \name saturate output modifier
690 * Only valid for opcodes that output floating-point numbers. Clamps the
691 * output to between 0.0 and 1.0 inclusive.
696 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
700 nir_type_invalid
= 0, /* Not a valid type */
705 nir_type_bool32
= 32 | nir_type_bool
,
706 nir_type_int8
= 8 | nir_type_int
,
707 nir_type_int16
= 16 | nir_type_int
,
708 nir_type_int32
= 32 | nir_type_int
,
709 nir_type_int64
= 64 | nir_type_int
,
710 nir_type_uint8
= 8 | nir_type_uint
,
711 nir_type_uint16
= 16 | nir_type_uint
,
712 nir_type_uint32
= 32 | nir_type_uint
,
713 nir_type_uint64
= 64 | nir_type_uint
,
714 nir_type_float16
= 16 | nir_type_float
,
715 nir_type_float32
= 32 | nir_type_float
,
716 nir_type_float64
= 64 | nir_type_float
,
719 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
720 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
722 static inline unsigned
723 nir_alu_type_get_type_size(nir_alu_type type
)
725 return type
& NIR_ALU_TYPE_SIZE_MASK
;
728 static inline unsigned
729 nir_alu_type_get_base_type(nir_alu_type type
)
731 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
734 static inline nir_alu_type
735 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
739 return nir_type_bool32
;
742 return nir_type_uint32
;
745 return nir_type_int32
;
747 case GLSL_TYPE_UINT16
:
748 return nir_type_uint16
;
750 case GLSL_TYPE_INT16
:
751 return nir_type_int16
;
753 case GLSL_TYPE_UINT8
:
754 return nir_type_uint8
;
756 return nir_type_int8
;
757 case GLSL_TYPE_UINT64
:
758 return nir_type_uint64
;
760 case GLSL_TYPE_INT64
:
761 return nir_type_int64
;
763 case GLSL_TYPE_FLOAT
:
764 return nir_type_float32
;
766 case GLSL_TYPE_FLOAT16
:
767 return nir_type_float16
;
769 case GLSL_TYPE_DOUBLE
:
770 return nir_type_float64
;
773 unreachable("unknown type");
777 static inline nir_alu_type
778 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
780 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
783 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
,
784 nir_rounding_mode rnd
);
787 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
788 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
789 } nir_op_algebraic_property
;
797 * The number of components in the output
799 * If non-zero, this is the size of the output and input sizes are
800 * explicitly given; swizzle and writemask are still in effect, but if
801 * the output component is masked out, then the input component may
804 * If zero, the opcode acts in the standard, per-component manner; the
805 * operation is performed on each component (except the ones that are
806 * masked out) with the input being taken from the input swizzle for
809 * The size of some of the inputs may be given (i.e. non-zero) even
810 * though output_size is zero; in that case, the inputs with a zero
811 * size act per-component, while the inputs with non-zero size don't.
813 unsigned output_size
;
816 * The type of vector that the instruction outputs. Note that the
817 * staurate modifier is only allowed on outputs with the float type.
820 nir_alu_type output_type
;
823 * The number of components in each input
825 unsigned input_sizes
[4];
828 * The type of vector that each input takes. Note that negate and
829 * absolute value are only allowed on inputs with int or float type and
830 * behave differently on the two.
832 nir_alu_type input_types
[4];
834 nir_op_algebraic_property algebraic_properties
;
837 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
839 typedef struct nir_alu_instr
{
843 /** Indicates that this ALU instruction generates an exact value
845 * This is kind of a mixture of GLSL "precise" and "invariant" and not
846 * really equivalent to either. This indicates that the value generated by
847 * this operation is high-precision and any code transformations that touch
848 * it must ensure that the resulting value is bit-for-bit identical to the
857 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
858 nir_alu_instr
*instr
);
859 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
860 nir_alu_instr
*instr
);
862 /* is this source channel used? */
864 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
867 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
868 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
870 return (instr
->dest
.write_mask
>> channel
) & 1;
874 * For instructions whose destinations are SSA, get the number of channels
877 static inline unsigned
878 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
880 assert(instr
->dest
.dest
.is_ssa
);
882 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
883 return nir_op_infos
[instr
->op
].input_sizes
[src
];
885 return instr
->dest
.dest
.ssa
.num_components
;
888 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
889 unsigned src1
, unsigned src2
);
893 nir_deref_type_array
,
894 nir_deref_type_struct
897 typedef struct nir_deref
{
898 nir_deref_type deref_type
;
899 struct nir_deref
*child
;
900 const struct glsl_type
*type
;
909 /* This enum describes how the array is referenced. If the deref is
910 * direct then the base_offset is used. If the deref is indirect then
911 * offset is given by base_offset + indirect. If the deref is a wildcard
912 * then the deref refers to all of the elements of the array at the same
913 * time. Wildcard dereferences are only ever allowed in copy_var
914 * intrinsics and the source and destination derefs must have matching
918 nir_deref_array_type_direct
,
919 nir_deref_array_type_indirect
,
920 nir_deref_array_type_wildcard
,
921 } nir_deref_array_type
;
926 nir_deref_array_type deref_array_type
;
927 unsigned base_offset
;
937 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
,
938 deref_type
, nir_deref_type_var
)
939 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
,
940 deref_type
, nir_deref_type_array
)
941 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
,
942 deref_type
, nir_deref_type_struct
)
944 /* Returns the last deref in the chain. */
945 static inline nir_deref
*
946 nir_deref_tail(nir_deref
*deref
)
949 deref
= deref
->child
;
957 nir_deref_var
**params
;
958 nir_deref_var
*return_deref
;
960 struct nir_function
*callee
;
963 #include "nir_intrinsics.h"
965 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
967 /** Represents an intrinsic
969 * An intrinsic is an instruction type for handling things that are
970 * more-or-less regular operations but don't just consume and produce SSA
971 * values like ALU operations do. Intrinsics are not for things that have
972 * special semantic meaning such as phi nodes and parallel copies.
973 * Examples of intrinsics include variable load/store operations, system
974 * value loads, and the like. Even though texturing more-or-less falls
975 * under this category, texturing is its own instruction type because
976 * trying to represent texturing with intrinsics would lead to a
977 * combinatorial explosion of intrinsic opcodes.
979 * By having a single instruction type for handling a lot of different
980 * cases, optimization passes can look for intrinsics and, for the most
981 * part, completely ignore them. Each intrinsic type also has a few
982 * possible flags that govern whether or not they can be reordered or
983 * eliminated. That way passes like dead code elimination can still work
984 * on intrisics without understanding the meaning of each.
986 * Each intrinsic has some number of constant indices, some number of
987 * variables, and some number of sources. What these sources, variables,
988 * and indices mean depends on the intrinsic and is documented with the
989 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
990 * instructions are the only types of instruction that can operate on
996 nir_intrinsic_op intrinsic
;
1000 /** number of components if this is a vectorized intrinsic
1002 * Similarly to ALU operations, some intrinsics are vectorized.
1003 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
1004 * For vectorized intrinsics, the num_components field specifies the
1005 * number of destination components and the number of source components
1006 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
1008 uint8_t num_components
;
1010 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
1012 nir_deref_var
*variables
[2];
1015 } nir_intrinsic_instr
;
1018 * \name NIR intrinsics semantic flags
1020 * information about what the compiler can do with the intrinsics.
1022 * \sa nir_intrinsic_info::flags
1026 * whether the intrinsic can be safely eliminated if none of its output
1027 * value is not being used.
1029 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
1032 * Whether the intrinsic can be reordered with respect to any other
1033 * intrinsic, i.e. whether the only reordering dependencies of the
1034 * intrinsic are due to the register reads/writes.
1036 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
1037 } nir_intrinsic_semantic_flag
;
1040 * \name NIR intrinsics const-index flag
1042 * Indicates the usage of a const_index slot.
1044 * \sa nir_intrinsic_info::index_map
1048 * Generally instructions that take a offset src argument, can encode
1049 * a constant 'base' value which is added to the offset.
1051 NIR_INTRINSIC_BASE
= 1,
1054 * For store instructions, a writemask for the store.
1056 NIR_INTRINSIC_WRMASK
= 2,
1059 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1061 NIR_INTRINSIC_STREAM_ID
= 3,
1064 * The clip-plane id for load_user_clip_plane intrinsic.
1066 NIR_INTRINSIC_UCP_ID
= 4,
1069 * The amount of data, starting from BASE, that this instruction may
1070 * access. This is used to provide bounds if the offset is not constant.
1072 NIR_INTRINSIC_RANGE
= 5,
1075 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1077 NIR_INTRINSIC_DESC_SET
= 6,
1080 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1082 NIR_INTRINSIC_BINDING
= 7,
1087 NIR_INTRINSIC_COMPONENT
= 8,
1090 * Interpolation mode (only meaningful for FS inputs).
1092 NIR_INTRINSIC_INTERP_MODE
= 9,
1095 * A binary nir_op to use when performing a reduction or scan operation
1097 NIR_INTRINSIC_REDUCTION_OP
= 10,
1100 * Cluster size for reduction operations
1102 NIR_INTRINSIC_CLUSTER_SIZE
= 11,
1104 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1106 } nir_intrinsic_index_flag
;
1108 #define NIR_INTRINSIC_MAX_INPUTS 4
1113 unsigned num_srcs
; /** < number of register/SSA inputs */
1115 /** number of components of each input register
1117 * If this value is 0, the number of components is given by the
1118 * num_components field of nir_intrinsic_instr.
1120 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1124 /** number of components of the output register
1126 * If this value is 0, the number of components is given by the
1127 * num_components field of nir_intrinsic_instr.
1129 unsigned dest_components
;
1131 /** the number of inputs/outputs that are variables */
1132 unsigned num_variables
;
1134 /** the number of constant indices used by the intrinsic */
1135 unsigned num_indices
;
1137 /** indicates the usage of intr->const_index[n] */
1138 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1140 /** semantic flags for calls to this intrinsic */
1141 nir_intrinsic_semantic_flag flags
;
1142 } nir_intrinsic_info
;
1144 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1146 static inline unsigned
1147 nir_intrinsic_src_components(nir_intrinsic_instr
*intr
, unsigned srcn
)
1149 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1150 assert(srcn
< info
->num_srcs
);
1151 if (info
->src_components
[srcn
])
1152 return info
->src_components
[srcn
];
1154 return intr
->num_components
;
1157 static inline unsigned
1158 nir_intrinsic_dest_components(nir_intrinsic_instr
*intr
)
1160 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1161 if (!info
->has_dest
)
1163 else if (info
->dest_components
)
1164 return info
->dest_components
;
1166 return intr
->num_components
;
1169 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1170 static inline type \
1171 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1173 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1174 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1175 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1177 static inline void \
1178 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1180 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1181 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1182 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1185 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1186 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1187 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1188 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1189 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1190 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1191 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1192 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1193 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1194 INTRINSIC_IDX_ACCESSORS(reduction_op
, REDUCTION_OP
, unsigned)
1195 INTRINSIC_IDX_ACCESSORS(cluster_size
, CLUSTER_SIZE
, unsigned)
1198 * \group texture information
1200 * This gives semantic information about textures which is useful to the
1201 * frontend, the backend, and lowering passes, but not the optimizer.
1206 nir_tex_src_projector
,
1207 nir_tex_src_comparator
, /* shadow comparator */
1211 nir_tex_src_ms_index
, /* MSAA sample index */
1212 nir_tex_src_ms_mcs
, /* MSAA compression value */
1215 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1216 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1217 nir_tex_src_plane
, /* < selects plane for planar textures */
1218 nir_num_tex_src_types
1223 nir_tex_src_type src_type
;
1227 nir_texop_tex
, /**< Regular texture look-up */
1228 nir_texop_txb
, /**< Texture look-up with LOD bias */
1229 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1230 nir_texop_txd
, /**< Texture look-up with partial derivatives */
1231 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1232 nir_texop_txf_ms
, /**< Multisample texture fetch */
1233 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1234 nir_texop_txs
, /**< Texture size */
1235 nir_texop_lod
, /**< Texture lod query */
1236 nir_texop_tg4
, /**< Texture gather */
1237 nir_texop_query_levels
, /**< Texture levels query */
1238 nir_texop_texture_samples
, /**< Texture samples query */
1239 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1247 enum glsl_sampler_dim sampler_dim
;
1248 nir_alu_type dest_type
;
1253 unsigned num_srcs
, coord_components
;
1254 bool is_array
, is_shadow
;
1257 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1258 * components or the new-style shadow that outputs 1 component.
1260 bool is_new_style_shadow
;
1262 /* gather component selector */
1263 unsigned component
: 2;
1265 /** The texture index
1267 * If this texture instruction has a nir_tex_src_texture_offset source,
1268 * then the texture index is given by texture_index + texture_offset.
1270 unsigned texture_index
;
1272 /** The size of the texture array or 0 if it's not an array */
1273 unsigned texture_array_size
;
1275 /** The texture deref
1277 * If this is null, use texture_index instead.
1279 nir_deref_var
*texture
;
1281 /** The sampler index
1283 * The following operations do not require a sampler and, as such, this
1284 * field should be ignored:
1286 * - nir_texop_txf_ms
1289 * - nir_texop_query_levels
1290 * - nir_texop_texture_samples
1291 * - nir_texop_samples_identical
1293 * If this texture instruction has a nir_tex_src_sampler_offset source,
1294 * then the sampler index is given by sampler_index + sampler_offset.
1296 unsigned sampler_index
;
1298 /** The sampler deref
1300 * If this is null, use sampler_index instead.
1302 nir_deref_var
*sampler
;
1305 static inline unsigned
1306 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1308 switch (instr
->op
) {
1309 case nir_texop_txs
: {
1311 switch (instr
->sampler_dim
) {
1312 case GLSL_SAMPLER_DIM_1D
:
1313 case GLSL_SAMPLER_DIM_BUF
:
1316 case GLSL_SAMPLER_DIM_2D
:
1317 case GLSL_SAMPLER_DIM_CUBE
:
1318 case GLSL_SAMPLER_DIM_MS
:
1319 case GLSL_SAMPLER_DIM_RECT
:
1320 case GLSL_SAMPLER_DIM_EXTERNAL
:
1321 case GLSL_SAMPLER_DIM_SUBPASS
:
1324 case GLSL_SAMPLER_DIM_3D
:
1328 unreachable("not reached");
1330 if (instr
->is_array
)
1338 case nir_texop_texture_samples
:
1339 case nir_texop_query_levels
:
1340 case nir_texop_samples_identical
:
1344 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1351 /* Returns true if this texture operation queries something about the texture
1352 * rather than actually sampling it.
1355 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1357 switch (instr
->op
) {
1360 case nir_texop_texture_samples
:
1361 case nir_texop_query_levels
:
1362 case nir_texop_txf_ms_mcs
:
1369 case nir_texop_txf_ms
:
1373 unreachable("Invalid texture opcode");
1378 nir_alu_instr_is_comparison(const nir_alu_instr
*instr
)
1380 switch (instr
->op
) {
1401 static inline nir_alu_type
1402 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1404 switch (instr
->src
[src
].src_type
) {
1405 case nir_tex_src_coord
:
1406 switch (instr
->op
) {
1408 case nir_texop_txf_ms
:
1409 case nir_texop_txf_ms_mcs
:
1410 case nir_texop_samples_identical
:
1411 return nir_type_int
;
1414 return nir_type_float
;
1417 case nir_tex_src_lod
:
1418 switch (instr
->op
) {
1421 return nir_type_int
;
1424 return nir_type_float
;
1427 case nir_tex_src_projector
:
1428 case nir_tex_src_comparator
:
1429 case nir_tex_src_bias
:
1430 case nir_tex_src_ddx
:
1431 case nir_tex_src_ddy
:
1432 return nir_type_float
;
1434 case nir_tex_src_offset
:
1435 case nir_tex_src_ms_index
:
1436 case nir_tex_src_texture_offset
:
1437 case nir_tex_src_sampler_offset
:
1438 return nir_type_int
;
1441 unreachable("Invalid texture source type");
1445 static inline unsigned
1446 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1448 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1449 return instr
->coord_components
;
1451 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1452 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1455 if (instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1456 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1457 if (instr
->is_array
)
1458 return instr
->coord_components
- 1;
1460 return instr
->coord_components
;
1463 /* Usual APIs don't allow cube + offset, but we allow it, with 2 coords for
1464 * the offset, since a cube maps to a single face.
1466 if (instr
->src
[src
].src_type
== nir_tex_src_offset
) {
1467 if (instr
->sampler_dim
== GLSL_SAMPLER_DIM_CUBE
)
1469 else if (instr
->is_array
)
1470 return instr
->coord_components
- 1;
1472 return instr
->coord_components
;
1479 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1481 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1482 if (instr
->src
[i
].src_type
== type
)
1488 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1489 nir_tex_src_type src_type
,
1492 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1497 nir_const_value value
;
1500 } nir_load_const_instr
;
1513 /* creates a new SSA variable in an undefined state */
1518 } nir_ssa_undef_instr
;
1521 struct exec_node node
;
1523 /* The predecessor block corresponding to this source */
1524 struct nir_block
*pred
;
1529 #define nir_foreach_phi_src(phi_src, phi) \
1530 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1531 #define nir_foreach_phi_src_safe(phi_src, phi) \
1532 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1537 struct exec_list srcs
; /** < list of nir_phi_src */
1543 struct exec_node node
;
1546 } nir_parallel_copy_entry
;
1548 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1549 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1554 /* A list of nir_parallel_copy_entrys. The sources of all of the
1555 * entries are copied to the corresponding destinations "in parallel".
1556 * In other words, if we have two entries: a -> b and b -> a, the values
1559 struct exec_list entries
;
1560 } nir_parallel_copy_instr
;
1562 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1563 type
, nir_instr_type_alu
)
1564 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1565 type
, nir_instr_type_call
)
1566 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1567 type
, nir_instr_type_jump
)
1568 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1569 type
, nir_instr_type_tex
)
1570 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1571 type
, nir_instr_type_intrinsic
)
1572 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1573 type
, nir_instr_type_load_const
)
1574 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1575 type
, nir_instr_type_ssa_undef
)
1576 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1577 type
, nir_instr_type_phi
)
1578 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1579 nir_parallel_copy_instr
, instr
,
1580 type
, nir_instr_type_parallel_copy
)
1585 * Control flow consists of a tree of control flow nodes, which include
1586 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1587 * instructions that always run start-to-finish. Each basic block also keeps
1588 * track of its successors (blocks which may run immediately after the current
1589 * block) and predecessors (blocks which could have run immediately before the
1590 * current block). Each function also has a start block and an end block which
1591 * all return statements point to (which is always empty). Together, all the
1592 * blocks with their predecessors and successors make up the control flow
1593 * graph (CFG) of the function. There are helpers that modify the tree of
1594 * control flow nodes while modifying the CFG appropriately; these should be
1595 * used instead of modifying the tree directly.
1602 nir_cf_node_function
1605 typedef struct nir_cf_node
{
1606 struct exec_node node
;
1607 nir_cf_node_type type
;
1608 struct nir_cf_node
*parent
;
1611 typedef struct nir_block
{
1612 nir_cf_node cf_node
;
1614 struct exec_list instr_list
; /** < list of nir_instr */
1616 /** generic block index; generated by nir_index_blocks */
1620 * Each block can only have up to 2 successors, so we put them in a simple
1621 * array - no need for anything more complicated.
1623 struct nir_block
*successors
[2];
1625 /* Set of nir_block predecessors in the CFG */
1626 struct set
*predecessors
;
1629 * this node's immediate dominator in the dominance tree - set to NULL for
1632 struct nir_block
*imm_dom
;
1634 /* This node's children in the dominance tree */
1635 unsigned num_dom_children
;
1636 struct nir_block
**dom_children
;
1638 /* Set of nir_blocks on the dominance frontier of this block */
1639 struct set
*dom_frontier
;
1642 * These two indices have the property that dom_{pre,post}_index for each
1643 * child of this block in the dominance tree will always be between
1644 * dom_pre_index and dom_post_index for this block, which makes testing if
1645 * a given block is dominated by another block an O(1) operation.
1647 unsigned dom_pre_index
, dom_post_index
;
1649 /* live in and out for this block; used for liveness analysis */
1650 BITSET_WORD
*live_in
;
1651 BITSET_WORD
*live_out
;
1654 static inline nir_instr
*
1655 nir_block_first_instr(nir_block
*block
)
1657 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1658 return exec_node_data(nir_instr
, head
, node
);
1661 static inline nir_instr
*
1662 nir_block_last_instr(nir_block
*block
)
1664 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1665 return exec_node_data(nir_instr
, tail
, node
);
1668 #define nir_foreach_instr(instr, block) \
1669 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1670 #define nir_foreach_instr_reverse(instr, block) \
1671 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1672 #define nir_foreach_instr_safe(instr, block) \
1673 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1674 #define nir_foreach_instr_reverse_safe(instr, block) \
1675 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1677 typedef struct nir_if
{
1678 nir_cf_node cf_node
;
1681 struct exec_list then_list
; /** < list of nir_cf_node */
1682 struct exec_list else_list
; /** < list of nir_cf_node */
1688 nir_instr
*conditional_instr
;
1690 nir_block
*break_block
;
1691 nir_block
*continue_from_block
;
1693 bool continue_from_then
;
1695 struct list_head loop_terminator_link
;
1696 } nir_loop_terminator
;
1699 /* Number of instructions in the loop */
1700 unsigned num_instructions
;
1702 /* How many times the loop is run (if known) */
1703 unsigned trip_count
;
1704 bool is_trip_count_known
;
1706 /* Unroll the loop regardless of its size */
1709 nir_loop_terminator
*limiting_terminator
;
1711 /* A list of loop_terminators terminating this loop. */
1712 struct list_head loop_terminator_list
;
1716 nir_cf_node cf_node
;
1718 struct exec_list body
; /** < list of nir_cf_node */
1720 nir_loop_info
*info
;
1724 * Various bits of metadata that can may be created or required by
1725 * optimization and analysis passes
1728 nir_metadata_none
= 0x0,
1729 nir_metadata_block_index
= 0x1,
1730 nir_metadata_dominance
= 0x2,
1731 nir_metadata_live_ssa_defs
= 0x4,
1732 nir_metadata_not_properly_reset
= 0x8,
1733 nir_metadata_loop_analysis
= 0x10,
1737 nir_cf_node cf_node
;
1739 /** pointer to the function of which this is an implementation */
1740 struct nir_function
*function
;
1742 struct exec_list body
; /** < list of nir_cf_node */
1744 nir_block
*end_block
;
1746 /** list for all local variables in the function */
1747 struct exec_list locals
;
1749 /** array of variables used as parameters */
1750 unsigned num_params
;
1751 nir_variable
**params
;
1753 /** variable used to hold the result of the function */
1754 nir_variable
*return_var
;
1756 /** list of local registers in the function */
1757 struct exec_list registers
;
1759 /** next available local register index */
1762 /** next available SSA value index */
1765 /* total number of basic blocks, only valid when block_index_dirty = false */
1766 unsigned num_blocks
;
1768 nir_metadata valid_metadata
;
1769 } nir_function_impl
;
1771 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1772 nir_start_block(nir_function_impl
*impl
)
1774 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1777 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1778 nir_impl_last_block(nir_function_impl
*impl
)
1780 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1783 static inline nir_cf_node
*
1784 nir_cf_node_next(nir_cf_node
*node
)
1786 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1787 if (exec_node_is_tail_sentinel(next
))
1790 return exec_node_data(nir_cf_node
, next
, node
);
1793 static inline nir_cf_node
*
1794 nir_cf_node_prev(nir_cf_node
*node
)
1796 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1797 if (exec_node_is_head_sentinel(prev
))
1800 return exec_node_data(nir_cf_node
, prev
, node
);
1804 nir_cf_node_is_first(const nir_cf_node
*node
)
1806 return exec_node_is_head_sentinel(node
->node
.prev
);
1810 nir_cf_node_is_last(const nir_cf_node
*node
)
1812 return exec_node_is_tail_sentinel(node
->node
.next
);
1815 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1816 type
, nir_cf_node_block
)
1817 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1818 type
, nir_cf_node_if
)
1819 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1820 type
, nir_cf_node_loop
)
1821 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1822 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1824 static inline nir_block
*
1825 nir_if_first_then_block(nir_if
*if_stmt
)
1827 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1828 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1831 static inline nir_block
*
1832 nir_if_last_then_block(nir_if
*if_stmt
)
1834 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1835 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1838 static inline nir_block
*
1839 nir_if_first_else_block(nir_if
*if_stmt
)
1841 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1842 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1845 static inline nir_block
*
1846 nir_if_last_else_block(nir_if
*if_stmt
)
1848 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1849 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1852 static inline nir_block
*
1853 nir_loop_first_block(nir_loop
*loop
)
1855 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1856 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1859 static inline nir_block
*
1860 nir_loop_last_block(nir_loop
*loop
)
1862 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1863 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1869 nir_parameter_inout
,
1870 } nir_parameter_type
;
1873 nir_parameter_type param_type
;
1874 const struct glsl_type
*type
;
1877 typedef struct nir_function
{
1878 struct exec_node node
;
1881 struct nir_shader
*shader
;
1883 unsigned num_params
;
1884 nir_parameter
*params
;
1885 const struct glsl_type
*return_type
;
1887 /** The implementation of this function.
1889 * If the function is only declared and not implemented, this is NULL.
1891 nir_function_impl
*impl
;
1894 typedef struct nir_shader_compiler_options
{
1899 /** Lowers flrp when it does not support doubles */
1906 /** Lowers ibitfield_extract/ubitfield_extract to ibfe/ubfe. */
1907 bool lower_bitfield_extract
;
1908 /** Lowers ibitfield_extract/ubitfield_extract to bfm, compares, shifts. */
1909 bool lower_bitfield_extract_to_shifts
;
1910 /** Lowers bitfield_insert to bfi/bfm */
1911 bool lower_bitfield_insert
;
1912 /** Lowers bitfield_insert to bfm, compares, and shifts. */
1913 bool lower_bitfield_insert_to_shifts
;
1914 /** Lowers bitfield_reverse to shifts. */
1915 bool lower_bitfield_reverse
;
1916 /** Lowers bfm to shifts and subtracts. */
1918 /** Lowers ifind_msb to compare and ufind_msb */
1919 bool lower_ifind_msb
;
1920 /** Lowers find_lsb to ufind_msb and logic ops */
1921 bool lower_find_lsb
;
1922 bool lower_uadd_carry
;
1923 bool lower_usub_borrow
;
1924 /** Lowers imul_high/umul_high to 16-bit multiplies and carry operations. */
1925 bool lower_mul_high
;
1926 /** lowers fneg and ineg to fsub and isub. */
1928 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1931 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1934 /** enables rules to lower idiv by power-of-two: */
1937 /* lower b2f to iand */
1940 /* Does the native fdot instruction replicate its result for four
1941 * components? If so, then opt_algebraic_late will turn all fdotN
1942 * instructions into fdot_replicatedN instructions.
1944 bool fdot_replicates
;
1946 /** lowers ffract to fsub+ffloor: */
1951 bool lower_pack_half_2x16
;
1952 bool lower_pack_unorm_2x16
;
1953 bool lower_pack_snorm_2x16
;
1954 bool lower_pack_unorm_4x8
;
1955 bool lower_pack_snorm_4x8
;
1956 bool lower_unpack_half_2x16
;
1957 bool lower_unpack_unorm_2x16
;
1958 bool lower_unpack_snorm_2x16
;
1959 bool lower_unpack_unorm_4x8
;
1960 bool lower_unpack_snorm_4x8
;
1962 bool lower_extract_byte
;
1963 bool lower_extract_word
;
1965 bool lower_all_io_to_temps
;
1968 * Does the driver support real 32-bit integers? (Otherwise, integers
1969 * are simulated by floats.)
1971 bool native_integers
;
1973 /* Indicates that the driver only has zero-based vertex id */
1974 bool vertex_id_zero_based
;
1977 * If enabled, gl_BaseVertex will be lowered as:
1978 * is_indexed_draw (~0/0) & firstvertex
1980 bool lower_base_vertex
;
1982 bool lower_cs_local_index_from_id
;
1984 bool lower_device_index_to_zero
;
1987 * Should nir_lower_io() create load_interpolated_input intrinsics?
1989 * If not, it generates regular load_input intrinsics and interpolation
1990 * information must be inferred from the list of input nir_variables.
1992 bool use_interpolated_input_intrinsics
;
1995 * Do vertex shader double inputs use two locations? The Vulkan spec
1996 * requires two locations to be used, OpenGL allows a single location.
1998 bool vs_inputs_dual_locations
;
2000 unsigned max_unroll_iterations
;
2001 } nir_shader_compiler_options
;
2003 typedef struct nir_shader
{
2004 /** list of uniforms (nir_variable) */
2005 struct exec_list uniforms
;
2007 /** list of inputs (nir_variable) */
2008 struct exec_list inputs
;
2010 /** list of outputs (nir_variable) */
2011 struct exec_list outputs
;
2013 /** list of shared compute variables (nir_variable) */
2014 struct exec_list shared
;
2016 /** Set of driver-specific options for the shader.
2018 * The memory for the options is expected to be kept in a single static
2019 * copy by the driver.
2021 const struct nir_shader_compiler_options
*options
;
2023 /** Various bits of compile-time information about a given shader */
2024 struct shader_info info
;
2026 /** list of global variables in the shader (nir_variable) */
2027 struct exec_list globals
;
2029 /** list of system value variables in the shader (nir_variable) */
2030 struct exec_list system_values
;
2032 struct exec_list functions
; /** < list of nir_function */
2034 /** list of global register in the shader */
2035 struct exec_list registers
;
2037 /** next available global register index */
2041 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
2044 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
2047 static inline nir_function_impl
*
2048 nir_shader_get_entrypoint(nir_shader
*shader
)
2050 assert(exec_list_length(&shader
->functions
) == 1);
2051 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
2052 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
2053 assert(func
->return_type
== glsl_void_type());
2054 assert(func
->num_params
== 0);
2059 #define nir_foreach_function(func, shader) \
2060 foreach_list_typed(nir_function, func, node, &(shader)->functions)
2062 nir_shader
*nir_shader_create(void *mem_ctx
,
2063 gl_shader_stage stage
,
2064 const nir_shader_compiler_options
*options
,
2067 /** creates a register, including assigning it an index and adding it to the list */
2068 nir_register
*nir_global_reg_create(nir_shader
*shader
);
2070 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
2072 void nir_reg_remove(nir_register
*reg
);
2074 /** Adds a variable to the appropriate list in nir_shader */
2075 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
2078 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
2080 assert(var
->data
.mode
== nir_var_local
);
2081 exec_list_push_tail(&impl
->locals
, &var
->node
);
2084 /** creates a variable, sets a few defaults, and adds it to the list */
2085 nir_variable
*nir_variable_create(nir_shader
*shader
,
2086 nir_variable_mode mode
,
2087 const struct glsl_type
*type
,
2089 /** creates a local variable and adds it to the list */
2090 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
2091 const struct glsl_type
*type
,
2094 /** creates a function and adds it to the shader's list of functions */
2095 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
2097 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
2098 /** creates a function_impl that isn't tied to any particular function */
2099 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
2101 nir_block
*nir_block_create(nir_shader
*shader
);
2102 nir_if
*nir_if_create(nir_shader
*shader
);
2103 nir_loop
*nir_loop_create(nir_shader
*shader
);
2105 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
2107 /** requests that the given pieces of metadata be generated */
2108 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
2109 /** dirties all but the preserved metadata */
2110 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
2112 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
2113 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
2115 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
2117 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
2118 unsigned num_components
,
2121 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
2122 nir_intrinsic_op op
);
2124 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
2125 nir_function
*callee
);
2127 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
2129 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
2131 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
2133 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
2134 unsigned num_components
,
2137 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
2138 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
2139 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
2141 typedef bool (*nir_deref_foreach_leaf_cb
)(nir_deref_var
*deref
, void *state
);
2142 bool nir_deref_foreach_leaf(nir_deref_var
*deref
,
2143 nir_deref_foreach_leaf_cb cb
, void *state
);
2145 nir_load_const_instr
*
2146 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
2148 nir_const_value
nir_alu_binop_identity(nir_op binop
, unsigned bit_size
);
2151 * NIR Cursors and Instruction Insertion API
2154 * A tiny struct representing a point to insert/extract instructions or
2155 * control flow nodes. Helps reduce the combinatorial explosion of possible
2156 * points to insert/extract.
2158 * \sa nir_control_flow.h
2161 nir_cursor_before_block
,
2162 nir_cursor_after_block
,
2163 nir_cursor_before_instr
,
2164 nir_cursor_after_instr
,
2165 } nir_cursor_option
;
2168 nir_cursor_option option
;
2175 static inline nir_block
*
2176 nir_cursor_current_block(nir_cursor cursor
)
2178 if (cursor
.option
== nir_cursor_before_instr
||
2179 cursor
.option
== nir_cursor_after_instr
) {
2180 return cursor
.instr
->block
;
2182 return cursor
.block
;
2186 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2188 static inline nir_cursor
2189 nir_before_block(nir_block
*block
)
2192 cursor
.option
= nir_cursor_before_block
;
2193 cursor
.block
= block
;
2197 static inline nir_cursor
2198 nir_after_block(nir_block
*block
)
2201 cursor
.option
= nir_cursor_after_block
;
2202 cursor
.block
= block
;
2206 static inline nir_cursor
2207 nir_before_instr(nir_instr
*instr
)
2210 cursor
.option
= nir_cursor_before_instr
;
2211 cursor
.instr
= instr
;
2215 static inline nir_cursor
2216 nir_after_instr(nir_instr
*instr
)
2219 cursor
.option
= nir_cursor_after_instr
;
2220 cursor
.instr
= instr
;
2224 static inline nir_cursor
2225 nir_after_block_before_jump(nir_block
*block
)
2227 nir_instr
*last_instr
= nir_block_last_instr(block
);
2228 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2229 return nir_before_instr(last_instr
);
2231 return nir_after_block(block
);
2235 static inline nir_cursor
2236 nir_before_cf_node(nir_cf_node
*node
)
2238 if (node
->type
== nir_cf_node_block
)
2239 return nir_before_block(nir_cf_node_as_block(node
));
2241 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2244 static inline nir_cursor
2245 nir_after_cf_node(nir_cf_node
*node
)
2247 if (node
->type
== nir_cf_node_block
)
2248 return nir_after_block(nir_cf_node_as_block(node
));
2250 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2253 static inline nir_cursor
2254 nir_after_phis(nir_block
*block
)
2256 nir_foreach_instr(instr
, block
) {
2257 if (instr
->type
!= nir_instr_type_phi
)
2258 return nir_before_instr(instr
);
2260 return nir_after_block(block
);
2263 static inline nir_cursor
2264 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2266 if (node
->type
== nir_cf_node_block
)
2267 return nir_after_block(nir_cf_node_as_block(node
));
2269 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2271 return nir_after_phis(block
);
2274 static inline nir_cursor
2275 nir_before_cf_list(struct exec_list
*cf_list
)
2277 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2278 exec_list_get_head(cf_list
), node
);
2279 return nir_before_cf_node(first_node
);
2282 static inline nir_cursor
2283 nir_after_cf_list(struct exec_list
*cf_list
)
2285 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2286 exec_list_get_tail(cf_list
), node
);
2287 return nir_after_cf_node(last_node
);
2291 * Insert a NIR instruction at the given cursor.
2293 * Note: This does not update the cursor.
2295 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2298 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2300 nir_instr_insert(nir_before_instr(instr
), before
);
2304 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2306 nir_instr_insert(nir_after_instr(instr
), after
);
2310 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2312 nir_instr_insert(nir_before_block(block
), before
);
2316 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2318 nir_instr_insert(nir_after_block(block
), after
);
2322 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2324 nir_instr_insert(nir_before_cf_node(node
), before
);
2328 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2330 nir_instr_insert(nir_after_cf_node(node
), after
);
2334 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2336 nir_instr_insert(nir_before_cf_list(list
), before
);
2340 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2342 nir_instr_insert(nir_after_cf_list(list
), after
);
2345 void nir_instr_remove_v(nir_instr
*instr
);
2347 static inline nir_cursor
2348 nir_instr_remove(nir_instr
*instr
)
2351 nir_instr
*prev
= nir_instr_prev(instr
);
2353 cursor
= nir_after_instr(prev
);
2355 cursor
= nir_before_block(instr
->block
);
2357 nir_instr_remove_v(instr
);
2363 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2364 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2365 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2366 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2368 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2369 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2371 nir_const_value
*nir_src_as_const_value(nir_src src
);
2372 bool nir_src_is_dynamically_uniform(nir_src src
);
2373 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2374 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2375 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2376 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2377 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2379 void nir_instr_rewrite_deref(nir_instr
*instr
, nir_deref_var
**deref
,
2380 nir_deref_var
*new_deref
);
2382 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2383 unsigned num_components
, unsigned bit_size
,
2385 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2386 unsigned num_components
, unsigned bit_size
,
2389 nir_ssa_dest_init_for_type(nir_instr
*instr
, nir_dest
*dest
,
2390 const struct glsl_type
*type
,
2393 assert(glsl_type_is_vector_or_scalar(type
));
2394 nir_ssa_dest_init(instr
, dest
, glsl_get_components(type
),
2395 glsl_get_bit_size(type
), name
);
2397 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2398 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2399 nir_instr
*after_me
);
2401 uint8_t nir_ssa_def_components_read(const nir_ssa_def
*def
);
2404 * finds the next basic block in source-code order, returns NULL if there is
2408 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2410 /* Performs the opposite of nir_block_cf_tree_next() */
2412 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2414 /* Gets the first block in a CF node in source-code order */
2416 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2418 /* Gets the last block in a CF node in source-code order */
2420 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2422 /* Gets the next block after a CF node in source-code order */
2424 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2426 /* Macros for loops that visit blocks in source-code order */
2428 #define nir_foreach_block(block, impl) \
2429 for (nir_block *block = nir_start_block(impl); block != NULL; \
2430 block = nir_block_cf_tree_next(block))
2432 #define nir_foreach_block_safe(block, impl) \
2433 for (nir_block *block = nir_start_block(impl), \
2434 *next = nir_block_cf_tree_next(block); \
2436 block = next, next = nir_block_cf_tree_next(block))
2438 #define nir_foreach_block_reverse(block, impl) \
2439 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2440 block = nir_block_cf_tree_prev(block))
2442 #define nir_foreach_block_reverse_safe(block, impl) \
2443 for (nir_block *block = nir_impl_last_block(impl), \
2444 *prev = nir_block_cf_tree_prev(block); \
2446 block = prev, prev = nir_block_cf_tree_prev(block))
2448 #define nir_foreach_block_in_cf_node(block, node) \
2449 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2450 block != nir_cf_node_cf_tree_next(node); \
2451 block = nir_block_cf_tree_next(block))
2453 /* If the following CF node is an if, this function returns that if.
2454 * Otherwise, it returns NULL.
2456 nir_if
*nir_block_get_following_if(nir_block
*block
);
2458 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2460 void nir_index_local_regs(nir_function_impl
*impl
);
2461 void nir_index_global_regs(nir_shader
*shader
);
2462 void nir_index_ssa_defs(nir_function_impl
*impl
);
2463 unsigned nir_index_instrs(nir_function_impl
*impl
);
2465 void nir_index_blocks(nir_function_impl
*impl
);
2467 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2468 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2469 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2471 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2472 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2473 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2474 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2475 nir_deref
*nir_deref_clone(const nir_deref
*deref
, void *mem_ctx
);
2476 nir_deref_var
*nir_deref_var_clone(const nir_deref_var
*deref
, void *mem_ctx
);
2478 nir_shader
*nir_shader_serialize_deserialize(void *mem_ctx
, nir_shader
*s
);
2481 void nir_validate_shader(nir_shader
*shader
);
2482 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2483 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2486 should_clone_nir(void)
2488 static int should_clone
= -1;
2489 if (should_clone
< 0)
2490 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2492 return should_clone
;
2496 should_serialize_deserialize_nir(void)
2498 static int test_serialize
= -1;
2499 if (test_serialize
< 0)
2500 test_serialize
= env_var_as_boolean("NIR_TEST_SERIALIZE", false);
2502 return test_serialize
;
2506 should_print_nir(void)
2508 static int should_print
= -1;
2509 if (should_print
< 0)
2510 should_print
= env_var_as_boolean("NIR_PRINT", false);
2512 return should_print
;
2515 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2516 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2517 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2518 static inline bool should_clone_nir(void) { return false; }
2519 static inline bool should_serialize_deserialize_nir(void) { return false; }
2520 static inline bool should_print_nir(void) { return false; }
2523 #define _PASS(nir, do_pass) do { \
2525 nir_validate_shader(nir); \
2526 if (should_clone_nir()) { \
2527 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2531 if (should_serialize_deserialize_nir()) { \
2532 void *mem_ctx = ralloc_parent(nir); \
2533 nir = nir_shader_serialize_deserialize(mem_ctx, nir); \
2537 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2538 nir_metadata_set_validation_flag(nir); \
2539 if (should_print_nir()) \
2540 printf("%s\n", #pass); \
2541 if (pass(nir, ##__VA_ARGS__)) { \
2543 if (should_print_nir()) \
2544 nir_print_shader(nir, stdout); \
2545 nir_metadata_check_validation_flag(nir); \
2549 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2550 if (should_print_nir()) \
2551 printf("%s\n", #pass); \
2552 pass(nir, ##__VA_ARGS__); \
2553 if (should_print_nir()) \
2554 nir_print_shader(nir, stdout); \
2557 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2558 void nir_calc_dominance(nir_shader
*shader
);
2560 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2561 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2563 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2564 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2566 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2567 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2569 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2570 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2572 int nir_gs_count_vertices(const nir_shader
*shader
);
2574 bool nir_split_var_copies(nir_shader
*shader
);
2576 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2577 bool nir_lower_returns(nir_shader
*shader
);
2579 bool nir_inline_functions(nir_shader
*shader
);
2581 bool nir_propagate_invariant(nir_shader
*shader
);
2583 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2584 bool nir_lower_var_copies(nir_shader
*shader
);
2586 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2588 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2590 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2592 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2593 nir_function_impl
*entrypoint
,
2594 bool outputs
, bool inputs
);
2596 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2598 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2599 int (*type_size
)(const struct glsl_type
*));
2601 /* Some helpers to do very simple linking */
2602 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2603 void nir_compact_varyings(nir_shader
*producer
, nir_shader
*consumer
,
2604 bool default_to_smooth_interp
);
2607 /* If set, this forces all non-flat fragment shader inputs to be
2608 * interpolated as if with the "sample" qualifier. This requires
2609 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2611 nir_lower_io_force_sample_interpolation
= (1 << 1),
2612 } nir_lower_io_options
;
2613 bool nir_lower_io(nir_shader
*shader
,
2614 nir_variable_mode modes
,
2615 int (*type_size
)(const struct glsl_type
*),
2616 nir_lower_io_options
);
2617 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2618 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2620 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2622 void nir_lower_io_types(nir_shader
*shader
);
2623 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2624 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2625 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2627 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2628 bool nir_lower_constant_initializers(nir_shader
*shader
,
2629 nir_variable_mode modes
);
2631 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2632 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2633 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
2635 bool nir_lower_alu(nir_shader
*shader
);
2636 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2637 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2638 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
2639 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2640 void nir_lower_io_arrays_to_elements(nir_shader
*producer
, nir_shader
*consumer
);
2641 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader
*shader
,
2643 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2644 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
2646 typedef struct nir_lower_subgroups_options
{
2647 uint8_t subgroup_size
;
2648 uint8_t ballot_bit_size
;
2649 bool lower_to_scalar
:1;
2650 bool lower_vote_trivial
:1;
2651 bool lower_vote_eq_to_ballot
:1;
2652 bool lower_subgroup_masks
:1;
2653 bool lower_shuffle
:1;
2654 bool lower_shuffle_to_32bit
:1;
2656 } nir_lower_subgroups_options
;
2658 bool nir_lower_subgroups(nir_shader
*shader
,
2659 const nir_lower_subgroups_options
*options
);
2661 bool nir_lower_system_values(nir_shader
*shader
);
2663 typedef struct nir_lower_tex_options
{
2665 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2666 * sampler types a texture projector is lowered.
2671 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2673 bool lower_txf_offset
;
2676 * If true, lower away nir_tex_src_offset for all rect textures.
2678 bool lower_rect_offset
;
2681 * If true, lower rect textures to 2D, using txs to fetch the
2682 * texture dimensions and dividing the texture coords by the
2683 * texture dims to normalize.
2688 * If true, convert yuv to rgb.
2690 unsigned lower_y_uv_external
;
2691 unsigned lower_y_u_v_external
;
2692 unsigned lower_yx_xuxv_external
;
2693 unsigned lower_xy_uxvx_external
;
2696 * To emulate certain texture wrap modes, this can be used
2697 * to saturate the specified tex coord to [0.0, 1.0]. The
2698 * bits are according to sampler #, ie. if, for example:
2700 * (conf->saturate_s & (1 << n))
2702 * is true, then the s coord for sampler n is saturated.
2704 * Note that clamping must happen *after* projector lowering
2705 * so any projected texture sample instruction with a clamped
2706 * coordinate gets automatically lowered, regardless of the
2707 * 'lower_txp' setting.
2709 unsigned saturate_s
;
2710 unsigned saturate_t
;
2711 unsigned saturate_r
;
2713 /* Bitmask of textures that need swizzling.
2715 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2716 * swizzles[texture_index] is applied to the result of the texturing
2719 unsigned swizzle_result
;
2721 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2722 * while 4 and 5 represent 0 and 1 respectively.
2724 uint8_t swizzles
[32][4];
2727 * Bitmap of textures that need srgb to linear conversion. If
2728 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2729 * of the texture are lowered to linear.
2731 unsigned lower_srgb
;
2734 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2736 bool lower_txd_cube_map
;
2739 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
2740 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
2741 * with lower_txd_cube_map.
2743 bool lower_txd_shadow
;
2746 * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
2747 * Implies lower_txd_cube_map and lower_txd_shadow.
2750 } nir_lower_tex_options
;
2752 bool nir_lower_tex(nir_shader
*shader
,
2753 const nir_lower_tex_options
*options
);
2755 bool nir_lower_idiv(nir_shader
*shader
);
2757 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2758 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2759 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2761 void nir_lower_two_sided_color(nir_shader
*shader
);
2763 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
2765 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2766 void nir_lower_tes_patch_vertices(nir_shader
*tes
, unsigned patch_vertices
);
2768 typedef struct nir_lower_wpos_ytransform_options
{
2769 gl_state_index16 state_tokens
[STATE_LENGTH
];
2770 bool fs_coord_origin_upper_left
:1;
2771 bool fs_coord_origin_lower_left
:1;
2772 bool fs_coord_pixel_center_integer
:1;
2773 bool fs_coord_pixel_center_half_integer
:1;
2774 } nir_lower_wpos_ytransform_options
;
2776 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2777 const nir_lower_wpos_ytransform_options
*options
);
2778 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
2780 typedef struct nir_lower_drawpixels_options
{
2781 gl_state_index16 texcoord_state_tokens
[STATE_LENGTH
];
2782 gl_state_index16 scale_state_tokens
[STATE_LENGTH
];
2783 gl_state_index16 bias_state_tokens
[STATE_LENGTH
];
2784 unsigned drawpix_sampler
;
2785 unsigned pixelmap_sampler
;
2787 bool scale_and_bias
:1;
2788 } nir_lower_drawpixels_options
;
2790 void nir_lower_drawpixels(nir_shader
*shader
,
2791 const nir_lower_drawpixels_options
*options
);
2793 typedef struct nir_lower_bitmap_options
{
2796 } nir_lower_bitmap_options
;
2798 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2800 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
2801 bool nir_lower_to_source_mods(nir_shader
*shader
);
2803 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2805 typedef unsigned (*nir_lower_bit_size_callback
)(const nir_alu_instr
*, void *);
2807 bool nir_lower_bit_size(nir_shader
*shader
,
2808 nir_lower_bit_size_callback callback
,
2809 void *callback_data
);
2812 nir_lower_imul64
= (1 << 0),
2813 nir_lower_isign64
= (1 << 1),
2814 /** Lower all int64 modulus and division opcodes */
2815 nir_lower_divmod64
= (1 << 2),
2816 } nir_lower_int64_options
;
2818 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
2821 nir_lower_drcp
= (1 << 0),
2822 nir_lower_dsqrt
= (1 << 1),
2823 nir_lower_drsq
= (1 << 2),
2824 nir_lower_dtrunc
= (1 << 3),
2825 nir_lower_dfloor
= (1 << 4),
2826 nir_lower_dceil
= (1 << 5),
2827 nir_lower_dfract
= (1 << 6),
2828 nir_lower_dround_even
= (1 << 7),
2829 nir_lower_dmod
= (1 << 8)
2830 } nir_lower_doubles_options
;
2832 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2833 bool nir_lower_pack(nir_shader
*shader
);
2835 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2837 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2839 void nir_loop_analyze_impl(nir_function_impl
*impl
,
2840 nir_variable_mode indirect_mask
);
2842 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2844 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2845 bool nir_repair_ssa(nir_shader
*shader
);
2847 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
2849 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2850 * registers. If false, convert all values (even those not involved in a phi
2851 * node) to registers.
2853 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2855 bool nir_lower_phis_to_regs_block(nir_block
*block
);
2856 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
2858 bool nir_opt_algebraic(nir_shader
*shader
);
2859 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
2860 bool nir_opt_algebraic_late(nir_shader
*shader
);
2861 bool nir_opt_constant_folding(nir_shader
*shader
);
2863 bool nir_opt_global_to_local(nir_shader
*shader
);
2865 bool nir_copy_prop(nir_shader
*shader
);
2867 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
2869 bool nir_opt_cse(nir_shader
*shader
);
2871 bool nir_opt_dce(nir_shader
*shader
);
2873 bool nir_opt_dead_cf(nir_shader
*shader
);
2875 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2877 bool nir_opt_if(nir_shader
*shader
);
2879 bool nir_opt_intrinsics(nir_shader
*shader
);
2881 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
2883 bool nir_opt_move_comparisons(nir_shader
*shader
);
2885 bool nir_opt_move_load_ubo(nir_shader
*shader
);
2887 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2889 bool nir_opt_remove_phis(nir_shader
*shader
);
2891 bool nir_opt_shrink_load(nir_shader
*shader
);
2893 bool nir_opt_trivial_continues(nir_shader
*shader
);
2895 bool nir_opt_undef(nir_shader
*shader
);
2897 bool nir_opt_conditional_discard(nir_shader
*shader
);
2899 void nir_sweep(nir_shader
*shader
);
2901 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2902 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);