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 * Was an explicit binding set in the shader?
258 unsigned explicit_binding
:1;
261 * \brief Layout qualifier for gl_FragDepth.
263 * This is not equal to \c ir_depth_layout_none if and only if this
264 * variable is \c gl_FragDepth and a layout qualifier is specified.
266 nir_depth_layout depth_layout
;
269 * Storage location of the base of this variable
271 * The precise meaning of this field depends on the nature of the variable.
273 * - Vertex shader input: one of the values from \c gl_vert_attrib.
274 * - Vertex shader output: one of the values from \c gl_varying_slot.
275 * - Geometry shader input: one of the values from \c gl_varying_slot.
276 * - Geometry shader output: one of the values from \c gl_varying_slot.
277 * - Fragment shader input: one of the values from \c gl_varying_slot.
278 * - Fragment shader output: one of the values from \c gl_frag_result.
279 * - Uniforms: Per-stage uniform slot number for default uniform block.
280 * - Uniforms: Index within the uniform block definition for UBO members.
281 * - Non-UBO Uniforms: uniform slot number.
282 * - Other: This field is not currently used.
284 * If the variable is a uniform, shader input, or shader output, and the
285 * slot has not been assigned, the value will be -1.
290 * The actual location of the variable in the IR. Only valid for inputs
293 unsigned int driver_location
;
296 * Vertex stream output identifier.
298 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
299 * stream of the i-th component.
304 * output index for dual source blending.
309 * Descriptor set binding for sampler or UBO.
314 * Initial binding point for a sampler or UBO.
316 * For array types, this represents the binding point for the first element.
321 * Location an atomic counter is stored at.
326 * ARB_shader_image_load_store qualifiers.
329 bool read_only
; /**< "readonly" qualifier. */
330 bool write_only
; /**< "writeonly" qualifier. */
335 /** Image internal format if specified explicitly, otherwise GL_NONE. */
341 * Built-in state that backs this uniform
343 * Once set at variable creation, \c state_slots must remain invariant.
344 * This is because, ideally, this array would be shared by all clones of
345 * this variable in the IR tree. In other words, we'd really like for it
346 * to be a fly-weight.
348 * If the variable is not a uniform, \c num_state_slots will be zero and
349 * \c state_slots will be \c NULL.
352 unsigned num_state_slots
; /**< Number of state slots used */
353 nir_state_slot
*state_slots
; /**< State descriptors. */
357 * Constant expression assigned in the initializer of the variable
359 * This field should only be used temporarily by creators of NIR shaders
360 * and then lower_constant_initializers can be used to get rid of them.
361 * Most of the rest of NIR ignores this field or asserts that it's NULL.
363 nir_constant
*constant_initializer
;
366 * For variables that are in an interface block or are an instance of an
367 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
369 * \sa ir_variable::location
371 const struct glsl_type
*interface_type
;
374 #define nir_foreach_variable(var, var_list) \
375 foreach_list_typed(nir_variable, var, node, var_list)
377 #define nir_foreach_variable_safe(var, var_list) \
378 foreach_list_typed_safe(nir_variable, var, node, var_list)
381 nir_variable_is_global(const nir_variable
*var
)
383 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
386 typedef struct nir_register
{
387 struct exec_node node
;
389 unsigned num_components
; /** < number of vector components */
390 unsigned num_array_elems
; /** < size of array (0 for no array) */
392 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
395 /** generic register index. */
398 /** only for debug purposes, can be NULL */
401 /** whether this register is local (per-function) or global (per-shader) */
405 * If this flag is set to true, then accessing channels >= num_components
406 * is well-defined, and simply spills over to the next array element. This
407 * is useful for backends that can do per-component accessing, in
408 * particular scalar backends. By setting this flag and making
409 * num_components equal to 1, structures can be packed tightly into
410 * registers and then registers can be accessed per-component to get to
411 * each structure member, even if it crosses vec4 boundaries.
415 /** set of nir_srcs where this register is used (read from) */
416 struct list_head uses
;
418 /** set of nir_dests where this register is defined (written to) */
419 struct list_head defs
;
421 /** set of nir_ifs where this register is used as a condition */
422 struct list_head if_uses
;
425 #define nir_foreach_register(reg, reg_list) \
426 foreach_list_typed(nir_register, reg, node, reg_list)
427 #define nir_foreach_register_safe(reg, reg_list) \
428 foreach_list_typed_safe(nir_register, reg, node, reg_list)
432 nir_instr_type_deref
,
435 nir_instr_type_intrinsic
,
436 nir_instr_type_load_const
,
438 nir_instr_type_ssa_undef
,
440 nir_instr_type_parallel_copy
,
443 typedef struct nir_instr
{
444 struct exec_node node
;
446 struct nir_block
*block
;
448 /** generic instruction index. */
451 /* A temporary for optimization and analysis passes to use for storing
452 * flags. For instance, DCE uses this to store the "dead/live" info.
457 static inline nir_instr
*
458 nir_instr_next(nir_instr
*instr
)
460 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
461 if (exec_node_is_tail_sentinel(next
))
464 return exec_node_data(nir_instr
, next
, node
);
467 static inline nir_instr
*
468 nir_instr_prev(nir_instr
*instr
)
470 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
471 if (exec_node_is_head_sentinel(prev
))
474 return exec_node_data(nir_instr
, prev
, node
);
478 nir_instr_is_first(const nir_instr
*instr
)
480 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
484 nir_instr_is_last(const nir_instr
*instr
)
486 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
489 typedef struct nir_ssa_def
{
490 /** for debugging only, can be NULL */
493 /** generic SSA definition index. */
496 /** Index into the live_in and live_out bitfields */
499 /** Instruction which produces this SSA value. */
500 nir_instr
*parent_instr
;
502 /** set of nir_instrs where this register is used (read from) */
503 struct list_head uses
;
505 /** set of nir_ifs where this register is used as a condition */
506 struct list_head if_uses
;
508 uint8_t num_components
;
510 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
518 struct nir_src
*indirect
; /** < NULL for no indirect offset */
519 unsigned base_offset
;
521 /* TODO use-def chain goes here */
525 nir_instr
*parent_instr
;
526 struct list_head def_link
;
529 struct nir_src
*indirect
; /** < NULL for no indirect offset */
530 unsigned base_offset
;
532 /* TODO def-use chain goes here */
537 typedef struct nir_src
{
539 /** Instruction that consumes this value as a source. */
540 nir_instr
*parent_instr
;
541 struct nir_if
*parent_if
;
544 struct list_head use_link
;
554 static inline nir_src
557 nir_src src
= { { NULL
} };
561 #define NIR_SRC_INIT nir_src_init()
563 #define nir_foreach_use(src, reg_or_ssa_def) \
564 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
566 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
567 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
569 #define nir_foreach_if_use(src, reg_or_ssa_def) \
570 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
572 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
573 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
584 static inline nir_dest
587 nir_dest dest
= { { { NULL
} } };
591 #define NIR_DEST_INIT nir_dest_init()
593 #define nir_foreach_def(dest, reg) \
594 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
596 #define nir_foreach_def_safe(dest, reg) \
597 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
599 static inline nir_src
600 nir_src_for_ssa(nir_ssa_def
*def
)
602 nir_src src
= NIR_SRC_INIT
;
610 static inline nir_src
611 nir_src_for_reg(nir_register
*reg
)
613 nir_src src
= NIR_SRC_INIT
;
617 src
.reg
.indirect
= NULL
;
618 src
.reg
.base_offset
= 0;
623 static inline nir_dest
624 nir_dest_for_reg(nir_register
*reg
)
626 nir_dest dest
= NIR_DEST_INIT
;
633 static inline unsigned
634 nir_src_bit_size(nir_src src
)
636 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
639 static inline unsigned
640 nir_src_num_components(nir_src src
)
642 return src
.is_ssa
? src
.ssa
->num_components
: src
.reg
.reg
->num_components
;
645 static inline unsigned
646 nir_dest_bit_size(nir_dest dest
)
648 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
651 static inline unsigned
652 nir_dest_num_components(nir_dest dest
)
654 return dest
.is_ssa
? dest
.ssa
.num_components
: dest
.reg
.reg
->num_components
;
657 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
658 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
664 * \name input modifiers
668 * For inputs interpreted as floating point, flips the sign bit. For
669 * inputs interpreted as integers, performs the two's complement negation.
674 * Clears the sign bit for floating point values, and computes the integer
675 * absolute value for integers. Note that the negate modifier acts after
676 * the absolute value modifier, therefore if both are set then all inputs
677 * will become negative.
683 * For each input component, says which component of the register it is
684 * chosen from. Note that which elements of the swizzle are used and which
685 * are ignored are based on the write mask for most opcodes - for example,
686 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
687 * a swizzle of {2, x, 1, 0} where x means "don't care."
696 * \name saturate output modifier
698 * Only valid for opcodes that output floating-point numbers. Clamps the
699 * output to between 0.0 and 1.0 inclusive.
704 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
708 nir_type_invalid
= 0, /* Not a valid type */
713 nir_type_bool32
= 32 | nir_type_bool
,
714 nir_type_int8
= 8 | nir_type_int
,
715 nir_type_int16
= 16 | nir_type_int
,
716 nir_type_int32
= 32 | nir_type_int
,
717 nir_type_int64
= 64 | nir_type_int
,
718 nir_type_uint8
= 8 | nir_type_uint
,
719 nir_type_uint16
= 16 | nir_type_uint
,
720 nir_type_uint32
= 32 | nir_type_uint
,
721 nir_type_uint64
= 64 | nir_type_uint
,
722 nir_type_float16
= 16 | nir_type_float
,
723 nir_type_float32
= 32 | nir_type_float
,
724 nir_type_float64
= 64 | nir_type_float
,
727 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
728 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
730 static inline unsigned
731 nir_alu_type_get_type_size(nir_alu_type type
)
733 return type
& NIR_ALU_TYPE_SIZE_MASK
;
736 static inline unsigned
737 nir_alu_type_get_base_type(nir_alu_type type
)
739 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
742 static inline nir_alu_type
743 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
747 return nir_type_bool32
;
750 return nir_type_uint32
;
753 return nir_type_int32
;
755 case GLSL_TYPE_UINT16
:
756 return nir_type_uint16
;
758 case GLSL_TYPE_INT16
:
759 return nir_type_int16
;
761 case GLSL_TYPE_UINT8
:
762 return nir_type_uint8
;
764 return nir_type_int8
;
765 case GLSL_TYPE_UINT64
:
766 return nir_type_uint64
;
768 case GLSL_TYPE_INT64
:
769 return nir_type_int64
;
771 case GLSL_TYPE_FLOAT
:
772 return nir_type_float32
;
774 case GLSL_TYPE_FLOAT16
:
775 return nir_type_float16
;
777 case GLSL_TYPE_DOUBLE
:
778 return nir_type_float64
;
781 unreachable("unknown type");
785 static inline nir_alu_type
786 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
788 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
791 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
,
792 nir_rounding_mode rnd
);
795 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
796 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
797 } nir_op_algebraic_property
;
805 * The number of components in the output
807 * If non-zero, this is the size of the output and input sizes are
808 * explicitly given; swizzle and writemask are still in effect, but if
809 * the output component is masked out, then the input component may
812 * If zero, the opcode acts in the standard, per-component manner; the
813 * operation is performed on each component (except the ones that are
814 * masked out) with the input being taken from the input swizzle for
817 * The size of some of the inputs may be given (i.e. non-zero) even
818 * though output_size is zero; in that case, the inputs with a zero
819 * size act per-component, while the inputs with non-zero size don't.
821 unsigned output_size
;
824 * The type of vector that the instruction outputs. Note that the
825 * staurate modifier is only allowed on outputs with the float type.
828 nir_alu_type output_type
;
831 * The number of components in each input
833 unsigned input_sizes
[4];
836 * The type of vector that each input takes. Note that negate and
837 * absolute value are only allowed on inputs with int or float type and
838 * behave differently on the two.
840 nir_alu_type input_types
[4];
842 nir_op_algebraic_property algebraic_properties
;
845 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
847 typedef struct nir_alu_instr
{
851 /** Indicates that this ALU instruction generates an exact value
853 * This is kind of a mixture of GLSL "precise" and "invariant" and not
854 * really equivalent to either. This indicates that the value generated by
855 * this operation is high-precision and any code transformations that touch
856 * it must ensure that the resulting value is bit-for-bit identical to the
865 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
866 nir_alu_instr
*instr
);
867 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
868 nir_alu_instr
*instr
);
870 /* is this source channel used? */
872 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
875 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
876 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
878 return (instr
->dest
.write_mask
>> channel
) & 1;
882 * For instructions whose destinations are SSA, get the number of channels
885 static inline unsigned
886 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
888 assert(instr
->dest
.dest
.is_ssa
);
890 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
891 return nir_op_infos
[instr
->op
].input_sizes
[src
];
893 return instr
->dest
.dest
.ssa
.num_components
;
896 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
897 unsigned src1
, unsigned src2
);
901 nir_deref_type_array
,
902 nir_deref_type_array_wildcard
,
903 nir_deref_type_struct
,
907 typedef struct nir_deref
{
908 nir_deref_type deref_type
;
909 struct nir_deref
*child
;
910 const struct glsl_type
*type
;
919 /* This enum describes how the array is referenced. If the deref is
920 * direct then the base_offset is used. If the deref is indirect then
921 * offset is given by base_offset + indirect. If the deref is a wildcard
922 * then the deref refers to all of the elements of the array at the same
923 * time. Wildcard dereferences are only ever allowed in copy_var
924 * intrinsics and the source and destination derefs must have matching
928 nir_deref_array_type_direct
,
929 nir_deref_array_type_indirect
,
930 nir_deref_array_type_wildcard
,
931 } nir_deref_array_type
;
936 nir_deref_array_type deref_array_type
;
937 unsigned base_offset
;
947 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
,
948 deref_type
, nir_deref_type_var
)
949 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
,
950 deref_type
, nir_deref_type_array
)
951 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
,
952 deref_type
, nir_deref_type_struct
)
954 /* Returns the last deref in the chain. */
955 static inline nir_deref
*
956 nir_deref_tail(nir_deref
*deref
)
959 deref
= deref
->child
;
966 /** The type of this deref instruction */
967 nir_deref_type deref_type
;
969 /** The mode of the underlying variable */
970 nir_variable_mode mode
;
972 /** The dereferenced type of the resulting pointer value */
973 const struct glsl_type
*type
;
976 /** Variable being dereferenced if deref_type is a deref_var */
979 /** Parent deref if deref_type is not deref_var */
983 /** Additional deref parameters */
994 /** Destination to store the resulting "pointer" */
998 NIR_DEFINE_CAST(nir_instr_as_deref
, nir_instr
, nir_deref_instr
, instr
,
999 type
, nir_instr_type_deref
)
1001 static inline nir_deref_instr
*
1002 nir_src_as_deref(nir_src src
)
1007 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_deref
)
1010 return nir_instr_as_deref(src
.ssa
->parent_instr
);
1013 static inline nir_deref_instr
*
1014 nir_deref_instr_parent(const nir_deref_instr
*instr
)
1016 if (instr
->deref_type
== nir_deref_type_var
)
1019 return nir_src_as_deref(instr
->parent
);
1022 static inline nir_variable
*
1023 nir_deref_instr_get_variable(const nir_deref_instr
*instr
)
1025 while (instr
->deref_type
!= nir_deref_type_var
) {
1026 if (instr
->deref_type
== nir_deref_type_cast
)
1029 instr
= nir_deref_instr_parent(instr
);
1035 bool nir_deref_instr_remove_if_unused(nir_deref_instr
*instr
);
1038 nir_deref_instr_to_deref(nir_deref_instr
*instr
, void *mem_ctx
);
1043 unsigned num_params
;
1044 nir_deref_var
**params
;
1045 nir_deref_var
*return_deref
;
1047 struct nir_function
*callee
;
1050 #include "nir_intrinsics.h"
1052 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
1054 /** Represents an intrinsic
1056 * An intrinsic is an instruction type for handling things that are
1057 * more-or-less regular operations but don't just consume and produce SSA
1058 * values like ALU operations do. Intrinsics are not for things that have
1059 * special semantic meaning such as phi nodes and parallel copies.
1060 * Examples of intrinsics include variable load/store operations, system
1061 * value loads, and the like. Even though texturing more-or-less falls
1062 * under this category, texturing is its own instruction type because
1063 * trying to represent texturing with intrinsics would lead to a
1064 * combinatorial explosion of intrinsic opcodes.
1066 * By having a single instruction type for handling a lot of different
1067 * cases, optimization passes can look for intrinsics and, for the most
1068 * part, completely ignore them. Each intrinsic type also has a few
1069 * possible flags that govern whether or not they can be reordered or
1070 * eliminated. That way passes like dead code elimination can still work
1071 * on intrisics without understanding the meaning of each.
1073 * Each intrinsic has some number of constant indices, some number of
1074 * variables, and some number of sources. What these sources, variables,
1075 * and indices mean depends on the intrinsic and is documented with the
1076 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
1077 * instructions are the only types of instruction that can operate on
1083 nir_intrinsic_op intrinsic
;
1087 /** number of components if this is a vectorized intrinsic
1089 * Similarly to ALU operations, some intrinsics are vectorized.
1090 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
1091 * For vectorized intrinsics, the num_components field specifies the
1092 * number of destination components and the number of source components
1093 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
1095 uint8_t num_components
;
1097 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
1099 nir_deref_var
*variables
[2];
1102 } nir_intrinsic_instr
;
1105 * \name NIR intrinsics semantic flags
1107 * information about what the compiler can do with the intrinsics.
1109 * \sa nir_intrinsic_info::flags
1113 * whether the intrinsic can be safely eliminated if none of its output
1114 * value is not being used.
1116 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
1119 * Whether the intrinsic can be reordered with respect to any other
1120 * intrinsic, i.e. whether the only reordering dependencies of the
1121 * intrinsic are due to the register reads/writes.
1123 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
1124 } nir_intrinsic_semantic_flag
;
1127 * \name NIR intrinsics const-index flag
1129 * Indicates the usage of a const_index slot.
1131 * \sa nir_intrinsic_info::index_map
1135 * Generally instructions that take a offset src argument, can encode
1136 * a constant 'base' value which is added to the offset.
1138 NIR_INTRINSIC_BASE
= 1,
1141 * For store instructions, a writemask for the store.
1143 NIR_INTRINSIC_WRMASK
= 2,
1146 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1148 NIR_INTRINSIC_STREAM_ID
= 3,
1151 * The clip-plane id for load_user_clip_plane intrinsic.
1153 NIR_INTRINSIC_UCP_ID
= 4,
1156 * The amount of data, starting from BASE, that this instruction may
1157 * access. This is used to provide bounds if the offset is not constant.
1159 NIR_INTRINSIC_RANGE
= 5,
1162 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1164 NIR_INTRINSIC_DESC_SET
= 6,
1167 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1169 NIR_INTRINSIC_BINDING
= 7,
1174 NIR_INTRINSIC_COMPONENT
= 8,
1177 * Interpolation mode (only meaningful for FS inputs).
1179 NIR_INTRINSIC_INTERP_MODE
= 9,
1182 * A binary nir_op to use when performing a reduction or scan operation
1184 NIR_INTRINSIC_REDUCTION_OP
= 10,
1187 * Cluster size for reduction operations
1189 NIR_INTRINSIC_CLUSTER_SIZE
= 11,
1191 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1193 } nir_intrinsic_index_flag
;
1195 #define NIR_INTRINSIC_MAX_INPUTS 5
1200 unsigned num_srcs
; /** < number of register/SSA inputs */
1202 /** number of components of each input register
1204 * If this value is 0, the number of components is given by the
1205 * num_components field of nir_intrinsic_instr.
1207 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1211 /** number of components of the output register
1213 * If this value is 0, the number of components is given by the
1214 * num_components field of nir_intrinsic_instr.
1216 unsigned dest_components
;
1218 /** the number of inputs/outputs that are variables */
1219 unsigned num_variables
;
1221 /** the number of constant indices used by the intrinsic */
1222 unsigned num_indices
;
1224 /** indicates the usage of intr->const_index[n] */
1225 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1227 /** semantic flags for calls to this intrinsic */
1228 nir_intrinsic_semantic_flag flags
;
1229 } nir_intrinsic_info
;
1231 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1233 static inline unsigned
1234 nir_intrinsic_src_components(nir_intrinsic_instr
*intr
, unsigned srcn
)
1236 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1237 assert(srcn
< info
->num_srcs
);
1238 if (info
->src_components
[srcn
])
1239 return info
->src_components
[srcn
];
1241 return intr
->num_components
;
1244 static inline unsigned
1245 nir_intrinsic_dest_components(nir_intrinsic_instr
*intr
)
1247 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1248 if (!info
->has_dest
)
1250 else if (info
->dest_components
)
1251 return info
->dest_components
;
1253 return intr
->num_components
;
1256 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1257 static inline type \
1258 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1260 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1261 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1262 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1264 static inline void \
1265 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1267 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1268 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1269 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1272 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1273 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1274 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1275 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1276 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1277 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1278 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1279 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1280 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1281 INTRINSIC_IDX_ACCESSORS(reduction_op
, REDUCTION_OP
, unsigned)
1282 INTRINSIC_IDX_ACCESSORS(cluster_size
, CLUSTER_SIZE
, unsigned)
1285 * \group texture information
1287 * This gives semantic information about textures which is useful to the
1288 * frontend, the backend, and lowering passes, but not the optimizer.
1293 nir_tex_src_projector
,
1294 nir_tex_src_comparator
, /* shadow comparator */
1298 nir_tex_src_ms_index
, /* MSAA sample index */
1299 nir_tex_src_ms_mcs
, /* MSAA compression value */
1302 nir_tex_src_texture_deref
, /* < deref pointing to the texture */
1303 nir_tex_src_sampler_deref
, /* < deref pointing to the sampler */
1304 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1305 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1306 nir_tex_src_plane
, /* < selects plane for planar textures */
1307 nir_num_tex_src_types
1312 nir_tex_src_type src_type
;
1316 nir_texop_tex
, /**< Regular texture look-up */
1317 nir_texop_txb
, /**< Texture look-up with LOD bias */
1318 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1319 nir_texop_txd
, /**< Texture look-up with partial derivatives */
1320 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1321 nir_texop_txf_ms
, /**< Multisample texture fetch */
1322 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1323 nir_texop_txs
, /**< Texture size */
1324 nir_texop_lod
, /**< Texture lod query */
1325 nir_texop_tg4
, /**< Texture gather */
1326 nir_texop_query_levels
, /**< Texture levels query */
1327 nir_texop_texture_samples
, /**< Texture samples query */
1328 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1336 enum glsl_sampler_dim sampler_dim
;
1337 nir_alu_type dest_type
;
1342 unsigned num_srcs
, coord_components
;
1343 bool is_array
, is_shadow
;
1346 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1347 * components or the new-style shadow that outputs 1 component.
1349 bool is_new_style_shadow
;
1351 /* gather component selector */
1352 unsigned component
: 2;
1354 /** The texture index
1356 * If this texture instruction has a nir_tex_src_texture_offset source,
1357 * then the texture index is given by texture_index + texture_offset.
1359 unsigned texture_index
;
1361 /** The size of the texture array or 0 if it's not an array */
1362 unsigned texture_array_size
;
1364 /** The texture deref
1366 * If this is null, use texture_index instead.
1368 nir_deref_var
*texture
;
1370 /** The sampler index
1372 * The following operations do not require a sampler and, as such, this
1373 * field should be ignored:
1375 * - nir_texop_txf_ms
1378 * - nir_texop_query_levels
1379 * - nir_texop_texture_samples
1380 * - nir_texop_samples_identical
1382 * If this texture instruction has a nir_tex_src_sampler_offset source,
1383 * then the sampler index is given by sampler_index + sampler_offset.
1385 unsigned sampler_index
;
1387 /** The sampler deref
1389 * If this is null, use sampler_index instead.
1391 nir_deref_var
*sampler
;
1394 static inline unsigned
1395 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1397 switch (instr
->op
) {
1398 case nir_texop_txs
: {
1400 switch (instr
->sampler_dim
) {
1401 case GLSL_SAMPLER_DIM_1D
:
1402 case GLSL_SAMPLER_DIM_BUF
:
1405 case GLSL_SAMPLER_DIM_2D
:
1406 case GLSL_SAMPLER_DIM_CUBE
:
1407 case GLSL_SAMPLER_DIM_MS
:
1408 case GLSL_SAMPLER_DIM_RECT
:
1409 case GLSL_SAMPLER_DIM_EXTERNAL
:
1410 case GLSL_SAMPLER_DIM_SUBPASS
:
1413 case GLSL_SAMPLER_DIM_3D
:
1417 unreachable("not reached");
1419 if (instr
->is_array
)
1427 case nir_texop_texture_samples
:
1428 case nir_texop_query_levels
:
1429 case nir_texop_samples_identical
:
1433 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1440 /* Returns true if this texture operation queries something about the texture
1441 * rather than actually sampling it.
1444 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1446 switch (instr
->op
) {
1449 case nir_texop_texture_samples
:
1450 case nir_texop_query_levels
:
1451 case nir_texop_txf_ms_mcs
:
1458 case nir_texop_txf_ms
:
1462 unreachable("Invalid texture opcode");
1467 nir_alu_instr_is_comparison(const nir_alu_instr
*instr
)
1469 switch (instr
->op
) {
1490 static inline nir_alu_type
1491 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1493 switch (instr
->src
[src
].src_type
) {
1494 case nir_tex_src_coord
:
1495 switch (instr
->op
) {
1497 case nir_texop_txf_ms
:
1498 case nir_texop_txf_ms_mcs
:
1499 case nir_texop_samples_identical
:
1500 return nir_type_int
;
1503 return nir_type_float
;
1506 case nir_tex_src_lod
:
1507 switch (instr
->op
) {
1510 return nir_type_int
;
1513 return nir_type_float
;
1516 case nir_tex_src_projector
:
1517 case nir_tex_src_comparator
:
1518 case nir_tex_src_bias
:
1519 case nir_tex_src_ddx
:
1520 case nir_tex_src_ddy
:
1521 return nir_type_float
;
1523 case nir_tex_src_offset
:
1524 case nir_tex_src_ms_index
:
1525 case nir_tex_src_texture_offset
:
1526 case nir_tex_src_sampler_offset
:
1527 return nir_type_int
;
1530 unreachable("Invalid texture source type");
1534 static inline unsigned
1535 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1537 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1538 return instr
->coord_components
;
1540 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1541 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1544 if (instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1545 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1546 if (instr
->is_array
)
1547 return instr
->coord_components
- 1;
1549 return instr
->coord_components
;
1552 /* Usual APIs don't allow cube + offset, but we allow it, with 2 coords for
1553 * the offset, since a cube maps to a single face.
1555 if (instr
->src
[src
].src_type
== nir_tex_src_offset
) {
1556 if (instr
->sampler_dim
== GLSL_SAMPLER_DIM_CUBE
)
1558 else if (instr
->is_array
)
1559 return instr
->coord_components
- 1;
1561 return instr
->coord_components
;
1568 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1570 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1571 if (instr
->src
[i
].src_type
== type
)
1577 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1578 nir_tex_src_type src_type
,
1581 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1586 nir_const_value value
;
1589 } nir_load_const_instr
;
1602 /* creates a new SSA variable in an undefined state */
1607 } nir_ssa_undef_instr
;
1610 struct exec_node node
;
1612 /* The predecessor block corresponding to this source */
1613 struct nir_block
*pred
;
1618 #define nir_foreach_phi_src(phi_src, phi) \
1619 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1620 #define nir_foreach_phi_src_safe(phi_src, phi) \
1621 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1626 struct exec_list srcs
; /** < list of nir_phi_src */
1632 struct exec_node node
;
1635 } nir_parallel_copy_entry
;
1637 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1638 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1643 /* A list of nir_parallel_copy_entrys. The sources of all of the
1644 * entries are copied to the corresponding destinations "in parallel".
1645 * In other words, if we have two entries: a -> b and b -> a, the values
1648 struct exec_list entries
;
1649 } nir_parallel_copy_instr
;
1651 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1652 type
, nir_instr_type_alu
)
1653 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1654 type
, nir_instr_type_call
)
1655 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1656 type
, nir_instr_type_jump
)
1657 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1658 type
, nir_instr_type_tex
)
1659 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1660 type
, nir_instr_type_intrinsic
)
1661 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1662 type
, nir_instr_type_load_const
)
1663 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1664 type
, nir_instr_type_ssa_undef
)
1665 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1666 type
, nir_instr_type_phi
)
1667 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1668 nir_parallel_copy_instr
, instr
,
1669 type
, nir_instr_type_parallel_copy
)
1674 * Control flow consists of a tree of control flow nodes, which include
1675 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1676 * instructions that always run start-to-finish. Each basic block also keeps
1677 * track of its successors (blocks which may run immediately after the current
1678 * block) and predecessors (blocks which could have run immediately before the
1679 * current block). Each function also has a start block and an end block which
1680 * all return statements point to (which is always empty). Together, all the
1681 * blocks with their predecessors and successors make up the control flow
1682 * graph (CFG) of the function. There are helpers that modify the tree of
1683 * control flow nodes while modifying the CFG appropriately; these should be
1684 * used instead of modifying the tree directly.
1691 nir_cf_node_function
1694 typedef struct nir_cf_node
{
1695 struct exec_node node
;
1696 nir_cf_node_type type
;
1697 struct nir_cf_node
*parent
;
1700 typedef struct nir_block
{
1701 nir_cf_node cf_node
;
1703 struct exec_list instr_list
; /** < list of nir_instr */
1705 /** generic block index; generated by nir_index_blocks */
1709 * Each block can only have up to 2 successors, so we put them in a simple
1710 * array - no need for anything more complicated.
1712 struct nir_block
*successors
[2];
1714 /* Set of nir_block predecessors in the CFG */
1715 struct set
*predecessors
;
1718 * this node's immediate dominator in the dominance tree - set to NULL for
1721 struct nir_block
*imm_dom
;
1723 /* This node's children in the dominance tree */
1724 unsigned num_dom_children
;
1725 struct nir_block
**dom_children
;
1727 /* Set of nir_blocks on the dominance frontier of this block */
1728 struct set
*dom_frontier
;
1731 * These two indices have the property that dom_{pre,post}_index for each
1732 * child of this block in the dominance tree will always be between
1733 * dom_pre_index and dom_post_index for this block, which makes testing if
1734 * a given block is dominated by another block an O(1) operation.
1736 unsigned dom_pre_index
, dom_post_index
;
1738 /* live in and out for this block; used for liveness analysis */
1739 BITSET_WORD
*live_in
;
1740 BITSET_WORD
*live_out
;
1743 static inline nir_instr
*
1744 nir_block_first_instr(nir_block
*block
)
1746 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1747 return exec_node_data(nir_instr
, head
, node
);
1750 static inline nir_instr
*
1751 nir_block_last_instr(nir_block
*block
)
1753 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1754 return exec_node_data(nir_instr
, tail
, node
);
1757 #define nir_foreach_instr(instr, block) \
1758 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1759 #define nir_foreach_instr_reverse(instr, block) \
1760 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1761 #define nir_foreach_instr_safe(instr, block) \
1762 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1763 #define nir_foreach_instr_reverse_safe(instr, block) \
1764 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1766 typedef struct nir_if
{
1767 nir_cf_node cf_node
;
1770 struct exec_list then_list
; /** < list of nir_cf_node */
1771 struct exec_list else_list
; /** < list of nir_cf_node */
1777 nir_instr
*conditional_instr
;
1779 nir_block
*break_block
;
1780 nir_block
*continue_from_block
;
1782 bool continue_from_then
;
1784 struct list_head loop_terminator_link
;
1785 } nir_loop_terminator
;
1788 /* Number of instructions in the loop */
1789 unsigned num_instructions
;
1791 /* How many times the loop is run (if known) */
1792 unsigned trip_count
;
1793 bool is_trip_count_known
;
1795 /* Unroll the loop regardless of its size */
1798 nir_loop_terminator
*limiting_terminator
;
1800 /* A list of loop_terminators terminating this loop. */
1801 struct list_head loop_terminator_list
;
1805 nir_cf_node cf_node
;
1807 struct exec_list body
; /** < list of nir_cf_node */
1809 nir_loop_info
*info
;
1813 * Various bits of metadata that can may be created or required by
1814 * optimization and analysis passes
1817 nir_metadata_none
= 0x0,
1818 nir_metadata_block_index
= 0x1,
1819 nir_metadata_dominance
= 0x2,
1820 nir_metadata_live_ssa_defs
= 0x4,
1821 nir_metadata_not_properly_reset
= 0x8,
1822 nir_metadata_loop_analysis
= 0x10,
1826 nir_cf_node cf_node
;
1828 /** pointer to the function of which this is an implementation */
1829 struct nir_function
*function
;
1831 struct exec_list body
; /** < list of nir_cf_node */
1833 nir_block
*end_block
;
1835 /** list for all local variables in the function */
1836 struct exec_list locals
;
1838 /** array of variables used as parameters */
1839 unsigned num_params
;
1840 nir_variable
**params
;
1842 /** variable used to hold the result of the function */
1843 nir_variable
*return_var
;
1845 /** list of local registers in the function */
1846 struct exec_list registers
;
1848 /** next available local register index */
1851 /** next available SSA value index */
1854 /* total number of basic blocks, only valid when block_index_dirty = false */
1855 unsigned num_blocks
;
1857 nir_metadata valid_metadata
;
1858 } nir_function_impl
;
1860 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1861 nir_start_block(nir_function_impl
*impl
)
1863 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1866 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1867 nir_impl_last_block(nir_function_impl
*impl
)
1869 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1872 static inline nir_cf_node
*
1873 nir_cf_node_next(nir_cf_node
*node
)
1875 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1876 if (exec_node_is_tail_sentinel(next
))
1879 return exec_node_data(nir_cf_node
, next
, node
);
1882 static inline nir_cf_node
*
1883 nir_cf_node_prev(nir_cf_node
*node
)
1885 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1886 if (exec_node_is_head_sentinel(prev
))
1889 return exec_node_data(nir_cf_node
, prev
, node
);
1893 nir_cf_node_is_first(const nir_cf_node
*node
)
1895 return exec_node_is_head_sentinel(node
->node
.prev
);
1899 nir_cf_node_is_last(const nir_cf_node
*node
)
1901 return exec_node_is_tail_sentinel(node
->node
.next
);
1904 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1905 type
, nir_cf_node_block
)
1906 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1907 type
, nir_cf_node_if
)
1908 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1909 type
, nir_cf_node_loop
)
1910 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1911 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1913 static inline nir_block
*
1914 nir_if_first_then_block(nir_if
*if_stmt
)
1916 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1917 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1920 static inline nir_block
*
1921 nir_if_last_then_block(nir_if
*if_stmt
)
1923 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1924 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1927 static inline nir_block
*
1928 nir_if_first_else_block(nir_if
*if_stmt
)
1930 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1931 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1934 static inline nir_block
*
1935 nir_if_last_else_block(nir_if
*if_stmt
)
1937 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1938 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1941 static inline nir_block
*
1942 nir_loop_first_block(nir_loop
*loop
)
1944 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1945 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1948 static inline nir_block
*
1949 nir_loop_last_block(nir_loop
*loop
)
1951 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1952 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1958 nir_parameter_inout
,
1959 } nir_parameter_type
;
1962 nir_parameter_type param_type
;
1963 const struct glsl_type
*type
;
1966 typedef struct nir_function
{
1967 struct exec_node node
;
1970 struct nir_shader
*shader
;
1972 unsigned num_params
;
1973 nir_parameter
*params
;
1974 const struct glsl_type
*return_type
;
1976 /** The implementation of this function.
1978 * If the function is only declared and not implemented, this is NULL.
1980 nir_function_impl
*impl
;
1983 typedef struct nir_shader_compiler_options
{
1988 /** Lowers flrp when it does not support doubles */
1995 /** Lowers ibitfield_extract/ubitfield_extract to ibfe/ubfe. */
1996 bool lower_bitfield_extract
;
1997 /** Lowers ibitfield_extract/ubitfield_extract to bfm, compares, shifts. */
1998 bool lower_bitfield_extract_to_shifts
;
1999 /** Lowers bitfield_insert to bfi/bfm */
2000 bool lower_bitfield_insert
;
2001 /** Lowers bitfield_insert to bfm, compares, and shifts. */
2002 bool lower_bitfield_insert_to_shifts
;
2003 /** Lowers bitfield_reverse to shifts. */
2004 bool lower_bitfield_reverse
;
2005 /** Lowers bit_count to shifts. */
2006 bool lower_bit_count
;
2007 /** Lowers bfm to shifts and subtracts. */
2009 /** Lowers ifind_msb to compare and ufind_msb */
2010 bool lower_ifind_msb
;
2011 /** Lowers find_lsb to ufind_msb and logic ops */
2012 bool lower_find_lsb
;
2013 bool lower_uadd_carry
;
2014 bool lower_usub_borrow
;
2015 /** Lowers imul_high/umul_high to 16-bit multiplies and carry operations. */
2016 bool lower_mul_high
;
2017 /** lowers fneg and ineg to fsub and isub. */
2019 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
2022 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
2025 /** enables rules to lower idiv by power-of-two: */
2028 /* lower b2f to iand */
2031 /* Does the native fdot instruction replicate its result for four
2032 * components? If so, then opt_algebraic_late will turn all fdotN
2033 * instructions into fdot_replicatedN instructions.
2035 bool fdot_replicates
;
2037 /** lowers ffract to fsub+ffloor: */
2042 bool lower_pack_half_2x16
;
2043 bool lower_pack_unorm_2x16
;
2044 bool lower_pack_snorm_2x16
;
2045 bool lower_pack_unorm_4x8
;
2046 bool lower_pack_snorm_4x8
;
2047 bool lower_unpack_half_2x16
;
2048 bool lower_unpack_unorm_2x16
;
2049 bool lower_unpack_snorm_2x16
;
2050 bool lower_unpack_unorm_4x8
;
2051 bool lower_unpack_snorm_4x8
;
2053 bool lower_extract_byte
;
2054 bool lower_extract_word
;
2056 bool lower_all_io_to_temps
;
2059 * Does the driver support real 32-bit integers? (Otherwise, integers
2060 * are simulated by floats.)
2062 bool native_integers
;
2064 /* Indicates that the driver only has zero-based vertex id */
2065 bool vertex_id_zero_based
;
2068 * If enabled, gl_BaseVertex will be lowered as:
2069 * is_indexed_draw (~0/0) & firstvertex
2071 bool lower_base_vertex
;
2073 bool lower_cs_local_index_from_id
;
2075 bool lower_device_index_to_zero
;
2078 * Should nir_lower_io() create load_interpolated_input intrinsics?
2080 * If not, it generates regular load_input intrinsics and interpolation
2081 * information must be inferred from the list of input nir_variables.
2083 bool use_interpolated_input_intrinsics
;
2086 * Do vertex shader double inputs use two locations? The Vulkan spec
2087 * requires two locations to be used, OpenGL allows a single location.
2089 bool vs_inputs_dual_locations
;
2091 unsigned max_unroll_iterations
;
2092 } nir_shader_compiler_options
;
2094 typedef struct nir_shader
{
2095 /** list of uniforms (nir_variable) */
2096 struct exec_list uniforms
;
2098 /** list of inputs (nir_variable) */
2099 struct exec_list inputs
;
2101 /** list of outputs (nir_variable) */
2102 struct exec_list outputs
;
2104 /** list of shared compute variables (nir_variable) */
2105 struct exec_list shared
;
2107 /** Set of driver-specific options for the shader.
2109 * The memory for the options is expected to be kept in a single static
2110 * copy by the driver.
2112 const struct nir_shader_compiler_options
*options
;
2114 /** Various bits of compile-time information about a given shader */
2115 struct shader_info info
;
2117 /** list of global variables in the shader (nir_variable) */
2118 struct exec_list globals
;
2120 /** list of system value variables in the shader (nir_variable) */
2121 struct exec_list system_values
;
2123 struct exec_list functions
; /** < list of nir_function */
2125 /** list of global register in the shader */
2126 struct exec_list registers
;
2128 /** next available global register index */
2132 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
2135 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
2138 static inline nir_function_impl
*
2139 nir_shader_get_entrypoint(nir_shader
*shader
)
2141 assert(exec_list_length(&shader
->functions
) == 1);
2142 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
2143 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
2144 assert(func
->return_type
== glsl_void_type());
2145 assert(func
->num_params
== 0);
2150 #define nir_foreach_function(func, shader) \
2151 foreach_list_typed(nir_function, func, node, &(shader)->functions)
2153 nir_shader
*nir_shader_create(void *mem_ctx
,
2154 gl_shader_stage stage
,
2155 const nir_shader_compiler_options
*options
,
2158 /** creates a register, including assigning it an index and adding it to the list */
2159 nir_register
*nir_global_reg_create(nir_shader
*shader
);
2161 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
2163 void nir_reg_remove(nir_register
*reg
);
2165 /** Adds a variable to the appropriate list in nir_shader */
2166 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
2169 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
2171 assert(var
->data
.mode
== nir_var_local
);
2172 exec_list_push_tail(&impl
->locals
, &var
->node
);
2175 /** creates a variable, sets a few defaults, and adds it to the list */
2176 nir_variable
*nir_variable_create(nir_shader
*shader
,
2177 nir_variable_mode mode
,
2178 const struct glsl_type
*type
,
2180 /** creates a local variable and adds it to the list */
2181 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
2182 const struct glsl_type
*type
,
2185 /** creates a function and adds it to the shader's list of functions */
2186 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
2188 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
2189 /** creates a function_impl that isn't tied to any particular function */
2190 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
2192 nir_block
*nir_block_create(nir_shader
*shader
);
2193 nir_if
*nir_if_create(nir_shader
*shader
);
2194 nir_loop
*nir_loop_create(nir_shader
*shader
);
2196 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
2198 /** requests that the given pieces of metadata be generated */
2199 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
2200 /** dirties all but the preserved metadata */
2201 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
2203 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
2204 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
2206 nir_deref_instr
*nir_deref_instr_create(nir_shader
*shader
,
2207 nir_deref_type deref_type
);
2209 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
2211 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
2212 unsigned num_components
,
2215 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
2216 nir_intrinsic_op op
);
2218 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
2219 nir_function
*callee
);
2221 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
2223 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
2225 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
2227 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
2228 unsigned num_components
,
2231 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
2232 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
2233 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
2235 typedef bool (*nir_deref_foreach_leaf_cb
)(nir_deref_var
*deref
, void *state
);
2236 bool nir_deref_foreach_leaf(nir_deref_var
*deref
,
2237 nir_deref_foreach_leaf_cb cb
, void *state
);
2239 nir_load_const_instr
*
2240 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
2242 nir_const_value
nir_alu_binop_identity(nir_op binop
, unsigned bit_size
);
2245 * NIR Cursors and Instruction Insertion API
2248 * A tiny struct representing a point to insert/extract instructions or
2249 * control flow nodes. Helps reduce the combinatorial explosion of possible
2250 * points to insert/extract.
2252 * \sa nir_control_flow.h
2255 nir_cursor_before_block
,
2256 nir_cursor_after_block
,
2257 nir_cursor_before_instr
,
2258 nir_cursor_after_instr
,
2259 } nir_cursor_option
;
2262 nir_cursor_option option
;
2269 static inline nir_block
*
2270 nir_cursor_current_block(nir_cursor cursor
)
2272 if (cursor
.option
== nir_cursor_before_instr
||
2273 cursor
.option
== nir_cursor_after_instr
) {
2274 return cursor
.instr
->block
;
2276 return cursor
.block
;
2280 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2282 static inline nir_cursor
2283 nir_before_block(nir_block
*block
)
2286 cursor
.option
= nir_cursor_before_block
;
2287 cursor
.block
= block
;
2291 static inline nir_cursor
2292 nir_after_block(nir_block
*block
)
2295 cursor
.option
= nir_cursor_after_block
;
2296 cursor
.block
= block
;
2300 static inline nir_cursor
2301 nir_before_instr(nir_instr
*instr
)
2304 cursor
.option
= nir_cursor_before_instr
;
2305 cursor
.instr
= instr
;
2309 static inline nir_cursor
2310 nir_after_instr(nir_instr
*instr
)
2313 cursor
.option
= nir_cursor_after_instr
;
2314 cursor
.instr
= instr
;
2318 static inline nir_cursor
2319 nir_after_block_before_jump(nir_block
*block
)
2321 nir_instr
*last_instr
= nir_block_last_instr(block
);
2322 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2323 return nir_before_instr(last_instr
);
2325 return nir_after_block(block
);
2329 static inline nir_cursor
2330 nir_before_cf_node(nir_cf_node
*node
)
2332 if (node
->type
== nir_cf_node_block
)
2333 return nir_before_block(nir_cf_node_as_block(node
));
2335 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2338 static inline nir_cursor
2339 nir_after_cf_node(nir_cf_node
*node
)
2341 if (node
->type
== nir_cf_node_block
)
2342 return nir_after_block(nir_cf_node_as_block(node
));
2344 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2347 static inline nir_cursor
2348 nir_after_phis(nir_block
*block
)
2350 nir_foreach_instr(instr
, block
) {
2351 if (instr
->type
!= nir_instr_type_phi
)
2352 return nir_before_instr(instr
);
2354 return nir_after_block(block
);
2357 static inline nir_cursor
2358 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2360 if (node
->type
== nir_cf_node_block
)
2361 return nir_after_block(nir_cf_node_as_block(node
));
2363 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2365 return nir_after_phis(block
);
2368 static inline nir_cursor
2369 nir_before_cf_list(struct exec_list
*cf_list
)
2371 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2372 exec_list_get_head(cf_list
), node
);
2373 return nir_before_cf_node(first_node
);
2376 static inline nir_cursor
2377 nir_after_cf_list(struct exec_list
*cf_list
)
2379 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2380 exec_list_get_tail(cf_list
), node
);
2381 return nir_after_cf_node(last_node
);
2385 * Insert a NIR instruction at the given cursor.
2387 * Note: This does not update the cursor.
2389 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2392 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2394 nir_instr_insert(nir_before_instr(instr
), before
);
2398 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2400 nir_instr_insert(nir_after_instr(instr
), after
);
2404 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2406 nir_instr_insert(nir_before_block(block
), before
);
2410 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2412 nir_instr_insert(nir_after_block(block
), after
);
2416 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2418 nir_instr_insert(nir_before_cf_node(node
), before
);
2422 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2424 nir_instr_insert(nir_after_cf_node(node
), after
);
2428 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2430 nir_instr_insert(nir_before_cf_list(list
), before
);
2434 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2436 nir_instr_insert(nir_after_cf_list(list
), after
);
2439 void nir_instr_remove_v(nir_instr
*instr
);
2441 static inline nir_cursor
2442 nir_instr_remove(nir_instr
*instr
)
2445 nir_instr
*prev
= nir_instr_prev(instr
);
2447 cursor
= nir_after_instr(prev
);
2449 cursor
= nir_before_block(instr
->block
);
2451 nir_instr_remove_v(instr
);
2457 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2458 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2459 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2460 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2462 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2463 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2465 nir_const_value
*nir_src_as_const_value(nir_src src
);
2466 bool nir_src_is_dynamically_uniform(nir_src src
);
2467 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2468 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2469 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2470 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2471 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2473 void nir_instr_rewrite_deref(nir_instr
*instr
, nir_deref_var
**deref
,
2474 nir_deref_var
*new_deref
);
2476 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2477 unsigned num_components
, unsigned bit_size
,
2479 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2480 unsigned num_components
, unsigned bit_size
,
2483 nir_ssa_dest_init_for_type(nir_instr
*instr
, nir_dest
*dest
,
2484 const struct glsl_type
*type
,
2487 assert(glsl_type_is_vector_or_scalar(type
));
2488 nir_ssa_dest_init(instr
, dest
, glsl_get_components(type
),
2489 glsl_get_bit_size(type
), name
);
2491 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2492 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2493 nir_instr
*after_me
);
2495 uint8_t nir_ssa_def_components_read(const nir_ssa_def
*def
);
2498 * finds the next basic block in source-code order, returns NULL if there is
2502 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2504 /* Performs the opposite of nir_block_cf_tree_next() */
2506 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2508 /* Gets the first block in a CF node in source-code order */
2510 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2512 /* Gets the last block in a CF node in source-code order */
2514 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2516 /* Gets the next block after a CF node in source-code order */
2518 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2520 /* Macros for loops that visit blocks in source-code order */
2522 #define nir_foreach_block(block, impl) \
2523 for (nir_block *block = nir_start_block(impl); block != NULL; \
2524 block = nir_block_cf_tree_next(block))
2526 #define nir_foreach_block_safe(block, impl) \
2527 for (nir_block *block = nir_start_block(impl), \
2528 *next = nir_block_cf_tree_next(block); \
2530 block = next, next = nir_block_cf_tree_next(block))
2532 #define nir_foreach_block_reverse(block, impl) \
2533 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2534 block = nir_block_cf_tree_prev(block))
2536 #define nir_foreach_block_reverse_safe(block, impl) \
2537 for (nir_block *block = nir_impl_last_block(impl), \
2538 *prev = nir_block_cf_tree_prev(block); \
2540 block = prev, prev = nir_block_cf_tree_prev(block))
2542 #define nir_foreach_block_in_cf_node(block, node) \
2543 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2544 block != nir_cf_node_cf_tree_next(node); \
2545 block = nir_block_cf_tree_next(block))
2547 /* If the following CF node is an if, this function returns that if.
2548 * Otherwise, it returns NULL.
2550 nir_if
*nir_block_get_following_if(nir_block
*block
);
2552 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2554 void nir_index_local_regs(nir_function_impl
*impl
);
2555 void nir_index_global_regs(nir_shader
*shader
);
2556 void nir_index_ssa_defs(nir_function_impl
*impl
);
2557 unsigned nir_index_instrs(nir_function_impl
*impl
);
2559 void nir_index_blocks(nir_function_impl
*impl
);
2561 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2562 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2563 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2565 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2566 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2567 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2568 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2569 nir_deref
*nir_deref_clone(const nir_deref
*deref
, void *mem_ctx
);
2570 nir_deref_var
*nir_deref_var_clone(const nir_deref_var
*deref
, void *mem_ctx
);
2572 nir_shader
*nir_shader_serialize_deserialize(void *mem_ctx
, nir_shader
*s
);
2575 void nir_validate_shader(nir_shader
*shader
);
2576 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2577 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2580 should_clone_nir(void)
2582 static int should_clone
= -1;
2583 if (should_clone
< 0)
2584 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2586 return should_clone
;
2590 should_serialize_deserialize_nir(void)
2592 static int test_serialize
= -1;
2593 if (test_serialize
< 0)
2594 test_serialize
= env_var_as_boolean("NIR_TEST_SERIALIZE", false);
2596 return test_serialize
;
2600 should_print_nir(void)
2602 static int should_print
= -1;
2603 if (should_print
< 0)
2604 should_print
= env_var_as_boolean("NIR_PRINT", false);
2606 return should_print
;
2609 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2610 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2611 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2612 static inline bool should_clone_nir(void) { return false; }
2613 static inline bool should_serialize_deserialize_nir(void) { return false; }
2614 static inline bool should_print_nir(void) { return false; }
2617 #define _PASS(nir, do_pass) do { \
2619 nir_validate_shader(nir); \
2620 if (should_clone_nir()) { \
2621 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2625 if (should_serialize_deserialize_nir()) { \
2626 void *mem_ctx = ralloc_parent(nir); \
2627 nir = nir_shader_serialize_deserialize(mem_ctx, nir); \
2631 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2632 nir_metadata_set_validation_flag(nir); \
2633 if (should_print_nir()) \
2634 printf("%s\n", #pass); \
2635 if (pass(nir, ##__VA_ARGS__)) { \
2637 if (should_print_nir()) \
2638 nir_print_shader(nir, stdout); \
2639 nir_metadata_check_validation_flag(nir); \
2643 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2644 if (should_print_nir()) \
2645 printf("%s\n", #pass); \
2646 pass(nir, ##__VA_ARGS__); \
2647 if (should_print_nir()) \
2648 nir_print_shader(nir, stdout); \
2651 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2652 void nir_calc_dominance(nir_shader
*shader
);
2654 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2655 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2657 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2658 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2660 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2661 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2663 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2664 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2666 int nir_gs_count_vertices(const nir_shader
*shader
);
2668 bool nir_split_var_copies(nir_shader
*shader
);
2670 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2671 bool nir_lower_returns(nir_shader
*shader
);
2673 bool nir_inline_functions(nir_shader
*shader
);
2675 bool nir_propagate_invariant(nir_shader
*shader
);
2677 enum nir_lower_deref_flags
{
2678 nir_lower_load_store_derefs
= (1 << 0),
2679 nir_lower_texture_derefs
= (1 << 1),
2680 nir_lower_interp_derefs
= (1 << 2),
2681 nir_lower_atomic_counter_derefs
= (1 << 3),
2682 nir_lower_atomic_derefs
= (1 << 4),
2683 nir_lower_image_derefs
= (1 << 5),
2686 bool nir_lower_deref_instrs(nir_shader
*shader
,
2687 enum nir_lower_deref_flags flags
);
2689 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2690 bool nir_lower_var_copies(nir_shader
*shader
);
2692 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2694 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2696 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2698 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2699 nir_function_impl
*entrypoint
,
2700 bool outputs
, bool inputs
);
2702 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2704 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2705 int (*type_size
)(const struct glsl_type
*));
2707 /* Some helpers to do very simple linking */
2708 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2709 void nir_compact_varyings(nir_shader
*producer
, nir_shader
*consumer
,
2710 bool default_to_smooth_interp
);
2713 /* If set, this forces all non-flat fragment shader inputs to be
2714 * interpolated as if with the "sample" qualifier. This requires
2715 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2717 nir_lower_io_force_sample_interpolation
= (1 << 1),
2718 } nir_lower_io_options
;
2719 bool nir_lower_io(nir_shader
*shader
,
2720 nir_variable_mode modes
,
2721 int (*type_size
)(const struct glsl_type
*),
2722 nir_lower_io_options
);
2723 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2724 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2726 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2728 void nir_lower_io_types(nir_shader
*shader
);
2729 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2730 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2731 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2733 bool nir_remove_dead_derefs(nir_shader
*shader
);
2734 bool nir_remove_dead_derefs_impl(nir_function_impl
*impl
);
2735 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2736 bool nir_lower_constant_initializers(nir_shader
*shader
,
2737 nir_variable_mode modes
);
2739 bool nir_move_load_const(nir_shader
*shader
);
2740 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2741 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2742 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
2744 bool nir_lower_alu(nir_shader
*shader
);
2745 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2746 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2747 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
2748 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2749 void nir_lower_io_arrays_to_elements(nir_shader
*producer
, nir_shader
*consumer
);
2750 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader
*shader
,
2752 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2753 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
2755 typedef struct nir_lower_subgroups_options
{
2756 uint8_t subgroup_size
;
2757 uint8_t ballot_bit_size
;
2758 bool lower_to_scalar
:1;
2759 bool lower_vote_trivial
:1;
2760 bool lower_vote_eq_to_ballot
:1;
2761 bool lower_subgroup_masks
:1;
2762 bool lower_shuffle
:1;
2763 bool lower_shuffle_to_32bit
:1;
2765 } nir_lower_subgroups_options
;
2767 bool nir_lower_subgroups(nir_shader
*shader
,
2768 const nir_lower_subgroups_options
*options
);
2770 bool nir_lower_system_values(nir_shader
*shader
);
2772 typedef struct nir_lower_tex_options
{
2774 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2775 * sampler types a texture projector is lowered.
2780 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2782 bool lower_txf_offset
;
2785 * If true, lower away nir_tex_src_offset for all rect textures.
2787 bool lower_rect_offset
;
2790 * If true, lower rect textures to 2D, using txs to fetch the
2791 * texture dimensions and dividing the texture coords by the
2792 * texture dims to normalize.
2797 * If true, convert yuv to rgb.
2799 unsigned lower_y_uv_external
;
2800 unsigned lower_y_u_v_external
;
2801 unsigned lower_yx_xuxv_external
;
2802 unsigned lower_xy_uxvx_external
;
2805 * To emulate certain texture wrap modes, this can be used
2806 * to saturate the specified tex coord to [0.0, 1.0]. The
2807 * bits are according to sampler #, ie. if, for example:
2809 * (conf->saturate_s & (1 << n))
2811 * is true, then the s coord for sampler n is saturated.
2813 * Note that clamping must happen *after* projector lowering
2814 * so any projected texture sample instruction with a clamped
2815 * coordinate gets automatically lowered, regardless of the
2816 * 'lower_txp' setting.
2818 unsigned saturate_s
;
2819 unsigned saturate_t
;
2820 unsigned saturate_r
;
2822 /* Bitmask of textures that need swizzling.
2824 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2825 * swizzles[texture_index] is applied to the result of the texturing
2828 unsigned swizzle_result
;
2830 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2831 * while 4 and 5 represent 0 and 1 respectively.
2833 uint8_t swizzles
[32][4];
2836 * Bitmap of textures that need srgb to linear conversion. If
2837 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2838 * of the texture are lowered to linear.
2840 unsigned lower_srgb
;
2843 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2845 bool lower_txd_cube_map
;
2848 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
2849 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
2850 * with lower_txd_cube_map.
2852 bool lower_txd_shadow
;
2855 * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
2856 * Implies lower_txd_cube_map and lower_txd_shadow.
2859 } nir_lower_tex_options
;
2861 bool nir_lower_tex(nir_shader
*shader
,
2862 const nir_lower_tex_options
*options
);
2864 bool nir_lower_idiv(nir_shader
*shader
);
2866 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2867 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2868 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2870 void nir_lower_two_sided_color(nir_shader
*shader
);
2872 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
2874 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2875 void nir_lower_tes_patch_vertices(nir_shader
*tes
, unsigned patch_vertices
);
2877 typedef struct nir_lower_wpos_ytransform_options
{
2878 gl_state_index16 state_tokens
[STATE_LENGTH
];
2879 bool fs_coord_origin_upper_left
:1;
2880 bool fs_coord_origin_lower_left
:1;
2881 bool fs_coord_pixel_center_integer
:1;
2882 bool fs_coord_pixel_center_half_integer
:1;
2883 } nir_lower_wpos_ytransform_options
;
2885 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2886 const nir_lower_wpos_ytransform_options
*options
);
2887 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
2889 typedef struct nir_lower_drawpixels_options
{
2890 gl_state_index16 texcoord_state_tokens
[STATE_LENGTH
];
2891 gl_state_index16 scale_state_tokens
[STATE_LENGTH
];
2892 gl_state_index16 bias_state_tokens
[STATE_LENGTH
];
2893 unsigned drawpix_sampler
;
2894 unsigned pixelmap_sampler
;
2896 bool scale_and_bias
:1;
2897 } nir_lower_drawpixels_options
;
2899 void nir_lower_drawpixels(nir_shader
*shader
,
2900 const nir_lower_drawpixels_options
*options
);
2902 typedef struct nir_lower_bitmap_options
{
2905 } nir_lower_bitmap_options
;
2907 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2909 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
2910 bool nir_lower_to_source_mods(nir_shader
*shader
);
2912 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2914 typedef unsigned (*nir_lower_bit_size_callback
)(const nir_alu_instr
*, void *);
2916 bool nir_lower_bit_size(nir_shader
*shader
,
2917 nir_lower_bit_size_callback callback
,
2918 void *callback_data
);
2921 nir_lower_imul64
= (1 << 0),
2922 nir_lower_isign64
= (1 << 1),
2923 /** Lower all int64 modulus and division opcodes */
2924 nir_lower_divmod64
= (1 << 2),
2925 } nir_lower_int64_options
;
2927 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
2930 nir_lower_drcp
= (1 << 0),
2931 nir_lower_dsqrt
= (1 << 1),
2932 nir_lower_drsq
= (1 << 2),
2933 nir_lower_dtrunc
= (1 << 3),
2934 nir_lower_dfloor
= (1 << 4),
2935 nir_lower_dceil
= (1 << 5),
2936 nir_lower_dfract
= (1 << 6),
2937 nir_lower_dround_even
= (1 << 7),
2938 nir_lower_dmod
= (1 << 8)
2939 } nir_lower_doubles_options
;
2941 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2942 bool nir_lower_pack(nir_shader
*shader
);
2944 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2946 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2948 void nir_loop_analyze_impl(nir_function_impl
*impl
,
2949 nir_variable_mode indirect_mask
);
2951 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2953 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2954 bool nir_repair_ssa(nir_shader
*shader
);
2956 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
2958 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2959 * registers. If false, convert all values (even those not involved in a phi
2960 * node) to registers.
2962 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2964 bool nir_lower_phis_to_regs_block(nir_block
*block
);
2965 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
2967 bool nir_opt_algebraic(nir_shader
*shader
);
2968 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
2969 bool nir_opt_algebraic_late(nir_shader
*shader
);
2970 bool nir_opt_constant_folding(nir_shader
*shader
);
2972 bool nir_opt_global_to_local(nir_shader
*shader
);
2974 bool nir_copy_prop(nir_shader
*shader
);
2976 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
2978 bool nir_opt_cse(nir_shader
*shader
);
2980 bool nir_opt_dce(nir_shader
*shader
);
2982 bool nir_opt_dead_cf(nir_shader
*shader
);
2984 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2986 bool nir_opt_if(nir_shader
*shader
);
2988 bool nir_opt_intrinsics(nir_shader
*shader
);
2990 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
2992 bool nir_opt_move_comparisons(nir_shader
*shader
);
2994 bool nir_opt_move_load_ubo(nir_shader
*shader
);
2996 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2998 bool nir_opt_remove_phis(nir_shader
*shader
);
3000 bool nir_opt_shrink_load(nir_shader
*shader
);
3002 bool nir_opt_trivial_continues(nir_shader
*shader
);
3004 bool nir_opt_undef(nir_shader
*shader
);
3006 bool nir_opt_conditional_discard(nir_shader
*shader
);
3008 void nir_sweep(nir_shader
*shader
);
3010 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
3011 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);