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/bitscan.h"
38 #include "util/bitset.h"
39 #include "util/macros.h"
40 #include "compiler/nir_types.h"
41 #include "compiler/shader_enums.h"
42 #include "compiler/shader_info.h"
46 #include "util/debug.h"
49 #include "nir_opcodes.h"
51 #if defined(_WIN32) && !defined(snprintf)
52 #define snprintf _snprintf
60 #define NIR_TRUE (~0u)
61 #define NIR_MAX_VEC_COMPONENTS 4
62 typedef uint8_t nir_component_mask_t
;
64 /** Defines a cast function
66 * This macro defines a cast function from in_type to out_type where
67 * out_type is some structure type that contains a field of type out_type.
69 * Note that you have to be a bit careful as the generated cast function
72 #define NIR_DEFINE_CAST(name, in_type, out_type, field, \
73 type_field, type_value) \
74 static inline out_type * \
75 name(const in_type *parent) \
77 assert(parent && parent->type_field == type_value); \
78 return exec_node_data(out_type, parent, field); \
88 * Description of built-in state associated with a uniform
90 * \sa nir_variable::state_slots
93 gl_state_index16 tokens
[STATE_LENGTH
];
98 nir_var_shader_in
= (1 << 0),
99 nir_var_shader_out
= (1 << 1),
100 nir_var_shader_temp
= (1 << 2),
101 nir_var_function_temp
= (1 << 3),
102 nir_var_uniform
= (1 << 4),
103 nir_var_mem_ubo
= (1 << 5),
104 nir_var_system_value
= (1 << 6),
105 nir_var_mem_ssbo
= (1 << 7),
106 nir_var_mem_shared
= (1 << 8),
107 nir_var_mem_global
= (1 << 9),
115 nir_rounding_mode_undef
= 0,
116 nir_rounding_mode_rtne
= 1, /* round to nearest even */
117 nir_rounding_mode_ru
= 2, /* round up */
118 nir_rounding_mode_rd
= 3, /* round down */
119 nir_rounding_mode_rtz
= 4, /* round towards zero */
123 bool b
[NIR_MAX_VEC_COMPONENTS
];
124 float f32
[NIR_MAX_VEC_COMPONENTS
];
125 double f64
[NIR_MAX_VEC_COMPONENTS
];
126 int8_t i8
[NIR_MAX_VEC_COMPONENTS
];
127 uint8_t u8
[NIR_MAX_VEC_COMPONENTS
];
128 int16_t i16
[NIR_MAX_VEC_COMPONENTS
];
129 uint16_t u16
[NIR_MAX_VEC_COMPONENTS
];
130 int32_t i32
[NIR_MAX_VEC_COMPONENTS
];
131 uint32_t u32
[NIR_MAX_VEC_COMPONENTS
];
132 int64_t i64
[NIR_MAX_VEC_COMPONENTS
];
133 uint64_t u64
[NIR_MAX_VEC_COMPONENTS
];
136 typedef struct nir_constant
{
138 * Value of the constant.
140 * The field used to back the values supplied by the constant is determined
141 * by the type associated with the \c nir_variable. Constants may be
142 * scalars, vectors, or matrices.
144 nir_const_value values
[NIR_MAX_VEC_COMPONENTS
];
146 /* we could get this from the var->type but makes clone *much* easier to
147 * not have to care about the type.
149 unsigned num_elements
;
151 /* Array elements / Structure Fields */
152 struct nir_constant
**elements
;
156 * \brief Layout qualifiers for gl_FragDepth.
158 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
159 * with a layout qualifier.
162 nir_depth_layout_none
, /**< No depth layout is specified. */
163 nir_depth_layout_any
,
164 nir_depth_layout_greater
,
165 nir_depth_layout_less
,
166 nir_depth_layout_unchanged
170 * Enum keeping track of how a variable was declared.
174 * Normal declaration.
176 nir_var_declared_normally
= 0,
179 * Variable is implicitly generated by the compiler and should not be
180 * visible via the API.
183 } nir_var_declaration_type
;
186 * Either a uniform, global variable, shader input, or shader output. Based on
187 * ir_variable - it should be easy to translate between the two.
190 typedef struct nir_variable
{
191 struct exec_node node
;
194 * Declared type of the variable
196 const struct glsl_type
*type
;
199 * Declared name of the variable
203 struct nir_variable_data
{
205 * Storage class of the variable.
207 * \sa nir_variable_mode
209 nir_variable_mode mode
;
212 * Is the variable read-only?
214 * This is set for variables declared as \c const, shader inputs,
217 unsigned read_only
:1;
221 unsigned invariant
:1;
224 * When separate shader programs are enabled, only input/outputs between
225 * the stages of a multi-stage separate program can be safely removed
226 * from the shader interface. Other input/outputs must remains active.
228 * This is also used to make sure xfb varyings that are unused by the
229 * fragment shader are not removed.
231 unsigned always_active_io
:1;
234 * Interpolation mode for shader inputs / outputs
236 * \sa glsl_interp_mode
238 unsigned interpolation
:2;
241 * \name ARB_fragment_coord_conventions
244 unsigned origin_upper_left
:1;
245 unsigned pixel_center_integer
:1;
249 * If non-zero, then this variable may be packed along with other variables
250 * into a single varying slot, so this offset should be applied when
251 * accessing components. For example, an offset of 1 means that the x
252 * component of this variable is actually stored in component y of the
253 * location specified by \c location.
255 unsigned location_frac
:2;
258 * If true, this variable represents an array of scalars that should
259 * be tightly packed. In other words, consecutive array elements
260 * should be stored one component apart, rather than one slot apart.
265 * Whether this is a fragment shader output implicitly initialized with
266 * the previous contents of the specified render target at the
267 * framebuffer location corresponding to this shader invocation.
269 unsigned fb_fetch_output
:1;
272 * Non-zero if this variable is considered bindless as defined by
273 * ARB_bindless_texture.
278 * Was an explicit binding set in the shader?
280 unsigned explicit_binding
:1;
283 * Was a transfer feedback buffer set in the shader?
285 unsigned explicit_xfb_buffer
:1;
288 * Was a transfer feedback stride set in the shader?
290 unsigned explicit_xfb_stride
:1;
293 * Was an explicit offset set in the shader?
295 unsigned explicit_offset
:1;
298 * \brief Layout qualifier for gl_FragDepth.
300 * This is not equal to \c ir_depth_layout_none if and only if this
301 * variable is \c gl_FragDepth and a layout qualifier is specified.
303 nir_depth_layout depth_layout
;
306 * Storage location of the base of this variable
308 * The precise meaning of this field depends on the nature of the variable.
310 * - Vertex shader input: one of the values from \c gl_vert_attrib.
311 * - Vertex shader output: one of the values from \c gl_varying_slot.
312 * - Geometry shader input: one of the values from \c gl_varying_slot.
313 * - Geometry shader output: one of the values from \c gl_varying_slot.
314 * - Fragment shader input: one of the values from \c gl_varying_slot.
315 * - Fragment shader output: one of the values from \c gl_frag_result.
316 * - Uniforms: Per-stage uniform slot number for default uniform block.
317 * - Uniforms: Index within the uniform block definition for UBO members.
318 * - Non-UBO Uniforms: uniform slot number.
319 * - Other: This field is not currently used.
321 * If the variable is a uniform, shader input, or shader output, and the
322 * slot has not been assigned, the value will be -1.
327 * The actual location of the variable in the IR. Only valid for inputs
330 unsigned int driver_location
;
333 * Vertex stream output identifier.
335 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
336 * stream of the i-th component.
341 * output index for dual source blending.
346 * Descriptor set binding for sampler or UBO.
351 * Initial binding point for a sampler or UBO.
353 * For array types, this represents the binding point for the first element.
358 * Location an atomic counter or transform feedback is stored at.
363 * Transform feedback buffer.
368 * Transform feedback stride.
373 * How the variable was declared. See nir_var_declaration_type.
375 * This is used to detect variables generated by the compiler, so should
376 * not be visible via the API.
378 unsigned how_declared
:2;
381 * ARB_shader_image_load_store qualifiers.
384 enum gl_access_qualifier access
;
386 /** Image internal format if specified explicitly, otherwise GL_NONE. */
392 * Built-in state that backs this uniform
394 * Once set at variable creation, \c state_slots must remain invariant.
395 * This is because, ideally, this array would be shared by all clones of
396 * this variable in the IR tree. In other words, we'd really like for it
397 * to be a fly-weight.
399 * If the variable is not a uniform, \c num_state_slots will be zero and
400 * \c state_slots will be \c NULL.
403 unsigned num_state_slots
; /**< Number of state slots used */
404 nir_state_slot
*state_slots
; /**< State descriptors. */
408 * Constant expression assigned in the initializer of the variable
410 * This field should only be used temporarily by creators of NIR shaders
411 * and then lower_constant_initializers can be used to get rid of them.
412 * Most of the rest of NIR ignores this field or asserts that it's NULL.
414 nir_constant
*constant_initializer
;
417 * For variables that are in an interface block or are an instance of an
418 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
420 * \sa ir_variable::location
422 const struct glsl_type
*interface_type
;
425 * Description of per-member data for per-member struct variables
427 * This is used for variables which are actually an amalgamation of
428 * multiple entities such as a struct of built-in values or a struct of
429 * inputs each with their own layout specifier. This is only allowed on
430 * variables with a struct or array of array of struct type.
432 unsigned num_members
;
433 struct nir_variable_data
*members
;
436 #define nir_foreach_variable(var, var_list) \
437 foreach_list_typed(nir_variable, var, node, var_list)
439 #define nir_foreach_variable_safe(var, var_list) \
440 foreach_list_typed_safe(nir_variable, var, node, var_list)
443 nir_variable_is_global(const nir_variable
*var
)
445 return var
->data
.mode
!= nir_var_function_temp
;
448 typedef struct nir_register
{
449 struct exec_node node
;
451 unsigned num_components
; /** < number of vector components */
452 unsigned num_array_elems
; /** < size of array (0 for no array) */
454 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
457 /** generic register index. */
460 /** only for debug purposes, can be NULL */
463 /** whether this register is local (per-function) or global (per-shader) */
467 * If this flag is set to true, then accessing channels >= num_components
468 * is well-defined, and simply spills over to the next array element. This
469 * is useful for backends that can do per-component accessing, in
470 * particular scalar backends. By setting this flag and making
471 * num_components equal to 1, structures can be packed tightly into
472 * registers and then registers can be accessed per-component to get to
473 * each structure member, even if it crosses vec4 boundaries.
477 /** set of nir_srcs where this register is used (read from) */
478 struct list_head uses
;
480 /** set of nir_dests where this register is defined (written to) */
481 struct list_head defs
;
483 /** set of nir_ifs where this register is used as a condition */
484 struct list_head if_uses
;
487 #define nir_foreach_register(reg, reg_list) \
488 foreach_list_typed(nir_register, reg, node, reg_list)
489 #define nir_foreach_register_safe(reg, reg_list) \
490 foreach_list_typed_safe(nir_register, reg, node, reg_list)
492 typedef enum PACKED
{
494 nir_instr_type_deref
,
497 nir_instr_type_intrinsic
,
498 nir_instr_type_load_const
,
500 nir_instr_type_ssa_undef
,
502 nir_instr_type_parallel_copy
,
505 typedef struct nir_instr
{
506 struct exec_node node
;
507 struct nir_block
*block
;
510 /* A temporary for optimization and analysis passes to use for storing
511 * flags. For instance, DCE uses this to store the "dead/live" info.
515 /** generic instruction index. */
519 static inline nir_instr
*
520 nir_instr_next(nir_instr
*instr
)
522 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
523 if (exec_node_is_tail_sentinel(next
))
526 return exec_node_data(nir_instr
, next
, node
);
529 static inline nir_instr
*
530 nir_instr_prev(nir_instr
*instr
)
532 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
533 if (exec_node_is_head_sentinel(prev
))
536 return exec_node_data(nir_instr
, prev
, node
);
540 nir_instr_is_first(const nir_instr
*instr
)
542 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
546 nir_instr_is_last(const nir_instr
*instr
)
548 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
551 typedef struct nir_ssa_def
{
552 /** for debugging only, can be NULL */
555 /** generic SSA definition index. */
558 /** Index into the live_in and live_out bitfields */
561 /** Instruction which produces this SSA value. */
562 nir_instr
*parent_instr
;
564 /** set of nir_instrs where this register is used (read from) */
565 struct list_head uses
;
567 /** set of nir_ifs where this register is used as a condition */
568 struct list_head if_uses
;
570 uint8_t num_components
;
572 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
580 struct nir_src
*indirect
; /** < NULL for no indirect offset */
581 unsigned base_offset
;
583 /* TODO use-def chain goes here */
587 nir_instr
*parent_instr
;
588 struct list_head def_link
;
591 struct nir_src
*indirect
; /** < NULL for no indirect offset */
592 unsigned base_offset
;
594 /* TODO def-use chain goes here */
599 typedef struct nir_src
{
601 /** Instruction that consumes this value as a source. */
602 nir_instr
*parent_instr
;
603 struct nir_if
*parent_if
;
606 struct list_head use_link
;
616 static inline nir_src
619 nir_src src
= { { NULL
} };
623 #define NIR_SRC_INIT nir_src_init()
625 #define nir_foreach_use(src, reg_or_ssa_def) \
626 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
628 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
629 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
631 #define nir_foreach_if_use(src, reg_or_ssa_def) \
632 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
634 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
635 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
646 static inline nir_dest
649 nir_dest dest
= { { { NULL
} } };
653 #define NIR_DEST_INIT nir_dest_init()
655 #define nir_foreach_def(dest, reg) \
656 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
658 #define nir_foreach_def_safe(dest, reg) \
659 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
661 static inline nir_src
662 nir_src_for_ssa(nir_ssa_def
*def
)
664 nir_src src
= NIR_SRC_INIT
;
672 static inline nir_src
673 nir_src_for_reg(nir_register
*reg
)
675 nir_src src
= NIR_SRC_INIT
;
679 src
.reg
.indirect
= NULL
;
680 src
.reg
.base_offset
= 0;
685 static inline nir_dest
686 nir_dest_for_reg(nir_register
*reg
)
688 nir_dest dest
= NIR_DEST_INIT
;
695 static inline unsigned
696 nir_src_bit_size(nir_src src
)
698 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
701 static inline unsigned
702 nir_src_num_components(nir_src src
)
704 return src
.is_ssa
? src
.ssa
->num_components
: src
.reg
.reg
->num_components
;
708 nir_src_is_const(nir_src src
)
711 src
.ssa
->parent_instr
->type
== nir_instr_type_load_const
;
714 int64_t nir_src_as_int(nir_src src
);
715 uint64_t nir_src_as_uint(nir_src src
);
716 bool nir_src_as_bool(nir_src src
);
717 double nir_src_as_float(nir_src src
);
718 int64_t nir_src_comp_as_int(nir_src src
, unsigned component
);
719 uint64_t nir_src_comp_as_uint(nir_src src
, unsigned component
);
720 bool nir_src_comp_as_bool(nir_src src
, unsigned component
);
721 double nir_src_comp_as_float(nir_src src
, unsigned component
);
723 static inline unsigned
724 nir_dest_bit_size(nir_dest dest
)
726 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
729 static inline unsigned
730 nir_dest_num_components(nir_dest dest
)
732 return dest
.is_ssa
? dest
.ssa
.num_components
: dest
.reg
.reg
->num_components
;
735 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
736 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
742 * \name input modifiers
746 * For inputs interpreted as floating point, flips the sign bit. For
747 * inputs interpreted as integers, performs the two's complement negation.
752 * Clears the sign bit for floating point values, and computes the integer
753 * absolute value for integers. Note that the negate modifier acts after
754 * the absolute value modifier, therefore if both are set then all inputs
755 * will become negative.
761 * For each input component, says which component of the register it is
762 * chosen from. Note that which elements of the swizzle are used and which
763 * are ignored are based on the write mask for most opcodes - for example,
764 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
765 * a swizzle of {2, x, 1, 0} where x means "don't care."
767 uint8_t swizzle
[NIR_MAX_VEC_COMPONENTS
];
774 * \name saturate output modifier
776 * Only valid for opcodes that output floating-point numbers. Clamps the
777 * output to between 0.0 and 1.0 inclusive.
782 unsigned write_mask
: NIR_MAX_VEC_COMPONENTS
; /* ignored if dest.is_ssa is true */
785 /** NIR sized and unsized types
787 * The values in this enum are carefully chosen so that the sized type is
788 * just the unsized type OR the number of bits.
791 nir_type_invalid
= 0, /* Not a valid type */
795 nir_type_float
= 128,
796 nir_type_bool1
= 1 | nir_type_bool
,
797 nir_type_bool32
= 32 | nir_type_bool
,
798 nir_type_int1
= 1 | nir_type_int
,
799 nir_type_int8
= 8 | nir_type_int
,
800 nir_type_int16
= 16 | nir_type_int
,
801 nir_type_int32
= 32 | nir_type_int
,
802 nir_type_int64
= 64 | nir_type_int
,
803 nir_type_uint1
= 1 | nir_type_uint
,
804 nir_type_uint8
= 8 | nir_type_uint
,
805 nir_type_uint16
= 16 | nir_type_uint
,
806 nir_type_uint32
= 32 | nir_type_uint
,
807 nir_type_uint64
= 64 | nir_type_uint
,
808 nir_type_float16
= 16 | nir_type_float
,
809 nir_type_float32
= 32 | nir_type_float
,
810 nir_type_float64
= 64 | nir_type_float
,
813 #define NIR_ALU_TYPE_SIZE_MASK 0x79
814 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x86
816 static inline unsigned
817 nir_alu_type_get_type_size(nir_alu_type type
)
819 return type
& NIR_ALU_TYPE_SIZE_MASK
;
822 static inline unsigned
823 nir_alu_type_get_base_type(nir_alu_type type
)
825 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
828 static inline nir_alu_type
829 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
833 return nir_type_bool1
;
836 return nir_type_uint32
;
839 return nir_type_int32
;
841 case GLSL_TYPE_UINT16
:
842 return nir_type_uint16
;
844 case GLSL_TYPE_INT16
:
845 return nir_type_int16
;
847 case GLSL_TYPE_UINT8
:
848 return nir_type_uint8
;
850 return nir_type_int8
;
851 case GLSL_TYPE_UINT64
:
852 return nir_type_uint64
;
854 case GLSL_TYPE_INT64
:
855 return nir_type_int64
;
857 case GLSL_TYPE_FLOAT
:
858 return nir_type_float32
;
860 case GLSL_TYPE_FLOAT16
:
861 return nir_type_float16
;
863 case GLSL_TYPE_DOUBLE
:
864 return nir_type_float64
;
867 unreachable("unknown type");
871 static inline nir_alu_type
872 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
874 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
877 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
,
878 nir_rounding_mode rnd
);
881 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
882 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
883 } nir_op_algebraic_property
;
891 * The number of components in the output
893 * If non-zero, this is the size of the output and input sizes are
894 * explicitly given; swizzle and writemask are still in effect, but if
895 * the output component is masked out, then the input component may
898 * If zero, the opcode acts in the standard, per-component manner; the
899 * operation is performed on each component (except the ones that are
900 * masked out) with the input being taken from the input swizzle for
903 * The size of some of the inputs may be given (i.e. non-zero) even
904 * though output_size is zero; in that case, the inputs with a zero
905 * size act per-component, while the inputs with non-zero size don't.
907 unsigned output_size
;
910 * The type of vector that the instruction outputs. Note that the
911 * staurate modifier is only allowed on outputs with the float type.
914 nir_alu_type output_type
;
917 * The number of components in each input
919 unsigned input_sizes
[NIR_MAX_VEC_COMPONENTS
];
922 * The type of vector that each input takes. Note that negate and
923 * absolute value are only allowed on inputs with int or float type and
924 * behave differently on the two.
926 nir_alu_type input_types
[NIR_MAX_VEC_COMPONENTS
];
928 nir_op_algebraic_property algebraic_properties
;
931 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
933 typedef struct nir_alu_instr
{
937 /** Indicates that this ALU instruction generates an exact value
939 * This is kind of a mixture of GLSL "precise" and "invariant" and not
940 * really equivalent to either. This indicates that the value generated by
941 * this operation is high-precision and any code transformations that touch
942 * it must ensure that the resulting value is bit-for-bit identical to the
951 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
952 nir_alu_instr
*instr
);
953 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
954 nir_alu_instr
*instr
);
956 /* is this source channel used? */
958 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
961 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
962 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
964 return (instr
->dest
.write_mask
>> channel
) & 1;
967 static inline nir_component_mask_t
968 nir_alu_instr_src_read_mask(const nir_alu_instr
*instr
, unsigned src
)
970 nir_component_mask_t read_mask
= 0;
971 for (unsigned c
= 0; c
< NIR_MAX_VEC_COMPONENTS
; c
++) {
972 if (!nir_alu_instr_channel_used(instr
, src
, c
))
975 read_mask
|= (1 << instr
->src
[src
].swizzle
[c
]);
981 * For instructions whose destinations are SSA, get the number of channels
984 static inline unsigned
985 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
987 assert(instr
->dest
.dest
.is_ssa
);
989 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
990 return nir_op_infos
[instr
->op
].input_sizes
[src
];
992 return instr
->dest
.dest
.ssa
.num_components
;
995 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
996 unsigned src1
, unsigned src2
);
1000 nir_deref_type_array
,
1001 nir_deref_type_array_wildcard
,
1002 nir_deref_type_ptr_as_array
,
1003 nir_deref_type_struct
,
1004 nir_deref_type_cast
,
1010 /** The type of this deref instruction */
1011 nir_deref_type deref_type
;
1013 /** The mode of the underlying variable */
1014 nir_variable_mode mode
;
1016 /** The dereferenced type of the resulting pointer value */
1017 const struct glsl_type
*type
;
1020 /** Variable being dereferenced if deref_type is a deref_var */
1023 /** Parent deref if deref_type is not deref_var */
1027 /** Additional deref parameters */
1038 unsigned ptr_stride
;
1042 /** Destination to store the resulting "pointer" */
1046 NIR_DEFINE_CAST(nir_instr_as_deref
, nir_instr
, nir_deref_instr
, instr
,
1047 type
, nir_instr_type_deref
)
1049 static inline nir_deref_instr
*
1050 nir_src_as_deref(nir_src src
)
1055 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_deref
)
1058 return nir_instr_as_deref(src
.ssa
->parent_instr
);
1061 static inline nir_deref_instr
*
1062 nir_deref_instr_parent(const nir_deref_instr
*instr
)
1064 if (instr
->deref_type
== nir_deref_type_var
)
1067 return nir_src_as_deref(instr
->parent
);
1070 static inline nir_variable
*
1071 nir_deref_instr_get_variable(const nir_deref_instr
*instr
)
1073 while (instr
->deref_type
!= nir_deref_type_var
) {
1074 if (instr
->deref_type
== nir_deref_type_cast
)
1077 instr
= nir_deref_instr_parent(instr
);
1083 bool nir_deref_instr_has_indirect(nir_deref_instr
*instr
);
1085 bool nir_deref_instr_remove_if_unused(nir_deref_instr
*instr
);
1087 unsigned nir_deref_instr_ptr_as_array_stride(nir_deref_instr
*instr
);
1092 struct nir_function
*callee
;
1094 unsigned num_params
;
1098 #include "nir_intrinsics.h"
1100 #define NIR_INTRINSIC_MAX_CONST_INDEX 4
1102 /** Represents an intrinsic
1104 * An intrinsic is an instruction type for handling things that are
1105 * more-or-less regular operations but don't just consume and produce SSA
1106 * values like ALU operations do. Intrinsics are not for things that have
1107 * special semantic meaning such as phi nodes and parallel copies.
1108 * Examples of intrinsics include variable load/store operations, system
1109 * value loads, and the like. Even though texturing more-or-less falls
1110 * under this category, texturing is its own instruction type because
1111 * trying to represent texturing with intrinsics would lead to a
1112 * combinatorial explosion of intrinsic opcodes.
1114 * By having a single instruction type for handling a lot of different
1115 * cases, optimization passes can look for intrinsics and, for the most
1116 * part, completely ignore them. Each intrinsic type also has a few
1117 * possible flags that govern whether or not they can be reordered or
1118 * eliminated. That way passes like dead code elimination can still work
1119 * on intrisics without understanding the meaning of each.
1121 * Each intrinsic has some number of constant indices, some number of
1122 * variables, and some number of sources. What these sources, variables,
1123 * and indices mean depends on the intrinsic and is documented with the
1124 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
1125 * instructions are the only types of instruction that can operate on
1131 nir_intrinsic_op intrinsic
;
1135 /** number of components if this is a vectorized intrinsic
1137 * Similarly to ALU operations, some intrinsics are vectorized.
1138 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
1139 * For vectorized intrinsics, the num_components field specifies the
1140 * number of destination components and the number of source components
1141 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
1143 uint8_t num_components
;
1145 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
1148 } nir_intrinsic_instr
;
1150 static inline nir_variable
*
1151 nir_intrinsic_get_var(nir_intrinsic_instr
*intrin
, unsigned i
)
1153 return nir_deref_instr_get_variable(nir_src_as_deref(intrin
->src
[i
]));
1157 * \name NIR intrinsics semantic flags
1159 * information about what the compiler can do with the intrinsics.
1161 * \sa nir_intrinsic_info::flags
1165 * whether the intrinsic can be safely eliminated if none of its output
1166 * value is not being used.
1168 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
1171 * Whether the intrinsic can be reordered with respect to any other
1172 * intrinsic, i.e. whether the only reordering dependencies of the
1173 * intrinsic are due to the register reads/writes.
1175 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
1176 } nir_intrinsic_semantic_flag
;
1179 * \name NIR intrinsics const-index flag
1181 * Indicates the usage of a const_index slot.
1183 * \sa nir_intrinsic_info::index_map
1187 * Generally instructions that take a offset src argument, can encode
1188 * a constant 'base' value which is added to the offset.
1190 NIR_INTRINSIC_BASE
= 1,
1193 * For store instructions, a writemask for the store.
1195 NIR_INTRINSIC_WRMASK
= 2,
1198 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1200 NIR_INTRINSIC_STREAM_ID
= 3,
1203 * The clip-plane id for load_user_clip_plane intrinsic.
1205 NIR_INTRINSIC_UCP_ID
= 4,
1208 * The amount of data, starting from BASE, that this instruction may
1209 * access. This is used to provide bounds if the offset is not constant.
1211 NIR_INTRINSIC_RANGE
= 5,
1214 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1216 NIR_INTRINSIC_DESC_SET
= 6,
1219 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1221 NIR_INTRINSIC_BINDING
= 7,
1226 NIR_INTRINSIC_COMPONENT
= 8,
1229 * Interpolation mode (only meaningful for FS inputs).
1231 NIR_INTRINSIC_INTERP_MODE
= 9,
1234 * A binary nir_op to use when performing a reduction or scan operation
1236 NIR_INTRINSIC_REDUCTION_OP
= 10,
1239 * Cluster size for reduction operations
1241 NIR_INTRINSIC_CLUSTER_SIZE
= 11,
1244 * Parameter index for a load_param intrinsic
1246 NIR_INTRINSIC_PARAM_IDX
= 12,
1249 * Image dimensionality for image intrinsics
1251 * One of GLSL_SAMPLER_DIM_*
1253 NIR_INTRINSIC_IMAGE_DIM
= 13,
1256 * Non-zero if we are accessing an array image
1258 NIR_INTRINSIC_IMAGE_ARRAY
= 14,
1261 * Image format for image intrinsics
1263 NIR_INTRINSIC_FORMAT
= 15,
1266 * Access qualifiers for image and memory access intrinsics
1268 NIR_INTRINSIC_ACCESS
= 16,
1271 * Alignment for offsets and addresses
1273 * These two parameters, specify an alignment in terms of a multiplier and
1274 * an offset. The offset or address parameter X of the intrinsic is
1275 * guaranteed to satisfy the following:
1277 * (X - align_offset) % align_mul == 0
1279 NIR_INTRINSIC_ALIGN_MUL
= 17,
1280 NIR_INTRINSIC_ALIGN_OFFSET
= 18,
1283 * The Vulkan descriptor type for a vulkan_resource_[re]index intrinsic.
1285 NIR_INTRINSIC_DESC_TYPE
= 19,
1287 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1289 } nir_intrinsic_index_flag
;
1291 #define NIR_INTRINSIC_MAX_INPUTS 5
1296 unsigned num_srcs
; /** < number of register/SSA inputs */
1298 /** number of components of each input register
1300 * If this value is 0, the number of components is given by the
1301 * num_components field of nir_intrinsic_instr. If this value is -1, the
1302 * intrinsic consumes however many components are provided and it is not
1305 int src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1309 /** number of components of the output register
1311 * If this value is 0, the number of components is given by the
1312 * num_components field of nir_intrinsic_instr.
1314 unsigned dest_components
;
1316 /** the number of constant indices used by the intrinsic */
1317 unsigned num_indices
;
1319 /** indicates the usage of intr->const_index[n] */
1320 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1322 /** semantic flags for calls to this intrinsic */
1323 nir_intrinsic_semantic_flag flags
;
1324 } nir_intrinsic_info
;
1326 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1328 static inline unsigned
1329 nir_intrinsic_src_components(nir_intrinsic_instr
*intr
, unsigned srcn
)
1331 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1332 assert(srcn
< info
->num_srcs
);
1333 if (info
->src_components
[srcn
] > 0)
1334 return info
->src_components
[srcn
];
1335 else if (info
->src_components
[srcn
] == 0)
1336 return intr
->num_components
;
1338 return nir_src_num_components(intr
->src
[srcn
]);
1341 static inline unsigned
1342 nir_intrinsic_dest_components(nir_intrinsic_instr
*intr
)
1344 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1345 if (!info
->has_dest
)
1347 else if (info
->dest_components
)
1348 return info
->dest_components
;
1350 return intr
->num_components
;
1353 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1354 static inline type \
1355 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1357 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1358 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1359 return (type)instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1361 static inline void \
1362 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1364 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1365 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1366 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1369 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1370 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1371 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1372 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1373 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1374 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1375 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1376 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1377 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1378 INTRINSIC_IDX_ACCESSORS(reduction_op
, REDUCTION_OP
, unsigned)
1379 INTRINSIC_IDX_ACCESSORS(cluster_size
, CLUSTER_SIZE
, unsigned)
1380 INTRINSIC_IDX_ACCESSORS(param_idx
, PARAM_IDX
, unsigned)
1381 INTRINSIC_IDX_ACCESSORS(image_dim
, IMAGE_DIM
, enum glsl_sampler_dim
)
1382 INTRINSIC_IDX_ACCESSORS(image_array
, IMAGE_ARRAY
, bool)
1383 INTRINSIC_IDX_ACCESSORS(access
, ACCESS
, enum gl_access_qualifier
)
1384 INTRINSIC_IDX_ACCESSORS(format
, FORMAT
, unsigned)
1385 INTRINSIC_IDX_ACCESSORS(align_mul
, ALIGN_MUL
, unsigned)
1386 INTRINSIC_IDX_ACCESSORS(align_offset
, ALIGN_OFFSET
, unsigned)
1387 INTRINSIC_IDX_ACCESSORS(desc_type
, DESC_TYPE
, unsigned)
1390 nir_intrinsic_set_align(nir_intrinsic_instr
*intrin
,
1391 unsigned align_mul
, unsigned align_offset
)
1393 assert(util_is_power_of_two_nonzero(align_mul
));
1394 assert(align_offset
< align_mul
);
1395 nir_intrinsic_set_align_mul(intrin
, align_mul
);
1396 nir_intrinsic_set_align_offset(intrin
, align_offset
);
1399 /** Returns a simple alignment for a load/store intrinsic offset
1401 * Instead of the full mul+offset alignment scheme provided by the ALIGN_MUL
1402 * and ALIGN_OFFSET parameters, this helper takes both into account and
1403 * provides a single simple alignment parameter. The offset X is guaranteed
1404 * to satisfy X % align == 0.
1406 static inline unsigned
1407 nir_intrinsic_align(const nir_intrinsic_instr
*intrin
)
1409 const unsigned align_mul
= nir_intrinsic_align_mul(intrin
);
1410 const unsigned align_offset
= nir_intrinsic_align_offset(intrin
);
1411 assert(align_offset
< align_mul
);
1412 return align_offset
? 1 << (ffs(align_offset
) - 1) : align_mul
;
1416 * \group texture information
1418 * This gives semantic information about textures which is useful to the
1419 * frontend, the backend, and lowering passes, but not the optimizer.
1424 nir_tex_src_projector
,
1425 nir_tex_src_comparator
, /* shadow comparator */
1429 nir_tex_src_min_lod
,
1430 nir_tex_src_ms_index
, /* MSAA sample index */
1431 nir_tex_src_ms_mcs
, /* MSAA compression value */
1434 nir_tex_src_texture_deref
, /* < deref pointing to the texture */
1435 nir_tex_src_sampler_deref
, /* < deref pointing to the sampler */
1436 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1437 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1438 nir_tex_src_plane
, /* < selects plane for planar textures */
1439 nir_num_tex_src_types
1444 nir_tex_src_type src_type
;
1448 nir_texop_tex
, /**< Regular texture look-up */
1449 nir_texop_txb
, /**< Texture look-up with LOD bias */
1450 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1451 nir_texop_txd
, /**< Texture look-up with partial derivatives */
1452 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1453 nir_texop_txf_ms
, /**< Multisample texture fetch */
1454 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1455 nir_texop_txs
, /**< Texture size */
1456 nir_texop_lod
, /**< Texture lod query */
1457 nir_texop_tg4
, /**< Texture gather */
1458 nir_texop_query_levels
, /**< Texture levels query */
1459 nir_texop_texture_samples
, /**< Texture samples query */
1460 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1468 enum glsl_sampler_dim sampler_dim
;
1469 nir_alu_type dest_type
;
1474 unsigned num_srcs
, coord_components
;
1475 bool is_array
, is_shadow
;
1478 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1479 * components or the new-style shadow that outputs 1 component.
1481 bool is_new_style_shadow
;
1483 /* gather component selector */
1484 unsigned component
: 2;
1486 /** The texture index
1488 * If this texture instruction has a nir_tex_src_texture_offset source,
1489 * then the texture index is given by texture_index + texture_offset.
1491 unsigned texture_index
;
1493 /** The size of the texture array or 0 if it's not an array */
1494 unsigned texture_array_size
;
1496 /** The sampler index
1498 * The following operations do not require a sampler and, as such, this
1499 * field should be ignored:
1501 * - nir_texop_txf_ms
1504 * - nir_texop_query_levels
1505 * - nir_texop_texture_samples
1506 * - nir_texop_samples_identical
1508 * If this texture instruction has a nir_tex_src_sampler_offset source,
1509 * then the sampler index is given by sampler_index + sampler_offset.
1511 unsigned sampler_index
;
1514 static inline unsigned
1515 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1517 switch (instr
->op
) {
1518 case nir_texop_txs
: {
1520 switch (instr
->sampler_dim
) {
1521 case GLSL_SAMPLER_DIM_1D
:
1522 case GLSL_SAMPLER_DIM_BUF
:
1525 case GLSL_SAMPLER_DIM_2D
:
1526 case GLSL_SAMPLER_DIM_CUBE
:
1527 case GLSL_SAMPLER_DIM_MS
:
1528 case GLSL_SAMPLER_DIM_RECT
:
1529 case GLSL_SAMPLER_DIM_EXTERNAL
:
1530 case GLSL_SAMPLER_DIM_SUBPASS
:
1533 case GLSL_SAMPLER_DIM_3D
:
1537 unreachable("not reached");
1539 if (instr
->is_array
)
1547 case nir_texop_texture_samples
:
1548 case nir_texop_query_levels
:
1549 case nir_texop_samples_identical
:
1553 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1560 /* Returns true if this texture operation queries something about the texture
1561 * rather than actually sampling it.
1564 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1566 switch (instr
->op
) {
1569 case nir_texop_texture_samples
:
1570 case nir_texop_query_levels
:
1571 case nir_texop_txf_ms_mcs
:
1578 case nir_texop_txf_ms
:
1582 unreachable("Invalid texture opcode");
1587 nir_alu_instr_is_comparison(const nir_alu_instr
*instr
)
1589 switch (instr
->op
) {
1610 static inline nir_alu_type
1611 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1613 switch (instr
->src
[src
].src_type
) {
1614 case nir_tex_src_coord
:
1615 switch (instr
->op
) {
1617 case nir_texop_txf_ms
:
1618 case nir_texop_txf_ms_mcs
:
1619 case nir_texop_samples_identical
:
1620 return nir_type_int
;
1623 return nir_type_float
;
1626 case nir_tex_src_lod
:
1627 switch (instr
->op
) {
1630 return nir_type_int
;
1633 return nir_type_float
;
1636 case nir_tex_src_projector
:
1637 case nir_tex_src_comparator
:
1638 case nir_tex_src_bias
:
1639 case nir_tex_src_ddx
:
1640 case nir_tex_src_ddy
:
1641 return nir_type_float
;
1643 case nir_tex_src_offset
:
1644 case nir_tex_src_ms_index
:
1645 case nir_tex_src_texture_offset
:
1646 case nir_tex_src_sampler_offset
:
1647 return nir_type_int
;
1650 unreachable("Invalid texture source type");
1654 static inline unsigned
1655 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1657 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1658 return instr
->coord_components
;
1660 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1661 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1664 if (instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1665 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1666 if (instr
->is_array
)
1667 return instr
->coord_components
- 1;
1669 return instr
->coord_components
;
1672 /* Usual APIs don't allow cube + offset, but we allow it, with 2 coords for
1673 * the offset, since a cube maps to a single face.
1675 if (instr
->src
[src
].src_type
== nir_tex_src_offset
) {
1676 if (instr
->sampler_dim
== GLSL_SAMPLER_DIM_CUBE
)
1678 else if (instr
->is_array
)
1679 return instr
->coord_components
- 1;
1681 return instr
->coord_components
;
1688 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1690 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1691 if (instr
->src
[i
].src_type
== type
)
1697 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1698 nir_tex_src_type src_type
,
1701 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1706 nir_const_value value
;
1709 } nir_load_const_instr
;
1722 /* creates a new SSA variable in an undefined state */
1727 } nir_ssa_undef_instr
;
1730 struct exec_node node
;
1732 /* The predecessor block corresponding to this source */
1733 struct nir_block
*pred
;
1738 #define nir_foreach_phi_src(phi_src, phi) \
1739 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1740 #define nir_foreach_phi_src_safe(phi_src, phi) \
1741 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1746 struct exec_list srcs
; /** < list of nir_phi_src */
1752 struct exec_node node
;
1755 } nir_parallel_copy_entry
;
1757 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1758 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1763 /* A list of nir_parallel_copy_entrys. The sources of all of the
1764 * entries are copied to the corresponding destinations "in parallel".
1765 * In other words, if we have two entries: a -> b and b -> a, the values
1768 struct exec_list entries
;
1769 } nir_parallel_copy_instr
;
1771 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1772 type
, nir_instr_type_alu
)
1773 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1774 type
, nir_instr_type_call
)
1775 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1776 type
, nir_instr_type_jump
)
1777 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1778 type
, nir_instr_type_tex
)
1779 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1780 type
, nir_instr_type_intrinsic
)
1781 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1782 type
, nir_instr_type_load_const
)
1783 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1784 type
, nir_instr_type_ssa_undef
)
1785 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1786 type
, nir_instr_type_phi
)
1787 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1788 nir_parallel_copy_instr
, instr
,
1789 type
, nir_instr_type_parallel_copy
)
1794 * Control flow consists of a tree of control flow nodes, which include
1795 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1796 * instructions that always run start-to-finish. Each basic block also keeps
1797 * track of its successors (blocks which may run immediately after the current
1798 * block) and predecessors (blocks which could have run immediately before the
1799 * current block). Each function also has a start block and an end block which
1800 * all return statements point to (which is always empty). Together, all the
1801 * blocks with their predecessors and successors make up the control flow
1802 * graph (CFG) of the function. There are helpers that modify the tree of
1803 * control flow nodes while modifying the CFG appropriately; these should be
1804 * used instead of modifying the tree directly.
1811 nir_cf_node_function
1814 typedef struct nir_cf_node
{
1815 struct exec_node node
;
1816 nir_cf_node_type type
;
1817 struct nir_cf_node
*parent
;
1820 typedef struct nir_block
{
1821 nir_cf_node cf_node
;
1823 struct exec_list instr_list
; /** < list of nir_instr */
1825 /** generic block index; generated by nir_index_blocks */
1829 * Each block can only have up to 2 successors, so we put them in a simple
1830 * array - no need for anything more complicated.
1832 struct nir_block
*successors
[2];
1834 /* Set of nir_block predecessors in the CFG */
1835 struct set
*predecessors
;
1838 * this node's immediate dominator in the dominance tree - set to NULL for
1841 struct nir_block
*imm_dom
;
1843 /* This node's children in the dominance tree */
1844 unsigned num_dom_children
;
1845 struct nir_block
**dom_children
;
1847 /* Set of nir_blocks on the dominance frontier of this block */
1848 struct set
*dom_frontier
;
1851 * These two indices have the property that dom_{pre,post}_index for each
1852 * child of this block in the dominance tree will always be between
1853 * dom_pre_index and dom_post_index for this block, which makes testing if
1854 * a given block is dominated by another block an O(1) operation.
1856 unsigned dom_pre_index
, dom_post_index
;
1858 /* live in and out for this block; used for liveness analysis */
1859 BITSET_WORD
*live_in
;
1860 BITSET_WORD
*live_out
;
1863 static inline nir_instr
*
1864 nir_block_first_instr(nir_block
*block
)
1866 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1867 return exec_node_data(nir_instr
, head
, node
);
1870 static inline nir_instr
*
1871 nir_block_last_instr(nir_block
*block
)
1873 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1874 return exec_node_data(nir_instr
, tail
, node
);
1878 nir_block_ends_in_jump(nir_block
*block
)
1880 return !exec_list_is_empty(&block
->instr_list
) &&
1881 nir_block_last_instr(block
)->type
== nir_instr_type_jump
;
1884 #define nir_foreach_instr(instr, block) \
1885 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1886 #define nir_foreach_instr_reverse(instr, block) \
1887 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1888 #define nir_foreach_instr_safe(instr, block) \
1889 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1890 #define nir_foreach_instr_reverse_safe(instr, block) \
1891 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1893 typedef struct nir_if
{
1894 nir_cf_node cf_node
;
1897 struct exec_list then_list
; /** < list of nir_cf_node */
1898 struct exec_list else_list
; /** < list of nir_cf_node */
1904 nir_instr
*conditional_instr
;
1906 nir_block
*break_block
;
1907 nir_block
*continue_from_block
;
1909 bool continue_from_then
;
1911 struct list_head loop_terminator_link
;
1912 } nir_loop_terminator
;
1915 /* Number of instructions in the loop */
1916 unsigned num_instructions
;
1918 /* Maximum number of times the loop is run (if known) */
1919 unsigned max_trip_count
;
1921 /* Do we know the exact number of times the loop will be run */
1922 bool exact_trip_count_known
;
1924 /* Unroll the loop regardless of its size */
1927 /* Does the loop contain complex loop terminators, continues or other
1928 * complex behaviours? If this is true we can't rely on
1929 * loop_terminator_list to be complete or accurate.
1933 nir_loop_terminator
*limiting_terminator
;
1935 /* A list of loop_terminators terminating this loop. */
1936 struct list_head loop_terminator_list
;
1940 nir_cf_node cf_node
;
1942 struct exec_list body
; /** < list of nir_cf_node */
1944 nir_loop_info
*info
;
1948 * Various bits of metadata that can may be created or required by
1949 * optimization and analysis passes
1952 nir_metadata_none
= 0x0,
1953 nir_metadata_block_index
= 0x1,
1954 nir_metadata_dominance
= 0x2,
1955 nir_metadata_live_ssa_defs
= 0x4,
1956 nir_metadata_not_properly_reset
= 0x8,
1957 nir_metadata_loop_analysis
= 0x10,
1961 nir_cf_node cf_node
;
1963 /** pointer to the function of which this is an implementation */
1964 struct nir_function
*function
;
1966 struct exec_list body
; /** < list of nir_cf_node */
1968 nir_block
*end_block
;
1970 /** list for all local variables in the function */
1971 struct exec_list locals
;
1973 /** list of local registers in the function */
1974 struct exec_list registers
;
1976 /** next available local register index */
1979 /** next available SSA value index */
1982 /* total number of basic blocks, only valid when block_index_dirty = false */
1983 unsigned num_blocks
;
1985 nir_metadata valid_metadata
;
1986 } nir_function_impl
;
1988 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1989 nir_start_block(nir_function_impl
*impl
)
1991 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1994 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1995 nir_impl_last_block(nir_function_impl
*impl
)
1997 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
2000 static inline nir_cf_node
*
2001 nir_cf_node_next(nir_cf_node
*node
)
2003 struct exec_node
*next
= exec_node_get_next(&node
->node
);
2004 if (exec_node_is_tail_sentinel(next
))
2007 return exec_node_data(nir_cf_node
, next
, node
);
2010 static inline nir_cf_node
*
2011 nir_cf_node_prev(nir_cf_node
*node
)
2013 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
2014 if (exec_node_is_head_sentinel(prev
))
2017 return exec_node_data(nir_cf_node
, prev
, node
);
2021 nir_cf_node_is_first(const nir_cf_node
*node
)
2023 return exec_node_is_head_sentinel(node
->node
.prev
);
2027 nir_cf_node_is_last(const nir_cf_node
*node
)
2029 return exec_node_is_tail_sentinel(node
->node
.next
);
2032 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
2033 type
, nir_cf_node_block
)
2034 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
2035 type
, nir_cf_node_if
)
2036 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
2037 type
, nir_cf_node_loop
)
2038 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
2039 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
2041 static inline nir_block
*
2042 nir_if_first_then_block(nir_if
*if_stmt
)
2044 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
2045 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2048 static inline nir_block
*
2049 nir_if_last_then_block(nir_if
*if_stmt
)
2051 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
2052 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2055 static inline nir_block
*
2056 nir_if_first_else_block(nir_if
*if_stmt
)
2058 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
2059 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2062 static inline nir_block
*
2063 nir_if_last_else_block(nir_if
*if_stmt
)
2065 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
2066 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2069 static inline nir_block
*
2070 nir_loop_first_block(nir_loop
*loop
)
2072 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
2073 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2076 static inline nir_block
*
2077 nir_loop_last_block(nir_loop
*loop
)
2079 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
2080 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2084 uint8_t num_components
;
2088 typedef struct nir_function
{
2089 struct exec_node node
;
2092 struct nir_shader
*shader
;
2094 unsigned num_params
;
2095 nir_parameter
*params
;
2097 /** The implementation of this function.
2099 * If the function is only declared and not implemented, this is NULL.
2101 nir_function_impl
*impl
;
2106 typedef struct nir_shader_compiler_options
{
2111 /** Lowers flrp when it does not support doubles */
2118 /** Lowers ibitfield_extract/ubitfield_extract to ibfe/ubfe. */
2119 bool lower_bitfield_extract
;
2120 /** Lowers ibitfield_extract/ubitfield_extract to bfm, compares, shifts. */
2121 bool lower_bitfield_extract_to_shifts
;
2122 /** Lowers bitfield_insert to bfi/bfm */
2123 bool lower_bitfield_insert
;
2124 /** Lowers bitfield_insert to bfm, compares, and shifts. */
2125 bool lower_bitfield_insert_to_shifts
;
2126 /** Lowers bitfield_reverse to shifts. */
2127 bool lower_bitfield_reverse
;
2128 /** Lowers bit_count to shifts. */
2129 bool lower_bit_count
;
2130 /** Lowers bfm to shifts and subtracts. */
2132 /** Lowers ifind_msb to compare and ufind_msb */
2133 bool lower_ifind_msb
;
2134 /** Lowers find_lsb to ufind_msb and logic ops */
2135 bool lower_find_lsb
;
2136 bool lower_uadd_carry
;
2137 bool lower_usub_borrow
;
2138 /** Lowers imul_high/umul_high to 16-bit multiplies and carry operations. */
2139 bool lower_mul_high
;
2140 /** lowers fneg and ineg to fsub and isub. */
2142 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
2145 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
2148 /** enables rules to lower idiv by power-of-two: */
2151 /* Does the native fdot instruction replicate its result for four
2152 * components? If so, then opt_algebraic_late will turn all fdotN
2153 * instructions into fdot_replicatedN instructions.
2155 bool fdot_replicates
;
2157 /** lowers ffloor to fsub+ffract: */
2160 /** lowers ffract to fsub+ffloor: */
2163 /** lowers fceil to fneg+ffloor+fneg: */
2168 bool lower_pack_half_2x16
;
2169 bool lower_pack_unorm_2x16
;
2170 bool lower_pack_snorm_2x16
;
2171 bool lower_pack_unorm_4x8
;
2172 bool lower_pack_snorm_4x8
;
2173 bool lower_unpack_half_2x16
;
2174 bool lower_unpack_unorm_2x16
;
2175 bool lower_unpack_snorm_2x16
;
2176 bool lower_unpack_unorm_4x8
;
2177 bool lower_unpack_snorm_4x8
;
2179 bool lower_extract_byte
;
2180 bool lower_extract_word
;
2182 bool lower_all_io_to_temps
;
2185 * Does the driver support real 32-bit integers? (Otherwise, integers
2186 * are simulated by floats.)
2188 bool native_integers
;
2190 /* Indicates that the driver only has zero-based vertex id */
2191 bool vertex_id_zero_based
;
2194 * If enabled, gl_BaseVertex will be lowered as:
2195 * is_indexed_draw (~0/0) & firstvertex
2197 bool lower_base_vertex
;
2200 * If enabled, gl_HelperInvocation will be lowered as:
2202 * !((1 << sample_id) & sample_mask_in))
2204 * This depends on some possibly hw implementation details, which may
2205 * not be true for all hw. In particular that the FS is only executed
2206 * for covered samples or for helper invocations. So, do not blindly
2207 * enable this option.
2209 * Note: See also issue #22 in ARB_shader_image_load_store
2211 bool lower_helper_invocation
;
2213 bool lower_cs_local_index_from_id
;
2214 bool lower_cs_local_id_from_index
;
2216 bool lower_device_index_to_zero
;
2218 /* Set if nir_lower_wpos_ytransform() should also invert gl_PointCoord. */
2219 bool lower_wpos_pntc
;
2222 * Should nir_lower_io() create load_interpolated_input intrinsics?
2224 * If not, it generates regular load_input intrinsics and interpolation
2225 * information must be inferred from the list of input nir_variables.
2227 bool use_interpolated_input_intrinsics
;
2229 unsigned max_unroll_iterations
;
2230 } nir_shader_compiler_options
;
2232 typedef struct nir_shader
{
2233 /** list of uniforms (nir_variable) */
2234 struct exec_list uniforms
;
2236 /** list of inputs (nir_variable) */
2237 struct exec_list inputs
;
2239 /** list of outputs (nir_variable) */
2240 struct exec_list outputs
;
2242 /** list of shared compute variables (nir_variable) */
2243 struct exec_list shared
;
2245 /** Set of driver-specific options for the shader.
2247 * The memory for the options is expected to be kept in a single static
2248 * copy by the driver.
2250 const struct nir_shader_compiler_options
*options
;
2252 /** Various bits of compile-time information about a given shader */
2253 struct shader_info info
;
2255 /** list of global variables in the shader (nir_variable) */
2256 struct exec_list globals
;
2258 /** list of system value variables in the shader (nir_variable) */
2259 struct exec_list system_values
;
2261 struct exec_list functions
; /** < list of nir_function */
2263 /** list of global register in the shader */
2264 struct exec_list registers
;
2266 /** next available global register index */
2270 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
2273 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
2275 /** Constant data associated with this shader.
2277 * Constant data is loaded through load_constant intrinsics. See also
2278 * nir_opt_large_constants.
2280 void *constant_data
;
2281 unsigned constant_data_size
;
2284 #define nir_foreach_function(func, shader) \
2285 foreach_list_typed(nir_function, func, node, &(shader)->functions)
2287 static inline nir_function_impl
*
2288 nir_shader_get_entrypoint(nir_shader
*shader
)
2290 nir_function
*func
= NULL
;
2292 nir_foreach_function(function
, shader
) {
2293 assert(func
== NULL
);
2294 if (function
->is_entrypoint
) {
2305 assert(func
->num_params
== 0);
2310 nir_shader
*nir_shader_create(void *mem_ctx
,
2311 gl_shader_stage stage
,
2312 const nir_shader_compiler_options
*options
,
2315 /** creates a register, including assigning it an index and adding it to the list */
2316 nir_register
*nir_global_reg_create(nir_shader
*shader
);
2318 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
2320 void nir_reg_remove(nir_register
*reg
);
2322 /** Adds a variable to the appropriate list in nir_shader */
2323 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
2326 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
2328 assert(var
->data
.mode
== nir_var_function_temp
);
2329 exec_list_push_tail(&impl
->locals
, &var
->node
);
2332 /** creates a variable, sets a few defaults, and adds it to the list */
2333 nir_variable
*nir_variable_create(nir_shader
*shader
,
2334 nir_variable_mode mode
,
2335 const struct glsl_type
*type
,
2337 /** creates a local variable and adds it to the list */
2338 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
2339 const struct glsl_type
*type
,
2342 /** creates a function and adds it to the shader's list of functions */
2343 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
2345 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
2346 /** creates a function_impl that isn't tied to any particular function */
2347 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
2349 nir_block
*nir_block_create(nir_shader
*shader
);
2350 nir_if
*nir_if_create(nir_shader
*shader
);
2351 nir_loop
*nir_loop_create(nir_shader
*shader
);
2353 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
2355 /** requests that the given pieces of metadata be generated */
2356 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
2357 /** dirties all but the preserved metadata */
2358 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
2360 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
2361 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
2363 nir_deref_instr
*nir_deref_instr_create(nir_shader
*shader
,
2364 nir_deref_type deref_type
);
2366 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
2368 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
2369 unsigned num_components
,
2372 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
2373 nir_intrinsic_op op
);
2375 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
2376 nir_function
*callee
);
2378 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
2380 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
2382 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
2384 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
2385 unsigned num_components
,
2388 nir_const_value
nir_alu_binop_identity(nir_op binop
, unsigned bit_size
);
2391 * NIR Cursors and Instruction Insertion API
2394 * A tiny struct representing a point to insert/extract instructions or
2395 * control flow nodes. Helps reduce the combinatorial explosion of possible
2396 * points to insert/extract.
2398 * \sa nir_control_flow.h
2401 nir_cursor_before_block
,
2402 nir_cursor_after_block
,
2403 nir_cursor_before_instr
,
2404 nir_cursor_after_instr
,
2405 } nir_cursor_option
;
2408 nir_cursor_option option
;
2415 static inline nir_block
*
2416 nir_cursor_current_block(nir_cursor cursor
)
2418 if (cursor
.option
== nir_cursor_before_instr
||
2419 cursor
.option
== nir_cursor_after_instr
) {
2420 return cursor
.instr
->block
;
2422 return cursor
.block
;
2426 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2428 static inline nir_cursor
2429 nir_before_block(nir_block
*block
)
2432 cursor
.option
= nir_cursor_before_block
;
2433 cursor
.block
= block
;
2437 static inline nir_cursor
2438 nir_after_block(nir_block
*block
)
2441 cursor
.option
= nir_cursor_after_block
;
2442 cursor
.block
= block
;
2446 static inline nir_cursor
2447 nir_before_instr(nir_instr
*instr
)
2450 cursor
.option
= nir_cursor_before_instr
;
2451 cursor
.instr
= instr
;
2455 static inline nir_cursor
2456 nir_after_instr(nir_instr
*instr
)
2459 cursor
.option
= nir_cursor_after_instr
;
2460 cursor
.instr
= instr
;
2464 static inline nir_cursor
2465 nir_after_block_before_jump(nir_block
*block
)
2467 nir_instr
*last_instr
= nir_block_last_instr(block
);
2468 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2469 return nir_before_instr(last_instr
);
2471 return nir_after_block(block
);
2475 static inline nir_cursor
2476 nir_before_src(nir_src
*src
, bool is_if_condition
)
2478 if (is_if_condition
) {
2479 nir_block
*prev_block
=
2480 nir_cf_node_as_block(nir_cf_node_prev(&src
->parent_if
->cf_node
));
2481 assert(!nir_block_ends_in_jump(prev_block
));
2482 return nir_after_block(prev_block
);
2483 } else if (src
->parent_instr
->type
== nir_instr_type_phi
) {
2485 nir_phi_instr
*cond_phi
= nir_instr_as_phi(src
->parent_instr
);
2487 nir_foreach_phi_src(phi_src
, cond_phi
) {
2488 if (phi_src
->src
.ssa
== src
->ssa
) {
2495 /* The LIST_ENTRY macro is a generic container-of macro, it just happens
2496 * to have a more specific name.
2498 nir_phi_src
*phi_src
= LIST_ENTRY(nir_phi_src
, src
, src
);
2499 return nir_after_block_before_jump(phi_src
->pred
);
2501 return nir_before_instr(src
->parent_instr
);
2505 static inline nir_cursor
2506 nir_before_cf_node(nir_cf_node
*node
)
2508 if (node
->type
== nir_cf_node_block
)
2509 return nir_before_block(nir_cf_node_as_block(node
));
2511 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2514 static inline nir_cursor
2515 nir_after_cf_node(nir_cf_node
*node
)
2517 if (node
->type
== nir_cf_node_block
)
2518 return nir_after_block(nir_cf_node_as_block(node
));
2520 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2523 static inline nir_cursor
2524 nir_after_phis(nir_block
*block
)
2526 nir_foreach_instr(instr
, block
) {
2527 if (instr
->type
!= nir_instr_type_phi
)
2528 return nir_before_instr(instr
);
2530 return nir_after_block(block
);
2533 static inline nir_cursor
2534 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2536 if (node
->type
== nir_cf_node_block
)
2537 return nir_after_block(nir_cf_node_as_block(node
));
2539 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2541 return nir_after_phis(block
);
2544 static inline nir_cursor
2545 nir_before_cf_list(struct exec_list
*cf_list
)
2547 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2548 exec_list_get_head(cf_list
), node
);
2549 return nir_before_cf_node(first_node
);
2552 static inline nir_cursor
2553 nir_after_cf_list(struct exec_list
*cf_list
)
2555 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2556 exec_list_get_tail(cf_list
), node
);
2557 return nir_after_cf_node(last_node
);
2561 * Insert a NIR instruction at the given cursor.
2563 * Note: This does not update the cursor.
2565 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2568 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2570 nir_instr_insert(nir_before_instr(instr
), before
);
2574 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2576 nir_instr_insert(nir_after_instr(instr
), after
);
2580 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2582 nir_instr_insert(nir_before_block(block
), before
);
2586 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2588 nir_instr_insert(nir_after_block(block
), after
);
2592 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2594 nir_instr_insert(nir_before_cf_node(node
), before
);
2598 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2600 nir_instr_insert(nir_after_cf_node(node
), after
);
2604 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2606 nir_instr_insert(nir_before_cf_list(list
), before
);
2610 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2612 nir_instr_insert(nir_after_cf_list(list
), after
);
2615 void nir_instr_remove_v(nir_instr
*instr
);
2617 static inline nir_cursor
2618 nir_instr_remove(nir_instr
*instr
)
2621 nir_instr
*prev
= nir_instr_prev(instr
);
2623 cursor
= nir_after_instr(prev
);
2625 cursor
= nir_before_block(instr
->block
);
2627 nir_instr_remove_v(instr
);
2633 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2634 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2635 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2636 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2638 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2639 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2641 nir_const_value
*nir_src_as_const_value(nir_src src
);
2643 static inline struct nir_instr
*
2644 nir_src_instr(const struct nir_src
*src
)
2646 return src
->is_ssa
? src
->ssa
->parent_instr
: NULL
;
2649 #define NIR_SRC_AS_(name, c_type, type_enum, cast_macro) \
2650 static inline c_type * \
2651 nir_src_as_ ## name (struct nir_src *src) \
2653 return src->is_ssa && src->ssa->parent_instr->type == type_enum \
2654 ? cast_macro(src->ssa->parent_instr) : NULL; \
2656 static inline const c_type * \
2657 nir_src_as_ ## name ## _const(const struct nir_src *src) \
2659 return src->is_ssa && src->ssa->parent_instr->type == type_enum \
2660 ? cast_macro(src->ssa->parent_instr) : NULL; \
2663 NIR_SRC_AS_(alu_instr
, nir_alu_instr
, nir_instr_type_alu
, nir_instr_as_alu
)
2665 bool nir_src_is_dynamically_uniform(nir_src src
);
2666 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2667 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2668 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2669 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2670 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2673 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2674 unsigned num_components
, unsigned bit_size
,
2676 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2677 unsigned num_components
, unsigned bit_size
,
2680 nir_ssa_dest_init_for_type(nir_instr
*instr
, nir_dest
*dest
,
2681 const struct glsl_type
*type
,
2684 assert(glsl_type_is_vector_or_scalar(type
));
2685 nir_ssa_dest_init(instr
, dest
, glsl_get_components(type
),
2686 glsl_get_bit_size(type
), name
);
2688 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2689 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2690 nir_instr
*after_me
);
2692 nir_component_mask_t
nir_ssa_def_components_read(const nir_ssa_def
*def
);
2695 * finds the next basic block in source-code order, returns NULL if there is
2699 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2701 /* Performs the opposite of nir_block_cf_tree_next() */
2703 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2705 /* Gets the first block in a CF node in source-code order */
2707 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2709 /* Gets the last block in a CF node in source-code order */
2711 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2713 /* Gets the next block after a CF node in source-code order */
2715 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2717 /* Macros for loops that visit blocks in source-code order */
2719 #define nir_foreach_block(block, impl) \
2720 for (nir_block *block = nir_start_block(impl); block != NULL; \
2721 block = nir_block_cf_tree_next(block))
2723 #define nir_foreach_block_safe(block, impl) \
2724 for (nir_block *block = nir_start_block(impl), \
2725 *next = nir_block_cf_tree_next(block); \
2727 block = next, next = nir_block_cf_tree_next(block))
2729 #define nir_foreach_block_reverse(block, impl) \
2730 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2731 block = nir_block_cf_tree_prev(block))
2733 #define nir_foreach_block_reverse_safe(block, impl) \
2734 for (nir_block *block = nir_impl_last_block(impl), \
2735 *prev = nir_block_cf_tree_prev(block); \
2737 block = prev, prev = nir_block_cf_tree_prev(block))
2739 #define nir_foreach_block_in_cf_node(block, node) \
2740 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2741 block != nir_cf_node_cf_tree_next(node); \
2742 block = nir_block_cf_tree_next(block))
2744 /* If the following CF node is an if, this function returns that if.
2745 * Otherwise, it returns NULL.
2747 nir_if
*nir_block_get_following_if(nir_block
*block
);
2749 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2751 void nir_index_local_regs(nir_function_impl
*impl
);
2752 void nir_index_global_regs(nir_shader
*shader
);
2753 void nir_index_ssa_defs(nir_function_impl
*impl
);
2754 unsigned nir_index_instrs(nir_function_impl
*impl
);
2756 void nir_index_blocks(nir_function_impl
*impl
);
2758 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2759 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2760 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2761 void nir_print_deref(const nir_deref_instr
*deref
, FILE *fp
);
2763 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2764 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2765 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2766 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2768 nir_shader
*nir_shader_serialize_deserialize(void *mem_ctx
, nir_shader
*s
);
2771 void nir_validate_shader(nir_shader
*shader
, const char *when
);
2772 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2773 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2776 should_skip_nir(const char *name
)
2778 static const char *list
= NULL
;
2780 /* Comma separated list of names to skip. */
2781 list
= getenv("NIR_SKIP");
2789 return comma_separated_list_contains(list
, name
);
2793 should_clone_nir(void)
2795 static int should_clone
= -1;
2796 if (should_clone
< 0)
2797 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2799 return should_clone
;
2803 should_serialize_deserialize_nir(void)
2805 static int test_serialize
= -1;
2806 if (test_serialize
< 0)
2807 test_serialize
= env_var_as_boolean("NIR_TEST_SERIALIZE", false);
2809 return test_serialize
;
2813 should_print_nir(void)
2815 static int should_print
= -1;
2816 if (should_print
< 0)
2817 should_print
= env_var_as_boolean("NIR_PRINT", false);
2819 return should_print
;
2822 static inline void nir_validate_shader(nir_shader
*shader
, const char *when
) { (void) shader
; (void)when
; }
2823 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2824 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2825 static inline bool should_skip_nir(const char *pass_name
) { return false; }
2826 static inline bool should_clone_nir(void) { return false; }
2827 static inline bool should_serialize_deserialize_nir(void) { return false; }
2828 static inline bool should_print_nir(void) { return false; }
2831 #define _PASS(pass, nir, do_pass) do { \
2832 if (should_skip_nir(#pass)) { \
2833 printf("skipping %s\n", #pass); \
2837 nir_validate_shader(nir, "after " #pass); \
2838 if (should_clone_nir()) { \
2839 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2843 if (should_serialize_deserialize_nir()) { \
2844 void *mem_ctx = ralloc_parent(nir); \
2845 nir = nir_shader_serialize_deserialize(mem_ctx, nir); \
2849 #define NIR_PASS(progress, nir, pass, ...) _PASS(pass, nir, \
2850 nir_metadata_set_validation_flag(nir); \
2851 if (should_print_nir()) \
2852 printf("%s\n", #pass); \
2853 if (pass(nir, ##__VA_ARGS__)) { \
2855 if (should_print_nir()) \
2856 nir_print_shader(nir, stdout); \
2857 nir_metadata_check_validation_flag(nir); \
2861 #define NIR_PASS_V(nir, pass, ...) _PASS(pass, nir, \
2862 if (should_print_nir()) \
2863 printf("%s\n", #pass); \
2864 pass(nir, ##__VA_ARGS__); \
2865 if (should_print_nir()) \
2866 nir_print_shader(nir, stdout); \
2869 #define NIR_SKIP(name) should_skip_nir(#name)
2871 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2872 void nir_calc_dominance(nir_shader
*shader
);
2874 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2875 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2877 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2878 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2880 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2881 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2883 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2884 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2886 int nir_gs_count_vertices(const nir_shader
*shader
);
2888 bool nir_shrink_vec_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2889 bool nir_split_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2890 bool nir_split_var_copies(nir_shader
*shader
);
2891 bool nir_split_per_member_structs(nir_shader
*shader
);
2892 bool nir_split_struct_vars(nir_shader
*shader
, nir_variable_mode modes
);
2894 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2895 bool nir_lower_returns(nir_shader
*shader
);
2897 bool nir_inline_functions(nir_shader
*shader
);
2899 bool nir_propagate_invariant(nir_shader
*shader
);
2901 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2902 void nir_lower_deref_copy_instr(struct nir_builder
*b
,
2903 nir_intrinsic_instr
*copy
);
2904 bool nir_lower_var_copies(nir_shader
*shader
);
2906 void nir_fixup_deref_modes(nir_shader
*shader
);
2908 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2910 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2912 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2914 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2915 nir_function_impl
*entrypoint
,
2916 bool outputs
, bool inputs
);
2918 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2920 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2921 int (*type_size
)(const struct glsl_type
*));
2923 /* Some helpers to do very simple linking */
2924 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2925 bool nir_remove_unused_io_vars(nir_shader
*shader
, struct exec_list
*var_list
,
2926 uint64_t *used_by_other_stage
,
2927 uint64_t *used_by_other_stage_patches
);
2928 void nir_compact_varyings(nir_shader
*producer
, nir_shader
*consumer
,
2929 bool default_to_smooth_interp
);
2930 void nir_link_xfb_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2931 bool nir_link_opt_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2934 /* If set, this forces all non-flat fragment shader inputs to be
2935 * interpolated as if with the "sample" qualifier. This requires
2936 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2938 nir_lower_io_force_sample_interpolation
= (1 << 1),
2939 } nir_lower_io_options
;
2940 bool nir_lower_io(nir_shader
*shader
,
2941 nir_variable_mode modes
,
2942 int (*type_size
)(const struct glsl_type
*),
2943 nir_lower_io_options
);
2947 * An address format which is comprised of a vec2 where the first
2948 * component is a vulkan descriptor index and the second is an offset.
2950 nir_address_format_vk_index_offset
,
2951 } nir_address_format
;
2952 bool nir_lower_explicit_io(nir_shader
*shader
,
2953 nir_variable_mode modes
,
2954 nir_address_format
);
2956 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2957 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2959 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2961 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2962 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2963 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2965 bool nir_remove_dead_derefs(nir_shader
*shader
);
2966 bool nir_remove_dead_derefs_impl(nir_function_impl
*impl
);
2967 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2968 bool nir_lower_constant_initializers(nir_shader
*shader
,
2969 nir_variable_mode modes
);
2971 bool nir_move_load_const(nir_shader
*shader
);
2972 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2973 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2974 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
2976 bool nir_lower_alu(nir_shader
*shader
);
2977 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2978 bool nir_lower_bool_to_float(nir_shader
*shader
);
2979 bool nir_lower_bool_to_int32(nir_shader
*shader
);
2980 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2981 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
2982 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2983 void nir_lower_io_arrays_to_elements(nir_shader
*producer
, nir_shader
*consumer
);
2984 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader
*shader
,
2986 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2987 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
2989 typedef struct nir_lower_subgroups_options
{
2990 uint8_t subgroup_size
;
2991 uint8_t ballot_bit_size
;
2992 bool lower_to_scalar
:1;
2993 bool lower_vote_trivial
:1;
2994 bool lower_vote_eq_to_ballot
:1;
2995 bool lower_subgroup_masks
:1;
2996 bool lower_shuffle
:1;
2997 bool lower_shuffle_to_32bit
:1;
2999 } nir_lower_subgroups_options
;
3001 bool nir_lower_subgroups(nir_shader
*shader
,
3002 const nir_lower_subgroups_options
*options
);
3004 bool nir_lower_system_values(nir_shader
*shader
);
3006 enum PACKED nir_lower_tex_packing
{
3007 nir_lower_tex_packing_none
= 0,
3008 /* The sampler returns up to 2 32-bit words of half floats or 16-bit signed
3009 * or unsigned ints based on the sampler type
3011 nir_lower_tex_packing_16
,
3012 /* The sampler returns 1 32-bit word of 4x8 unorm */
3013 nir_lower_tex_packing_8
,
3016 typedef struct nir_lower_tex_options
{
3018 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
3019 * sampler types a texture projector is lowered.
3024 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
3026 bool lower_txf_offset
;
3029 * If true, lower away nir_tex_src_offset for all rect textures.
3031 bool lower_rect_offset
;
3034 * If true, lower rect textures to 2D, using txs to fetch the
3035 * texture dimensions and dividing the texture coords by the
3036 * texture dims to normalize.
3041 * If true, convert yuv to rgb.
3043 unsigned lower_y_uv_external
;
3044 unsigned lower_y_u_v_external
;
3045 unsigned lower_yx_xuxv_external
;
3046 unsigned lower_xy_uxvx_external
;
3047 unsigned lower_ayuv_external
;
3050 * To emulate certain texture wrap modes, this can be used
3051 * to saturate the specified tex coord to [0.0, 1.0]. The
3052 * bits are according to sampler #, ie. if, for example:
3054 * (conf->saturate_s & (1 << n))
3056 * is true, then the s coord for sampler n is saturated.
3058 * Note that clamping must happen *after* projector lowering
3059 * so any projected texture sample instruction with a clamped
3060 * coordinate gets automatically lowered, regardless of the
3061 * 'lower_txp' setting.
3063 unsigned saturate_s
;
3064 unsigned saturate_t
;
3065 unsigned saturate_r
;
3067 /* Bitmask of textures that need swizzling.
3069 * If (swizzle_result & (1 << texture_index)), then the swizzle in
3070 * swizzles[texture_index] is applied to the result of the texturing
3073 unsigned swizzle_result
;
3075 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
3076 * while 4 and 5 represent 0 and 1 respectively.
3078 uint8_t swizzles
[32][4];
3081 * Bitmap of textures that need srgb to linear conversion. If
3082 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
3083 * of the texture are lowered to linear.
3085 unsigned lower_srgb
;
3088 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
3090 bool lower_txd_cube_map
;
3093 * If true, lower nir_texop_txd on 3D surfaces with nir_texop_txl.
3098 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
3099 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
3100 * with lower_txd_cube_map.
3102 bool lower_txd_shadow
;
3105 * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
3106 * Implies lower_txd_cube_map and lower_txd_shadow.
3111 * If true, lower nir_texop_txb that try to use shadow compare and min_lod
3112 * at the same time to a nir_texop_lod, some math, and nir_texop_tex.
3114 bool lower_txb_shadow_clamp
;
3117 * If true, lower nir_texop_txd on shadow samplers when it uses min_lod
3118 * with nir_texop_txl. This includes cube maps.
3120 bool lower_txd_shadow_clamp
;
3123 * If true, lower nir_texop_txd on when it uses both offset and min_lod
3124 * with nir_texop_txl. This includes cube maps.
3126 bool lower_txd_offset_clamp
;
3129 * If true, apply a .bagr swizzle on tg4 results to handle Broadcom's
3130 * mixed-up tg4 locations.
3132 bool lower_tg4_broadcom_swizzle
;
3134 enum nir_lower_tex_packing lower_tex_packing
[32];
3135 } nir_lower_tex_options
;
3137 bool nir_lower_tex(nir_shader
*shader
,
3138 const nir_lower_tex_options
*options
);
3140 bool nir_lower_idiv(nir_shader
*shader
);
3142 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
, bool use_vars
);
3143 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
3144 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
3146 void nir_lower_two_sided_color(nir_shader
*shader
);
3148 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
3150 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
3151 bool nir_lower_patch_vertices(nir_shader
*nir
, unsigned static_count
,
3152 const gl_state_index16
*uniform_state_tokens
);
3154 typedef struct nir_lower_wpos_ytransform_options
{
3155 gl_state_index16 state_tokens
[STATE_LENGTH
];
3156 bool fs_coord_origin_upper_left
:1;
3157 bool fs_coord_origin_lower_left
:1;
3158 bool fs_coord_pixel_center_integer
:1;
3159 bool fs_coord_pixel_center_half_integer
:1;
3160 } nir_lower_wpos_ytransform_options
;
3162 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
3163 const nir_lower_wpos_ytransform_options
*options
);
3164 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
3166 typedef struct nir_lower_drawpixels_options
{
3167 gl_state_index16 texcoord_state_tokens
[STATE_LENGTH
];
3168 gl_state_index16 scale_state_tokens
[STATE_LENGTH
];
3169 gl_state_index16 bias_state_tokens
[STATE_LENGTH
];
3170 unsigned drawpix_sampler
;
3171 unsigned pixelmap_sampler
;
3173 bool scale_and_bias
:1;
3174 } nir_lower_drawpixels_options
;
3176 void nir_lower_drawpixels(nir_shader
*shader
,
3177 const nir_lower_drawpixels_options
*options
);
3179 typedef struct nir_lower_bitmap_options
{
3182 } nir_lower_bitmap_options
;
3184 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
3186 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
3189 nir_lower_int_source_mods
= 1 << 0,
3190 nir_lower_float_source_mods
= 1 << 1,
3191 nir_lower_all_source_mods
= (1 << 2) - 1
3192 } nir_lower_to_source_mods_flags
;
3195 bool nir_lower_to_source_mods(nir_shader
*shader
, nir_lower_to_source_mods_flags options
);
3197 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
3199 typedef unsigned (*nir_lower_bit_size_callback
)(const nir_alu_instr
*, void *);
3201 bool nir_lower_bit_size(nir_shader
*shader
,
3202 nir_lower_bit_size_callback callback
,
3203 void *callback_data
);
3206 nir_lower_imul64
= (1 << 0),
3207 nir_lower_isign64
= (1 << 1),
3208 /** Lower all int64 modulus and division opcodes */
3209 nir_lower_divmod64
= (1 << 2),
3210 /** Lower all 64-bit umul_high and imul_high opcodes */
3211 nir_lower_imul_high64
= (1 << 3),
3212 nir_lower_mov64
= (1 << 4),
3213 nir_lower_icmp64
= (1 << 5),
3214 nir_lower_iadd64
= (1 << 6),
3215 nir_lower_iabs64
= (1 << 7),
3216 nir_lower_ineg64
= (1 << 8),
3217 nir_lower_logic64
= (1 << 9),
3218 nir_lower_minmax64
= (1 << 10),
3219 nir_lower_shift64
= (1 << 11),
3220 } nir_lower_int64_options
;
3222 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
3225 nir_lower_drcp
= (1 << 0),
3226 nir_lower_dsqrt
= (1 << 1),
3227 nir_lower_drsq
= (1 << 2),
3228 nir_lower_dtrunc
= (1 << 3),
3229 nir_lower_dfloor
= (1 << 4),
3230 nir_lower_dceil
= (1 << 5),
3231 nir_lower_dfract
= (1 << 6),
3232 nir_lower_dround_even
= (1 << 7),
3233 nir_lower_dmod
= (1 << 8),
3234 nir_lower_fp64_full_software
= (1 << 9),
3235 } nir_lower_doubles_options
;
3237 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
3238 bool nir_lower_pack(nir_shader
*shader
);
3240 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
3242 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
3244 void nir_loop_analyze_impl(nir_function_impl
*impl
,
3245 nir_variable_mode indirect_mask
);
3247 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
3249 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
3250 bool nir_repair_ssa(nir_shader
*shader
);
3252 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
3254 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
3255 * registers. If false, convert all values (even those not involved in a phi
3256 * node) to registers.
3258 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
3260 bool nir_lower_phis_to_regs_block(nir_block
*block
);
3261 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
3262 bool nir_rematerialize_derefs_in_use_blocks_impl(nir_function_impl
*impl
);
3264 bool nir_opt_algebraic(nir_shader
*shader
);
3265 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
3266 bool nir_opt_algebraic_late(nir_shader
*shader
);
3267 bool nir_opt_constant_folding(nir_shader
*shader
);
3269 bool nir_opt_global_to_local(nir_shader
*shader
);
3271 bool nir_copy_prop(nir_shader
*shader
);
3273 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
3275 bool nir_opt_cse(nir_shader
*shader
);
3277 bool nir_opt_dce(nir_shader
*shader
);
3279 bool nir_opt_dead_cf(nir_shader
*shader
);
3281 bool nir_opt_dead_write_vars(nir_shader
*shader
);
3283 bool nir_opt_deref(nir_shader
*shader
);
3285 bool nir_opt_find_array_copies(nir_shader
*shader
);
3287 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
3289 bool nir_opt_idiv_const(nir_shader
*shader
, unsigned min_bit_size
);
3291 bool nir_opt_if(nir_shader
*shader
);
3293 bool nir_opt_intrinsics(nir_shader
*shader
);
3295 bool nir_opt_large_constants(nir_shader
*shader
,
3296 glsl_type_size_align_func size_align
,
3297 unsigned threshold
);
3299 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
3301 bool nir_opt_move_comparisons(nir_shader
*shader
);
3303 bool nir_opt_move_load_ubo(nir_shader
*shader
);
3305 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
,
3306 bool indirect_load_ok
, bool expensive_alu_ok
);
3308 bool nir_opt_remove_phis(nir_shader
*shader
);
3310 bool nir_opt_shrink_load(nir_shader
*shader
);
3312 bool nir_opt_trivial_continues(nir_shader
*shader
);
3314 bool nir_opt_undef(nir_shader
*shader
);
3316 bool nir_opt_conditional_discard(nir_shader
*shader
);
3318 void nir_sweep(nir_shader
*shader
);
3320 void nir_remap_dual_slot_attributes(nir_shader
*shader
,
3321 uint64_t *dual_slot_inputs
);
3322 uint64_t nir_get_single_slot_attribs_mask(uint64_t attribs
, uint64_t dual_slot
);
3324 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
3325 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);