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); \
85 * Description of built-in state associated with a uniform
87 * \sa nir_variable::state_slots
90 gl_state_index16 tokens
[STATE_LENGTH
];
95 nir_var_shader_in
= (1 << 0),
96 nir_var_shader_out
= (1 << 1),
97 nir_var_global
= (1 << 2),
98 nir_var_local
= (1 << 3),
99 nir_var_uniform
= (1 << 4),
100 nir_var_shader_storage
= (1 << 5),
101 nir_var_system_value
= (1 << 6),
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 * Description of per-member data for per-member struct variables
376 * This is used for variables which are actually an amalgamation of
377 * multiple entities such as a struct of built-in values or a struct of
378 * inputs each with their own layout specifier. This is only allowed on
379 * variables with a struct or array of array of struct type.
381 unsigned num_members
;
382 struct nir_variable_data
*members
;
385 #define nir_foreach_variable(var, var_list) \
386 foreach_list_typed(nir_variable, var, node, var_list)
388 #define nir_foreach_variable_safe(var, var_list) \
389 foreach_list_typed_safe(nir_variable, var, node, var_list)
392 nir_variable_is_global(const nir_variable
*var
)
394 return var
->data
.mode
!= nir_var_local
;
397 typedef struct nir_register
{
398 struct exec_node node
;
400 unsigned num_components
; /** < number of vector components */
401 unsigned num_array_elems
; /** < size of array (0 for no array) */
403 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
406 /** generic register index. */
409 /** only for debug purposes, can be NULL */
412 /** whether this register is local (per-function) or global (per-shader) */
416 * If this flag is set to true, then accessing channels >= num_components
417 * is well-defined, and simply spills over to the next array element. This
418 * is useful for backends that can do per-component accessing, in
419 * particular scalar backends. By setting this flag and making
420 * num_components equal to 1, structures can be packed tightly into
421 * registers and then registers can be accessed per-component to get to
422 * each structure member, even if it crosses vec4 boundaries.
426 /** set of nir_srcs where this register is used (read from) */
427 struct list_head uses
;
429 /** set of nir_dests where this register is defined (written to) */
430 struct list_head defs
;
432 /** set of nir_ifs where this register is used as a condition */
433 struct list_head if_uses
;
436 #define nir_foreach_register(reg, reg_list) \
437 foreach_list_typed(nir_register, reg, node, reg_list)
438 #define nir_foreach_register_safe(reg, reg_list) \
439 foreach_list_typed_safe(nir_register, reg, node, reg_list)
443 nir_instr_type_deref
,
446 nir_instr_type_intrinsic
,
447 nir_instr_type_load_const
,
449 nir_instr_type_ssa_undef
,
451 nir_instr_type_parallel_copy
,
454 typedef struct nir_instr
{
455 struct exec_node node
;
457 struct nir_block
*block
;
459 /** generic instruction index. */
462 /* A temporary for optimization and analysis passes to use for storing
463 * flags. For instance, DCE uses this to store the "dead/live" info.
468 static inline nir_instr
*
469 nir_instr_next(nir_instr
*instr
)
471 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
472 if (exec_node_is_tail_sentinel(next
))
475 return exec_node_data(nir_instr
, next
, node
);
478 static inline nir_instr
*
479 nir_instr_prev(nir_instr
*instr
)
481 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
482 if (exec_node_is_head_sentinel(prev
))
485 return exec_node_data(nir_instr
, prev
, node
);
489 nir_instr_is_first(const nir_instr
*instr
)
491 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
495 nir_instr_is_last(const nir_instr
*instr
)
497 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
500 typedef struct nir_ssa_def
{
501 /** for debugging only, can be NULL */
504 /** generic SSA definition index. */
507 /** Index into the live_in and live_out bitfields */
510 /** Instruction which produces this SSA value. */
511 nir_instr
*parent_instr
;
513 /** set of nir_instrs where this register is used (read from) */
514 struct list_head uses
;
516 /** set of nir_ifs where this register is used as a condition */
517 struct list_head if_uses
;
519 uint8_t num_components
;
521 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
529 struct nir_src
*indirect
; /** < NULL for no indirect offset */
530 unsigned base_offset
;
532 /* TODO use-def chain goes here */
536 nir_instr
*parent_instr
;
537 struct list_head def_link
;
540 struct nir_src
*indirect
; /** < NULL for no indirect offset */
541 unsigned base_offset
;
543 /* TODO def-use chain goes here */
548 typedef struct nir_src
{
550 /** Instruction that consumes this value as a source. */
551 nir_instr
*parent_instr
;
552 struct nir_if
*parent_if
;
555 struct list_head use_link
;
565 static inline nir_src
568 nir_src src
= { { NULL
} };
572 #define NIR_SRC_INIT nir_src_init()
574 #define nir_foreach_use(src, reg_or_ssa_def) \
575 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
577 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
578 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
580 #define nir_foreach_if_use(src, reg_or_ssa_def) \
581 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
583 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
584 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
595 static inline nir_dest
598 nir_dest dest
= { { { NULL
} } };
602 #define NIR_DEST_INIT nir_dest_init()
604 #define nir_foreach_def(dest, reg) \
605 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
607 #define nir_foreach_def_safe(dest, reg) \
608 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
610 static inline nir_src
611 nir_src_for_ssa(nir_ssa_def
*def
)
613 nir_src src
= NIR_SRC_INIT
;
621 static inline nir_src
622 nir_src_for_reg(nir_register
*reg
)
624 nir_src src
= NIR_SRC_INIT
;
628 src
.reg
.indirect
= NULL
;
629 src
.reg
.base_offset
= 0;
634 static inline nir_dest
635 nir_dest_for_reg(nir_register
*reg
)
637 nir_dest dest
= NIR_DEST_INIT
;
644 static inline unsigned
645 nir_src_bit_size(nir_src src
)
647 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
650 static inline unsigned
651 nir_src_num_components(nir_src src
)
653 return src
.is_ssa
? src
.ssa
->num_components
: src
.reg
.reg
->num_components
;
656 static inline unsigned
657 nir_dest_bit_size(nir_dest dest
)
659 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
662 static inline unsigned
663 nir_dest_num_components(nir_dest dest
)
665 return dest
.is_ssa
? dest
.ssa
.num_components
: dest
.reg
.reg
->num_components
;
668 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
669 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
675 * \name input modifiers
679 * For inputs interpreted as floating point, flips the sign bit. For
680 * inputs interpreted as integers, performs the two's complement negation.
685 * Clears the sign bit for floating point values, and computes the integer
686 * absolute value for integers. Note that the negate modifier acts after
687 * the absolute value modifier, therefore if both are set then all inputs
688 * will become negative.
694 * For each input component, says which component of the register it is
695 * chosen from. Note that which elements of the swizzle are used and which
696 * are ignored are based on the write mask for most opcodes - for example,
697 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
698 * a swizzle of {2, x, 1, 0} where x means "don't care."
707 * \name saturate output modifier
709 * Only valid for opcodes that output floating-point numbers. Clamps the
710 * output to between 0.0 and 1.0 inclusive.
715 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
719 nir_type_invalid
= 0, /* Not a valid type */
724 nir_type_bool32
= 32 | nir_type_bool
,
725 nir_type_int8
= 8 | nir_type_int
,
726 nir_type_int16
= 16 | nir_type_int
,
727 nir_type_int32
= 32 | nir_type_int
,
728 nir_type_int64
= 64 | nir_type_int
,
729 nir_type_uint8
= 8 | nir_type_uint
,
730 nir_type_uint16
= 16 | nir_type_uint
,
731 nir_type_uint32
= 32 | nir_type_uint
,
732 nir_type_uint64
= 64 | nir_type_uint
,
733 nir_type_float16
= 16 | nir_type_float
,
734 nir_type_float32
= 32 | nir_type_float
,
735 nir_type_float64
= 64 | nir_type_float
,
738 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
739 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
741 static inline unsigned
742 nir_alu_type_get_type_size(nir_alu_type type
)
744 return type
& NIR_ALU_TYPE_SIZE_MASK
;
747 static inline unsigned
748 nir_alu_type_get_base_type(nir_alu_type type
)
750 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
753 static inline nir_alu_type
754 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
758 return nir_type_bool32
;
761 return nir_type_uint32
;
764 return nir_type_int32
;
766 case GLSL_TYPE_UINT16
:
767 return nir_type_uint16
;
769 case GLSL_TYPE_INT16
:
770 return nir_type_int16
;
772 case GLSL_TYPE_UINT8
:
773 return nir_type_uint8
;
775 return nir_type_int8
;
776 case GLSL_TYPE_UINT64
:
777 return nir_type_uint64
;
779 case GLSL_TYPE_INT64
:
780 return nir_type_int64
;
782 case GLSL_TYPE_FLOAT
:
783 return nir_type_float32
;
785 case GLSL_TYPE_FLOAT16
:
786 return nir_type_float16
;
788 case GLSL_TYPE_DOUBLE
:
789 return nir_type_float64
;
792 unreachable("unknown type");
796 static inline nir_alu_type
797 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
799 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
802 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
,
803 nir_rounding_mode rnd
);
806 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
807 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
808 } nir_op_algebraic_property
;
816 * The number of components in the output
818 * If non-zero, this is the size of the output and input sizes are
819 * explicitly given; swizzle and writemask are still in effect, but if
820 * the output component is masked out, then the input component may
823 * If zero, the opcode acts in the standard, per-component manner; the
824 * operation is performed on each component (except the ones that are
825 * masked out) with the input being taken from the input swizzle for
828 * The size of some of the inputs may be given (i.e. non-zero) even
829 * though output_size is zero; in that case, the inputs with a zero
830 * size act per-component, while the inputs with non-zero size don't.
832 unsigned output_size
;
835 * The type of vector that the instruction outputs. Note that the
836 * staurate modifier is only allowed on outputs with the float type.
839 nir_alu_type output_type
;
842 * The number of components in each input
844 unsigned input_sizes
[4];
847 * The type of vector that each input takes. Note that negate and
848 * absolute value are only allowed on inputs with int or float type and
849 * behave differently on the two.
851 nir_alu_type input_types
[4];
853 nir_op_algebraic_property algebraic_properties
;
856 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
858 typedef struct nir_alu_instr
{
862 /** Indicates that this ALU instruction generates an exact value
864 * This is kind of a mixture of GLSL "precise" and "invariant" and not
865 * really equivalent to either. This indicates that the value generated by
866 * this operation is high-precision and any code transformations that touch
867 * it must ensure that the resulting value is bit-for-bit identical to the
876 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
877 nir_alu_instr
*instr
);
878 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
879 nir_alu_instr
*instr
);
881 /* is this source channel used? */
883 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
886 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
887 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
889 return (instr
->dest
.write_mask
>> channel
) & 1;
893 * For instructions whose destinations are SSA, get the number of channels
896 static inline unsigned
897 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
899 assert(instr
->dest
.dest
.is_ssa
);
901 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
902 return nir_op_infos
[instr
->op
].input_sizes
[src
];
904 return instr
->dest
.dest
.ssa
.num_components
;
907 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
908 unsigned src1
, unsigned src2
);
912 nir_deref_type_array
,
913 nir_deref_type_array_wildcard
,
914 nir_deref_type_struct
,
921 /** The type of this deref instruction */
922 nir_deref_type deref_type
;
924 /** The mode of the underlying variable */
925 nir_variable_mode mode
;
927 /** The dereferenced type of the resulting pointer value */
928 const struct glsl_type
*type
;
931 /** Variable being dereferenced if deref_type is a deref_var */
934 /** Parent deref if deref_type is not deref_var */
938 /** Additional deref parameters */
949 /** Destination to store the resulting "pointer" */
953 NIR_DEFINE_CAST(nir_instr_as_deref
, nir_instr
, nir_deref_instr
, instr
,
954 type
, nir_instr_type_deref
)
956 static inline nir_deref_instr
*
957 nir_src_as_deref(nir_src src
)
962 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_deref
)
965 return nir_instr_as_deref(src
.ssa
->parent_instr
);
968 static inline nir_deref_instr
*
969 nir_deref_instr_parent(const nir_deref_instr
*instr
)
971 if (instr
->deref_type
== nir_deref_type_var
)
974 return nir_src_as_deref(instr
->parent
);
977 static inline nir_variable
*
978 nir_deref_instr_get_variable(const nir_deref_instr
*instr
)
980 while (instr
->deref_type
!= nir_deref_type_var
) {
981 if (instr
->deref_type
== nir_deref_type_cast
)
984 instr
= nir_deref_instr_parent(instr
);
990 bool nir_deref_instr_remove_if_unused(nir_deref_instr
*instr
);
995 struct nir_function
*callee
;
1001 #include "nir_intrinsics.h"
1003 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
1005 /** Represents an intrinsic
1007 * An intrinsic is an instruction type for handling things that are
1008 * more-or-less regular operations but don't just consume and produce SSA
1009 * values like ALU operations do. Intrinsics are not for things that have
1010 * special semantic meaning such as phi nodes and parallel copies.
1011 * Examples of intrinsics include variable load/store operations, system
1012 * value loads, and the like. Even though texturing more-or-less falls
1013 * under this category, texturing is its own instruction type because
1014 * trying to represent texturing with intrinsics would lead to a
1015 * combinatorial explosion of intrinsic opcodes.
1017 * By having a single instruction type for handling a lot of different
1018 * cases, optimization passes can look for intrinsics and, for the most
1019 * part, completely ignore them. Each intrinsic type also has a few
1020 * possible flags that govern whether or not they can be reordered or
1021 * eliminated. That way passes like dead code elimination can still work
1022 * on intrisics without understanding the meaning of each.
1024 * Each intrinsic has some number of constant indices, some number of
1025 * variables, and some number of sources. What these sources, variables,
1026 * and indices mean depends on the intrinsic and is documented with the
1027 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
1028 * instructions are the only types of instruction that can operate on
1034 nir_intrinsic_op intrinsic
;
1038 /** number of components if this is a vectorized intrinsic
1040 * Similarly to ALU operations, some intrinsics are vectorized.
1041 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
1042 * For vectorized intrinsics, the num_components field specifies the
1043 * number of destination components and the number of source components
1044 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
1046 uint8_t num_components
;
1048 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
1051 } nir_intrinsic_instr
;
1053 static inline nir_variable
*
1054 nir_intrinsic_get_var(nir_intrinsic_instr
*intrin
, unsigned i
)
1056 return nir_deref_instr_get_variable(nir_src_as_deref(intrin
->src
[i
]));
1060 * \name NIR intrinsics semantic flags
1062 * information about what the compiler can do with the intrinsics.
1064 * \sa nir_intrinsic_info::flags
1068 * whether the intrinsic can be safely eliminated if none of its output
1069 * value is not being used.
1071 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
1074 * Whether the intrinsic can be reordered with respect to any other
1075 * intrinsic, i.e. whether the only reordering dependencies of the
1076 * intrinsic are due to the register reads/writes.
1078 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
1079 } nir_intrinsic_semantic_flag
;
1082 * \name NIR intrinsics const-index flag
1084 * Indicates the usage of a const_index slot.
1086 * \sa nir_intrinsic_info::index_map
1090 * Generally instructions that take a offset src argument, can encode
1091 * a constant 'base' value which is added to the offset.
1093 NIR_INTRINSIC_BASE
= 1,
1096 * For store instructions, a writemask for the store.
1098 NIR_INTRINSIC_WRMASK
= 2,
1101 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1103 NIR_INTRINSIC_STREAM_ID
= 3,
1106 * The clip-plane id for load_user_clip_plane intrinsic.
1108 NIR_INTRINSIC_UCP_ID
= 4,
1111 * The amount of data, starting from BASE, that this instruction may
1112 * access. This is used to provide bounds if the offset is not constant.
1114 NIR_INTRINSIC_RANGE
= 5,
1117 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1119 NIR_INTRINSIC_DESC_SET
= 6,
1122 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1124 NIR_INTRINSIC_BINDING
= 7,
1129 NIR_INTRINSIC_COMPONENT
= 8,
1132 * Interpolation mode (only meaningful for FS inputs).
1134 NIR_INTRINSIC_INTERP_MODE
= 9,
1137 * A binary nir_op to use when performing a reduction or scan operation
1139 NIR_INTRINSIC_REDUCTION_OP
= 10,
1142 * Cluster size for reduction operations
1144 NIR_INTRINSIC_CLUSTER_SIZE
= 11,
1147 * Parameter index for a load_param intrinsic
1149 NIR_INTRINSIC_PARAM_IDX
= 12,
1151 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1153 } nir_intrinsic_index_flag
;
1155 #define NIR_INTRINSIC_MAX_INPUTS 5
1160 unsigned num_srcs
; /** < number of register/SSA inputs */
1162 /** number of components of each input register
1164 * If this value is 0, the number of components is given by the
1165 * num_components field of nir_intrinsic_instr.
1167 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1171 /** number of components of the output register
1173 * If this value is 0, the number of components is given by the
1174 * num_components field of nir_intrinsic_instr.
1176 unsigned dest_components
;
1178 /** the number of constant indices used by the intrinsic */
1179 unsigned num_indices
;
1181 /** indicates the usage of intr->const_index[n] */
1182 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1184 /** semantic flags for calls to this intrinsic */
1185 nir_intrinsic_semantic_flag flags
;
1186 } nir_intrinsic_info
;
1188 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1190 static inline unsigned
1191 nir_intrinsic_src_components(nir_intrinsic_instr
*intr
, unsigned srcn
)
1193 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1194 assert(srcn
< info
->num_srcs
);
1195 if (info
->src_components
[srcn
])
1196 return info
->src_components
[srcn
];
1198 return intr
->num_components
;
1201 static inline unsigned
1202 nir_intrinsic_dest_components(nir_intrinsic_instr
*intr
)
1204 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1205 if (!info
->has_dest
)
1207 else if (info
->dest_components
)
1208 return info
->dest_components
;
1210 return intr
->num_components
;
1213 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1214 static inline type \
1215 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1217 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1218 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1219 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1221 static inline void \
1222 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1224 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1225 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1226 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1229 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1230 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1231 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1232 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1233 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1234 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1235 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1236 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1237 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1238 INTRINSIC_IDX_ACCESSORS(reduction_op
, REDUCTION_OP
, unsigned)
1239 INTRINSIC_IDX_ACCESSORS(cluster_size
, CLUSTER_SIZE
, unsigned)
1240 INTRINSIC_IDX_ACCESSORS(param_idx
, PARAM_IDX
, unsigned)
1243 * \group texture information
1245 * This gives semantic information about textures which is useful to the
1246 * frontend, the backend, and lowering passes, but not the optimizer.
1251 nir_tex_src_projector
,
1252 nir_tex_src_comparator
, /* shadow comparator */
1256 nir_tex_src_ms_index
, /* MSAA sample index */
1257 nir_tex_src_ms_mcs
, /* MSAA compression value */
1260 nir_tex_src_texture_deref
, /* < deref pointing to the texture */
1261 nir_tex_src_sampler_deref
, /* < deref pointing to the sampler */
1262 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1263 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1264 nir_tex_src_plane
, /* < selects plane for planar textures */
1265 nir_num_tex_src_types
1270 nir_tex_src_type src_type
;
1274 nir_texop_tex
, /**< Regular texture look-up */
1275 nir_texop_txb
, /**< Texture look-up with LOD bias */
1276 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1277 nir_texop_txd
, /**< Texture look-up with partial derivatives */
1278 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1279 nir_texop_txf_ms
, /**< Multisample texture fetch */
1280 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1281 nir_texop_txs
, /**< Texture size */
1282 nir_texop_lod
, /**< Texture lod query */
1283 nir_texop_tg4
, /**< Texture gather */
1284 nir_texop_query_levels
, /**< Texture levels query */
1285 nir_texop_texture_samples
, /**< Texture samples query */
1286 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1294 enum glsl_sampler_dim sampler_dim
;
1295 nir_alu_type dest_type
;
1300 unsigned num_srcs
, coord_components
;
1301 bool is_array
, is_shadow
;
1304 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1305 * components or the new-style shadow that outputs 1 component.
1307 bool is_new_style_shadow
;
1309 /* gather component selector */
1310 unsigned component
: 2;
1312 /** The texture index
1314 * If this texture instruction has a nir_tex_src_texture_offset source,
1315 * then the texture index is given by texture_index + texture_offset.
1317 unsigned texture_index
;
1319 /** The size of the texture array or 0 if it's not an array */
1320 unsigned texture_array_size
;
1322 /** The sampler index
1324 * The following operations do not require a sampler and, as such, this
1325 * field should be ignored:
1327 * - nir_texop_txf_ms
1330 * - nir_texop_query_levels
1331 * - nir_texop_texture_samples
1332 * - nir_texop_samples_identical
1334 * If this texture instruction has a nir_tex_src_sampler_offset source,
1335 * then the sampler index is given by sampler_index + sampler_offset.
1337 unsigned sampler_index
;
1340 static inline unsigned
1341 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1343 switch (instr
->op
) {
1344 case nir_texop_txs
: {
1346 switch (instr
->sampler_dim
) {
1347 case GLSL_SAMPLER_DIM_1D
:
1348 case GLSL_SAMPLER_DIM_BUF
:
1351 case GLSL_SAMPLER_DIM_2D
:
1352 case GLSL_SAMPLER_DIM_CUBE
:
1353 case GLSL_SAMPLER_DIM_MS
:
1354 case GLSL_SAMPLER_DIM_RECT
:
1355 case GLSL_SAMPLER_DIM_EXTERNAL
:
1356 case GLSL_SAMPLER_DIM_SUBPASS
:
1359 case GLSL_SAMPLER_DIM_3D
:
1363 unreachable("not reached");
1365 if (instr
->is_array
)
1373 case nir_texop_texture_samples
:
1374 case nir_texop_query_levels
:
1375 case nir_texop_samples_identical
:
1379 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1386 /* Returns true if this texture operation queries something about the texture
1387 * rather than actually sampling it.
1390 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1392 switch (instr
->op
) {
1395 case nir_texop_texture_samples
:
1396 case nir_texop_query_levels
:
1397 case nir_texop_txf_ms_mcs
:
1404 case nir_texop_txf_ms
:
1408 unreachable("Invalid texture opcode");
1413 nir_alu_instr_is_comparison(const nir_alu_instr
*instr
)
1415 switch (instr
->op
) {
1436 static inline nir_alu_type
1437 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1439 switch (instr
->src
[src
].src_type
) {
1440 case nir_tex_src_coord
:
1441 switch (instr
->op
) {
1443 case nir_texop_txf_ms
:
1444 case nir_texop_txf_ms_mcs
:
1445 case nir_texop_samples_identical
:
1446 return nir_type_int
;
1449 return nir_type_float
;
1452 case nir_tex_src_lod
:
1453 switch (instr
->op
) {
1456 return nir_type_int
;
1459 return nir_type_float
;
1462 case nir_tex_src_projector
:
1463 case nir_tex_src_comparator
:
1464 case nir_tex_src_bias
:
1465 case nir_tex_src_ddx
:
1466 case nir_tex_src_ddy
:
1467 return nir_type_float
;
1469 case nir_tex_src_offset
:
1470 case nir_tex_src_ms_index
:
1471 case nir_tex_src_texture_offset
:
1472 case nir_tex_src_sampler_offset
:
1473 return nir_type_int
;
1476 unreachable("Invalid texture source type");
1480 static inline unsigned
1481 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1483 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1484 return instr
->coord_components
;
1486 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1487 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1490 if (instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1491 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1492 if (instr
->is_array
)
1493 return instr
->coord_components
- 1;
1495 return instr
->coord_components
;
1498 /* Usual APIs don't allow cube + offset, but we allow it, with 2 coords for
1499 * the offset, since a cube maps to a single face.
1501 if (instr
->src
[src
].src_type
== nir_tex_src_offset
) {
1502 if (instr
->sampler_dim
== GLSL_SAMPLER_DIM_CUBE
)
1504 else if (instr
->is_array
)
1505 return instr
->coord_components
- 1;
1507 return instr
->coord_components
;
1514 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1516 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1517 if (instr
->src
[i
].src_type
== type
)
1523 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1524 nir_tex_src_type src_type
,
1527 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1532 nir_const_value value
;
1535 } nir_load_const_instr
;
1548 /* creates a new SSA variable in an undefined state */
1553 } nir_ssa_undef_instr
;
1556 struct exec_node node
;
1558 /* The predecessor block corresponding to this source */
1559 struct nir_block
*pred
;
1564 #define nir_foreach_phi_src(phi_src, phi) \
1565 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1566 #define nir_foreach_phi_src_safe(phi_src, phi) \
1567 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1572 struct exec_list srcs
; /** < list of nir_phi_src */
1578 struct exec_node node
;
1581 } nir_parallel_copy_entry
;
1583 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1584 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1589 /* A list of nir_parallel_copy_entrys. The sources of all of the
1590 * entries are copied to the corresponding destinations "in parallel".
1591 * In other words, if we have two entries: a -> b and b -> a, the values
1594 struct exec_list entries
;
1595 } nir_parallel_copy_instr
;
1597 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1598 type
, nir_instr_type_alu
)
1599 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1600 type
, nir_instr_type_call
)
1601 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1602 type
, nir_instr_type_jump
)
1603 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1604 type
, nir_instr_type_tex
)
1605 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1606 type
, nir_instr_type_intrinsic
)
1607 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1608 type
, nir_instr_type_load_const
)
1609 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1610 type
, nir_instr_type_ssa_undef
)
1611 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1612 type
, nir_instr_type_phi
)
1613 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1614 nir_parallel_copy_instr
, instr
,
1615 type
, nir_instr_type_parallel_copy
)
1620 * Control flow consists of a tree of control flow nodes, which include
1621 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1622 * instructions that always run start-to-finish. Each basic block also keeps
1623 * track of its successors (blocks which may run immediately after the current
1624 * block) and predecessors (blocks which could have run immediately before the
1625 * current block). Each function also has a start block and an end block which
1626 * all return statements point to (which is always empty). Together, all the
1627 * blocks with their predecessors and successors make up the control flow
1628 * graph (CFG) of the function. There are helpers that modify the tree of
1629 * control flow nodes while modifying the CFG appropriately; these should be
1630 * used instead of modifying the tree directly.
1637 nir_cf_node_function
1640 typedef struct nir_cf_node
{
1641 struct exec_node node
;
1642 nir_cf_node_type type
;
1643 struct nir_cf_node
*parent
;
1646 typedef struct nir_block
{
1647 nir_cf_node cf_node
;
1649 struct exec_list instr_list
; /** < list of nir_instr */
1651 /** generic block index; generated by nir_index_blocks */
1655 * Each block can only have up to 2 successors, so we put them in a simple
1656 * array - no need for anything more complicated.
1658 struct nir_block
*successors
[2];
1660 /* Set of nir_block predecessors in the CFG */
1661 struct set
*predecessors
;
1664 * this node's immediate dominator in the dominance tree - set to NULL for
1667 struct nir_block
*imm_dom
;
1669 /* This node's children in the dominance tree */
1670 unsigned num_dom_children
;
1671 struct nir_block
**dom_children
;
1673 /* Set of nir_blocks on the dominance frontier of this block */
1674 struct set
*dom_frontier
;
1677 * These two indices have the property that dom_{pre,post}_index for each
1678 * child of this block in the dominance tree will always be between
1679 * dom_pre_index and dom_post_index for this block, which makes testing if
1680 * a given block is dominated by another block an O(1) operation.
1682 unsigned dom_pre_index
, dom_post_index
;
1684 /* live in and out for this block; used for liveness analysis */
1685 BITSET_WORD
*live_in
;
1686 BITSET_WORD
*live_out
;
1689 static inline nir_instr
*
1690 nir_block_first_instr(nir_block
*block
)
1692 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1693 return exec_node_data(nir_instr
, head
, node
);
1696 static inline nir_instr
*
1697 nir_block_last_instr(nir_block
*block
)
1699 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1700 return exec_node_data(nir_instr
, tail
, node
);
1703 #define nir_foreach_instr(instr, block) \
1704 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1705 #define nir_foreach_instr_reverse(instr, block) \
1706 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1707 #define nir_foreach_instr_safe(instr, block) \
1708 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1709 #define nir_foreach_instr_reverse_safe(instr, block) \
1710 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1712 typedef struct nir_if
{
1713 nir_cf_node cf_node
;
1716 struct exec_list then_list
; /** < list of nir_cf_node */
1717 struct exec_list else_list
; /** < list of nir_cf_node */
1723 nir_instr
*conditional_instr
;
1725 nir_block
*break_block
;
1726 nir_block
*continue_from_block
;
1728 bool continue_from_then
;
1730 struct list_head loop_terminator_link
;
1731 } nir_loop_terminator
;
1734 /* Number of instructions in the loop */
1735 unsigned num_instructions
;
1737 /* How many times the loop is run (if known) */
1738 unsigned trip_count
;
1739 bool is_trip_count_known
;
1741 /* Unroll the loop regardless of its size */
1744 nir_loop_terminator
*limiting_terminator
;
1746 /* A list of loop_terminators terminating this loop. */
1747 struct list_head loop_terminator_list
;
1751 nir_cf_node cf_node
;
1753 struct exec_list body
; /** < list of nir_cf_node */
1755 nir_loop_info
*info
;
1759 * Various bits of metadata that can may be created or required by
1760 * optimization and analysis passes
1763 nir_metadata_none
= 0x0,
1764 nir_metadata_block_index
= 0x1,
1765 nir_metadata_dominance
= 0x2,
1766 nir_metadata_live_ssa_defs
= 0x4,
1767 nir_metadata_not_properly_reset
= 0x8,
1768 nir_metadata_loop_analysis
= 0x10,
1772 nir_cf_node cf_node
;
1774 /** pointer to the function of which this is an implementation */
1775 struct nir_function
*function
;
1777 struct exec_list body
; /** < list of nir_cf_node */
1779 nir_block
*end_block
;
1781 /** list for all local variables in the function */
1782 struct exec_list locals
;
1784 /** list of local registers in the function */
1785 struct exec_list registers
;
1787 /** next available local register index */
1790 /** next available SSA value index */
1793 /* total number of basic blocks, only valid when block_index_dirty = false */
1794 unsigned num_blocks
;
1796 nir_metadata valid_metadata
;
1797 } nir_function_impl
;
1799 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1800 nir_start_block(nir_function_impl
*impl
)
1802 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1805 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1806 nir_impl_last_block(nir_function_impl
*impl
)
1808 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1811 static inline nir_cf_node
*
1812 nir_cf_node_next(nir_cf_node
*node
)
1814 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1815 if (exec_node_is_tail_sentinel(next
))
1818 return exec_node_data(nir_cf_node
, next
, node
);
1821 static inline nir_cf_node
*
1822 nir_cf_node_prev(nir_cf_node
*node
)
1824 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1825 if (exec_node_is_head_sentinel(prev
))
1828 return exec_node_data(nir_cf_node
, prev
, node
);
1832 nir_cf_node_is_first(const nir_cf_node
*node
)
1834 return exec_node_is_head_sentinel(node
->node
.prev
);
1838 nir_cf_node_is_last(const nir_cf_node
*node
)
1840 return exec_node_is_tail_sentinel(node
->node
.next
);
1843 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1844 type
, nir_cf_node_block
)
1845 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1846 type
, nir_cf_node_if
)
1847 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1848 type
, nir_cf_node_loop
)
1849 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1850 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1852 static inline nir_block
*
1853 nir_if_first_then_block(nir_if
*if_stmt
)
1855 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1856 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1859 static inline nir_block
*
1860 nir_if_last_then_block(nir_if
*if_stmt
)
1862 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1863 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1866 static inline nir_block
*
1867 nir_if_first_else_block(nir_if
*if_stmt
)
1869 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1870 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1873 static inline nir_block
*
1874 nir_if_last_else_block(nir_if
*if_stmt
)
1876 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1877 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1880 static inline nir_block
*
1881 nir_loop_first_block(nir_loop
*loop
)
1883 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1884 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1887 static inline nir_block
*
1888 nir_loop_last_block(nir_loop
*loop
)
1890 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1891 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1895 uint8_t num_components
;
1899 typedef struct nir_function
{
1900 struct exec_node node
;
1903 struct nir_shader
*shader
;
1905 unsigned num_params
;
1906 nir_parameter
*params
;
1908 /** The implementation of this function.
1910 * If the function is only declared and not implemented, this is NULL.
1912 nir_function_impl
*impl
;
1915 typedef struct nir_shader_compiler_options
{
1920 /** Lowers flrp when it does not support doubles */
1927 /** Lowers ibitfield_extract/ubitfield_extract to ibfe/ubfe. */
1928 bool lower_bitfield_extract
;
1929 /** Lowers ibitfield_extract/ubitfield_extract to bfm, compares, shifts. */
1930 bool lower_bitfield_extract_to_shifts
;
1931 /** Lowers bitfield_insert to bfi/bfm */
1932 bool lower_bitfield_insert
;
1933 /** Lowers bitfield_insert to bfm, compares, and shifts. */
1934 bool lower_bitfield_insert_to_shifts
;
1935 /** Lowers bitfield_reverse to shifts. */
1936 bool lower_bitfield_reverse
;
1937 /** Lowers bit_count to shifts. */
1938 bool lower_bit_count
;
1939 /** Lowers bfm to shifts and subtracts. */
1941 /** Lowers ifind_msb to compare and ufind_msb */
1942 bool lower_ifind_msb
;
1943 /** Lowers find_lsb to ufind_msb and logic ops */
1944 bool lower_find_lsb
;
1945 bool lower_uadd_carry
;
1946 bool lower_usub_borrow
;
1947 /** Lowers imul_high/umul_high to 16-bit multiplies and carry operations. */
1948 bool lower_mul_high
;
1949 /** lowers fneg and ineg to fsub and isub. */
1951 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1954 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1957 /** enables rules to lower idiv by power-of-two: */
1960 /* lower b2f to iand */
1963 /* Does the native fdot instruction replicate its result for four
1964 * components? If so, then opt_algebraic_late will turn all fdotN
1965 * instructions into fdot_replicatedN instructions.
1967 bool fdot_replicates
;
1969 /** lowers ffract to fsub+ffloor: */
1974 bool lower_pack_half_2x16
;
1975 bool lower_pack_unorm_2x16
;
1976 bool lower_pack_snorm_2x16
;
1977 bool lower_pack_unorm_4x8
;
1978 bool lower_pack_snorm_4x8
;
1979 bool lower_unpack_half_2x16
;
1980 bool lower_unpack_unorm_2x16
;
1981 bool lower_unpack_snorm_2x16
;
1982 bool lower_unpack_unorm_4x8
;
1983 bool lower_unpack_snorm_4x8
;
1985 bool lower_extract_byte
;
1986 bool lower_extract_word
;
1988 bool lower_all_io_to_temps
;
1991 * Does the driver support real 32-bit integers? (Otherwise, integers
1992 * are simulated by floats.)
1994 bool native_integers
;
1996 /* Indicates that the driver only has zero-based vertex id */
1997 bool vertex_id_zero_based
;
2000 * If enabled, gl_BaseVertex will be lowered as:
2001 * is_indexed_draw (~0/0) & firstvertex
2003 bool lower_base_vertex
;
2005 bool lower_cs_local_index_from_id
;
2007 bool lower_device_index_to_zero
;
2010 * Should nir_lower_io() create load_interpolated_input intrinsics?
2012 * If not, it generates regular load_input intrinsics and interpolation
2013 * information must be inferred from the list of input nir_variables.
2015 bool use_interpolated_input_intrinsics
;
2018 * Do vertex shader double inputs use two locations? The Vulkan spec
2019 * requires two locations to be used, OpenGL allows a single location.
2021 bool vs_inputs_dual_locations
;
2023 unsigned max_unroll_iterations
;
2024 } nir_shader_compiler_options
;
2026 typedef struct nir_shader
{
2027 /** list of uniforms (nir_variable) */
2028 struct exec_list uniforms
;
2030 /** list of inputs (nir_variable) */
2031 struct exec_list inputs
;
2033 /** list of outputs (nir_variable) */
2034 struct exec_list outputs
;
2036 /** list of shared compute variables (nir_variable) */
2037 struct exec_list shared
;
2039 /** Set of driver-specific options for the shader.
2041 * The memory for the options is expected to be kept in a single static
2042 * copy by the driver.
2044 const struct nir_shader_compiler_options
*options
;
2046 /** Various bits of compile-time information about a given shader */
2047 struct shader_info info
;
2049 /** list of global variables in the shader (nir_variable) */
2050 struct exec_list globals
;
2052 /** list of system value variables in the shader (nir_variable) */
2053 struct exec_list system_values
;
2055 struct exec_list functions
; /** < list of nir_function */
2057 /** list of global register in the shader */
2058 struct exec_list registers
;
2060 /** next available global register index */
2064 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
2067 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
2070 static inline nir_function_impl
*
2071 nir_shader_get_entrypoint(nir_shader
*shader
)
2073 assert(exec_list_length(&shader
->functions
) == 1);
2074 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
2075 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
2076 assert(func
->num_params
== 0);
2081 #define nir_foreach_function(func, shader) \
2082 foreach_list_typed(nir_function, func, node, &(shader)->functions)
2084 nir_shader
*nir_shader_create(void *mem_ctx
,
2085 gl_shader_stage stage
,
2086 const nir_shader_compiler_options
*options
,
2089 /** creates a register, including assigning it an index and adding it to the list */
2090 nir_register
*nir_global_reg_create(nir_shader
*shader
);
2092 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
2094 void nir_reg_remove(nir_register
*reg
);
2096 /** Adds a variable to the appropriate list in nir_shader */
2097 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
2100 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
2102 assert(var
->data
.mode
== nir_var_local
);
2103 exec_list_push_tail(&impl
->locals
, &var
->node
);
2106 /** creates a variable, sets a few defaults, and adds it to the list */
2107 nir_variable
*nir_variable_create(nir_shader
*shader
,
2108 nir_variable_mode mode
,
2109 const struct glsl_type
*type
,
2111 /** creates a local variable and adds it to the list */
2112 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
2113 const struct glsl_type
*type
,
2116 /** creates a function and adds it to the shader's list of functions */
2117 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
2119 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
2120 /** creates a function_impl that isn't tied to any particular function */
2121 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
2123 nir_block
*nir_block_create(nir_shader
*shader
);
2124 nir_if
*nir_if_create(nir_shader
*shader
);
2125 nir_loop
*nir_loop_create(nir_shader
*shader
);
2127 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
2129 /** requests that the given pieces of metadata be generated */
2130 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
2131 /** dirties all but the preserved metadata */
2132 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
2134 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
2135 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
2137 nir_deref_instr
*nir_deref_instr_create(nir_shader
*shader
,
2138 nir_deref_type deref_type
);
2140 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
2142 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
2143 unsigned num_components
,
2146 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
2147 nir_intrinsic_op op
);
2149 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
2150 nir_function
*callee
);
2152 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
2154 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
2156 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
2158 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
2159 unsigned num_components
,
2162 nir_const_value
nir_alu_binop_identity(nir_op binop
, unsigned bit_size
);
2165 * NIR Cursors and Instruction Insertion API
2168 * A tiny struct representing a point to insert/extract instructions or
2169 * control flow nodes. Helps reduce the combinatorial explosion of possible
2170 * points to insert/extract.
2172 * \sa nir_control_flow.h
2175 nir_cursor_before_block
,
2176 nir_cursor_after_block
,
2177 nir_cursor_before_instr
,
2178 nir_cursor_after_instr
,
2179 } nir_cursor_option
;
2182 nir_cursor_option option
;
2189 static inline nir_block
*
2190 nir_cursor_current_block(nir_cursor cursor
)
2192 if (cursor
.option
== nir_cursor_before_instr
||
2193 cursor
.option
== nir_cursor_after_instr
) {
2194 return cursor
.instr
->block
;
2196 return cursor
.block
;
2200 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2202 static inline nir_cursor
2203 nir_before_block(nir_block
*block
)
2206 cursor
.option
= nir_cursor_before_block
;
2207 cursor
.block
= block
;
2211 static inline nir_cursor
2212 nir_after_block(nir_block
*block
)
2215 cursor
.option
= nir_cursor_after_block
;
2216 cursor
.block
= block
;
2220 static inline nir_cursor
2221 nir_before_instr(nir_instr
*instr
)
2224 cursor
.option
= nir_cursor_before_instr
;
2225 cursor
.instr
= instr
;
2229 static inline nir_cursor
2230 nir_after_instr(nir_instr
*instr
)
2233 cursor
.option
= nir_cursor_after_instr
;
2234 cursor
.instr
= instr
;
2238 static inline nir_cursor
2239 nir_after_block_before_jump(nir_block
*block
)
2241 nir_instr
*last_instr
= nir_block_last_instr(block
);
2242 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2243 return nir_before_instr(last_instr
);
2245 return nir_after_block(block
);
2249 static inline nir_cursor
2250 nir_before_cf_node(nir_cf_node
*node
)
2252 if (node
->type
== nir_cf_node_block
)
2253 return nir_before_block(nir_cf_node_as_block(node
));
2255 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2258 static inline nir_cursor
2259 nir_after_cf_node(nir_cf_node
*node
)
2261 if (node
->type
== nir_cf_node_block
)
2262 return nir_after_block(nir_cf_node_as_block(node
));
2264 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2267 static inline nir_cursor
2268 nir_after_phis(nir_block
*block
)
2270 nir_foreach_instr(instr
, block
) {
2271 if (instr
->type
!= nir_instr_type_phi
)
2272 return nir_before_instr(instr
);
2274 return nir_after_block(block
);
2277 static inline nir_cursor
2278 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2280 if (node
->type
== nir_cf_node_block
)
2281 return nir_after_block(nir_cf_node_as_block(node
));
2283 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2285 return nir_after_phis(block
);
2288 static inline nir_cursor
2289 nir_before_cf_list(struct exec_list
*cf_list
)
2291 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2292 exec_list_get_head(cf_list
), node
);
2293 return nir_before_cf_node(first_node
);
2296 static inline nir_cursor
2297 nir_after_cf_list(struct exec_list
*cf_list
)
2299 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2300 exec_list_get_tail(cf_list
), node
);
2301 return nir_after_cf_node(last_node
);
2305 * Insert a NIR instruction at the given cursor.
2307 * Note: This does not update the cursor.
2309 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2312 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2314 nir_instr_insert(nir_before_instr(instr
), before
);
2318 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2320 nir_instr_insert(nir_after_instr(instr
), after
);
2324 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2326 nir_instr_insert(nir_before_block(block
), before
);
2330 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2332 nir_instr_insert(nir_after_block(block
), after
);
2336 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2338 nir_instr_insert(nir_before_cf_node(node
), before
);
2342 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2344 nir_instr_insert(nir_after_cf_node(node
), after
);
2348 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2350 nir_instr_insert(nir_before_cf_list(list
), before
);
2354 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2356 nir_instr_insert(nir_after_cf_list(list
), after
);
2359 void nir_instr_remove_v(nir_instr
*instr
);
2361 static inline nir_cursor
2362 nir_instr_remove(nir_instr
*instr
)
2365 nir_instr
*prev
= nir_instr_prev(instr
);
2367 cursor
= nir_after_instr(prev
);
2369 cursor
= nir_before_block(instr
->block
);
2371 nir_instr_remove_v(instr
);
2377 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2378 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2379 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2380 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2382 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2383 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2385 nir_const_value
*nir_src_as_const_value(nir_src src
);
2386 bool nir_src_is_dynamically_uniform(nir_src src
);
2387 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2388 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2389 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2390 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2391 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2394 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2395 unsigned num_components
, unsigned bit_size
,
2397 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2398 unsigned num_components
, unsigned bit_size
,
2401 nir_ssa_dest_init_for_type(nir_instr
*instr
, nir_dest
*dest
,
2402 const struct glsl_type
*type
,
2405 assert(glsl_type_is_vector_or_scalar(type
));
2406 nir_ssa_dest_init(instr
, dest
, glsl_get_components(type
),
2407 glsl_get_bit_size(type
), name
);
2409 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2410 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2411 nir_instr
*after_me
);
2413 uint8_t nir_ssa_def_components_read(const nir_ssa_def
*def
);
2416 * finds the next basic block in source-code order, returns NULL if there is
2420 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2422 /* Performs the opposite of nir_block_cf_tree_next() */
2424 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2426 /* Gets the first block in a CF node in source-code order */
2428 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2430 /* Gets the last block in a CF node in source-code order */
2432 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2434 /* Gets the next block after a CF node in source-code order */
2436 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2438 /* Macros for loops that visit blocks in source-code order */
2440 #define nir_foreach_block(block, impl) \
2441 for (nir_block *block = nir_start_block(impl); block != NULL; \
2442 block = nir_block_cf_tree_next(block))
2444 #define nir_foreach_block_safe(block, impl) \
2445 for (nir_block *block = nir_start_block(impl), \
2446 *next = nir_block_cf_tree_next(block); \
2448 block = next, next = nir_block_cf_tree_next(block))
2450 #define nir_foreach_block_reverse(block, impl) \
2451 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2452 block = nir_block_cf_tree_prev(block))
2454 #define nir_foreach_block_reverse_safe(block, impl) \
2455 for (nir_block *block = nir_impl_last_block(impl), \
2456 *prev = nir_block_cf_tree_prev(block); \
2458 block = prev, prev = nir_block_cf_tree_prev(block))
2460 #define nir_foreach_block_in_cf_node(block, node) \
2461 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2462 block != nir_cf_node_cf_tree_next(node); \
2463 block = nir_block_cf_tree_next(block))
2465 /* If the following CF node is an if, this function returns that if.
2466 * Otherwise, it returns NULL.
2468 nir_if
*nir_block_get_following_if(nir_block
*block
);
2470 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2472 void nir_index_local_regs(nir_function_impl
*impl
);
2473 void nir_index_global_regs(nir_shader
*shader
);
2474 void nir_index_ssa_defs(nir_function_impl
*impl
);
2475 unsigned nir_index_instrs(nir_function_impl
*impl
);
2477 void nir_index_blocks(nir_function_impl
*impl
);
2479 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2480 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2481 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2483 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2484 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2485 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2486 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2488 nir_shader
*nir_shader_serialize_deserialize(void *mem_ctx
, nir_shader
*s
);
2491 void nir_validate_shader(nir_shader
*shader
);
2492 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2493 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2496 should_clone_nir(void)
2498 static int should_clone
= -1;
2499 if (should_clone
< 0)
2500 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2502 return should_clone
;
2506 should_serialize_deserialize_nir(void)
2508 static int test_serialize
= -1;
2509 if (test_serialize
< 0)
2510 test_serialize
= env_var_as_boolean("NIR_TEST_SERIALIZE", false);
2512 return test_serialize
;
2516 should_print_nir(void)
2518 static int should_print
= -1;
2519 if (should_print
< 0)
2520 should_print
= env_var_as_boolean("NIR_PRINT", false);
2522 return should_print
;
2525 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2526 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2527 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2528 static inline bool should_clone_nir(void) { return false; }
2529 static inline bool should_serialize_deserialize_nir(void) { return false; }
2530 static inline bool should_print_nir(void) { return false; }
2533 #define _PASS(nir, do_pass) do { \
2535 nir_validate_shader(nir); \
2536 if (should_clone_nir()) { \
2537 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2541 if (should_serialize_deserialize_nir()) { \
2542 void *mem_ctx = ralloc_parent(nir); \
2543 nir = nir_shader_serialize_deserialize(mem_ctx, nir); \
2547 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2548 nir_metadata_set_validation_flag(nir); \
2549 if (should_print_nir()) \
2550 printf("%s\n", #pass); \
2551 if (pass(nir, ##__VA_ARGS__)) { \
2553 if (should_print_nir()) \
2554 nir_print_shader(nir, stdout); \
2555 nir_metadata_check_validation_flag(nir); \
2559 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2560 if (should_print_nir()) \
2561 printf("%s\n", #pass); \
2562 pass(nir, ##__VA_ARGS__); \
2563 if (should_print_nir()) \
2564 nir_print_shader(nir, stdout); \
2567 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2568 void nir_calc_dominance(nir_shader
*shader
);
2570 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2571 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2573 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2574 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2576 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2577 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2579 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2580 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2582 int nir_gs_count_vertices(const nir_shader
*shader
);
2584 bool nir_split_var_copies(nir_shader
*shader
);
2585 bool nir_split_per_member_structs(nir_shader
*shader
);
2587 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2588 bool nir_lower_returns(nir_shader
*shader
);
2590 bool nir_inline_functions(nir_shader
*shader
);
2592 bool nir_propagate_invariant(nir_shader
*shader
);
2594 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2595 void nir_lower_deref_copy_instr(struct nir_builder
*b
,
2596 nir_intrinsic_instr
*copy
);
2597 bool nir_lower_var_copies(nir_shader
*shader
);
2599 void nir_fixup_deref_modes(nir_shader
*shader
);
2601 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2603 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2605 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2607 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2608 nir_function_impl
*entrypoint
,
2609 bool outputs
, bool inputs
);
2611 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2613 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2614 int (*type_size
)(const struct glsl_type
*));
2616 /* Some helpers to do very simple linking */
2617 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2618 void nir_compact_varyings(nir_shader
*producer
, nir_shader
*consumer
,
2619 bool default_to_smooth_interp
);
2622 /* If set, this forces all non-flat fragment shader inputs to be
2623 * interpolated as if with the "sample" qualifier. This requires
2624 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2626 nir_lower_io_force_sample_interpolation
= (1 << 1),
2627 } nir_lower_io_options
;
2628 bool nir_lower_io(nir_shader
*shader
,
2629 nir_variable_mode modes
,
2630 int (*type_size
)(const struct glsl_type
*),
2631 nir_lower_io_options
);
2632 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2633 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2635 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2637 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2638 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2639 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2641 bool nir_remove_dead_derefs(nir_shader
*shader
);
2642 bool nir_remove_dead_derefs_impl(nir_function_impl
*impl
);
2643 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2644 bool nir_lower_constant_initializers(nir_shader
*shader
,
2645 nir_variable_mode modes
);
2647 bool nir_move_load_const(nir_shader
*shader
);
2648 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2649 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2650 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
2652 bool nir_lower_alu(nir_shader
*shader
);
2653 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2654 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2655 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
2656 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2657 void nir_lower_io_arrays_to_elements(nir_shader
*producer
, nir_shader
*consumer
);
2658 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader
*shader
,
2660 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2661 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
2663 typedef struct nir_lower_subgroups_options
{
2664 uint8_t subgroup_size
;
2665 uint8_t ballot_bit_size
;
2666 bool lower_to_scalar
:1;
2667 bool lower_vote_trivial
:1;
2668 bool lower_vote_eq_to_ballot
:1;
2669 bool lower_subgroup_masks
:1;
2670 bool lower_shuffle
:1;
2671 bool lower_shuffle_to_32bit
:1;
2673 } nir_lower_subgroups_options
;
2675 bool nir_lower_subgroups(nir_shader
*shader
,
2676 const nir_lower_subgroups_options
*options
);
2678 bool nir_lower_system_values(nir_shader
*shader
);
2680 typedef struct nir_lower_tex_options
{
2682 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2683 * sampler types a texture projector is lowered.
2688 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2690 bool lower_txf_offset
;
2693 * If true, lower away nir_tex_src_offset for all rect textures.
2695 bool lower_rect_offset
;
2698 * If true, lower rect textures to 2D, using txs to fetch the
2699 * texture dimensions and dividing the texture coords by the
2700 * texture dims to normalize.
2705 * If true, convert yuv to rgb.
2707 unsigned lower_y_uv_external
;
2708 unsigned lower_y_u_v_external
;
2709 unsigned lower_yx_xuxv_external
;
2710 unsigned lower_xy_uxvx_external
;
2713 * To emulate certain texture wrap modes, this can be used
2714 * to saturate the specified tex coord to [0.0, 1.0]. The
2715 * bits are according to sampler #, ie. if, for example:
2717 * (conf->saturate_s & (1 << n))
2719 * is true, then the s coord for sampler n is saturated.
2721 * Note that clamping must happen *after* projector lowering
2722 * so any projected texture sample instruction with a clamped
2723 * coordinate gets automatically lowered, regardless of the
2724 * 'lower_txp' setting.
2726 unsigned saturate_s
;
2727 unsigned saturate_t
;
2728 unsigned saturate_r
;
2730 /* Bitmask of textures that need swizzling.
2732 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2733 * swizzles[texture_index] is applied to the result of the texturing
2736 unsigned swizzle_result
;
2738 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2739 * while 4 and 5 represent 0 and 1 respectively.
2741 uint8_t swizzles
[32][4];
2744 * Bitmap of textures that need srgb to linear conversion. If
2745 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2746 * of the texture are lowered to linear.
2748 unsigned lower_srgb
;
2751 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2753 bool lower_txd_cube_map
;
2756 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
2757 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
2758 * with lower_txd_cube_map.
2760 bool lower_txd_shadow
;
2763 * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
2764 * Implies lower_txd_cube_map and lower_txd_shadow.
2767 } nir_lower_tex_options
;
2769 bool nir_lower_tex(nir_shader
*shader
,
2770 const nir_lower_tex_options
*options
);
2772 bool nir_lower_idiv(nir_shader
*shader
);
2774 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2775 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2776 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2778 void nir_lower_two_sided_color(nir_shader
*shader
);
2780 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
2782 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2783 void nir_lower_tes_patch_vertices(nir_shader
*tes
, unsigned patch_vertices
);
2785 typedef struct nir_lower_wpos_ytransform_options
{
2786 gl_state_index16 state_tokens
[STATE_LENGTH
];
2787 bool fs_coord_origin_upper_left
:1;
2788 bool fs_coord_origin_lower_left
:1;
2789 bool fs_coord_pixel_center_integer
:1;
2790 bool fs_coord_pixel_center_half_integer
:1;
2791 } nir_lower_wpos_ytransform_options
;
2793 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2794 const nir_lower_wpos_ytransform_options
*options
);
2795 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
2797 typedef struct nir_lower_drawpixels_options
{
2798 gl_state_index16 texcoord_state_tokens
[STATE_LENGTH
];
2799 gl_state_index16 scale_state_tokens
[STATE_LENGTH
];
2800 gl_state_index16 bias_state_tokens
[STATE_LENGTH
];
2801 unsigned drawpix_sampler
;
2802 unsigned pixelmap_sampler
;
2804 bool scale_and_bias
:1;
2805 } nir_lower_drawpixels_options
;
2807 void nir_lower_drawpixels(nir_shader
*shader
,
2808 const nir_lower_drawpixels_options
*options
);
2810 typedef struct nir_lower_bitmap_options
{
2813 } nir_lower_bitmap_options
;
2815 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2817 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
2818 bool nir_lower_to_source_mods(nir_shader
*shader
);
2820 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2822 typedef unsigned (*nir_lower_bit_size_callback
)(const nir_alu_instr
*, void *);
2824 bool nir_lower_bit_size(nir_shader
*shader
,
2825 nir_lower_bit_size_callback callback
,
2826 void *callback_data
);
2829 nir_lower_imul64
= (1 << 0),
2830 nir_lower_isign64
= (1 << 1),
2831 /** Lower all int64 modulus and division opcodes */
2832 nir_lower_divmod64
= (1 << 2),
2833 } nir_lower_int64_options
;
2835 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
2838 nir_lower_drcp
= (1 << 0),
2839 nir_lower_dsqrt
= (1 << 1),
2840 nir_lower_drsq
= (1 << 2),
2841 nir_lower_dtrunc
= (1 << 3),
2842 nir_lower_dfloor
= (1 << 4),
2843 nir_lower_dceil
= (1 << 5),
2844 nir_lower_dfract
= (1 << 6),
2845 nir_lower_dround_even
= (1 << 7),
2846 nir_lower_dmod
= (1 << 8)
2847 } nir_lower_doubles_options
;
2849 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2850 bool nir_lower_pack(nir_shader
*shader
);
2852 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2854 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2856 void nir_loop_analyze_impl(nir_function_impl
*impl
,
2857 nir_variable_mode indirect_mask
);
2859 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2861 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2862 bool nir_repair_ssa(nir_shader
*shader
);
2864 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
2866 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2867 * registers. If false, convert all values (even those not involved in a phi
2868 * node) to registers.
2870 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2872 bool nir_lower_phis_to_regs_block(nir_block
*block
);
2873 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
2875 bool nir_opt_algebraic(nir_shader
*shader
);
2876 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
2877 bool nir_opt_algebraic_late(nir_shader
*shader
);
2878 bool nir_opt_constant_folding(nir_shader
*shader
);
2880 bool nir_opt_global_to_local(nir_shader
*shader
);
2882 bool nir_copy_prop(nir_shader
*shader
);
2884 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
2886 bool nir_opt_cse(nir_shader
*shader
);
2888 bool nir_opt_dce(nir_shader
*shader
);
2890 bool nir_opt_dead_cf(nir_shader
*shader
);
2892 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2894 bool nir_opt_if(nir_shader
*shader
);
2896 bool nir_opt_intrinsics(nir_shader
*shader
);
2898 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
2900 bool nir_opt_move_comparisons(nir_shader
*shader
);
2902 bool nir_opt_move_load_ubo(nir_shader
*shader
);
2904 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2906 bool nir_opt_remove_phis(nir_shader
*shader
);
2908 bool nir_opt_shrink_load(nir_shader
*shader
);
2910 bool nir_opt_trivial_continues(nir_shader
*shader
);
2912 bool nir_opt_undef(nir_shader
*shader
);
2914 bool nir_opt_conditional_discard(nir_shader
*shader
);
2916 void nir_sweep(nir_shader
*shader
);
2918 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2919 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);