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 struct gl_shader_program
;
62 #define NIR_TRUE (~0u)
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); \
87 * Description of built-in state associated with a uniform
89 * \sa nir_variable::state_slots
97 nir_var_shader_in
= (1 << 0),
98 nir_var_shader_out
= (1 << 1),
99 nir_var_global
= (1 << 2),
100 nir_var_local
= (1 << 3),
101 nir_var_uniform
= (1 << 4),
102 nir_var_shader_storage
= (1 << 5),
103 nir_var_system_value
= (1 << 6),
104 nir_var_param
= (1 << 7),
105 nir_var_shared
= (1 << 8),
123 typedef struct nir_constant
{
125 * Value of the constant.
127 * The field used to back the values supplied by the constant is determined
128 * by the type associated with the \c nir_variable. Constants may be
129 * scalars, vectors, or matrices.
131 nir_const_value values
[4];
133 /* we could get this from the var->type but makes clone *much* easier to
134 * not have to care about the type.
136 unsigned num_elements
;
138 /* Array elements / Structure Fields */
139 struct nir_constant
**elements
;
143 * \brief Layout qualifiers for gl_FragDepth.
145 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
146 * with a layout qualifier.
149 nir_depth_layout_none
, /**< No depth layout is specified. */
150 nir_depth_layout_any
,
151 nir_depth_layout_greater
,
152 nir_depth_layout_less
,
153 nir_depth_layout_unchanged
157 * Either a uniform, global variable, shader input, or shader output. Based on
158 * ir_variable - it should be easy to translate between the two.
161 typedef struct nir_variable
{
162 struct exec_node node
;
165 * Declared type of the variable
167 const struct glsl_type
*type
;
170 * Declared name of the variable
174 struct nir_variable_data
{
176 * Storage class of the variable.
178 * \sa nir_variable_mode
180 nir_variable_mode mode
;
183 * Is the variable read-only?
185 * This is set for variables declared as \c const, shader inputs,
188 unsigned read_only
:1;
192 unsigned invariant
:1;
195 * When separate shader programs are enabled, only input/outputs between
196 * the stages of a multi-stage separate program can be safely removed
197 * from the shader interface. Other input/outputs must remains active.
199 * This is also used to make sure xfb varyings that are unused by the
200 * fragment shader are not removed.
202 unsigned always_active_io
:1;
205 * Interpolation mode for shader inputs / outputs
207 * \sa glsl_interp_mode
209 unsigned interpolation
:2;
212 * \name ARB_fragment_coord_conventions
215 unsigned origin_upper_left
:1;
216 unsigned pixel_center_integer
:1;
220 * If non-zero, then this variable may be packed along with other variables
221 * into a single varying slot, so this offset should be applied when
222 * accessing components. For example, an offset of 1 means that the x
223 * component of this variable is actually stored in component y of the
224 * location specified by \c location.
226 unsigned location_frac
:2;
229 * If true, this variable represents an array of scalars that should
230 * be tightly packed. In other words, consecutive array elements
231 * should be stored one component apart, rather than one slot apart.
236 * Whether this is a fragment shader output implicitly initialized with
237 * the previous contents of the specified render target at the
238 * framebuffer location corresponding to this shader invocation.
240 unsigned fb_fetch_output
:1;
243 * \brief Layout qualifier for gl_FragDepth.
245 * This is not equal to \c ir_depth_layout_none if and only if this
246 * variable is \c gl_FragDepth and a layout qualifier is specified.
248 nir_depth_layout depth_layout
;
251 * Storage location of the base of this variable
253 * The precise meaning of this field depends on the nature of the variable.
255 * - Vertex shader input: one of the values from \c gl_vert_attrib.
256 * - Vertex shader output: one of the values from \c gl_varying_slot.
257 * - Geometry shader input: one of the values from \c gl_varying_slot.
258 * - Geometry shader output: one of the values from \c gl_varying_slot.
259 * - Fragment shader input: one of the values from \c gl_varying_slot.
260 * - Fragment shader output: one of the values from \c gl_frag_result.
261 * - Uniforms: Per-stage uniform slot number for default uniform block.
262 * - Uniforms: Index within the uniform block definition for UBO members.
263 * - Non-UBO Uniforms: uniform slot number.
264 * - Other: This field is not currently used.
266 * If the variable is a uniform, shader input, or shader output, and the
267 * slot has not been assigned, the value will be -1.
272 * The actual location of the variable in the IR. Only valid for inputs
275 unsigned int driver_location
;
278 * output index for dual source blending.
283 * Descriptor set binding for sampler or UBO.
288 * Initial binding point for a sampler or UBO.
290 * For array types, this represents the binding point for the first element.
295 * Location an atomic counter is stored at.
300 * ARB_shader_image_load_store qualifiers.
303 bool read_only
; /**< "readonly" qualifier. */
304 bool write_only
; /**< "writeonly" qualifier. */
309 /** Image internal format if specified explicitly, otherwise GL_NONE. */
315 * Built-in state that backs this uniform
317 * Once set at variable creation, \c state_slots must remain invariant.
318 * This is because, ideally, this array would be shared by all clones of
319 * this variable in the IR tree. In other words, we'd really like for it
320 * to be a fly-weight.
322 * If the variable is not a uniform, \c num_state_slots will be zero and
323 * \c state_slots will be \c NULL.
326 unsigned num_state_slots
; /**< Number of state slots used */
327 nir_state_slot
*state_slots
; /**< State descriptors. */
331 * Constant expression assigned in the initializer of the variable
333 * This field should only be used temporarily by creators of NIR shaders
334 * and then lower_constant_initializers can be used to get rid of them.
335 * Most of the rest of NIR ignores this field or asserts that it's NULL.
337 nir_constant
*constant_initializer
;
340 * For variables that are in an interface block or are an instance of an
341 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
343 * \sa ir_variable::location
345 const struct glsl_type
*interface_type
;
348 #define nir_foreach_variable(var, var_list) \
349 foreach_list_typed(nir_variable, var, node, var_list)
351 #define nir_foreach_variable_safe(var, var_list) \
352 foreach_list_typed_safe(nir_variable, var, node, var_list)
355 nir_variable_is_global(const nir_variable
*var
)
357 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
360 typedef struct nir_register
{
361 struct exec_node node
;
363 unsigned num_components
; /** < number of vector components */
364 unsigned num_array_elems
; /** < size of array (0 for no array) */
366 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
369 /** generic register index. */
372 /** only for debug purposes, can be NULL */
375 /** whether this register is local (per-function) or global (per-shader) */
379 * If this flag is set to true, then accessing channels >= num_components
380 * is well-defined, and simply spills over to the next array element. This
381 * is useful for backends that can do per-component accessing, in
382 * particular scalar backends. By setting this flag and making
383 * num_components equal to 1, structures can be packed tightly into
384 * registers and then registers can be accessed per-component to get to
385 * each structure member, even if it crosses vec4 boundaries.
389 /** set of nir_srcs where this register is used (read from) */
390 struct list_head uses
;
392 /** set of nir_dests where this register is defined (written to) */
393 struct list_head defs
;
395 /** set of nir_ifs where this register is used as a condition */
396 struct list_head if_uses
;
399 #define nir_foreach_register(reg, reg_list) \
400 foreach_list_typed(nir_register, reg, node, reg_list)
401 #define nir_foreach_register_safe(reg, reg_list) \
402 foreach_list_typed_safe(nir_register, reg, node, reg_list)
408 nir_instr_type_intrinsic
,
409 nir_instr_type_load_const
,
411 nir_instr_type_ssa_undef
,
413 nir_instr_type_parallel_copy
,
416 typedef struct nir_instr
{
417 struct exec_node node
;
419 struct nir_block
*block
;
421 /** generic instruction index. */
424 /* A temporary for optimization and analysis passes to use for storing
425 * flags. For instance, DCE uses this to store the "dead/live" info.
430 static inline nir_instr
*
431 nir_instr_next(nir_instr
*instr
)
433 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
434 if (exec_node_is_tail_sentinel(next
))
437 return exec_node_data(nir_instr
, next
, node
);
440 static inline nir_instr
*
441 nir_instr_prev(nir_instr
*instr
)
443 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
444 if (exec_node_is_head_sentinel(prev
))
447 return exec_node_data(nir_instr
, prev
, node
);
451 nir_instr_is_first(const nir_instr
*instr
)
453 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
457 nir_instr_is_last(const nir_instr
*instr
)
459 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
462 typedef struct nir_ssa_def
{
463 /** for debugging only, can be NULL */
466 /** generic SSA definition index. */
469 /** Index into the live_in and live_out bitfields */
472 nir_instr
*parent_instr
;
474 /** set of nir_instrs where this register is used (read from) */
475 struct list_head uses
;
477 /** set of nir_ifs where this register is used as a condition */
478 struct list_head if_uses
;
480 uint8_t num_components
;
482 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
490 struct nir_src
*indirect
; /** < NULL for no indirect offset */
491 unsigned base_offset
;
493 /* TODO use-def chain goes here */
497 nir_instr
*parent_instr
;
498 struct list_head def_link
;
501 struct nir_src
*indirect
; /** < NULL for no indirect offset */
502 unsigned base_offset
;
504 /* TODO def-use chain goes here */
509 typedef struct nir_src
{
511 nir_instr
*parent_instr
;
512 struct nir_if
*parent_if
;
515 struct list_head use_link
;
525 static inline nir_src
528 nir_src src
= { { NULL
} };
532 #define NIR_SRC_INIT nir_src_init()
534 #define nir_foreach_use(src, reg_or_ssa_def) \
535 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
537 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
538 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
540 #define nir_foreach_if_use(src, reg_or_ssa_def) \
541 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
543 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
544 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
555 static inline nir_dest
558 nir_dest dest
= { { { NULL
} } };
562 #define NIR_DEST_INIT nir_dest_init()
564 #define nir_foreach_def(dest, reg) \
565 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
567 #define nir_foreach_def_safe(dest, reg) \
568 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
570 static inline nir_src
571 nir_src_for_ssa(nir_ssa_def
*def
)
573 nir_src src
= NIR_SRC_INIT
;
581 static inline nir_src
582 nir_src_for_reg(nir_register
*reg
)
584 nir_src src
= NIR_SRC_INIT
;
588 src
.reg
.indirect
= NULL
;
589 src
.reg
.base_offset
= 0;
594 static inline nir_dest
595 nir_dest_for_reg(nir_register
*reg
)
597 nir_dest dest
= NIR_DEST_INIT
;
604 static inline unsigned
605 nir_src_bit_size(nir_src src
)
607 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
610 static inline unsigned
611 nir_dest_bit_size(nir_dest dest
)
613 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
616 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
617 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
623 * \name input modifiers
627 * For inputs interpreted as floating point, flips the sign bit. For
628 * inputs interpreted as integers, performs the two's complement negation.
633 * Clears the sign bit for floating point values, and computes the integer
634 * absolute value for integers. Note that the negate modifier acts after
635 * the absolute value modifier, therefore if both are set then all inputs
636 * will become negative.
642 * For each input component, says which component of the register it is
643 * chosen from. Note that which elements of the swizzle are used and which
644 * are ignored are based on the write mask for most opcodes - for example,
645 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
646 * a swizzle of {2, x, 1, 0} where x means "don't care."
655 * \name saturate output modifier
657 * Only valid for opcodes that output floating-point numbers. Clamps the
658 * output to between 0.0 and 1.0 inclusive.
663 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
667 nir_type_invalid
= 0, /* Not a valid type */
672 nir_type_bool32
= 32 | nir_type_bool
,
673 nir_type_int8
= 8 | nir_type_int
,
674 nir_type_int16
= 16 | nir_type_int
,
675 nir_type_int32
= 32 | nir_type_int
,
676 nir_type_int64
= 64 | nir_type_int
,
677 nir_type_uint8
= 8 | nir_type_uint
,
678 nir_type_uint16
= 16 | nir_type_uint
,
679 nir_type_uint32
= 32 | nir_type_uint
,
680 nir_type_uint64
= 64 | nir_type_uint
,
681 nir_type_float16
= 16 | nir_type_float
,
682 nir_type_float32
= 32 | nir_type_float
,
683 nir_type_float64
= 64 | nir_type_float
,
686 #define NIR_ALU_TYPE_SIZE_MASK 0xfffffff8
687 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x00000007
689 static inline unsigned
690 nir_alu_type_get_type_size(nir_alu_type type
)
692 return type
& NIR_ALU_TYPE_SIZE_MASK
;
695 static inline unsigned
696 nir_alu_type_get_base_type(nir_alu_type type
)
698 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
701 static inline nir_alu_type
702 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
706 return nir_type_bool32
;
709 return nir_type_uint32
;
712 return nir_type_int32
;
714 case GLSL_TYPE_UINT64
:
715 return nir_type_uint64
;
717 case GLSL_TYPE_INT64
:
718 return nir_type_int64
;
720 case GLSL_TYPE_FLOAT
:
721 return nir_type_float32
;
723 case GLSL_TYPE_DOUBLE
:
724 return nir_type_float64
;
727 unreachable("unknown type");
731 static inline nir_alu_type
732 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
734 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
737 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
);
740 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
741 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
742 } nir_op_algebraic_property
;
750 * The number of components in the output
752 * If non-zero, this is the size of the output and input sizes are
753 * explicitly given; swizzle and writemask are still in effect, but if
754 * the output component is masked out, then the input component may
757 * If zero, the opcode acts in the standard, per-component manner; the
758 * operation is performed on each component (except the ones that are
759 * masked out) with the input being taken from the input swizzle for
762 * The size of some of the inputs may be given (i.e. non-zero) even
763 * though output_size is zero; in that case, the inputs with a zero
764 * size act per-component, while the inputs with non-zero size don't.
766 unsigned output_size
;
769 * The type of vector that the instruction outputs. Note that the
770 * staurate modifier is only allowed on outputs with the float type.
773 nir_alu_type output_type
;
776 * The number of components in each input
778 unsigned input_sizes
[4];
781 * The type of vector that each input takes. Note that negate and
782 * absolute value are only allowed on inputs with int or float type and
783 * behave differently on the two.
785 nir_alu_type input_types
[4];
787 nir_op_algebraic_property algebraic_properties
;
790 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
792 typedef struct nir_alu_instr
{
796 /** Indicates that this ALU instruction generates an exact value
798 * This is kind of a mixture of GLSL "precise" and "invariant" and not
799 * really equivalent to either. This indicates that the value generated by
800 * this operation is high-precision and any code transformations that touch
801 * it must ensure that the resulting value is bit-for-bit identical to the
810 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
811 nir_alu_instr
*instr
);
812 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
813 nir_alu_instr
*instr
);
815 /* is this source channel used? */
817 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
820 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
821 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
823 return (instr
->dest
.write_mask
>> channel
) & 1;
827 * For instructions whose destinations are SSA, get the number of channels
830 static inline unsigned
831 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
833 assert(instr
->dest
.dest
.is_ssa
);
835 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
836 return nir_op_infos
[instr
->op
].input_sizes
[src
];
838 return instr
->dest
.dest
.ssa
.num_components
;
841 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
842 unsigned src1
, unsigned src2
);
846 nir_deref_type_array
,
847 nir_deref_type_struct
850 typedef struct nir_deref
{
851 nir_deref_type deref_type
;
852 struct nir_deref
*child
;
853 const struct glsl_type
*type
;
862 /* This enum describes how the array is referenced. If the deref is
863 * direct then the base_offset is used. If the deref is indirect then
864 * offset is given by base_offset + indirect. If the deref is a wildcard
865 * then the deref refers to all of the elements of the array at the same
866 * time. Wildcard dereferences are only ever allowed in copy_var
867 * intrinsics and the source and destination derefs must have matching
871 nir_deref_array_type_direct
,
872 nir_deref_array_type_indirect
,
873 nir_deref_array_type_wildcard
,
874 } nir_deref_array_type
;
879 nir_deref_array_type deref_array_type
;
880 unsigned base_offset
;
890 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
,
891 deref_type
, nir_deref_type_var
)
892 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
,
893 deref_type
, nir_deref_type_array
)
894 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
,
895 deref_type
, nir_deref_type_struct
)
897 /* Returns the last deref in the chain. */
898 static inline nir_deref
*
899 nir_deref_tail(nir_deref
*deref
)
902 deref
= deref
->child
;
910 nir_deref_var
**params
;
911 nir_deref_var
*return_deref
;
913 struct nir_function
*callee
;
916 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
917 num_variables, num_indices, idx0, idx1, idx2, flags) \
918 nir_intrinsic_##name,
920 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
923 #include "nir_intrinsics.h"
924 nir_num_intrinsics
= nir_last_intrinsic
+ 1
927 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
929 /** Represents an intrinsic
931 * An intrinsic is an instruction type for handling things that are
932 * more-or-less regular operations but don't just consume and produce SSA
933 * values like ALU operations do. Intrinsics are not for things that have
934 * special semantic meaning such as phi nodes and parallel copies.
935 * Examples of intrinsics include variable load/store operations, system
936 * value loads, and the like. Even though texturing more-or-less falls
937 * under this category, texturing is its own instruction type because
938 * trying to represent texturing with intrinsics would lead to a
939 * combinatorial explosion of intrinsic opcodes.
941 * By having a single instruction type for handling a lot of different
942 * cases, optimization passes can look for intrinsics and, for the most
943 * part, completely ignore them. Each intrinsic type also has a few
944 * possible flags that govern whether or not they can be reordered or
945 * eliminated. That way passes like dead code elimination can still work
946 * on intrisics without understanding the meaning of each.
948 * Each intrinsic has some number of constant indices, some number of
949 * variables, and some number of sources. What these sources, variables,
950 * and indices mean depends on the intrinsic and is documented with the
951 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
952 * instructions are the only types of instruction that can operate on
958 nir_intrinsic_op intrinsic
;
962 /** number of components if this is a vectorized intrinsic
964 * Similarly to ALU operations, some intrinsics are vectorized.
965 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
966 * For vectorized intrinsics, the num_components field specifies the
967 * number of destination components and the number of source components
968 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
970 uint8_t num_components
;
972 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
974 nir_deref_var
*variables
[2];
977 } nir_intrinsic_instr
;
980 * \name NIR intrinsics semantic flags
982 * information about what the compiler can do with the intrinsics.
984 * \sa nir_intrinsic_info::flags
988 * whether the intrinsic can be safely eliminated if none of its output
989 * value is not being used.
991 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
994 * Whether the intrinsic can be reordered with respect to any other
995 * intrinsic, i.e. whether the only reordering dependencies of the
996 * intrinsic are due to the register reads/writes.
998 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
999 } nir_intrinsic_semantic_flag
;
1002 * \name NIR intrinsics const-index flag
1004 * Indicates the usage of a const_index slot.
1006 * \sa nir_intrinsic_info::index_map
1010 * Generally instructions that take a offset src argument, can encode
1011 * a constant 'base' value which is added to the offset.
1013 NIR_INTRINSIC_BASE
= 1,
1016 * For store instructions, a writemask for the store.
1018 NIR_INTRINSIC_WRMASK
= 2,
1021 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1023 NIR_INTRINSIC_STREAM_ID
= 3,
1026 * The clip-plane id for load_user_clip_plane intrinsic.
1028 NIR_INTRINSIC_UCP_ID
= 4,
1031 * The amount of data, starting from BASE, that this instruction may
1032 * access. This is used to provide bounds if the offset is not constant.
1034 NIR_INTRINSIC_RANGE
= 5,
1037 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1039 NIR_INTRINSIC_DESC_SET
= 6,
1042 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1044 NIR_INTRINSIC_BINDING
= 7,
1049 NIR_INTRINSIC_COMPONENT
= 8,
1052 * Interpolation mode (only meaningful for FS inputs).
1054 NIR_INTRINSIC_INTERP_MODE
= 9,
1056 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1058 } nir_intrinsic_index_flag
;
1060 #define NIR_INTRINSIC_MAX_INPUTS 4
1065 unsigned num_srcs
; /** < number of register/SSA inputs */
1067 /** number of components of each input register
1069 * If this value is 0, the number of components is given by the
1070 * num_components field of nir_intrinsic_instr.
1072 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1076 /** number of components of the output register
1078 * If this value is 0, the number of components is given by the
1079 * num_components field of nir_intrinsic_instr.
1081 unsigned dest_components
;
1083 /** the number of inputs/outputs that are variables */
1084 unsigned num_variables
;
1086 /** the number of constant indices used by the intrinsic */
1087 unsigned num_indices
;
1089 /** indicates the usage of intr->const_index[n] */
1090 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1092 /** semantic flags for calls to this intrinsic */
1093 nir_intrinsic_semantic_flag flags
;
1094 } nir_intrinsic_info
;
1096 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1099 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1100 static inline type \
1101 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1103 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1104 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1105 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1107 static inline void \
1108 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1110 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1111 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1112 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1115 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1116 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1117 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1118 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1119 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1120 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1121 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1122 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1123 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1126 * \group texture information
1128 * This gives semantic information about textures which is useful to the
1129 * frontend, the backend, and lowering passes, but not the optimizer.
1134 nir_tex_src_projector
,
1135 nir_tex_src_comparator
, /* shadow comparator */
1139 nir_tex_src_ms_index
, /* MSAA sample index */
1140 nir_tex_src_ms_mcs
, /* MSAA compression value */
1143 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1144 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1145 nir_tex_src_plane
, /* < selects plane for planar textures */
1146 nir_num_tex_src_types
1151 nir_tex_src_type src_type
;
1155 nir_texop_tex
, /**< Regular texture look-up */
1156 nir_texop_txb
, /**< Texture look-up with LOD bias */
1157 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1158 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1159 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1160 nir_texop_txf_ms
, /**< Multisample texture fetch */
1161 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1162 nir_texop_txs
, /**< Texture size */
1163 nir_texop_lod
, /**< Texture lod query */
1164 nir_texop_tg4
, /**< Texture gather */
1165 nir_texop_query_levels
, /**< Texture levels query */
1166 nir_texop_texture_samples
, /**< Texture samples query */
1167 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1175 enum glsl_sampler_dim sampler_dim
;
1176 nir_alu_type dest_type
;
1181 unsigned num_srcs
, coord_components
;
1182 bool is_array
, is_shadow
;
1185 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1186 * components or the new-style shadow that outputs 1 component.
1188 bool is_new_style_shadow
;
1190 /* gather component selector */
1191 unsigned component
: 2;
1193 /** The texture index
1195 * If this texture instruction has a nir_tex_src_texture_offset source,
1196 * then the texture index is given by texture_index + texture_offset.
1198 unsigned texture_index
;
1200 /** The size of the texture array or 0 if it's not an array */
1201 unsigned texture_array_size
;
1203 /** The texture deref
1205 * If this is null, use texture_index instead.
1207 nir_deref_var
*texture
;
1209 /** The sampler index
1211 * The following operations do not require a sampler and, as such, this
1212 * field should be ignored:
1214 * - nir_texop_txf_ms
1218 * - nir_texop_query_levels
1219 * - nir_texop_texture_samples
1220 * - nir_texop_samples_identical
1222 * If this texture instruction has a nir_tex_src_sampler_offset source,
1223 * then the sampler index is given by sampler_index + sampler_offset.
1225 unsigned sampler_index
;
1227 /** The sampler deref
1229 * If this is null, use sampler_index instead.
1231 nir_deref_var
*sampler
;
1234 static inline unsigned
1235 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1237 switch (instr
->op
) {
1238 case nir_texop_txs
: {
1240 switch (instr
->sampler_dim
) {
1241 case GLSL_SAMPLER_DIM_1D
:
1242 case GLSL_SAMPLER_DIM_BUF
:
1245 case GLSL_SAMPLER_DIM_2D
:
1246 case GLSL_SAMPLER_DIM_CUBE
:
1247 case GLSL_SAMPLER_DIM_MS
:
1248 case GLSL_SAMPLER_DIM_RECT
:
1249 case GLSL_SAMPLER_DIM_EXTERNAL
:
1250 case GLSL_SAMPLER_DIM_SUBPASS
:
1253 case GLSL_SAMPLER_DIM_3D
:
1257 unreachable("not reached");
1259 if (instr
->is_array
)
1267 case nir_texop_texture_samples
:
1268 case nir_texop_query_levels
:
1269 case nir_texop_samples_identical
:
1273 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1280 /* Returns true if this texture operation queries something about the texture
1281 * rather than actually sampling it.
1284 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1286 switch (instr
->op
) {
1289 case nir_texop_texture_samples
:
1290 case nir_texop_query_levels
:
1291 case nir_texop_txf_ms_mcs
:
1298 case nir_texop_txf_ms
:
1302 unreachable("Invalid texture opcode");
1306 static inline nir_alu_type
1307 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1309 switch (instr
->src
[src
].src_type
) {
1310 case nir_tex_src_coord
:
1311 switch (instr
->op
) {
1313 case nir_texop_txf_ms
:
1314 case nir_texop_txf_ms_mcs
:
1315 case nir_texop_samples_identical
:
1316 return nir_type_int
;
1319 return nir_type_float
;
1322 case nir_tex_src_lod
:
1323 switch (instr
->op
) {
1326 return nir_type_int
;
1329 return nir_type_float
;
1332 case nir_tex_src_projector
:
1333 case nir_tex_src_comparator
:
1334 case nir_tex_src_bias
:
1335 case nir_tex_src_ddx
:
1336 case nir_tex_src_ddy
:
1337 return nir_type_float
;
1339 case nir_tex_src_offset
:
1340 case nir_tex_src_ms_index
:
1341 case nir_tex_src_texture_offset
:
1342 case nir_tex_src_sampler_offset
:
1343 return nir_type_int
;
1346 unreachable("Invalid texture source type");
1350 static inline unsigned
1351 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1353 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1354 return instr
->coord_components
;
1356 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1357 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1360 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1361 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1362 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1363 if (instr
->is_array
)
1364 return instr
->coord_components
- 1;
1366 return instr
->coord_components
;
1373 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1375 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1376 if (instr
->src
[i
].src_type
== type
)
1382 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1383 nir_tex_src_type src_type
,
1386 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1391 nir_const_value value
;
1394 } nir_load_const_instr
;
1407 /* creates a new SSA variable in an undefined state */
1412 } nir_ssa_undef_instr
;
1415 struct exec_node node
;
1417 /* The predecessor block corresponding to this source */
1418 struct nir_block
*pred
;
1423 #define nir_foreach_phi_src(phi_src, phi) \
1424 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1425 #define nir_foreach_phi_src_safe(phi_src, phi) \
1426 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1431 struct exec_list srcs
; /** < list of nir_phi_src */
1437 struct exec_node node
;
1440 } nir_parallel_copy_entry
;
1442 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1443 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1448 /* A list of nir_parallel_copy_entrys. The sources of all of the
1449 * entries are copied to the corresponding destinations "in parallel".
1450 * In other words, if we have two entries: a -> b and b -> a, the values
1453 struct exec_list entries
;
1454 } nir_parallel_copy_instr
;
1456 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1457 type
, nir_instr_type_alu
)
1458 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1459 type
, nir_instr_type_call
)
1460 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1461 type
, nir_instr_type_jump
)
1462 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1463 type
, nir_instr_type_tex
)
1464 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1465 type
, nir_instr_type_intrinsic
)
1466 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1467 type
, nir_instr_type_load_const
)
1468 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1469 type
, nir_instr_type_ssa_undef
)
1470 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1471 type
, nir_instr_type_phi
)
1472 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1473 nir_parallel_copy_instr
, instr
,
1474 type
, nir_instr_type_parallel_copy
)
1479 * Control flow consists of a tree of control flow nodes, which include
1480 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1481 * instructions that always run start-to-finish. Each basic block also keeps
1482 * track of its successors (blocks which may run immediately after the current
1483 * block) and predecessors (blocks which could have run immediately before the
1484 * current block). Each function also has a start block and an end block which
1485 * all return statements point to (which is always empty). Together, all the
1486 * blocks with their predecessors and successors make up the control flow
1487 * graph (CFG) of the function. There are helpers that modify the tree of
1488 * control flow nodes while modifying the CFG appropriately; these should be
1489 * used instead of modifying the tree directly.
1496 nir_cf_node_function
1499 typedef struct nir_cf_node
{
1500 struct exec_node node
;
1501 nir_cf_node_type type
;
1502 struct nir_cf_node
*parent
;
1505 typedef struct nir_block
{
1506 nir_cf_node cf_node
;
1508 struct exec_list instr_list
; /** < list of nir_instr */
1510 /** generic block index; generated by nir_index_blocks */
1514 * Each block can only have up to 2 successors, so we put them in a simple
1515 * array - no need for anything more complicated.
1517 struct nir_block
*successors
[2];
1519 /* Set of nir_block predecessors in the CFG */
1520 struct set
*predecessors
;
1523 * this node's immediate dominator in the dominance tree - set to NULL for
1526 struct nir_block
*imm_dom
;
1528 /* This node's children in the dominance tree */
1529 unsigned num_dom_children
;
1530 struct nir_block
**dom_children
;
1532 /* Set of nir_blocks on the dominance frontier of this block */
1533 struct set
*dom_frontier
;
1536 * These two indices have the property that dom_{pre,post}_index for each
1537 * child of this block in the dominance tree will always be between
1538 * dom_pre_index and dom_post_index for this block, which makes testing if
1539 * a given block is dominated by another block an O(1) operation.
1541 unsigned dom_pre_index
, dom_post_index
;
1543 /* live in and out for this block; used for liveness analysis */
1544 BITSET_WORD
*live_in
;
1545 BITSET_WORD
*live_out
;
1548 static inline nir_instr
*
1549 nir_block_first_instr(nir_block
*block
)
1551 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1552 return exec_node_data(nir_instr
, head
, node
);
1555 static inline nir_instr
*
1556 nir_block_last_instr(nir_block
*block
)
1558 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1559 return exec_node_data(nir_instr
, tail
, node
);
1562 #define nir_foreach_instr(instr, block) \
1563 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1564 #define nir_foreach_instr_reverse(instr, block) \
1565 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1566 #define nir_foreach_instr_safe(instr, block) \
1567 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1568 #define nir_foreach_instr_reverse_safe(instr, block) \
1569 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1571 typedef struct nir_if
{
1572 nir_cf_node cf_node
;
1575 struct exec_list then_list
; /** < list of nir_cf_node */
1576 struct exec_list else_list
; /** < list of nir_cf_node */
1582 nir_instr
*conditional_instr
;
1584 nir_block
*break_block
;
1585 nir_block
*continue_from_block
;
1587 bool continue_from_then
;
1589 struct list_head loop_terminator_link
;
1590 } nir_loop_terminator
;
1593 /* Number of instructions in the loop */
1594 unsigned num_instructions
;
1596 /* How many times the loop is run (if known) */
1597 unsigned trip_count
;
1598 bool is_trip_count_known
;
1600 /* Unroll the loop regardless of its size */
1603 nir_loop_terminator
*limiting_terminator
;
1605 /* A list of loop_terminators terminating this loop. */
1606 struct list_head loop_terminator_list
;
1610 nir_cf_node cf_node
;
1612 struct exec_list body
; /** < list of nir_cf_node */
1614 nir_loop_info
*info
;
1618 * Various bits of metadata that can may be created or required by
1619 * optimization and analysis passes
1622 nir_metadata_none
= 0x0,
1623 nir_metadata_block_index
= 0x1,
1624 nir_metadata_dominance
= 0x2,
1625 nir_metadata_live_ssa_defs
= 0x4,
1626 nir_metadata_not_properly_reset
= 0x8,
1627 nir_metadata_loop_analysis
= 0x10,
1631 nir_cf_node cf_node
;
1633 /** pointer to the function of which this is an implementation */
1634 struct nir_function
*function
;
1636 struct exec_list body
; /** < list of nir_cf_node */
1638 nir_block
*end_block
;
1640 /** list for all local variables in the function */
1641 struct exec_list locals
;
1643 /** array of variables used as parameters */
1644 unsigned num_params
;
1645 nir_variable
**params
;
1647 /** variable used to hold the result of the function */
1648 nir_variable
*return_var
;
1650 /** list of local registers in the function */
1651 struct exec_list registers
;
1653 /** next available local register index */
1656 /** next available SSA value index */
1659 /* total number of basic blocks, only valid when block_index_dirty = false */
1660 unsigned num_blocks
;
1662 nir_metadata valid_metadata
;
1663 } nir_function_impl
;
1665 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1666 nir_start_block(nir_function_impl
*impl
)
1668 return (nir_block
*) impl
->body
.head_sentinel
.next
;
1671 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
1672 nir_impl_last_block(nir_function_impl
*impl
)
1674 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
1677 static inline nir_cf_node
*
1678 nir_cf_node_next(nir_cf_node
*node
)
1680 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1681 if (exec_node_is_tail_sentinel(next
))
1684 return exec_node_data(nir_cf_node
, next
, node
);
1687 static inline nir_cf_node
*
1688 nir_cf_node_prev(nir_cf_node
*node
)
1690 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1691 if (exec_node_is_head_sentinel(prev
))
1694 return exec_node_data(nir_cf_node
, prev
, node
);
1698 nir_cf_node_is_first(const nir_cf_node
*node
)
1700 return exec_node_is_head_sentinel(node
->node
.prev
);
1704 nir_cf_node_is_last(const nir_cf_node
*node
)
1706 return exec_node_is_tail_sentinel(node
->node
.next
);
1709 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
1710 type
, nir_cf_node_block
)
1711 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
1712 type
, nir_cf_node_if
)
1713 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
1714 type
, nir_cf_node_loop
)
1715 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
1716 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
1718 static inline nir_block
*
1719 nir_if_first_then_block(nir_if
*if_stmt
)
1721 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1722 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1725 static inline nir_block
*
1726 nir_if_last_then_block(nir_if
*if_stmt
)
1728 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1729 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1732 static inline nir_block
*
1733 nir_if_first_else_block(nir_if
*if_stmt
)
1735 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1736 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1739 static inline nir_block
*
1740 nir_if_last_else_block(nir_if
*if_stmt
)
1742 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1743 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1746 static inline nir_block
*
1747 nir_loop_first_block(nir_loop
*loop
)
1749 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
1750 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
1753 static inline nir_block
*
1754 nir_loop_last_block(nir_loop
*loop
)
1756 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
1757 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
1763 nir_parameter_inout
,
1764 } nir_parameter_type
;
1767 nir_parameter_type param_type
;
1768 const struct glsl_type
*type
;
1771 typedef struct nir_function
{
1772 struct exec_node node
;
1775 struct nir_shader
*shader
;
1777 unsigned num_params
;
1778 nir_parameter
*params
;
1779 const struct glsl_type
*return_type
;
1781 /** The implementation of this function.
1783 * If the function is only declared and not implemented, this is NULL.
1785 nir_function_impl
*impl
;
1788 typedef struct nir_shader_compiler_options
{
1793 /** Lowers flrp when it does not support doubles */
1800 bool lower_bitfield_extract
;
1801 bool lower_bitfield_insert
;
1802 bool lower_uadd_carry
;
1803 bool lower_usub_borrow
;
1804 /** lowers fneg and ineg to fsub and isub. */
1806 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1809 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1812 /** enables rules to lower idiv by power-of-two: */
1815 /* Does the native fdot instruction replicate its result for four
1816 * components? If so, then opt_algebraic_late will turn all fdotN
1817 * instructions into fdot_replicatedN instructions.
1819 bool fdot_replicates
;
1821 /** lowers ffract to fsub+ffloor: */
1824 bool lower_pack_half_2x16
;
1825 bool lower_pack_unorm_2x16
;
1826 bool lower_pack_snorm_2x16
;
1827 bool lower_pack_unorm_4x8
;
1828 bool lower_pack_snorm_4x8
;
1829 bool lower_unpack_half_2x16
;
1830 bool lower_unpack_unorm_2x16
;
1831 bool lower_unpack_snorm_2x16
;
1832 bool lower_unpack_unorm_4x8
;
1833 bool lower_unpack_snorm_4x8
;
1835 bool lower_extract_byte
;
1836 bool lower_extract_word
;
1838 bool lower_vote_trivial
;
1839 bool lower_subgroup_masks
;
1842 * Does the driver support real 32-bit integers? (Otherwise, integers
1843 * are simulated by floats.)
1845 bool native_integers
;
1847 /* Indicates that the driver only has zero-based vertex id */
1848 bool vertex_id_zero_based
;
1850 bool lower_cs_local_index_from_id
;
1853 * Should nir_lower_io() create load_interpolated_input intrinsics?
1855 * If not, it generates regular load_input intrinsics and interpolation
1856 * information must be inferred from the list of input nir_variables.
1858 bool use_interpolated_input_intrinsics
;
1860 unsigned max_subgroup_size
;
1862 unsigned max_unroll_iterations
;
1863 } nir_shader_compiler_options
;
1865 typedef struct nir_shader
{
1866 /** list of uniforms (nir_variable) */
1867 struct exec_list uniforms
;
1869 /** list of inputs (nir_variable) */
1870 struct exec_list inputs
;
1872 /** list of outputs (nir_variable) */
1873 struct exec_list outputs
;
1875 /** list of shared compute variables (nir_variable) */
1876 struct exec_list shared
;
1878 /** Set of driver-specific options for the shader.
1880 * The memory for the options is expected to be kept in a single static
1881 * copy by the driver.
1883 const struct nir_shader_compiler_options
*options
;
1885 /** Various bits of compile-time information about a given shader */
1886 struct shader_info info
;
1888 /** list of global variables in the shader (nir_variable) */
1889 struct exec_list globals
;
1891 /** list of system value variables in the shader (nir_variable) */
1892 struct exec_list system_values
;
1894 struct exec_list functions
; /** < list of nir_function */
1896 /** list of global register in the shader */
1897 struct exec_list registers
;
1899 /** next available global register index */
1903 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1906 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1909 static inline nir_function_impl
*
1910 nir_shader_get_entrypoint(nir_shader
*shader
)
1912 assert(exec_list_length(&shader
->functions
) == 1);
1913 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1914 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1915 assert(func
->return_type
== glsl_void_type());
1916 assert(func
->num_params
== 0);
1921 #define nir_foreach_function(func, shader) \
1922 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1924 nir_shader
*nir_shader_create(void *mem_ctx
,
1925 gl_shader_stage stage
,
1926 const nir_shader_compiler_options
*options
,
1929 /** creates a register, including assigning it an index and adding it to the list */
1930 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1932 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1934 void nir_reg_remove(nir_register
*reg
);
1936 /** Adds a variable to the appropriate list in nir_shader */
1937 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1940 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1942 assert(var
->data
.mode
== nir_var_local
);
1943 exec_list_push_tail(&impl
->locals
, &var
->node
);
1946 /** creates a variable, sets a few defaults, and adds it to the list */
1947 nir_variable
*nir_variable_create(nir_shader
*shader
,
1948 nir_variable_mode mode
,
1949 const struct glsl_type
*type
,
1951 /** creates a local variable and adds it to the list */
1952 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1953 const struct glsl_type
*type
,
1956 /** creates a function and adds it to the shader's list of functions */
1957 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1959 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1960 /** creates a function_impl that isn't tied to any particular function */
1961 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1963 nir_block
*nir_block_create(nir_shader
*shader
);
1964 nir_if
*nir_if_create(nir_shader
*shader
);
1965 nir_loop
*nir_loop_create(nir_shader
*shader
);
1967 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1969 /** requests that the given pieces of metadata be generated */
1970 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
1971 /** dirties all but the preserved metadata */
1972 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1974 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1975 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1977 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1979 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1980 unsigned num_components
,
1983 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1984 nir_intrinsic_op op
);
1986 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1987 nir_function
*callee
);
1989 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1991 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1993 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1995 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1996 unsigned num_components
,
1999 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
2000 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
2001 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
2003 typedef bool (*nir_deref_foreach_leaf_cb
)(nir_deref_var
*deref
, void *state
);
2004 bool nir_deref_foreach_leaf(nir_deref_var
*deref
,
2005 nir_deref_foreach_leaf_cb cb
, void *state
);
2007 nir_load_const_instr
*
2008 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
2011 * NIR Cursors and Instruction Insertion API
2014 * A tiny struct representing a point to insert/extract instructions or
2015 * control flow nodes. Helps reduce the combinatorial explosion of possible
2016 * points to insert/extract.
2018 * \sa nir_control_flow.h
2021 nir_cursor_before_block
,
2022 nir_cursor_after_block
,
2023 nir_cursor_before_instr
,
2024 nir_cursor_after_instr
,
2025 } nir_cursor_option
;
2028 nir_cursor_option option
;
2035 static inline nir_block
*
2036 nir_cursor_current_block(nir_cursor cursor
)
2038 if (cursor
.option
== nir_cursor_before_instr
||
2039 cursor
.option
== nir_cursor_after_instr
) {
2040 return cursor
.instr
->block
;
2042 return cursor
.block
;
2046 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2048 static inline nir_cursor
2049 nir_before_block(nir_block
*block
)
2052 cursor
.option
= nir_cursor_before_block
;
2053 cursor
.block
= block
;
2057 static inline nir_cursor
2058 nir_after_block(nir_block
*block
)
2061 cursor
.option
= nir_cursor_after_block
;
2062 cursor
.block
= block
;
2066 static inline nir_cursor
2067 nir_before_instr(nir_instr
*instr
)
2070 cursor
.option
= nir_cursor_before_instr
;
2071 cursor
.instr
= instr
;
2075 static inline nir_cursor
2076 nir_after_instr(nir_instr
*instr
)
2079 cursor
.option
= nir_cursor_after_instr
;
2080 cursor
.instr
= instr
;
2084 static inline nir_cursor
2085 nir_after_block_before_jump(nir_block
*block
)
2087 nir_instr
*last_instr
= nir_block_last_instr(block
);
2088 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2089 return nir_before_instr(last_instr
);
2091 return nir_after_block(block
);
2095 static inline nir_cursor
2096 nir_before_cf_node(nir_cf_node
*node
)
2098 if (node
->type
== nir_cf_node_block
)
2099 return nir_before_block(nir_cf_node_as_block(node
));
2101 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2104 static inline nir_cursor
2105 nir_after_cf_node(nir_cf_node
*node
)
2107 if (node
->type
== nir_cf_node_block
)
2108 return nir_after_block(nir_cf_node_as_block(node
));
2110 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2113 static inline nir_cursor
2114 nir_after_phis(nir_block
*block
)
2116 nir_foreach_instr(instr
, block
) {
2117 if (instr
->type
!= nir_instr_type_phi
)
2118 return nir_before_instr(instr
);
2120 return nir_after_block(block
);
2123 static inline nir_cursor
2124 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2126 if (node
->type
== nir_cf_node_block
)
2127 return nir_after_block(nir_cf_node_as_block(node
));
2129 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2131 return nir_after_phis(block
);
2134 static inline nir_cursor
2135 nir_before_cf_list(struct exec_list
*cf_list
)
2137 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2138 exec_list_get_head(cf_list
), node
);
2139 return nir_before_cf_node(first_node
);
2142 static inline nir_cursor
2143 nir_after_cf_list(struct exec_list
*cf_list
)
2145 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2146 exec_list_get_tail(cf_list
), node
);
2147 return nir_after_cf_node(last_node
);
2151 * Insert a NIR instruction at the given cursor.
2153 * Note: This does not update the cursor.
2155 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2158 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2160 nir_instr_insert(nir_before_instr(instr
), before
);
2164 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2166 nir_instr_insert(nir_after_instr(instr
), after
);
2170 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2172 nir_instr_insert(nir_before_block(block
), before
);
2176 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2178 nir_instr_insert(nir_after_block(block
), after
);
2182 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2184 nir_instr_insert(nir_before_cf_node(node
), before
);
2188 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2190 nir_instr_insert(nir_after_cf_node(node
), after
);
2194 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2196 nir_instr_insert(nir_before_cf_list(list
), before
);
2200 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2202 nir_instr_insert(nir_after_cf_list(list
), after
);
2205 void nir_instr_remove(nir_instr
*instr
);
2209 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2210 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2211 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2212 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2214 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2215 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2217 nir_const_value
*nir_src_as_const_value(nir_src src
);
2218 bool nir_src_is_dynamically_uniform(nir_src src
);
2219 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2220 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2221 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2222 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2223 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2225 void nir_instr_rewrite_deref(nir_instr
*instr
, nir_deref_var
**deref
,
2226 nir_deref_var
*new_deref
);
2228 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2229 unsigned num_components
, unsigned bit_size
,
2231 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2232 unsigned num_components
, unsigned bit_size
,
2234 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2235 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2236 nir_instr
*after_me
);
2238 uint8_t nir_ssa_def_components_read(nir_ssa_def
*def
);
2241 * finds the next basic block in source-code order, returns NULL if there is
2245 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2247 /* Performs the opposite of nir_block_cf_tree_next() */
2249 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2251 /* Gets the first block in a CF node in source-code order */
2253 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2255 /* Gets the last block in a CF node in source-code order */
2257 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2259 /* Gets the next block after a CF node in source-code order */
2261 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2263 /* Macros for loops that visit blocks in source-code order */
2265 #define nir_foreach_block(block, impl) \
2266 for (nir_block *block = nir_start_block(impl); block != NULL; \
2267 block = nir_block_cf_tree_next(block))
2269 #define nir_foreach_block_safe(block, impl) \
2270 for (nir_block *block = nir_start_block(impl), \
2271 *next = nir_block_cf_tree_next(block); \
2273 block = next, next = nir_block_cf_tree_next(block))
2275 #define nir_foreach_block_reverse(block, impl) \
2276 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2277 block = nir_block_cf_tree_prev(block))
2279 #define nir_foreach_block_reverse_safe(block, impl) \
2280 for (nir_block *block = nir_impl_last_block(impl), \
2281 *prev = nir_block_cf_tree_prev(block); \
2283 block = prev, prev = nir_block_cf_tree_prev(block))
2285 #define nir_foreach_block_in_cf_node(block, node) \
2286 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2287 block != nir_cf_node_cf_tree_next(node); \
2288 block = nir_block_cf_tree_next(block))
2290 /* If the following CF node is an if, this function returns that if.
2291 * Otherwise, it returns NULL.
2293 nir_if
*nir_block_get_following_if(nir_block
*block
);
2295 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2297 void nir_index_local_regs(nir_function_impl
*impl
);
2298 void nir_index_global_regs(nir_shader
*shader
);
2299 void nir_index_ssa_defs(nir_function_impl
*impl
);
2300 unsigned nir_index_instrs(nir_function_impl
*impl
);
2302 void nir_index_blocks(nir_function_impl
*impl
);
2304 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2305 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2306 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2308 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2309 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2310 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2311 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2312 nir_deref
*nir_deref_clone(const nir_deref
*deref
, void *mem_ctx
);
2313 nir_deref_var
*nir_deref_var_clone(const nir_deref_var
*deref
, void *mem_ctx
);
2316 void nir_validate_shader(nir_shader
*shader
);
2317 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2318 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2321 should_clone_nir(void)
2323 static int should_clone
= -1;
2324 if (should_clone
< 0)
2325 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2327 return should_clone
;
2331 should_print_nir(void)
2333 static int should_print
= -1;
2334 if (should_print
< 0)
2335 should_print
= env_var_as_boolean("NIR_PRINT", false);
2337 return should_print
;
2340 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2341 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2342 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2343 static inline bool should_clone_nir(void) { return false; }
2344 static inline bool should_print_nir(void) { return false; }
2347 #define _PASS(nir, do_pass) do { \
2349 nir_validate_shader(nir); \
2350 if (should_clone_nir()) { \
2351 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2357 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2358 nir_metadata_set_validation_flag(nir); \
2359 if (should_print_nir()) \
2360 printf("%s\n", #pass); \
2361 if (pass(nir, ##__VA_ARGS__)) { \
2363 if (should_print_nir()) \
2364 nir_print_shader(nir, stdout); \
2365 nir_metadata_check_validation_flag(nir); \
2369 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2370 if (should_print_nir()) \
2371 printf("%s\n", #pass); \
2372 pass(nir, ##__VA_ARGS__); \
2373 if (should_print_nir()) \
2374 nir_print_shader(nir, stdout); \
2377 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2378 void nir_calc_dominance(nir_shader
*shader
);
2380 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2381 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2383 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2384 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2386 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2387 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2389 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2390 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2392 int nir_gs_count_vertices(const nir_shader
*shader
);
2394 bool nir_split_var_copies(nir_shader
*shader
);
2396 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2397 bool nir_lower_returns(nir_shader
*shader
);
2399 bool nir_inline_functions(nir_shader
*shader
);
2401 bool nir_propagate_invariant(nir_shader
*shader
);
2403 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
2404 bool nir_lower_var_copies(nir_shader
*shader
);
2406 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2408 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
2410 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2412 void nir_lower_io_to_temporaries(nir_shader
*shader
,
2413 nir_function_impl
*entrypoint
,
2414 bool outputs
, bool inputs
);
2416 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2418 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
2419 int (*type_size
)(const struct glsl_type
*));
2421 /* Some helpers to do very simple linking */
2422 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
2425 /* If set, this forces all non-flat fragment shader inputs to be
2426 * interpolated as if with the "sample" qualifier. This requires
2427 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
2429 nir_lower_io_force_sample_interpolation
= (1 << 1),
2430 } nir_lower_io_options
;
2431 bool nir_lower_io(nir_shader
*shader
,
2432 nir_variable_mode modes
,
2433 int (*type_size
)(const struct glsl_type
*),
2434 nir_lower_io_options
);
2435 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2436 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2438 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
2440 void nir_lower_io_types(nir_shader
*shader
);
2441 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
2442 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
2443 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
2445 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
2446 bool nir_lower_constant_initializers(nir_shader
*shader
,
2447 nir_variable_mode modes
);
2449 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2450 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2451 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
2453 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
2454 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
2455 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
2456 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
2457 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
2458 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
2460 bool nir_lower_samplers(nir_shader
*shader
,
2461 const struct gl_shader_program
*shader_program
);
2462 bool nir_lower_samplers_as_deref(nir_shader
*shader
,
2463 const struct gl_shader_program
*shader_program
);
2465 bool nir_lower_system_values(nir_shader
*shader
);
2467 typedef struct nir_lower_tex_options
{
2469 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2470 * sampler types a texture projector is lowered.
2475 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
2477 bool lower_txf_offset
;
2480 * If true, lower away nir_tex_src_offset for all rect textures.
2482 bool lower_rect_offset
;
2485 * If true, lower rect textures to 2D, using txs to fetch the
2486 * texture dimensions and dividing the texture coords by the
2487 * texture dims to normalize.
2492 * If true, convert yuv to rgb.
2494 unsigned lower_y_uv_external
;
2495 unsigned lower_y_u_v_external
;
2496 unsigned lower_yx_xuxv_external
;
2497 unsigned lower_xy_uxvx_external
;
2500 * To emulate certain texture wrap modes, this can be used
2501 * to saturate the specified tex coord to [0.0, 1.0]. The
2502 * bits are according to sampler #, ie. if, for example:
2504 * (conf->saturate_s & (1 << n))
2506 * is true, then the s coord for sampler n is saturated.
2508 * Note that clamping must happen *after* projector lowering
2509 * so any projected texture sample instruction with a clamped
2510 * coordinate gets automatically lowered, regardless of the
2511 * 'lower_txp' setting.
2513 unsigned saturate_s
;
2514 unsigned saturate_t
;
2515 unsigned saturate_r
;
2517 /* Bitmask of textures that need swizzling.
2519 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2520 * swizzles[texture_index] is applied to the result of the texturing
2523 unsigned swizzle_result
;
2525 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2526 * while 4 and 5 represent 0 and 1 respectively.
2528 uint8_t swizzles
[32][4];
2531 * Bitmap of textures that need srgb to linear conversion. If
2532 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
2533 * of the texture are lowered to linear.
2535 unsigned lower_srgb
;
2538 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
2540 bool lower_txd_cube_map
;
2543 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
2544 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
2545 * with lower_txd_cube_map.
2547 bool lower_txd_shadow
;
2548 } nir_lower_tex_options
;
2550 bool nir_lower_tex(nir_shader
*shader
,
2551 const nir_lower_tex_options
*options
);
2553 bool nir_lower_idiv(nir_shader
*shader
);
2555 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2556 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2557 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
2559 void nir_lower_two_sided_color(nir_shader
*shader
);
2561 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
2563 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
2564 void nir_lower_tes_patch_vertices(nir_shader
*tes
, unsigned patch_vertices
);
2566 typedef struct nir_lower_wpos_ytransform_options
{
2567 int state_tokens
[5];
2568 bool fs_coord_origin_upper_left
:1;
2569 bool fs_coord_origin_lower_left
:1;
2570 bool fs_coord_pixel_center_integer
:1;
2571 bool fs_coord_pixel_center_half_integer
:1;
2572 } nir_lower_wpos_ytransform_options
;
2574 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
2575 const nir_lower_wpos_ytransform_options
*options
);
2576 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
2578 typedef struct nir_lower_drawpixels_options
{
2579 int texcoord_state_tokens
[5];
2580 int scale_state_tokens
[5];
2581 int bias_state_tokens
[5];
2582 unsigned drawpix_sampler
;
2583 unsigned pixelmap_sampler
;
2585 bool scale_and_bias
:1;
2586 } nir_lower_drawpixels_options
;
2588 void nir_lower_drawpixels(nir_shader
*shader
,
2589 const nir_lower_drawpixels_options
*options
);
2591 typedef struct nir_lower_bitmap_options
{
2594 } nir_lower_bitmap_options
;
2596 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
2598 bool nir_lower_atomics(nir_shader
*shader
,
2599 const struct gl_shader_program
*shader_program
);
2600 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
2601 bool nir_lower_uniforms_to_ubo(nir_shader
*shader
);
2602 bool nir_lower_to_source_mods(nir_shader
*shader
);
2604 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2607 nir_lower_imul64
= (1 << 0),
2608 nir_lower_isign64
= (1 << 1),
2609 /** Lower all int64 modulus and division opcodes */
2610 nir_lower_divmod64
= (1 << 2),
2611 } nir_lower_int64_options
;
2613 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
2616 nir_lower_drcp
= (1 << 0),
2617 nir_lower_dsqrt
= (1 << 1),
2618 nir_lower_drsq
= (1 << 2),
2619 nir_lower_dtrunc
= (1 << 3),
2620 nir_lower_dfloor
= (1 << 4),
2621 nir_lower_dceil
= (1 << 5),
2622 nir_lower_dfract
= (1 << 6),
2623 nir_lower_dround_even
= (1 << 7),
2624 nir_lower_dmod
= (1 << 8)
2625 } nir_lower_doubles_options
;
2627 bool nir_lower_doubles(nir_shader
*shader
, nir_lower_doubles_options options
);
2628 bool nir_lower_64bit_pack(nir_shader
*shader
);
2630 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2632 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2634 void nir_loop_analyze_impl(nir_function_impl
*impl
,
2635 nir_variable_mode indirect_mask
);
2637 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2639 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2640 bool nir_repair_ssa(nir_shader
*shader
);
2642 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
2644 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2645 * registers. If false, convert all values (even those not involved in a phi
2646 * node) to registers.
2648 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2650 bool nir_lower_phis_to_regs_block(nir_block
*block
);
2651 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
2653 bool nir_opt_algebraic(nir_shader
*shader
);
2654 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
2655 bool nir_opt_algebraic_late(nir_shader
*shader
);
2656 bool nir_opt_constant_folding(nir_shader
*shader
);
2658 bool nir_opt_global_to_local(nir_shader
*shader
);
2660 bool nir_copy_prop(nir_shader
*shader
);
2662 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
2664 bool nir_opt_cse(nir_shader
*shader
);
2666 bool nir_opt_dce(nir_shader
*shader
);
2668 bool nir_opt_dead_cf(nir_shader
*shader
);
2670 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
2672 bool nir_opt_if(nir_shader
*shader
);
2674 bool nir_opt_intrinsics(nir_shader
*shader
);
2676 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
2678 bool nir_opt_move_comparisons(nir_shader
*shader
);
2680 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
);
2682 bool nir_opt_remove_phis(nir_shader
*shader
);
2684 bool nir_opt_trivial_continues(nir_shader
*shader
);
2686 bool nir_opt_undef(nir_shader
*shader
);
2688 bool nir_opt_conditional_discard(nir_shader
*shader
);
2690 void nir_sweep(nir_shader
*shader
);
2692 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2693 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);