2 * Copyright © 2014 Connor Abbott
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Connor Abbott (cwabbott0@gmail.com)
31 #include "util/hash_table.h"
32 #include "compiler/glsl/list.h"
33 #include "GL/gl.h" /* GLenum */
34 #include "util/list.h"
35 #include "util/ralloc.h"
37 #include "util/bitscan.h"
38 #include "util/bitset.h"
39 #include "util/macros.h"
40 #include "compiler/nir_types.h"
41 #include "compiler/shader_enums.h"
42 #include "compiler/shader_info.h"
46 #include "util/debug.h"
49 #include "nir_opcodes.h"
51 #if defined(_WIN32) && !defined(snprintf)
52 #define snprintf _snprintf
60 #define NIR_TRUE (~0u)
61 #define NIR_MAX_VEC_COMPONENTS 4
62 #define NIR_MAX_MATRIX_COLUMNS 4
63 typedef uint8_t nir_component_mask_t
;
65 /** Defines a cast function
67 * This macro defines a cast function from in_type to out_type where
68 * out_type is some structure type that contains a field of type out_type.
70 * Note that you have to be a bit careful as the generated cast function
73 #define NIR_DEFINE_CAST(name, in_type, out_type, field, \
74 type_field, type_value) \
75 static inline out_type * \
76 name(const in_type *parent) \
78 assert(parent && parent->type_field == type_value); \
79 return exec_node_data(out_type, parent, field); \
89 * Description of built-in state associated with a uniform
91 * \sa nir_variable::state_slots
94 gl_state_index16 tokens
[STATE_LENGTH
];
99 nir_var_shader_in
= (1 << 0),
100 nir_var_shader_out
= (1 << 1),
101 nir_var_shader_temp
= (1 << 2),
102 nir_var_function_temp
= (1 << 3),
103 nir_var_uniform
= (1 << 4),
104 nir_var_mem_ubo
= (1 << 5),
105 nir_var_system_value
= (1 << 6),
106 nir_var_mem_ssbo
= (1 << 7),
107 nir_var_mem_shared
= (1 << 8),
108 nir_var_mem_global
= (1 << 9),
116 nir_rounding_mode_undef
= 0,
117 nir_rounding_mode_rtne
= 1, /* round to nearest even */
118 nir_rounding_mode_ru
= 2, /* round up */
119 nir_rounding_mode_rd
= 3, /* round down */
120 nir_rounding_mode_rtz
= 4, /* round towards zero */
124 bool b
[NIR_MAX_VEC_COMPONENTS
];
125 float f32
[NIR_MAX_VEC_COMPONENTS
];
126 double f64
[NIR_MAX_VEC_COMPONENTS
];
127 int8_t i8
[NIR_MAX_VEC_COMPONENTS
];
128 uint8_t u8
[NIR_MAX_VEC_COMPONENTS
];
129 int16_t i16
[NIR_MAX_VEC_COMPONENTS
];
130 uint16_t u16
[NIR_MAX_VEC_COMPONENTS
];
131 int32_t i32
[NIR_MAX_VEC_COMPONENTS
];
132 uint32_t u32
[NIR_MAX_VEC_COMPONENTS
];
133 int64_t i64
[NIR_MAX_VEC_COMPONENTS
];
134 uint64_t u64
[NIR_MAX_VEC_COMPONENTS
];
137 typedef struct nir_constant
{
139 * Value of the constant.
141 * The field used to back the values supplied by the constant is determined
142 * by the type associated with the \c nir_variable. Constants may be
143 * scalars, vectors, or matrices.
145 nir_const_value values
[NIR_MAX_MATRIX_COLUMNS
];
147 /* we could get this from the var->type but makes clone *much* easier to
148 * not have to care about the type.
150 unsigned num_elements
;
152 /* Array elements / Structure Fields */
153 struct nir_constant
**elements
;
157 * \brief Layout qualifiers for gl_FragDepth.
159 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
160 * with a layout qualifier.
163 nir_depth_layout_none
, /**< No depth layout is specified. */
164 nir_depth_layout_any
,
165 nir_depth_layout_greater
,
166 nir_depth_layout_less
,
167 nir_depth_layout_unchanged
171 * Enum keeping track of how a variable was declared.
175 * Normal declaration.
177 nir_var_declared_normally
= 0,
180 * Variable is implicitly generated by the compiler and should not be
181 * visible via the API.
184 } nir_var_declaration_type
;
187 * Either a uniform, global variable, shader input, or shader output. Based on
188 * ir_variable - it should be easy to translate between the two.
191 typedef struct nir_variable
{
192 struct exec_node node
;
195 * Declared type of the variable
197 const struct glsl_type
*type
;
200 * Declared name of the variable
204 struct nir_variable_data
{
206 * Storage class of the variable.
208 * \sa nir_variable_mode
210 nir_variable_mode mode
;
213 * Is the variable read-only?
215 * This is set for variables declared as \c const, shader inputs,
218 unsigned read_only
:1;
222 unsigned invariant
:1;
225 * When separate shader programs are enabled, only input/outputs between
226 * the stages of a multi-stage separate program can be safely removed
227 * from the shader interface. Other input/outputs must remains active.
229 * This is also used to make sure xfb varyings that are unused by the
230 * fragment shader are not removed.
232 unsigned always_active_io
:1;
235 * Interpolation mode for shader inputs / outputs
237 * \sa glsl_interp_mode
239 unsigned interpolation
:2;
242 * If non-zero, then this variable may be packed along with other variables
243 * into a single varying slot, so this offset should be applied when
244 * accessing components. For example, an offset of 1 means that the x
245 * component of this variable is actually stored in component y of the
246 * location specified by \c location.
248 unsigned location_frac
:2;
251 * If true, this variable represents an array of scalars that should
252 * be tightly packed. In other words, consecutive array elements
253 * should be stored one component apart, rather than one slot apart.
258 * Whether this is a fragment shader output implicitly initialized with
259 * the previous contents of the specified render target at the
260 * framebuffer location corresponding to this shader invocation.
262 unsigned fb_fetch_output
:1;
265 * Non-zero if this variable is considered bindless as defined by
266 * ARB_bindless_texture.
271 * Was an explicit binding set in the shader?
273 unsigned explicit_binding
:1;
276 * Was a transfer feedback buffer set in the shader?
278 unsigned explicit_xfb_buffer
:1;
281 * Was a transfer feedback stride set in the shader?
283 unsigned explicit_xfb_stride
:1;
286 * Was an explicit offset set in the shader?
288 unsigned explicit_offset
:1;
291 * \brief Layout qualifier for gl_FragDepth.
293 * This is not equal to \c ir_depth_layout_none if and only if this
294 * variable is \c gl_FragDepth and a layout qualifier is specified.
296 nir_depth_layout depth_layout
;
299 * Storage location of the base of this variable
301 * The precise meaning of this field depends on the nature of the variable.
303 * - Vertex shader input: one of the values from \c gl_vert_attrib.
304 * - Vertex shader output: one of the values from \c gl_varying_slot.
305 * - Geometry shader input: one of the values from \c gl_varying_slot.
306 * - Geometry shader output: one of the values from \c gl_varying_slot.
307 * - Fragment shader input: one of the values from \c gl_varying_slot.
308 * - Fragment shader output: one of the values from \c gl_frag_result.
309 * - Uniforms: Per-stage uniform slot number for default uniform block.
310 * - Uniforms: Index within the uniform block definition for UBO members.
311 * - Non-UBO Uniforms: uniform slot number.
312 * - Other: This field is not currently used.
314 * If the variable is a uniform, shader input, or shader output, and the
315 * slot has not been assigned, the value will be -1.
320 * The actual location of the variable in the IR. Only valid for inputs
323 unsigned int driver_location
;
326 * Vertex stream output identifier.
328 * For packed outputs, bit 31 is set and bits [2*i+1,2*i] indicate the
329 * stream of the i-th component.
334 * output index for dual source blending.
339 * Descriptor set binding for sampler or UBO.
344 * Initial binding point for a sampler or UBO.
346 * For array types, this represents the binding point for the first element.
351 * Location an atomic counter or transform feedback is stored at.
356 * Transform feedback buffer.
361 * Transform feedback stride.
366 * How the variable was declared. See nir_var_declaration_type.
368 * This is used to detect variables generated by the compiler, so should
369 * not be visible via the API.
371 unsigned how_declared
:2;
374 * ARB_shader_image_load_store qualifiers.
377 enum gl_access_qualifier access
;
379 /** Image internal format if specified explicitly, otherwise GL_NONE. */
385 * Built-in state that backs this uniform
387 * Once set at variable creation, \c state_slots must remain invariant.
388 * This is because, ideally, this array would be shared by all clones of
389 * this variable in the IR tree. In other words, we'd really like for it
390 * to be a fly-weight.
392 * If the variable is not a uniform, \c num_state_slots will be zero and
393 * \c state_slots will be \c NULL.
396 unsigned num_state_slots
; /**< Number of state slots used */
397 nir_state_slot
*state_slots
; /**< State descriptors. */
401 * Constant expression assigned in the initializer of the variable
403 * This field should only be used temporarily by creators of NIR shaders
404 * and then lower_constant_initializers can be used to get rid of them.
405 * Most of the rest of NIR ignores this field or asserts that it's NULL.
407 nir_constant
*constant_initializer
;
410 * For variables that are in an interface block or are an instance of an
411 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
413 * \sa ir_variable::location
415 const struct glsl_type
*interface_type
;
418 * Description of per-member data for per-member struct variables
420 * This is used for variables which are actually an amalgamation of
421 * multiple entities such as a struct of built-in values or a struct of
422 * inputs each with their own layout specifier. This is only allowed on
423 * variables with a struct or array of array of struct type.
425 unsigned num_members
;
426 struct nir_variable_data
*members
;
429 #define nir_foreach_variable(var, var_list) \
430 foreach_list_typed(nir_variable, var, node, var_list)
432 #define nir_foreach_variable_safe(var, var_list) \
433 foreach_list_typed_safe(nir_variable, var, node, var_list)
436 nir_variable_is_global(const nir_variable
*var
)
438 return var
->data
.mode
!= nir_var_function_temp
;
441 typedef struct nir_register
{
442 struct exec_node node
;
444 unsigned num_components
; /** < number of vector components */
445 unsigned num_array_elems
; /** < size of array (0 for no array) */
447 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
450 /** generic register index. */
453 /** only for debug purposes, can be NULL */
456 /** whether this register is local (per-function) or global (per-shader) */
460 * If this flag is set to true, then accessing channels >= num_components
461 * is well-defined, and simply spills over to the next array element. This
462 * is useful for backends that can do per-component accessing, in
463 * particular scalar backends. By setting this flag and making
464 * num_components equal to 1, structures can be packed tightly into
465 * registers and then registers can be accessed per-component to get to
466 * each structure member, even if it crosses vec4 boundaries.
470 /** set of nir_srcs where this register is used (read from) */
471 struct list_head uses
;
473 /** set of nir_dests where this register is defined (written to) */
474 struct list_head defs
;
476 /** set of nir_ifs where this register is used as a condition */
477 struct list_head if_uses
;
480 #define nir_foreach_register(reg, reg_list) \
481 foreach_list_typed(nir_register, reg, node, reg_list)
482 #define nir_foreach_register_safe(reg, reg_list) \
483 foreach_list_typed_safe(nir_register, reg, node, reg_list)
485 typedef enum PACKED
{
487 nir_instr_type_deref
,
490 nir_instr_type_intrinsic
,
491 nir_instr_type_load_const
,
493 nir_instr_type_ssa_undef
,
495 nir_instr_type_parallel_copy
,
498 typedef struct nir_instr
{
499 struct exec_node node
;
500 struct nir_block
*block
;
503 /* A temporary for optimization and analysis passes to use for storing
504 * flags. For instance, DCE uses this to store the "dead/live" info.
508 /** generic instruction index. */
512 static inline nir_instr
*
513 nir_instr_next(nir_instr
*instr
)
515 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
516 if (exec_node_is_tail_sentinel(next
))
519 return exec_node_data(nir_instr
, next
, node
);
522 static inline nir_instr
*
523 nir_instr_prev(nir_instr
*instr
)
525 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
526 if (exec_node_is_head_sentinel(prev
))
529 return exec_node_data(nir_instr
, prev
, node
);
533 nir_instr_is_first(const nir_instr
*instr
)
535 return exec_node_is_head_sentinel(exec_node_get_prev_const(&instr
->node
));
539 nir_instr_is_last(const nir_instr
*instr
)
541 return exec_node_is_tail_sentinel(exec_node_get_next_const(&instr
->node
));
544 typedef struct nir_ssa_def
{
545 /** for debugging only, can be NULL */
548 /** generic SSA definition index. */
551 /** Index into the live_in and live_out bitfields */
554 /** Instruction which produces this SSA value. */
555 nir_instr
*parent_instr
;
557 /** set of nir_instrs where this register is used (read from) */
558 struct list_head uses
;
560 /** set of nir_ifs where this register is used as a condition */
561 struct list_head if_uses
;
563 uint8_t num_components
;
565 /* The bit-size of each channel; must be one of 8, 16, 32, or 64 */
573 struct nir_src
*indirect
; /** < NULL for no indirect offset */
574 unsigned base_offset
;
576 /* TODO use-def chain goes here */
580 nir_instr
*parent_instr
;
581 struct list_head def_link
;
584 struct nir_src
*indirect
; /** < NULL for no indirect offset */
585 unsigned base_offset
;
587 /* TODO def-use chain goes here */
592 typedef struct nir_src
{
594 /** Instruction that consumes this value as a source. */
595 nir_instr
*parent_instr
;
596 struct nir_if
*parent_if
;
599 struct list_head use_link
;
609 static inline nir_src
612 nir_src src
= { { NULL
} };
616 #define NIR_SRC_INIT nir_src_init()
618 #define nir_foreach_use(src, reg_or_ssa_def) \
619 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
621 #define nir_foreach_use_safe(src, reg_or_ssa_def) \
622 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
624 #define nir_foreach_if_use(src, reg_or_ssa_def) \
625 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
627 #define nir_foreach_if_use_safe(src, reg_or_ssa_def) \
628 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
639 static inline nir_dest
642 nir_dest dest
= { { { NULL
} } };
646 #define NIR_DEST_INIT nir_dest_init()
648 #define nir_foreach_def(dest, reg) \
649 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
651 #define nir_foreach_def_safe(dest, reg) \
652 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
654 static inline nir_src
655 nir_src_for_ssa(nir_ssa_def
*def
)
657 nir_src src
= NIR_SRC_INIT
;
665 static inline nir_src
666 nir_src_for_reg(nir_register
*reg
)
668 nir_src src
= NIR_SRC_INIT
;
672 src
.reg
.indirect
= NULL
;
673 src
.reg
.base_offset
= 0;
678 static inline nir_dest
679 nir_dest_for_reg(nir_register
*reg
)
681 nir_dest dest
= NIR_DEST_INIT
;
688 static inline unsigned
689 nir_src_bit_size(nir_src src
)
691 return src
.is_ssa
? src
.ssa
->bit_size
: src
.reg
.reg
->bit_size
;
694 static inline unsigned
695 nir_src_num_components(nir_src src
)
697 return src
.is_ssa
? src
.ssa
->num_components
: src
.reg
.reg
->num_components
;
701 nir_src_is_const(nir_src src
)
704 src
.ssa
->parent_instr
->type
== nir_instr_type_load_const
;
707 int64_t nir_src_as_int(nir_src src
);
708 uint64_t nir_src_as_uint(nir_src src
);
709 bool nir_src_as_bool(nir_src src
);
710 double nir_src_as_float(nir_src src
);
711 int64_t nir_src_comp_as_int(nir_src src
, unsigned component
);
712 uint64_t nir_src_comp_as_uint(nir_src src
, unsigned component
);
713 bool nir_src_comp_as_bool(nir_src src
, unsigned component
);
714 double nir_src_comp_as_float(nir_src src
, unsigned component
);
716 static inline unsigned
717 nir_dest_bit_size(nir_dest dest
)
719 return dest
.is_ssa
? dest
.ssa
.bit_size
: dest
.reg
.reg
->bit_size
;
722 static inline unsigned
723 nir_dest_num_components(nir_dest dest
)
725 return dest
.is_ssa
? dest
.ssa
.num_components
: dest
.reg
.reg
->num_components
;
728 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
729 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
735 * \name input modifiers
739 * For inputs interpreted as floating point, flips the sign bit. For
740 * inputs interpreted as integers, performs the two's complement negation.
745 * Clears the sign bit for floating point values, and computes the integer
746 * absolute value for integers. Note that the negate modifier acts after
747 * the absolute value modifier, therefore if both are set then all inputs
748 * will become negative.
754 * For each input component, says which component of the register it is
755 * chosen from. Note that which elements of the swizzle are used and which
756 * are ignored are based on the write mask for most opcodes - for example,
757 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
758 * a swizzle of {2, x, 1, 0} where x means "don't care."
760 uint8_t swizzle
[NIR_MAX_VEC_COMPONENTS
];
767 * \name saturate output modifier
769 * Only valid for opcodes that output floating-point numbers. Clamps the
770 * output to between 0.0 and 1.0 inclusive.
775 unsigned write_mask
: NIR_MAX_VEC_COMPONENTS
; /* ignored if dest.is_ssa is true */
778 /** NIR sized and unsized types
780 * The values in this enum are carefully chosen so that the sized type is
781 * just the unsized type OR the number of bits.
784 nir_type_invalid
= 0, /* Not a valid type */
788 nir_type_float
= 128,
789 nir_type_bool1
= 1 | nir_type_bool
,
790 nir_type_bool32
= 32 | nir_type_bool
,
791 nir_type_int1
= 1 | nir_type_int
,
792 nir_type_int8
= 8 | nir_type_int
,
793 nir_type_int16
= 16 | nir_type_int
,
794 nir_type_int32
= 32 | nir_type_int
,
795 nir_type_int64
= 64 | nir_type_int
,
796 nir_type_uint1
= 1 | nir_type_uint
,
797 nir_type_uint8
= 8 | nir_type_uint
,
798 nir_type_uint16
= 16 | nir_type_uint
,
799 nir_type_uint32
= 32 | nir_type_uint
,
800 nir_type_uint64
= 64 | nir_type_uint
,
801 nir_type_float16
= 16 | nir_type_float
,
802 nir_type_float32
= 32 | nir_type_float
,
803 nir_type_float64
= 64 | nir_type_float
,
806 #define NIR_ALU_TYPE_SIZE_MASK 0x79
807 #define NIR_ALU_TYPE_BASE_TYPE_MASK 0x86
809 static inline unsigned
810 nir_alu_type_get_type_size(nir_alu_type type
)
812 return type
& NIR_ALU_TYPE_SIZE_MASK
;
815 static inline unsigned
816 nir_alu_type_get_base_type(nir_alu_type type
)
818 return type
& NIR_ALU_TYPE_BASE_TYPE_MASK
;
821 static inline nir_alu_type
822 nir_get_nir_type_for_glsl_base_type(enum glsl_base_type base_type
)
826 return nir_type_bool1
;
829 return nir_type_uint32
;
832 return nir_type_int32
;
834 case GLSL_TYPE_UINT16
:
835 return nir_type_uint16
;
837 case GLSL_TYPE_INT16
:
838 return nir_type_int16
;
840 case GLSL_TYPE_UINT8
:
841 return nir_type_uint8
;
843 return nir_type_int8
;
844 case GLSL_TYPE_UINT64
:
845 return nir_type_uint64
;
847 case GLSL_TYPE_INT64
:
848 return nir_type_int64
;
850 case GLSL_TYPE_FLOAT
:
851 return nir_type_float32
;
853 case GLSL_TYPE_FLOAT16
:
854 return nir_type_float16
;
856 case GLSL_TYPE_DOUBLE
:
857 return nir_type_float64
;
860 unreachable("unknown type");
864 static inline nir_alu_type
865 nir_get_nir_type_for_glsl_type(const struct glsl_type
*type
)
867 return nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(type
));
870 nir_op
nir_type_conversion_op(nir_alu_type src
, nir_alu_type dst
,
871 nir_rounding_mode rnd
);
874 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
875 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
876 } nir_op_algebraic_property
;
884 * The number of components in the output
886 * If non-zero, this is the size of the output and input sizes are
887 * explicitly given; swizzle and writemask are still in effect, but if
888 * the output component is masked out, then the input component may
891 * If zero, the opcode acts in the standard, per-component manner; the
892 * operation is performed on each component (except the ones that are
893 * masked out) with the input being taken from the input swizzle for
896 * The size of some of the inputs may be given (i.e. non-zero) even
897 * though output_size is zero; in that case, the inputs with a zero
898 * size act per-component, while the inputs with non-zero size don't.
900 unsigned output_size
;
903 * The type of vector that the instruction outputs. Note that the
904 * staurate modifier is only allowed on outputs with the float type.
907 nir_alu_type output_type
;
910 * The number of components in each input
912 unsigned input_sizes
[NIR_MAX_VEC_COMPONENTS
];
915 * The type of vector that each input takes. Note that negate and
916 * absolute value are only allowed on inputs with int or float type and
917 * behave differently on the two.
919 nir_alu_type input_types
[NIR_MAX_VEC_COMPONENTS
];
921 nir_op_algebraic_property algebraic_properties
;
923 /* Whether this represents a numeric conversion opcode */
927 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
929 typedef struct nir_alu_instr
{
933 /** Indicates that this ALU instruction generates an exact value
935 * This is kind of a mixture of GLSL "precise" and "invariant" and not
936 * really equivalent to either. This indicates that the value generated by
937 * this operation is high-precision and any code transformations that touch
938 * it must ensure that the resulting value is bit-for-bit identical to the
947 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
948 nir_alu_instr
*instr
);
949 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
950 nir_alu_instr
*instr
);
952 /* is this source channel used? */
954 nir_alu_instr_channel_used(const nir_alu_instr
*instr
, unsigned src
,
957 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
958 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
960 return (instr
->dest
.write_mask
>> channel
) & 1;
963 static inline nir_component_mask_t
964 nir_alu_instr_src_read_mask(const nir_alu_instr
*instr
, unsigned src
)
966 nir_component_mask_t read_mask
= 0;
967 for (unsigned c
= 0; c
< NIR_MAX_VEC_COMPONENTS
; c
++) {
968 if (!nir_alu_instr_channel_used(instr
, src
, c
))
971 read_mask
|= (1 << instr
->src
[src
].swizzle
[c
]);
977 * For instructions whose destinations are SSA, get the number of channels
980 static inline unsigned
981 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
983 assert(instr
->dest
.dest
.is_ssa
);
985 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
986 return nir_op_infos
[instr
->op
].input_sizes
[src
];
988 return instr
->dest
.dest
.ssa
.num_components
;
991 bool nir_const_value_negative_equal(const nir_const_value
*c1
,
992 const nir_const_value
*c2
,
994 nir_alu_type base_type
,
997 bool nir_alu_srcs_equal(const nir_alu_instr
*alu1
, const nir_alu_instr
*alu2
,
998 unsigned src1
, unsigned src2
);
1002 nir_deref_type_array
,
1003 nir_deref_type_array_wildcard
,
1004 nir_deref_type_ptr_as_array
,
1005 nir_deref_type_struct
,
1006 nir_deref_type_cast
,
1012 /** The type of this deref instruction */
1013 nir_deref_type deref_type
;
1015 /** The mode of the underlying variable */
1016 nir_variable_mode mode
;
1018 /** The dereferenced type of the resulting pointer value */
1019 const struct glsl_type
*type
;
1022 /** Variable being dereferenced if deref_type is a deref_var */
1025 /** Parent deref if deref_type is not deref_var */
1029 /** Additional deref parameters */
1040 unsigned ptr_stride
;
1044 /** Destination to store the resulting "pointer" */
1048 NIR_DEFINE_CAST(nir_instr_as_deref
, nir_instr
, nir_deref_instr
, instr
,
1049 type
, nir_instr_type_deref
)
1051 static inline nir_deref_instr
*
1052 nir_src_as_deref(nir_src src
)
1057 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_deref
)
1060 return nir_instr_as_deref(src
.ssa
->parent_instr
);
1063 static inline nir_deref_instr
*
1064 nir_deref_instr_parent(const nir_deref_instr
*instr
)
1066 if (instr
->deref_type
== nir_deref_type_var
)
1069 return nir_src_as_deref(instr
->parent
);
1072 static inline nir_variable
*
1073 nir_deref_instr_get_variable(const nir_deref_instr
*instr
)
1075 while (instr
->deref_type
!= nir_deref_type_var
) {
1076 if (instr
->deref_type
== nir_deref_type_cast
)
1079 instr
= nir_deref_instr_parent(instr
);
1085 bool nir_deref_instr_has_indirect(nir_deref_instr
*instr
);
1087 bool nir_deref_instr_remove_if_unused(nir_deref_instr
*instr
);
1089 unsigned nir_deref_instr_ptr_as_array_stride(nir_deref_instr
*instr
);
1094 struct nir_function
*callee
;
1096 unsigned num_params
;
1100 #include "nir_intrinsics.h"
1102 #define NIR_INTRINSIC_MAX_CONST_INDEX 4
1104 /** Represents an intrinsic
1106 * An intrinsic is an instruction type for handling things that are
1107 * more-or-less regular operations but don't just consume and produce SSA
1108 * values like ALU operations do. Intrinsics are not for things that have
1109 * special semantic meaning such as phi nodes and parallel copies.
1110 * Examples of intrinsics include variable load/store operations, system
1111 * value loads, and the like. Even though texturing more-or-less falls
1112 * under this category, texturing is its own instruction type because
1113 * trying to represent texturing with intrinsics would lead to a
1114 * combinatorial explosion of intrinsic opcodes.
1116 * By having a single instruction type for handling a lot of different
1117 * cases, optimization passes can look for intrinsics and, for the most
1118 * part, completely ignore them. Each intrinsic type also has a few
1119 * possible flags that govern whether or not they can be reordered or
1120 * eliminated. That way passes like dead code elimination can still work
1121 * on intrisics without understanding the meaning of each.
1123 * Each intrinsic has some number of constant indices, some number of
1124 * variables, and some number of sources. What these sources, variables,
1125 * and indices mean depends on the intrinsic and is documented with the
1126 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
1127 * instructions are the only types of instruction that can operate on
1133 nir_intrinsic_op intrinsic
;
1137 /** number of components if this is a vectorized intrinsic
1139 * Similarly to ALU operations, some intrinsics are vectorized.
1140 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
1141 * For vectorized intrinsics, the num_components field specifies the
1142 * number of destination components and the number of source components
1143 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
1145 uint8_t num_components
;
1147 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
1150 } nir_intrinsic_instr
;
1152 static inline nir_variable
*
1153 nir_intrinsic_get_var(nir_intrinsic_instr
*intrin
, unsigned i
)
1155 return nir_deref_instr_get_variable(nir_src_as_deref(intrin
->src
[i
]));
1159 * \name NIR intrinsics semantic flags
1161 * information about what the compiler can do with the intrinsics.
1163 * \sa nir_intrinsic_info::flags
1167 * whether the intrinsic can be safely eliminated if none of its output
1168 * value is not being used.
1170 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
1173 * Whether the intrinsic can be reordered with respect to any other
1174 * intrinsic, i.e. whether the only reordering dependencies of the
1175 * intrinsic are due to the register reads/writes.
1177 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
1178 } nir_intrinsic_semantic_flag
;
1181 * \name NIR intrinsics const-index flag
1183 * Indicates the usage of a const_index slot.
1185 * \sa nir_intrinsic_info::index_map
1189 * Generally instructions that take a offset src argument, can encode
1190 * a constant 'base' value which is added to the offset.
1192 NIR_INTRINSIC_BASE
= 1,
1195 * For store instructions, a writemask for the store.
1197 NIR_INTRINSIC_WRMASK
= 2,
1200 * The stream-id for GS emit_vertex/end_primitive intrinsics.
1202 NIR_INTRINSIC_STREAM_ID
= 3,
1205 * The clip-plane id for load_user_clip_plane intrinsic.
1207 NIR_INTRINSIC_UCP_ID
= 4,
1210 * The amount of data, starting from BASE, that this instruction may
1211 * access. This is used to provide bounds if the offset is not constant.
1213 NIR_INTRINSIC_RANGE
= 5,
1216 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
1218 NIR_INTRINSIC_DESC_SET
= 6,
1221 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
1223 NIR_INTRINSIC_BINDING
= 7,
1228 NIR_INTRINSIC_COMPONENT
= 8,
1231 * Interpolation mode (only meaningful for FS inputs).
1233 NIR_INTRINSIC_INTERP_MODE
= 9,
1236 * A binary nir_op to use when performing a reduction or scan operation
1238 NIR_INTRINSIC_REDUCTION_OP
= 10,
1241 * Cluster size for reduction operations
1243 NIR_INTRINSIC_CLUSTER_SIZE
= 11,
1246 * Parameter index for a load_param intrinsic
1248 NIR_INTRINSIC_PARAM_IDX
= 12,
1251 * Image dimensionality for image intrinsics
1253 * One of GLSL_SAMPLER_DIM_*
1255 NIR_INTRINSIC_IMAGE_DIM
= 13,
1258 * Non-zero if we are accessing an array image
1260 NIR_INTRINSIC_IMAGE_ARRAY
= 14,
1263 * Image format for image intrinsics
1265 NIR_INTRINSIC_FORMAT
= 15,
1268 * Access qualifiers for image and memory access intrinsics
1270 NIR_INTRINSIC_ACCESS
= 16,
1273 * Alignment for offsets and addresses
1275 * These two parameters, specify an alignment in terms of a multiplier and
1276 * an offset. The offset or address parameter X of the intrinsic is
1277 * guaranteed to satisfy the following:
1279 * (X - align_offset) % align_mul == 0
1281 NIR_INTRINSIC_ALIGN_MUL
= 17,
1282 NIR_INTRINSIC_ALIGN_OFFSET
= 18,
1285 * The Vulkan descriptor type for a vulkan_resource_[re]index intrinsic.
1287 NIR_INTRINSIC_DESC_TYPE
= 19,
1289 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
1291 } nir_intrinsic_index_flag
;
1293 #define NIR_INTRINSIC_MAX_INPUTS 5
1298 unsigned num_srcs
; /** < number of register/SSA inputs */
1300 /** number of components of each input register
1302 * If this value is 0, the number of components is given by the
1303 * num_components field of nir_intrinsic_instr. If this value is -1, the
1304 * intrinsic consumes however many components are provided and it is not
1307 int src_components
[NIR_INTRINSIC_MAX_INPUTS
];
1311 /** number of components of the output register
1313 * If this value is 0, the number of components is given by the
1314 * num_components field of nir_intrinsic_instr.
1316 unsigned dest_components
;
1318 /** bitfield of legal bit sizes */
1319 unsigned dest_bit_sizes
;
1321 /** the number of constant indices used by the intrinsic */
1322 unsigned num_indices
;
1324 /** indicates the usage of intr->const_index[n] */
1325 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
1327 /** semantic flags for calls to this intrinsic */
1328 nir_intrinsic_semantic_flag flags
;
1329 } nir_intrinsic_info
;
1331 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
1333 static inline unsigned
1334 nir_intrinsic_src_components(nir_intrinsic_instr
*intr
, unsigned srcn
)
1336 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1337 assert(srcn
< info
->num_srcs
);
1338 if (info
->src_components
[srcn
] > 0)
1339 return info
->src_components
[srcn
];
1340 else if (info
->src_components
[srcn
] == 0)
1341 return intr
->num_components
;
1343 return nir_src_num_components(intr
->src
[srcn
]);
1346 static inline unsigned
1347 nir_intrinsic_dest_components(nir_intrinsic_instr
*intr
)
1349 const nir_intrinsic_info
*info
= &nir_intrinsic_infos
[intr
->intrinsic
];
1350 if (!info
->has_dest
)
1352 else if (info
->dest_components
)
1353 return info
->dest_components
;
1355 return intr
->num_components
;
1358 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
1359 static inline type \
1360 nir_intrinsic_##name(const nir_intrinsic_instr *instr) \
1362 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1363 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1364 return (type)instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
1366 static inline void \
1367 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1369 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1370 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1371 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1374 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1375 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1376 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1377 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1378 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1379 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1380 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1381 INTRINSIC_IDX_ACCESSORS(component
, COMPONENT
, unsigned)
1382 INTRINSIC_IDX_ACCESSORS(interp_mode
, INTERP_MODE
, unsigned)
1383 INTRINSIC_IDX_ACCESSORS(reduction_op
, REDUCTION_OP
, unsigned)
1384 INTRINSIC_IDX_ACCESSORS(cluster_size
, CLUSTER_SIZE
, unsigned)
1385 INTRINSIC_IDX_ACCESSORS(param_idx
, PARAM_IDX
, unsigned)
1386 INTRINSIC_IDX_ACCESSORS(image_dim
, IMAGE_DIM
, enum glsl_sampler_dim
)
1387 INTRINSIC_IDX_ACCESSORS(image_array
, IMAGE_ARRAY
, bool)
1388 INTRINSIC_IDX_ACCESSORS(access
, ACCESS
, enum gl_access_qualifier
)
1389 INTRINSIC_IDX_ACCESSORS(format
, FORMAT
, unsigned)
1390 INTRINSIC_IDX_ACCESSORS(align_mul
, ALIGN_MUL
, unsigned)
1391 INTRINSIC_IDX_ACCESSORS(align_offset
, ALIGN_OFFSET
, unsigned)
1392 INTRINSIC_IDX_ACCESSORS(desc_type
, DESC_TYPE
, unsigned)
1395 nir_intrinsic_set_align(nir_intrinsic_instr
*intrin
,
1396 unsigned align_mul
, unsigned align_offset
)
1398 assert(util_is_power_of_two_nonzero(align_mul
));
1399 assert(align_offset
< align_mul
);
1400 nir_intrinsic_set_align_mul(intrin
, align_mul
);
1401 nir_intrinsic_set_align_offset(intrin
, align_offset
);
1404 /** Returns a simple alignment for a load/store intrinsic offset
1406 * Instead of the full mul+offset alignment scheme provided by the ALIGN_MUL
1407 * and ALIGN_OFFSET parameters, this helper takes both into account and
1408 * provides a single simple alignment parameter. The offset X is guaranteed
1409 * to satisfy X % align == 0.
1411 static inline unsigned
1412 nir_intrinsic_align(const nir_intrinsic_instr
*intrin
)
1414 const unsigned align_mul
= nir_intrinsic_align_mul(intrin
);
1415 const unsigned align_offset
= nir_intrinsic_align_offset(intrin
);
1416 assert(align_offset
< align_mul
);
1417 return align_offset
? 1 << (ffs(align_offset
) - 1) : align_mul
;
1421 * \group texture information
1423 * This gives semantic information about textures which is useful to the
1424 * frontend, the backend, and lowering passes, but not the optimizer.
1429 nir_tex_src_projector
,
1430 nir_tex_src_comparator
, /* shadow comparator */
1434 nir_tex_src_min_lod
,
1435 nir_tex_src_ms_index
, /* MSAA sample index */
1436 nir_tex_src_ms_mcs
, /* MSAA compression value */
1439 nir_tex_src_texture_deref
, /* < deref pointing to the texture */
1440 nir_tex_src_sampler_deref
, /* < deref pointing to the sampler */
1441 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1442 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1443 nir_tex_src_texture_handle
, /* < bindless texture handle */
1444 nir_tex_src_sampler_handle
, /* < bindless sampler handle */
1445 nir_tex_src_plane
, /* < selects plane for planar textures */
1446 nir_num_tex_src_types
1451 nir_tex_src_type src_type
;
1455 nir_texop_tex
, /**< Regular texture look-up */
1456 nir_texop_txb
, /**< Texture look-up with LOD bias */
1457 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1458 nir_texop_txd
, /**< Texture look-up with partial derivatives */
1459 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1460 nir_texop_txf_ms
, /**< Multisample texture fetch */
1461 nir_texop_txf_ms_mcs
, /**< Multisample compression value fetch */
1462 nir_texop_txs
, /**< Texture size */
1463 nir_texop_lod
, /**< Texture lod query */
1464 nir_texop_tg4
, /**< Texture gather */
1465 nir_texop_query_levels
, /**< Texture levels query */
1466 nir_texop_texture_samples
, /**< Texture samples query */
1467 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1475 enum glsl_sampler_dim sampler_dim
;
1476 nir_alu_type dest_type
;
1481 unsigned num_srcs
, coord_components
;
1482 bool is_array
, is_shadow
;
1485 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1486 * components or the new-style shadow that outputs 1 component.
1488 bool is_new_style_shadow
;
1490 /* gather component selector */
1491 unsigned component
: 2;
1493 /* gather offsets */
1494 int8_t tg4_offsets
[4][2];
1496 /* True if the texture index or handle is not dynamically uniform */
1497 bool texture_non_uniform
;
1499 /* True if the sampler index or handle is not dynamically uniform */
1500 bool sampler_non_uniform
;
1502 /** The texture index
1504 * If this texture instruction has a nir_tex_src_texture_offset source,
1505 * then the texture index is given by texture_index + texture_offset.
1507 unsigned texture_index
;
1509 /** The size of the texture array or 0 if it's not an array */
1510 unsigned texture_array_size
;
1512 /** The sampler index
1514 * The following operations do not require a sampler and, as such, this
1515 * field should be ignored:
1517 * - nir_texop_txf_ms
1520 * - nir_texop_query_levels
1521 * - nir_texop_texture_samples
1522 * - nir_texop_samples_identical
1524 * If this texture instruction has a nir_tex_src_sampler_offset source,
1525 * then the sampler index is given by sampler_index + sampler_offset.
1527 unsigned sampler_index
;
1530 static inline unsigned
1531 nir_tex_instr_dest_size(const nir_tex_instr
*instr
)
1533 switch (instr
->op
) {
1534 case nir_texop_txs
: {
1536 switch (instr
->sampler_dim
) {
1537 case GLSL_SAMPLER_DIM_1D
:
1538 case GLSL_SAMPLER_DIM_BUF
:
1541 case GLSL_SAMPLER_DIM_2D
:
1542 case GLSL_SAMPLER_DIM_CUBE
:
1543 case GLSL_SAMPLER_DIM_MS
:
1544 case GLSL_SAMPLER_DIM_RECT
:
1545 case GLSL_SAMPLER_DIM_EXTERNAL
:
1546 case GLSL_SAMPLER_DIM_SUBPASS
:
1549 case GLSL_SAMPLER_DIM_3D
:
1553 unreachable("not reached");
1555 if (instr
->is_array
)
1563 case nir_texop_texture_samples
:
1564 case nir_texop_query_levels
:
1565 case nir_texop_samples_identical
:
1569 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1576 /* Returns true if this texture operation queries something about the texture
1577 * rather than actually sampling it.
1580 nir_tex_instr_is_query(const nir_tex_instr
*instr
)
1582 switch (instr
->op
) {
1585 case nir_texop_texture_samples
:
1586 case nir_texop_query_levels
:
1587 case nir_texop_txf_ms_mcs
:
1594 case nir_texop_txf_ms
:
1598 unreachable("Invalid texture opcode");
1603 nir_alu_instr_is_comparison(const nir_alu_instr
*instr
)
1605 switch (instr
->op
) {
1626 static inline nir_alu_type
1627 nir_tex_instr_src_type(const nir_tex_instr
*instr
, unsigned src
)
1629 switch (instr
->src
[src
].src_type
) {
1630 case nir_tex_src_coord
:
1631 switch (instr
->op
) {
1633 case nir_texop_txf_ms
:
1634 case nir_texop_txf_ms_mcs
:
1635 case nir_texop_samples_identical
:
1636 return nir_type_int
;
1639 return nir_type_float
;
1642 case nir_tex_src_lod
:
1643 switch (instr
->op
) {
1646 return nir_type_int
;
1649 return nir_type_float
;
1652 case nir_tex_src_projector
:
1653 case nir_tex_src_comparator
:
1654 case nir_tex_src_bias
:
1655 case nir_tex_src_ddx
:
1656 case nir_tex_src_ddy
:
1657 return nir_type_float
;
1659 case nir_tex_src_offset
:
1660 case nir_tex_src_ms_index
:
1661 case nir_tex_src_texture_offset
:
1662 case nir_tex_src_sampler_offset
:
1663 return nir_type_int
;
1666 unreachable("Invalid texture source type");
1670 static inline unsigned
1671 nir_tex_instr_src_size(const nir_tex_instr
*instr
, unsigned src
)
1673 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1674 return instr
->coord_components
;
1676 /* The MCS value is expected to be a vec4 returned by a txf_ms_mcs */
1677 if (instr
->src
[src
].src_type
== nir_tex_src_ms_mcs
)
1680 if (instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1681 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1682 if (instr
->is_array
)
1683 return instr
->coord_components
- 1;
1685 return instr
->coord_components
;
1688 /* Usual APIs don't allow cube + offset, but we allow it, with 2 coords for
1689 * the offset, since a cube maps to a single face.
1691 if (instr
->src
[src
].src_type
== nir_tex_src_offset
) {
1692 if (instr
->sampler_dim
== GLSL_SAMPLER_DIM_CUBE
)
1694 else if (instr
->is_array
)
1695 return instr
->coord_components
- 1;
1697 return instr
->coord_components
;
1704 nir_tex_instr_src_index(const nir_tex_instr
*instr
, nir_tex_src_type type
)
1706 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1707 if (instr
->src
[i
].src_type
== type
)
1713 void nir_tex_instr_add_src(nir_tex_instr
*tex
,
1714 nir_tex_src_type src_type
,
1717 void nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
);
1719 bool nir_tex_instr_has_explicit_tg4_offsets(nir_tex_instr
*tex
);
1724 nir_const_value value
;
1727 } nir_load_const_instr
;
1740 /* creates a new SSA variable in an undefined state */
1745 } nir_ssa_undef_instr
;
1748 struct exec_node node
;
1750 /* The predecessor block corresponding to this source */
1751 struct nir_block
*pred
;
1756 #define nir_foreach_phi_src(phi_src, phi) \
1757 foreach_list_typed(nir_phi_src, phi_src, node, &(phi)->srcs)
1758 #define nir_foreach_phi_src_safe(phi_src, phi) \
1759 foreach_list_typed_safe(nir_phi_src, phi_src, node, &(phi)->srcs)
1764 struct exec_list srcs
; /** < list of nir_phi_src */
1770 struct exec_node node
;
1773 } nir_parallel_copy_entry
;
1775 #define nir_foreach_parallel_copy_entry(entry, pcopy) \
1776 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1781 /* A list of nir_parallel_copy_entrys. The sources of all of the
1782 * entries are copied to the corresponding destinations "in parallel".
1783 * In other words, if we have two entries: a -> b and b -> a, the values
1786 struct exec_list entries
;
1787 } nir_parallel_copy_instr
;
1789 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
,
1790 type
, nir_instr_type_alu
)
1791 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
,
1792 type
, nir_instr_type_call
)
1793 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
,
1794 type
, nir_instr_type_jump
)
1795 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
,
1796 type
, nir_instr_type_tex
)
1797 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
,
1798 type
, nir_instr_type_intrinsic
)
1799 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
,
1800 type
, nir_instr_type_load_const
)
1801 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
,
1802 type
, nir_instr_type_ssa_undef
)
1803 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
,
1804 type
, nir_instr_type_phi
)
1805 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1806 nir_parallel_copy_instr
, instr
,
1807 type
, nir_instr_type_parallel_copy
)
1812 * Control flow consists of a tree of control flow nodes, which include
1813 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1814 * instructions that always run start-to-finish. Each basic block also keeps
1815 * track of its successors (blocks which may run immediately after the current
1816 * block) and predecessors (blocks which could have run immediately before the
1817 * current block). Each function also has a start block and an end block which
1818 * all return statements point to (which is always empty). Together, all the
1819 * blocks with their predecessors and successors make up the control flow
1820 * graph (CFG) of the function. There are helpers that modify the tree of
1821 * control flow nodes while modifying the CFG appropriately; these should be
1822 * used instead of modifying the tree directly.
1829 nir_cf_node_function
1832 typedef struct nir_cf_node
{
1833 struct exec_node node
;
1834 nir_cf_node_type type
;
1835 struct nir_cf_node
*parent
;
1838 typedef struct nir_block
{
1839 nir_cf_node cf_node
;
1841 struct exec_list instr_list
; /** < list of nir_instr */
1843 /** generic block index; generated by nir_index_blocks */
1847 * Each block can only have up to 2 successors, so we put them in a simple
1848 * array - no need for anything more complicated.
1850 struct nir_block
*successors
[2];
1852 /* Set of nir_block predecessors in the CFG */
1853 struct set
*predecessors
;
1856 * this node's immediate dominator in the dominance tree - set to NULL for
1859 struct nir_block
*imm_dom
;
1861 /* This node's children in the dominance tree */
1862 unsigned num_dom_children
;
1863 struct nir_block
**dom_children
;
1865 /* Set of nir_blocks on the dominance frontier of this block */
1866 struct set
*dom_frontier
;
1869 * These two indices have the property that dom_{pre,post}_index for each
1870 * child of this block in the dominance tree will always be between
1871 * dom_pre_index and dom_post_index for this block, which makes testing if
1872 * a given block is dominated by another block an O(1) operation.
1874 unsigned dom_pre_index
, dom_post_index
;
1876 /* live in and out for this block; used for liveness analysis */
1877 BITSET_WORD
*live_in
;
1878 BITSET_WORD
*live_out
;
1881 static inline nir_instr
*
1882 nir_block_first_instr(nir_block
*block
)
1884 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1885 return exec_node_data(nir_instr
, head
, node
);
1888 static inline nir_instr
*
1889 nir_block_last_instr(nir_block
*block
)
1891 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1892 return exec_node_data(nir_instr
, tail
, node
);
1896 nir_block_ends_in_jump(nir_block
*block
)
1898 return !exec_list_is_empty(&block
->instr_list
) &&
1899 nir_block_last_instr(block
)->type
== nir_instr_type_jump
;
1902 #define nir_foreach_instr(instr, block) \
1903 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1904 #define nir_foreach_instr_reverse(instr, block) \
1905 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1906 #define nir_foreach_instr_safe(instr, block) \
1907 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1908 #define nir_foreach_instr_reverse_safe(instr, block) \
1909 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1912 nir_selection_control_none
= 0x0,
1913 nir_selection_control_flatten
= 0x1,
1914 nir_selection_control_dont_flatten
= 0x2,
1915 } nir_selection_control
;
1917 typedef struct nir_if
{
1918 nir_cf_node cf_node
;
1920 nir_selection_control control
;
1922 struct exec_list then_list
; /** < list of nir_cf_node */
1923 struct exec_list else_list
; /** < list of nir_cf_node */
1929 /** Instruction that generates nif::condition. */
1930 nir_instr
*conditional_instr
;
1932 /** Block within ::nif that has the break instruction. */
1933 nir_block
*break_block
;
1935 /** Last block for the then- or else-path that does not contain the break. */
1936 nir_block
*continue_from_block
;
1938 /** True when ::break_block is in the else-path of ::nif. */
1939 bool continue_from_then
;
1942 /* This is true if the terminators exact trip count is unknown. For
1945 * for (int i = 0; i < imin(x, 4); i++)
1948 * Here loop analysis would have set a max_trip_count of 4 however we dont
1949 * know for sure that this is the exact trip count.
1951 bool exact_trip_count_unknown
;
1953 struct list_head loop_terminator_link
;
1954 } nir_loop_terminator
;
1957 /* Estimated cost (in number of instructions) of the loop */
1958 unsigned instr_cost
;
1960 /* Guessed trip count based on array indexing */
1961 unsigned guessed_trip_count
;
1963 /* Maximum number of times the loop is run (if known) */
1964 unsigned max_trip_count
;
1966 /* Do we know the exact number of times the loop will be run */
1967 bool exact_trip_count_known
;
1969 /* Unroll the loop regardless of its size */
1972 /* Does the loop contain complex loop terminators, continues or other
1973 * complex behaviours? If this is true we can't rely on
1974 * loop_terminator_list to be complete or accurate.
1978 nir_loop_terminator
*limiting_terminator
;
1980 /* A list of loop_terminators terminating this loop. */
1981 struct list_head loop_terminator_list
;
1985 nir_loop_control_none
= 0x0,
1986 nir_loop_control_unroll
= 0x1,
1987 nir_loop_control_dont_unroll
= 0x2,
1991 nir_cf_node cf_node
;
1993 struct exec_list body
; /** < list of nir_cf_node */
1995 nir_loop_info
*info
;
1996 nir_loop_control control
;
1997 bool partially_unrolled
;
2001 * Various bits of metadata that can may be created or required by
2002 * optimization and analysis passes
2005 nir_metadata_none
= 0x0,
2006 nir_metadata_block_index
= 0x1,
2007 nir_metadata_dominance
= 0x2,
2008 nir_metadata_live_ssa_defs
= 0x4,
2009 nir_metadata_not_properly_reset
= 0x8,
2010 nir_metadata_loop_analysis
= 0x10,
2014 nir_cf_node cf_node
;
2016 /** pointer to the function of which this is an implementation */
2017 struct nir_function
*function
;
2019 struct exec_list body
; /** < list of nir_cf_node */
2021 nir_block
*end_block
;
2023 /** list for all local variables in the function */
2024 struct exec_list locals
;
2026 /** list of local registers in the function */
2027 struct exec_list registers
;
2029 /** next available local register index */
2032 /** next available SSA value index */
2035 /* total number of basic blocks, only valid when block_index_dirty = false */
2036 unsigned num_blocks
;
2038 nir_metadata valid_metadata
;
2039 } nir_function_impl
;
2041 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
2042 nir_start_block(nir_function_impl
*impl
)
2044 return (nir_block
*) impl
->body
.head_sentinel
.next
;
2047 ATTRIBUTE_RETURNS_NONNULL
static inline nir_block
*
2048 nir_impl_last_block(nir_function_impl
*impl
)
2050 return (nir_block
*) impl
->body
.tail_sentinel
.prev
;
2053 static inline nir_cf_node
*
2054 nir_cf_node_next(nir_cf_node
*node
)
2056 struct exec_node
*next
= exec_node_get_next(&node
->node
);
2057 if (exec_node_is_tail_sentinel(next
))
2060 return exec_node_data(nir_cf_node
, next
, node
);
2063 static inline nir_cf_node
*
2064 nir_cf_node_prev(nir_cf_node
*node
)
2066 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
2067 if (exec_node_is_head_sentinel(prev
))
2070 return exec_node_data(nir_cf_node
, prev
, node
);
2074 nir_cf_node_is_first(const nir_cf_node
*node
)
2076 return exec_node_is_head_sentinel(node
->node
.prev
);
2080 nir_cf_node_is_last(const nir_cf_node
*node
)
2082 return exec_node_is_tail_sentinel(node
->node
.next
);
2085 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
,
2086 type
, nir_cf_node_block
)
2087 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
,
2088 type
, nir_cf_node_if
)
2089 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
,
2090 type
, nir_cf_node_loop
)
2091 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
,
2092 nir_function_impl
, cf_node
, type
, nir_cf_node_function
)
2094 static inline nir_block
*
2095 nir_if_first_then_block(nir_if
*if_stmt
)
2097 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
2098 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2101 static inline nir_block
*
2102 nir_if_last_then_block(nir_if
*if_stmt
)
2104 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
2105 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2108 static inline nir_block
*
2109 nir_if_first_else_block(nir_if
*if_stmt
)
2111 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
2112 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2115 static inline nir_block
*
2116 nir_if_last_else_block(nir_if
*if_stmt
)
2118 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
2119 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2122 static inline nir_block
*
2123 nir_loop_first_block(nir_loop
*loop
)
2125 struct exec_node
*head
= exec_list_get_head(&loop
->body
);
2126 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, head
, node
));
2129 static inline nir_block
*
2130 nir_loop_last_block(nir_loop
*loop
)
2132 struct exec_node
*tail
= exec_list_get_tail(&loop
->body
);
2133 return nir_cf_node_as_block(exec_node_data(nir_cf_node
, tail
, node
));
2137 uint8_t num_components
;
2141 typedef struct nir_function
{
2142 struct exec_node node
;
2145 struct nir_shader
*shader
;
2147 unsigned num_params
;
2148 nir_parameter
*params
;
2150 /** The implementation of this function.
2152 * If the function is only declared and not implemented, this is NULL.
2154 nir_function_impl
*impl
;
2160 nir_lower_imul64
= (1 << 0),
2161 nir_lower_isign64
= (1 << 1),
2162 /** Lower all int64 modulus and division opcodes */
2163 nir_lower_divmod64
= (1 << 2),
2164 /** Lower all 64-bit umul_high and imul_high opcodes */
2165 nir_lower_imul_high64
= (1 << 3),
2166 nir_lower_mov64
= (1 << 4),
2167 nir_lower_icmp64
= (1 << 5),
2168 nir_lower_iadd64
= (1 << 6),
2169 nir_lower_iabs64
= (1 << 7),
2170 nir_lower_ineg64
= (1 << 8),
2171 nir_lower_logic64
= (1 << 9),
2172 nir_lower_minmax64
= (1 << 10),
2173 nir_lower_shift64
= (1 << 11),
2174 nir_lower_imul_2x32_64
= (1 << 12),
2175 } nir_lower_int64_options
;
2178 nir_lower_drcp
= (1 << 0),
2179 nir_lower_dsqrt
= (1 << 1),
2180 nir_lower_drsq
= (1 << 2),
2181 nir_lower_dtrunc
= (1 << 3),
2182 nir_lower_dfloor
= (1 << 4),
2183 nir_lower_dceil
= (1 << 5),
2184 nir_lower_dfract
= (1 << 6),
2185 nir_lower_dround_even
= (1 << 7),
2186 nir_lower_dmod
= (1 << 8),
2187 nir_lower_fp64_full_software
= (1 << 9),
2188 } nir_lower_doubles_options
;
2190 typedef struct nir_shader_compiler_options
{
2196 /** Lowers flrp when it does not support doubles */
2204 /** Lowers ibitfield_extract/ubitfield_extract to ibfe/ubfe. */
2205 bool lower_bitfield_extract
;
2206 /** Lowers ibitfield_extract/ubitfield_extract to bfm, compares, shifts. */
2207 bool lower_bitfield_extract_to_shifts
;
2208 /** Lowers bitfield_insert to bfi/bfm */
2209 bool lower_bitfield_insert
;
2210 /** Lowers bitfield_insert to bfm, compares, and shifts. */
2211 bool lower_bitfield_insert_to_shifts
;
2212 /** Lowers bitfield_reverse to shifts. */
2213 bool lower_bitfield_reverse
;
2214 /** Lowers bit_count to shifts. */
2215 bool lower_bit_count
;
2216 /** Lowers bfm to shifts and subtracts. */
2218 /** Lowers ifind_msb to compare and ufind_msb */
2219 bool lower_ifind_msb
;
2220 /** Lowers find_lsb to ufind_msb and logic ops */
2221 bool lower_find_lsb
;
2222 bool lower_uadd_carry
;
2223 bool lower_usub_borrow
;
2224 /** Lowers imul_high/umul_high to 16-bit multiplies and carry operations. */
2225 bool lower_mul_high
;
2226 /** lowers fneg and ineg to fsub and isub. */
2228 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
2231 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
2234 /** enables rules to lower idiv by power-of-two: */
2237 /** enables rules to lower isign to imin+imax */
2240 /* Does the native fdot instruction replicate its result for four
2241 * components? If so, then opt_algebraic_late will turn all fdotN
2242 * instructions into fdot_replicatedN instructions.
2244 bool fdot_replicates
;
2246 /** lowers ffloor to fsub+ffract: */
2249 /** lowers ffract to fsub+ffloor: */
2252 /** lowers fceil to fneg+ffloor+fneg: */
2257 bool lower_pack_half_2x16
;
2258 bool lower_pack_unorm_2x16
;
2259 bool lower_pack_snorm_2x16
;
2260 bool lower_pack_unorm_4x8
;
2261 bool lower_pack_snorm_4x8
;
2262 bool lower_unpack_half_2x16
;
2263 bool lower_unpack_unorm_2x16
;
2264 bool lower_unpack_snorm_2x16
;
2265 bool lower_unpack_unorm_4x8
;
2266 bool lower_unpack_snorm_4x8
;
2268 bool lower_extract_byte
;
2269 bool lower_extract_word
;
2271 bool lower_all_io_to_temps
;
2274 * Does the driver support real 32-bit integers? (Otherwise, integers
2275 * are simulated by floats.)
2277 bool native_integers
;
2279 /* Indicates that the driver only has zero-based vertex id */
2280 bool vertex_id_zero_based
;
2283 * If enabled, gl_BaseVertex will be lowered as:
2284 * is_indexed_draw (~0/0) & firstvertex
2286 bool lower_base_vertex
;
2289 * If enabled, gl_HelperInvocation will be lowered as:
2291 * !((1 << sample_id) & sample_mask_in))
2293 * This depends on some possibly hw implementation details, which may
2294 * not be true for all hw. In particular that the FS is only executed
2295 * for covered samples or for helper invocations. So, do not blindly
2296 * enable this option.
2298 * Note: See also issue #22 in ARB_shader_image_load_store
2300 bool lower_helper_invocation
;
2302 bool lower_cs_local_index_from_id
;
2303 bool lower_cs_local_id_from_index
;
2305 bool lower_device_index_to_zero
;
2307 /* Set if nir_lower_wpos_ytransform() should also invert gl_PointCoord. */
2308 bool lower_wpos_pntc
;
2314 * Should nir_lower_io() create load_interpolated_input intrinsics?
2316 * If not, it generates regular load_input intrinsics and interpolation
2317 * information must be inferred from the list of input nir_variables.
2319 bool use_interpolated_input_intrinsics
;
2321 /* Lowers when 32x32->64 bit multiplication is not supported */
2322 bool lower_mul_2x32_64
;
2324 unsigned max_unroll_iterations
;
2326 nir_lower_int64_options lower_int64_options
;
2327 nir_lower_doubles_options lower_doubles_options
;
2328 } nir_shader_compiler_options
;
2330 typedef struct nir_shader
{
2331 /** list of uniforms (nir_variable) */
2332 struct exec_list uniforms
;
2334 /** list of inputs (nir_variable) */
2335 struct exec_list inputs
;
2337 /** list of outputs (nir_variable) */
2338 struct exec_list outputs
;
2340 /** list of shared compute variables (nir_variable) */
2341 struct exec_list shared
;
2343 /** Set of driver-specific options for the shader.
2345 * The memory for the options is expected to be kept in a single static
2346 * copy by the driver.
2348 const struct nir_shader_compiler_options
*options
;
2350 /** Various bits of compile-time information about a given shader */
2351 struct shader_info info
;
2353 /** list of global variables in the shader (nir_variable) */
2354 struct exec_list globals
;
2356 /** list of system value variables in the shader (nir_variable) */
2357 struct exec_list system_values
;
2359 struct exec_list functions
; /** < list of nir_function */
2361 /** list of global register in the shader */
2362 struct exec_list registers
;
2364 /** next available global register index */
2368 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
2371 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
2373 /** Constant data associated with this shader.
2375 * Constant data is loaded through load_constant intrinsics. See also
2376 * nir_opt_large_constants.
2378 void *constant_data
;
2379 unsigned constant_data_size
;
2382 #define nir_foreach_function(func, shader) \
2383 foreach_list_typed(nir_function, func, node, &(shader)->functions)
2385 static inline nir_function_impl
*
2386 nir_shader_get_entrypoint(nir_shader
*shader
)
2388 nir_function
*func
= NULL
;
2390 nir_foreach_function(function
, shader
) {
2391 assert(func
== NULL
);
2392 if (function
->is_entrypoint
) {
2403 assert(func
->num_params
== 0);
2408 nir_shader
*nir_shader_create(void *mem_ctx
,
2409 gl_shader_stage stage
,
2410 const nir_shader_compiler_options
*options
,
2413 /** creates a register, including assigning it an index and adding it to the list */
2414 nir_register
*nir_global_reg_create(nir_shader
*shader
);
2416 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
2418 void nir_reg_remove(nir_register
*reg
);
2420 /** Adds a variable to the appropriate list in nir_shader */
2421 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
2424 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
2426 assert(var
->data
.mode
== nir_var_function_temp
);
2427 exec_list_push_tail(&impl
->locals
, &var
->node
);
2430 /** creates a variable, sets a few defaults, and adds it to the list */
2431 nir_variable
*nir_variable_create(nir_shader
*shader
,
2432 nir_variable_mode mode
,
2433 const struct glsl_type
*type
,
2435 /** creates a local variable and adds it to the list */
2436 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
2437 const struct glsl_type
*type
,
2440 /** creates a function and adds it to the shader's list of functions */
2441 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
2443 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
2444 /** creates a function_impl that isn't tied to any particular function */
2445 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
2447 nir_block
*nir_block_create(nir_shader
*shader
);
2448 nir_if
*nir_if_create(nir_shader
*shader
);
2449 nir_loop
*nir_loop_create(nir_shader
*shader
);
2451 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
2453 /** requests that the given pieces of metadata be generated */
2454 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
, ...);
2455 /** dirties all but the preserved metadata */
2456 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
2458 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
2459 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
2461 nir_deref_instr
*nir_deref_instr_create(nir_shader
*shader
,
2462 nir_deref_type deref_type
);
2464 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
2466 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
2467 unsigned num_components
,
2470 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
2471 nir_intrinsic_op op
);
2473 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
2474 nir_function
*callee
);
2476 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
2478 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
2480 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
2482 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
2483 unsigned num_components
,
2486 nir_const_value
nir_alu_binop_identity(nir_op binop
, unsigned bit_size
);
2489 * NIR Cursors and Instruction Insertion API
2492 * A tiny struct representing a point to insert/extract instructions or
2493 * control flow nodes. Helps reduce the combinatorial explosion of possible
2494 * points to insert/extract.
2496 * \sa nir_control_flow.h
2499 nir_cursor_before_block
,
2500 nir_cursor_after_block
,
2501 nir_cursor_before_instr
,
2502 nir_cursor_after_instr
,
2503 } nir_cursor_option
;
2506 nir_cursor_option option
;
2513 static inline nir_block
*
2514 nir_cursor_current_block(nir_cursor cursor
)
2516 if (cursor
.option
== nir_cursor_before_instr
||
2517 cursor
.option
== nir_cursor_after_instr
) {
2518 return cursor
.instr
->block
;
2520 return cursor
.block
;
2524 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
2526 static inline nir_cursor
2527 nir_before_block(nir_block
*block
)
2530 cursor
.option
= nir_cursor_before_block
;
2531 cursor
.block
= block
;
2535 static inline nir_cursor
2536 nir_after_block(nir_block
*block
)
2539 cursor
.option
= nir_cursor_after_block
;
2540 cursor
.block
= block
;
2544 static inline nir_cursor
2545 nir_before_instr(nir_instr
*instr
)
2548 cursor
.option
= nir_cursor_before_instr
;
2549 cursor
.instr
= instr
;
2553 static inline nir_cursor
2554 nir_after_instr(nir_instr
*instr
)
2557 cursor
.option
= nir_cursor_after_instr
;
2558 cursor
.instr
= instr
;
2562 static inline nir_cursor
2563 nir_after_block_before_jump(nir_block
*block
)
2565 nir_instr
*last_instr
= nir_block_last_instr(block
);
2566 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
2567 return nir_before_instr(last_instr
);
2569 return nir_after_block(block
);
2573 static inline nir_cursor
2574 nir_before_src(nir_src
*src
, bool is_if_condition
)
2576 if (is_if_condition
) {
2577 nir_block
*prev_block
=
2578 nir_cf_node_as_block(nir_cf_node_prev(&src
->parent_if
->cf_node
));
2579 assert(!nir_block_ends_in_jump(prev_block
));
2580 return nir_after_block(prev_block
);
2581 } else if (src
->parent_instr
->type
== nir_instr_type_phi
) {
2583 nir_phi_instr
*cond_phi
= nir_instr_as_phi(src
->parent_instr
);
2585 nir_foreach_phi_src(phi_src
, cond_phi
) {
2586 if (phi_src
->src
.ssa
== src
->ssa
) {
2593 /* The LIST_ENTRY macro is a generic container-of macro, it just happens
2594 * to have a more specific name.
2596 nir_phi_src
*phi_src
= LIST_ENTRY(nir_phi_src
, src
, src
);
2597 return nir_after_block_before_jump(phi_src
->pred
);
2599 return nir_before_instr(src
->parent_instr
);
2603 static inline nir_cursor
2604 nir_before_cf_node(nir_cf_node
*node
)
2606 if (node
->type
== nir_cf_node_block
)
2607 return nir_before_block(nir_cf_node_as_block(node
));
2609 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
2612 static inline nir_cursor
2613 nir_after_cf_node(nir_cf_node
*node
)
2615 if (node
->type
== nir_cf_node_block
)
2616 return nir_after_block(nir_cf_node_as_block(node
));
2618 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
2621 static inline nir_cursor
2622 nir_after_phis(nir_block
*block
)
2624 nir_foreach_instr(instr
, block
) {
2625 if (instr
->type
!= nir_instr_type_phi
)
2626 return nir_before_instr(instr
);
2628 return nir_after_block(block
);
2631 static inline nir_cursor
2632 nir_after_cf_node_and_phis(nir_cf_node
*node
)
2634 if (node
->type
== nir_cf_node_block
)
2635 return nir_after_block(nir_cf_node_as_block(node
));
2637 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
2639 return nir_after_phis(block
);
2642 static inline nir_cursor
2643 nir_before_cf_list(struct exec_list
*cf_list
)
2645 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
2646 exec_list_get_head(cf_list
), node
);
2647 return nir_before_cf_node(first_node
);
2650 static inline nir_cursor
2651 nir_after_cf_list(struct exec_list
*cf_list
)
2653 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
2654 exec_list_get_tail(cf_list
), node
);
2655 return nir_after_cf_node(last_node
);
2659 * Insert a NIR instruction at the given cursor.
2661 * Note: This does not update the cursor.
2663 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
2666 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
2668 nir_instr_insert(nir_before_instr(instr
), before
);
2672 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2674 nir_instr_insert(nir_after_instr(instr
), after
);
2678 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2680 nir_instr_insert(nir_before_block(block
), before
);
2684 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2686 nir_instr_insert(nir_after_block(block
), after
);
2690 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2692 nir_instr_insert(nir_before_cf_node(node
), before
);
2696 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2698 nir_instr_insert(nir_after_cf_node(node
), after
);
2702 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2704 nir_instr_insert(nir_before_cf_list(list
), before
);
2708 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2710 nir_instr_insert(nir_after_cf_list(list
), after
);
2713 void nir_instr_remove_v(nir_instr
*instr
);
2715 static inline nir_cursor
2716 nir_instr_remove(nir_instr
*instr
)
2719 nir_instr
*prev
= nir_instr_prev(instr
);
2721 cursor
= nir_after_instr(prev
);
2723 cursor
= nir_before_block(instr
->block
);
2725 nir_instr_remove_v(instr
);
2731 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2732 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2733 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2734 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2736 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2737 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2739 nir_const_value
*nir_src_as_const_value(nir_src src
);
2741 static inline struct nir_instr
*
2742 nir_src_instr(const struct nir_src
*src
)
2744 return src
->is_ssa
? src
->ssa
->parent_instr
: NULL
;
2747 #define NIR_SRC_AS_(name, c_type, type_enum, cast_macro) \
2748 static inline c_type * \
2749 nir_src_as_ ## name (struct nir_src *src) \
2751 return src->is_ssa && src->ssa->parent_instr->type == type_enum \
2752 ? cast_macro(src->ssa->parent_instr) : NULL; \
2754 static inline const c_type * \
2755 nir_src_as_ ## name ## _const(const struct nir_src *src) \
2757 return src->is_ssa && src->ssa->parent_instr->type == type_enum \
2758 ? cast_macro(src->ssa->parent_instr) : NULL; \
2761 NIR_SRC_AS_(alu_instr
, nir_alu_instr
, nir_instr_type_alu
, nir_instr_as_alu
)
2763 bool nir_src_is_dynamically_uniform(nir_src src
);
2764 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2765 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2766 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2767 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2768 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2771 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2772 unsigned num_components
, unsigned bit_size
,
2774 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2775 unsigned num_components
, unsigned bit_size
,
2778 nir_ssa_dest_init_for_type(nir_instr
*instr
, nir_dest
*dest
,
2779 const struct glsl_type
*type
,
2782 assert(glsl_type_is_vector_or_scalar(type
));
2783 nir_ssa_dest_init(instr
, dest
, glsl_get_components(type
),
2784 glsl_get_bit_size(type
), name
);
2786 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2787 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2788 nir_instr
*after_me
);
2790 nir_component_mask_t
nir_ssa_def_components_read(const nir_ssa_def
*def
);
2793 * finds the next basic block in source-code order, returns NULL if there is
2797 nir_block
*nir_block_cf_tree_next(nir_block
*block
);
2799 /* Performs the opposite of nir_block_cf_tree_next() */
2801 nir_block
*nir_block_cf_tree_prev(nir_block
*block
);
2803 /* Gets the first block in a CF node in source-code order */
2805 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
);
2807 /* Gets the last block in a CF node in source-code order */
2809 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
);
2811 /* Gets the next block after a CF node in source-code order */
2813 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
);
2815 /* Macros for loops that visit blocks in source-code order */
2817 #define nir_foreach_block(block, impl) \
2818 for (nir_block *block = nir_start_block(impl); block != NULL; \
2819 block = nir_block_cf_tree_next(block))
2821 #define nir_foreach_block_safe(block, impl) \
2822 for (nir_block *block = nir_start_block(impl), \
2823 *next = nir_block_cf_tree_next(block); \
2825 block = next, next = nir_block_cf_tree_next(block))
2827 #define nir_foreach_block_reverse(block, impl) \
2828 for (nir_block *block = nir_impl_last_block(impl); block != NULL; \
2829 block = nir_block_cf_tree_prev(block))
2831 #define nir_foreach_block_reverse_safe(block, impl) \
2832 for (nir_block *block = nir_impl_last_block(impl), \
2833 *prev = nir_block_cf_tree_prev(block); \
2835 block = prev, prev = nir_block_cf_tree_prev(block))
2837 #define nir_foreach_block_in_cf_node(block, node) \
2838 for (nir_block *block = nir_cf_node_cf_tree_first(node); \
2839 block != nir_cf_node_cf_tree_next(node); \
2840 block = nir_block_cf_tree_next(block))
2842 /* If the following CF node is an if, this function returns that if.
2843 * Otherwise, it returns NULL.
2845 nir_if
*nir_block_get_following_if(nir_block
*block
);
2847 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2849 void nir_index_local_regs(nir_function_impl
*impl
);
2850 void nir_index_global_regs(nir_shader
*shader
);
2851 void nir_index_ssa_defs(nir_function_impl
*impl
);
2852 unsigned nir_index_instrs(nir_function_impl
*impl
);
2854 void nir_index_blocks(nir_function_impl
*impl
);
2856 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2857 void nir_print_shader_annotated(nir_shader
*shader
, FILE *fp
, struct hash_table
*errors
);
2858 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2859 void nir_print_deref(const nir_deref_instr
*deref
, FILE *fp
);
2861 nir_shader
*nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2862 nir_function_impl
*nir_function_impl_clone(nir_shader
*shader
,
2863 const nir_function_impl
*fi
);
2864 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2865 nir_variable
*nir_variable_clone(const nir_variable
*c
, nir_shader
*shader
);
2867 nir_shader
*nir_shader_serialize_deserialize(void *mem_ctx
, nir_shader
*s
);
2870 void nir_validate_shader(nir_shader
*shader
, const char *when
);
2871 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2872 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2875 should_skip_nir(const char *name
)
2877 static const char *list
= NULL
;
2879 /* Comma separated list of names to skip. */
2880 list
= getenv("NIR_SKIP");
2888 return comma_separated_list_contains(list
, name
);
2892 should_clone_nir(void)
2894 static int should_clone
= -1;
2895 if (should_clone
< 0)
2896 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2898 return should_clone
;
2902 should_serialize_deserialize_nir(void)
2904 static int test_serialize
= -1;
2905 if (test_serialize
< 0)
2906 test_serialize
= env_var_as_boolean("NIR_TEST_SERIALIZE", false);
2908 return test_serialize
;
2912 should_print_nir(void)
2914 static int should_print
= -1;
2915 if (should_print
< 0)
2916 should_print
= env_var_as_boolean("NIR_PRINT", false);
2918 return should_print
;
2921 static inline void nir_validate_shader(nir_shader
*shader
, const char *when
) { (void) shader
; (void)when
; }
2922 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2923 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2924 static inline bool should_skip_nir(UNUSED
const char *pass_name
) { return false; }
2925 static inline bool should_clone_nir(void) { return false; }
2926 static inline bool should_serialize_deserialize_nir(void) { return false; }
2927 static inline bool should_print_nir(void) { return false; }
2930 #define _PASS(pass, nir, do_pass) do { \
2931 if (should_skip_nir(#pass)) { \
2932 printf("skipping %s\n", #pass); \
2936 nir_validate_shader(nir, "after " #pass); \
2937 if (should_clone_nir()) { \
2938 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2942 if (should_serialize_deserialize_nir()) { \
2943 void *mem_ctx = ralloc_parent(nir); \
2944 nir = nir_shader_serialize_deserialize(mem_ctx, nir); \
2948 #define NIR_PASS(progress, nir, pass, ...) _PASS(pass, nir, \
2949 nir_metadata_set_validation_flag(nir); \
2950 if (should_print_nir()) \
2951 printf("%s\n", #pass); \
2952 if (pass(nir, ##__VA_ARGS__)) { \
2954 if (should_print_nir()) \
2955 nir_print_shader(nir, stdout); \
2956 nir_metadata_check_validation_flag(nir); \
2960 #define NIR_PASS_V(nir, pass, ...) _PASS(pass, nir, \
2961 if (should_print_nir()) \
2962 printf("%s\n", #pass); \
2963 pass(nir, ##__VA_ARGS__); \
2964 if (should_print_nir()) \
2965 nir_print_shader(nir, stdout); \
2968 #define NIR_SKIP(name) should_skip_nir(#name)
2970 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2971 void nir_calc_dominance(nir_shader
*shader
);
2973 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2974 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2976 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2977 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2979 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2980 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2982 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2983 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2985 int nir_gs_count_vertices(const nir_shader
*shader
);
2987 bool nir_shrink_vec_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2988 bool nir_split_array_vars(nir_shader
*shader
, nir_variable_mode modes
);
2989 bool nir_split_var_copies(nir_shader
*shader
);
2990 bool nir_split_per_member_structs(nir_shader
*shader
);
2991 bool nir_split_struct_vars(nir_shader
*shader
, nir_variable_mode modes
);
2993 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2994 bool nir_lower_returns(nir_shader
*shader
);
2996 void nir_inline_function_impl(struct nir_builder
*b
,
2997 const nir_function_impl
*impl
,
2998 nir_ssa_def
**params
);
2999 bool nir_inline_functions(nir_shader
*shader
);
3001 bool nir_propagate_invariant(nir_shader
*shader
);
3003 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, nir_shader
*shader
);
3004 void nir_lower_deref_copy_instr(struct nir_builder
*b
,
3005 nir_intrinsic_instr
*copy
);
3006 bool nir_lower_var_copies(nir_shader
*shader
);
3008 void nir_fixup_deref_modes(nir_shader
*shader
);
3010 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
3013 nir_lower_direct_array_deref_of_vec_load
= (1 << 0),
3014 nir_lower_indirect_array_deref_of_vec_load
= (1 << 1),
3015 nir_lower_direct_array_deref_of_vec_store
= (1 << 2),
3016 nir_lower_indirect_array_deref_of_vec_store
= (1 << 3),
3017 } nir_lower_array_deref_of_vec_options
;
3019 bool nir_lower_array_deref_of_vec(nir_shader
*shader
, nir_variable_mode modes
,
3020 nir_lower_array_deref_of_vec_options options
);
3022 bool nir_lower_indirect_derefs(nir_shader
*shader
, nir_variable_mode modes
);
3024 bool nir_lower_locals_to_regs(nir_shader
*shader
);
3026 void nir_lower_io_to_temporaries(nir_shader
*shader
,
3027 nir_function_impl
*entrypoint
,
3028 bool outputs
, bool inputs
);
3030 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
3032 void nir_assign_var_locations(struct exec_list
*var_list
, unsigned *size
,
3033 int (*type_size
)(const struct glsl_type
*));
3035 /* Some helpers to do very simple linking */
3036 bool nir_remove_unused_varyings(nir_shader
*producer
, nir_shader
*consumer
);
3037 bool nir_remove_unused_io_vars(nir_shader
*shader
, struct exec_list
*var_list
,
3038 uint64_t *used_by_other_stage
,
3039 uint64_t *used_by_other_stage_patches
);
3040 void nir_compact_varyings(nir_shader
*producer
, nir_shader
*consumer
,
3041 bool default_to_smooth_interp
);
3042 void nir_link_xfb_varyings(nir_shader
*producer
, nir_shader
*consumer
);
3043 bool nir_link_opt_varyings(nir_shader
*producer
, nir_shader
*consumer
);
3046 /* If set, this forces all non-flat fragment shader inputs to be
3047 * interpolated as if with the "sample" qualifier. This requires
3048 * nir_shader_compiler_options::use_interpolated_input_intrinsics.
3050 nir_lower_io_force_sample_interpolation
= (1 << 1),
3051 } nir_lower_io_options
;
3052 bool nir_lower_io(nir_shader
*shader
,
3053 nir_variable_mode modes
,
3054 int (*type_size
)(const struct glsl_type
*),
3055 nir_lower_io_options
);
3059 * An address format which is a simple 32-bit global GPU address.
3061 nir_address_format_32bit_global
,
3064 * An address format which is a simple 64-bit global GPU address.
3066 nir_address_format_64bit_global
,
3069 * An address format which is a bounds-checked 64-bit global GPU address.
3071 * The address is comprised as a 32-bit vec4 where .xy are a uint64_t base
3072 * address stored with the low bits in .x and high bits in .y, .z is a
3073 * size, and .w is an offset. When the final I/O operation is lowered, .w
3074 * is checked against .z and the operation is predicated on the result.
3076 nir_address_format_64bit_bounded_global
,
3079 * An address format which is comprised of a vec2 where the first
3080 * component is a buffer index and the second is an offset.
3082 nir_address_format_32bit_index_offset
,
3083 } nir_address_format
;
3084 bool nir_lower_explicit_io(nir_shader
*shader
,
3085 nir_variable_mode modes
,
3086 nir_address_format
);
3088 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
3089 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
3091 bool nir_is_per_vertex_io(const nir_variable
*var
, gl_shader_stage stage
);
3093 bool nir_lower_regs_to_ssa_impl(nir_function_impl
*impl
);
3094 bool nir_lower_regs_to_ssa(nir_shader
*shader
);
3095 bool nir_lower_vars_to_ssa(nir_shader
*shader
);
3097 bool nir_remove_dead_derefs(nir_shader
*shader
);
3098 bool nir_remove_dead_derefs_impl(nir_function_impl
*impl
);
3099 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode modes
);
3100 bool nir_lower_constant_initializers(nir_shader
*shader
,
3101 nir_variable_mode modes
);
3103 bool nir_move_load_const(nir_shader
*shader
);
3104 bool nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
3105 bool nir_lower_vec_to_movs(nir_shader
*shader
);
3106 void nir_lower_alpha_test(nir_shader
*shader
, enum compare_func func
,
3108 bool nir_lower_alu(nir_shader
*shader
);
3109 bool nir_lower_alu_to_scalar(nir_shader
*shader
);
3110 bool nir_lower_bool_to_float(nir_shader
*shader
);
3111 bool nir_lower_bool_to_int32(nir_shader
*shader
);
3112 bool nir_lower_load_const_to_scalar(nir_shader
*shader
);
3113 bool nir_lower_read_invocation_to_scalar(nir_shader
*shader
);
3114 bool nir_lower_phis_to_scalar(nir_shader
*shader
);
3115 void nir_lower_io_arrays_to_elements(nir_shader
*producer
, nir_shader
*consumer
);
3116 void nir_lower_io_arrays_to_elements_no_indirects(nir_shader
*shader
,
3118 void nir_lower_io_to_scalar(nir_shader
*shader
, nir_variable_mode mask
);
3119 void nir_lower_io_to_scalar_early(nir_shader
*shader
, nir_variable_mode mask
);
3120 bool nir_lower_io_to_vector(nir_shader
*shader
, nir_variable_mode mask
);
3122 bool nir_lower_uniforms_to_ubo(nir_shader
*shader
, int multiplier
);
3124 typedef struct nir_lower_subgroups_options
{
3125 uint8_t subgroup_size
;
3126 uint8_t ballot_bit_size
;
3127 bool lower_to_scalar
:1;
3128 bool lower_vote_trivial
:1;
3129 bool lower_vote_eq_to_ballot
:1;
3130 bool lower_subgroup_masks
:1;
3131 bool lower_shuffle
:1;
3132 bool lower_shuffle_to_32bit
:1;
3134 } nir_lower_subgroups_options
;
3136 bool nir_lower_subgroups(nir_shader
*shader
,
3137 const nir_lower_subgroups_options
*options
);
3139 bool nir_lower_system_values(nir_shader
*shader
);
3141 enum PACKED nir_lower_tex_packing
{
3142 nir_lower_tex_packing_none
= 0,
3143 /* The sampler returns up to 2 32-bit words of half floats or 16-bit signed
3144 * or unsigned ints based on the sampler type
3146 nir_lower_tex_packing_16
,
3147 /* The sampler returns 1 32-bit word of 4x8 unorm */
3148 nir_lower_tex_packing_8
,
3151 typedef struct nir_lower_tex_options
{
3153 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
3154 * sampler types a texture projector is lowered.
3159 * If true, lower away nir_tex_src_offset for all texelfetch instructions.
3161 bool lower_txf_offset
;
3164 * If true, lower away nir_tex_src_offset for all rect textures.
3166 bool lower_rect_offset
;
3169 * If true, lower rect textures to 2D, using txs to fetch the
3170 * texture dimensions and dividing the texture coords by the
3171 * texture dims to normalize.
3176 * If true, convert yuv to rgb.
3178 unsigned lower_y_uv_external
;
3179 unsigned lower_y_u_v_external
;
3180 unsigned lower_yx_xuxv_external
;
3181 unsigned lower_xy_uxvx_external
;
3182 unsigned lower_ayuv_external
;
3183 unsigned lower_xyuv_external
;
3186 * To emulate certain texture wrap modes, this can be used
3187 * to saturate the specified tex coord to [0.0, 1.0]. The
3188 * bits are according to sampler #, ie. if, for example:
3190 * (conf->saturate_s & (1 << n))
3192 * is true, then the s coord for sampler n is saturated.
3194 * Note that clamping must happen *after* projector lowering
3195 * so any projected texture sample instruction with a clamped
3196 * coordinate gets automatically lowered, regardless of the
3197 * 'lower_txp' setting.
3199 unsigned saturate_s
;
3200 unsigned saturate_t
;
3201 unsigned saturate_r
;
3203 /* Bitmask of textures that need swizzling.
3205 * If (swizzle_result & (1 << texture_index)), then the swizzle in
3206 * swizzles[texture_index] is applied to the result of the texturing
3209 unsigned swizzle_result
;
3211 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
3212 * while 4 and 5 represent 0 and 1 respectively.
3214 uint8_t swizzles
[32][4];
3216 /* Can be used to scale sampled values in range required by the format. */
3217 float scale_factors
[32];
3220 * Bitmap of textures that need srgb to linear conversion. If
3221 * (lower_srgb & (1 << texture_index)) then the rgb (xyz) components
3222 * of the texture are lowered to linear.
3224 unsigned lower_srgb
;
3227 * If true, lower nir_texop_txd on cube maps with nir_texop_txl.
3229 bool lower_txd_cube_map
;
3232 * If true, lower nir_texop_txd on 3D surfaces with nir_texop_txl.
3237 * If true, lower nir_texop_txd on shadow samplers (except cube maps)
3238 * with nir_texop_txl. Notice that cube map shadow samplers are lowered
3239 * with lower_txd_cube_map.
3241 bool lower_txd_shadow
;
3244 * If true, lower nir_texop_txd on all samplers to a nir_texop_txl.
3245 * Implies lower_txd_cube_map and lower_txd_shadow.
3250 * If true, lower nir_texop_txb that try to use shadow compare and min_lod
3251 * at the same time to a nir_texop_lod, some math, and nir_texop_tex.
3253 bool lower_txb_shadow_clamp
;
3256 * If true, lower nir_texop_txd on shadow samplers when it uses min_lod
3257 * with nir_texop_txl. This includes cube maps.
3259 bool lower_txd_shadow_clamp
;
3262 * If true, lower nir_texop_txd on when it uses both offset and min_lod
3263 * with nir_texop_txl. This includes cube maps.
3265 bool lower_txd_offset_clamp
;
3268 * If true, lower nir_texop_txd with min_lod to a nir_texop_txl if the
3269 * sampler index is not statically determinable to be less than 16.
3271 bool lower_txd_clamp_if_sampler_index_not_lt_16
;
3274 * If true, apply a .bagr swizzle on tg4 results to handle Broadcom's
3275 * mixed-up tg4 locations.
3277 bool lower_tg4_broadcom_swizzle
;
3280 * If true, lowers tg4 with 4 constant offsets to 4 tg4 calls
3282 bool lower_tg4_offsets
;
3284 enum nir_lower_tex_packing lower_tex_packing
[32];
3285 } nir_lower_tex_options
;
3287 bool nir_lower_tex(nir_shader
*shader
,
3288 const nir_lower_tex_options
*options
);
3290 enum nir_lower_non_uniform_access_type
{
3291 nir_lower_non_uniform_ubo_access
= (1 << 0),
3292 nir_lower_non_uniform_ssbo_access
= (1 << 1),
3293 nir_lower_non_uniform_texture_access
= (1 << 2),
3294 nir_lower_non_uniform_image_access
= (1 << 3),
3297 bool nir_lower_non_uniform_access(nir_shader
*shader
,
3298 enum nir_lower_non_uniform_access_type
);
3300 bool nir_lower_idiv(nir_shader
*shader
);
3302 bool nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
, bool use_vars
);
3303 bool nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
3304 bool nir_lower_clip_cull_distance_arrays(nir_shader
*nir
);
3306 bool nir_lower_frexp(nir_shader
*nir
);
3308 void nir_lower_two_sided_color(nir_shader
*shader
);
3310 bool nir_lower_clamp_color_outputs(nir_shader
*shader
);
3312 void nir_lower_passthrough_edgeflags(nir_shader
*shader
);
3313 bool nir_lower_patch_vertices(nir_shader
*nir
, unsigned static_count
,
3314 const gl_state_index16
*uniform_state_tokens
);
3316 typedef struct nir_lower_wpos_ytransform_options
{
3317 gl_state_index16 state_tokens
[STATE_LENGTH
];
3318 bool fs_coord_origin_upper_left
:1;
3319 bool fs_coord_origin_lower_left
:1;
3320 bool fs_coord_pixel_center_integer
:1;
3321 bool fs_coord_pixel_center_half_integer
:1;
3322 } nir_lower_wpos_ytransform_options
;
3324 bool nir_lower_wpos_ytransform(nir_shader
*shader
,
3325 const nir_lower_wpos_ytransform_options
*options
);
3326 bool nir_lower_wpos_center(nir_shader
*shader
, const bool for_sample_shading
);
3328 typedef struct nir_lower_drawpixels_options
{
3329 gl_state_index16 texcoord_state_tokens
[STATE_LENGTH
];
3330 gl_state_index16 scale_state_tokens
[STATE_LENGTH
];
3331 gl_state_index16 bias_state_tokens
[STATE_LENGTH
];
3332 unsigned drawpix_sampler
;
3333 unsigned pixelmap_sampler
;
3335 bool scale_and_bias
:1;
3336 } nir_lower_drawpixels_options
;
3338 void nir_lower_drawpixels(nir_shader
*shader
,
3339 const nir_lower_drawpixels_options
*options
);
3341 typedef struct nir_lower_bitmap_options
{
3344 } nir_lower_bitmap_options
;
3346 void nir_lower_bitmap(nir_shader
*shader
, const nir_lower_bitmap_options
*options
);
3348 bool nir_lower_atomics_to_ssbo(nir_shader
*shader
, unsigned ssbo_offset
);
3351 nir_lower_int_source_mods
= 1 << 0,
3352 nir_lower_float_source_mods
= 1 << 1,
3353 nir_lower_triop_abs
= 1 << 2,
3354 nir_lower_all_source_mods
= (1 << 3) - 1
3355 } nir_lower_to_source_mods_flags
;
3358 bool nir_lower_to_source_mods(nir_shader
*shader
, nir_lower_to_source_mods_flags options
);
3360 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
3362 typedef unsigned (*nir_lower_bit_size_callback
)(const nir_alu_instr
*, void *);
3364 bool nir_lower_bit_size(nir_shader
*shader
,
3365 nir_lower_bit_size_callback callback
,
3366 void *callback_data
);
3368 nir_lower_int64_options
nir_lower_int64_op_to_options_mask(nir_op opcode
);
3369 bool nir_lower_int64(nir_shader
*shader
, nir_lower_int64_options options
);
3371 nir_lower_doubles_options
nir_lower_doubles_op_to_options_mask(nir_op opcode
);
3372 bool nir_lower_doubles(nir_shader
*shader
, const nir_shader
*softfp64
,
3373 nir_lower_doubles_options options
);
3374 bool nir_lower_pack(nir_shader
*shader
);
3376 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
3378 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
3380 void nir_loop_analyze_impl(nir_function_impl
*impl
,
3381 nir_variable_mode indirect_mask
);
3383 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
3385 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
3386 bool nir_repair_ssa(nir_shader
*shader
);
3388 void nir_convert_loop_to_lcssa(nir_loop
*loop
);
3390 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
3391 * registers. If false, convert all values (even those not involved in a phi
3392 * node) to registers.
3394 bool nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
3396 bool nir_lower_phis_to_regs_block(nir_block
*block
);
3397 bool nir_lower_ssa_defs_to_regs_block(nir_block
*block
);
3398 bool nir_rematerialize_derefs_in_use_blocks_impl(nir_function_impl
*impl
);
3400 bool nir_opt_algebraic(nir_shader
*shader
);
3401 bool nir_opt_algebraic_before_ffma(nir_shader
*shader
);
3402 bool nir_opt_algebraic_late(nir_shader
*shader
);
3403 bool nir_opt_constant_folding(nir_shader
*shader
);
3405 bool nir_opt_combine_stores(nir_shader
*shader
, nir_variable_mode modes
);
3407 bool nir_opt_global_to_local(nir_shader
*shader
);
3409 bool nir_copy_prop(nir_shader
*shader
);
3411 bool nir_opt_copy_prop_vars(nir_shader
*shader
);
3413 bool nir_opt_cse(nir_shader
*shader
);
3415 bool nir_opt_dce(nir_shader
*shader
);
3417 bool nir_opt_dead_cf(nir_shader
*shader
);
3419 bool nir_opt_dead_write_vars(nir_shader
*shader
);
3421 bool nir_opt_deref_impl(nir_function_impl
*impl
);
3422 bool nir_opt_deref(nir_shader
*shader
);
3424 bool nir_opt_find_array_copies(nir_shader
*shader
);
3426 bool nir_opt_gcm(nir_shader
*shader
, bool value_number
);
3428 bool nir_opt_idiv_const(nir_shader
*shader
, unsigned min_bit_size
);
3430 bool nir_opt_if(nir_shader
*shader
);
3432 bool nir_opt_intrinsics(nir_shader
*shader
);
3434 bool nir_opt_large_constants(nir_shader
*shader
,
3435 glsl_type_size_align_func size_align
,
3436 unsigned threshold
);
3438 bool nir_opt_loop_unroll(nir_shader
*shader
, nir_variable_mode indirect_mask
);
3440 bool nir_opt_move_comparisons(nir_shader
*shader
);
3442 bool nir_opt_move_load_ubo(nir_shader
*shader
);
3444 bool nir_opt_peephole_select(nir_shader
*shader
, unsigned limit
,
3445 bool indirect_load_ok
, bool expensive_alu_ok
);
3447 bool nir_opt_remove_phis(nir_shader
*shader
);
3449 bool nir_opt_shrink_load(nir_shader
*shader
);
3451 bool nir_opt_trivial_continues(nir_shader
*shader
);
3453 bool nir_opt_undef(nir_shader
*shader
);
3455 bool nir_opt_conditional_discard(nir_shader
*shader
);
3457 void nir_strip(nir_shader
*shader
);
3459 void nir_sweep(nir_shader
*shader
);
3461 void nir_remap_dual_slot_attributes(nir_shader
*shader
,
3462 uint64_t *dual_slot_inputs
);
3463 uint64_t nir_get_single_slot_attribs_mask(uint64_t attribs
, uint64_t dual_slot
);
3465 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
3466 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);