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)
30 #include "util/hash_table.h"
31 #include "compiler/glsl/list.h"
32 #include "GL/gl.h" /* GLenum */
33 #include "util/list.h"
34 #include "util/ralloc.h"
36 #include "util/bitset.h"
37 #include "compiler/nir_types.h"
38 #include "compiler/shader_enums.h"
41 #include "nir_opcodes.h"
48 struct gl_shader_program
;
51 #define NIR_TRUE (~0u)
53 /** Defines a cast function
55 * This macro defines a cast function from in_type to out_type where
56 * out_type is some structure type that contains a field of type out_type.
58 * Note that you have to be a bit careful as the generated cast function
61 #define NIR_DEFINE_CAST(name, in_type, out_type, field) \
62 static inline out_type * \
63 name(const in_type *parent) \
65 return exec_node_data(out_type, parent, field); \
74 * Description of built-in state associated with a uniform
76 * \sa nir_variable::state_slots
90 nir_var_shader_storage
,
97 * Data stored in an nir_constant
99 union nir_constant_data
{
106 typedef struct nir_constant
{
108 * Value of the constant.
110 * The field used to back the values supplied by the constant is determined
111 * by the type associated with the \c nir_variable. Constants may be
112 * scalars, vectors, or matrices.
114 union nir_constant_data value
;
116 /* we could get this from the var->type but makes clone *much* easier to
117 * not have to care about the type.
119 unsigned num_elements
;
121 /* Array elements / Structure Fields */
122 struct nir_constant
**elements
;
126 * \brief Layout qualifiers for gl_FragDepth.
128 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
129 * with a layout qualifier.
132 nir_depth_layout_none
, /**< No depth layout is specified. */
133 nir_depth_layout_any
,
134 nir_depth_layout_greater
,
135 nir_depth_layout_less
,
136 nir_depth_layout_unchanged
140 * Either a uniform, global variable, shader input, or shader output. Based on
141 * ir_variable - it should be easy to translate between the two.
144 typedef struct nir_variable
{
145 struct exec_node node
;
148 * Declared type of the variable
150 const struct glsl_type
*type
;
153 * Declared name of the variable
157 struct nir_variable_data
{
160 * Is the variable read-only?
162 * This is set for variables declared as \c const, shader inputs,
165 unsigned read_only
:1;
169 unsigned invariant
:1;
172 * Storage class of the variable.
174 * \sa nir_variable_mode
176 nir_variable_mode mode
:5;
179 * Interpolation mode for shader inputs / outputs
181 * \sa glsl_interp_qualifier
183 unsigned interpolation
:2;
186 * \name ARB_fragment_coord_conventions
189 unsigned origin_upper_left
:1;
190 unsigned pixel_center_integer
:1;
194 * Was the location explicitly set in the shader?
196 * If the location is explicitly set in the shader, it \b cannot be changed
197 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
200 unsigned explicit_location
:1;
201 unsigned explicit_index
:1;
204 * Was an initial binding explicitly set in the shader?
206 * If so, constant_initializer contains an integer nir_constant
207 * representing the initial binding point.
209 unsigned explicit_binding
:1;
212 * Does this variable have an initializer?
214 * This is used by the linker to cross-validiate initializers of global
217 unsigned has_initializer
:1;
220 * If non-zero, then this variable may be packed along with other variables
221 * into a single varying slot, so this offset should be applied when
222 * accessing components. For example, an offset of 1 means that the x
223 * component of this variable is actually stored in component y of the
224 * location specified by \c location.
226 unsigned location_frac
:2;
229 * \brief Layout qualifier for gl_FragDepth.
231 * This is not equal to \c ir_depth_layout_none if and only if this
232 * variable is \c gl_FragDepth and a layout qualifier is specified.
234 nir_depth_layout depth_layout
;
237 * Storage location of the base of this variable
239 * The precise meaning of this field depends on the nature of the variable.
241 * - Vertex shader input: one of the values from \c gl_vert_attrib.
242 * - Vertex shader output: one of the values from \c gl_varying_slot.
243 * - Geometry shader input: one of the values from \c gl_varying_slot.
244 * - Geometry shader output: one of the values from \c gl_varying_slot.
245 * - Fragment shader input: one of the values from \c gl_varying_slot.
246 * - Fragment shader output: one of the values from \c gl_frag_result.
247 * - Uniforms: Per-stage uniform slot number for default uniform block.
248 * - Uniforms: Index within the uniform block definition for UBO members.
249 * - Non-UBO Uniforms: uniform slot number.
250 * - Other: This field is not currently used.
252 * If the variable is a uniform, shader input, or shader output, and the
253 * slot has not been assigned, the value will be -1.
258 * The actual location of the variable in the IR. Only valid for inputs
261 unsigned int driver_location
;
264 * output index for dual source blending.
269 * Descriptor set binding for sampler or UBO.
274 * Initial binding point for a sampler or UBO.
276 * For array types, this represents the binding point for the first element.
281 * Location an atomic counter is stored at.
286 * ARB_shader_image_load_store qualifiers.
289 bool read_only
; /**< "readonly" qualifier. */
290 bool write_only
; /**< "writeonly" qualifier. */
295 /** Image internal format if specified explicitly, otherwise GL_NONE. */
300 * Highest element accessed with a constant expression array index
302 * Not used for non-array variables.
304 unsigned max_array_access
;
309 * Built-in state that backs this uniform
311 * Once set at variable creation, \c state_slots must remain invariant.
312 * This is because, ideally, this array would be shared by all clones of
313 * this variable in the IR tree. In other words, we'd really like for it
314 * to be a fly-weight.
316 * If the variable is not a uniform, \c num_state_slots will be zero and
317 * \c state_slots will be \c NULL.
320 unsigned num_state_slots
; /**< Number of state slots used */
321 nir_state_slot
*state_slots
; /**< State descriptors. */
325 * Constant expression assigned in the initializer of the variable
327 nir_constant
*constant_initializer
;
330 * For variables that are in an interface block or are an instance of an
331 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
333 * \sa ir_variable::location
335 const struct glsl_type
*interface_type
;
338 #define nir_foreach_variable(var, var_list) \
339 foreach_list_typed(nir_variable, var, node, var_list)
341 #define nir_foreach_variable_safe(var, var_list) \
342 foreach_list_typed_safe(nir_variable, var, node, var_list)
345 nir_variable_is_global(const nir_variable
*var
)
347 return var
->data
.mode
!= nir_var_local
&& var
->data
.mode
!= nir_var_param
;
351 * Returns the bits in the inputs_read, outputs_written, or
352 * system_values_read bitfield corresponding to this variable.
354 static inline uint64_t
355 nir_variable_get_io_mask(nir_variable
*var
, gl_shader_stage stage
)
357 assert(var
->data
.mode
== nir_var_shader_in
||
358 var
->data
.mode
== nir_var_shader_out
||
359 var
->data
.mode
== nir_var_system_value
);
360 assert(var
->data
.location
>= 0);
362 const struct glsl_type
*var_type
= var
->type
;
363 if (stage
== MESA_SHADER_GEOMETRY
&& var
->data
.mode
== nir_var_shader_in
) {
364 /* Most geometry shader inputs are per-vertex arrays */
365 if (var
->data
.location
>= VARYING_SLOT_VAR0
)
366 assert(glsl_type_is_array(var_type
));
368 if (glsl_type_is_array(var_type
))
369 var_type
= glsl_get_array_element(var_type
);
372 bool is_vertex_input
= (var
->data
.mode
== nir_var_shader_in
&&
373 stage
== MESA_SHADER_VERTEX
);
374 unsigned slots
= glsl_count_attribute_slots(var_type
, is_vertex_input
);
375 return ((1ull << slots
) - 1) << var
->data
.location
;
378 typedef struct nir_register
{
379 struct exec_node node
;
381 unsigned num_components
; /** < number of vector components */
382 unsigned num_array_elems
; /** < size of array (0 for no array) */
384 /** generic register index. */
387 /** only for debug purposes, can be NULL */
390 /** whether this register is local (per-function) or global (per-shader) */
394 * If this flag is set to true, then accessing channels >= num_components
395 * is well-defined, and simply spills over to the next array element. This
396 * is useful for backends that can do per-component accessing, in
397 * particular scalar backends. By setting this flag and making
398 * num_components equal to 1, structures can be packed tightly into
399 * registers and then registers can be accessed per-component to get to
400 * each structure member, even if it crosses vec4 boundaries.
404 /** set of nir_src's where this register is used (read from) */
405 struct list_head uses
;
407 /** set of nir_dest's where this register is defined (written to) */
408 struct list_head defs
;
410 /** set of nir_if's where this register is used as a condition */
411 struct list_head if_uses
;
418 nir_instr_type_intrinsic
,
419 nir_instr_type_load_const
,
421 nir_instr_type_ssa_undef
,
423 nir_instr_type_parallel_copy
,
426 typedef struct nir_instr
{
427 struct exec_node node
;
429 struct nir_block
*block
;
431 /** generic instruction index. */
434 /* A temporary for optimization and analysis passes to use for storing
435 * flags. For instance, DCE uses this to store the "dead/live" info.
440 static inline nir_instr
*
441 nir_instr_next(nir_instr
*instr
)
443 struct exec_node
*next
= exec_node_get_next(&instr
->node
);
444 if (exec_node_is_tail_sentinel(next
))
447 return exec_node_data(nir_instr
, next
, node
);
450 static inline nir_instr
*
451 nir_instr_prev(nir_instr
*instr
)
453 struct exec_node
*prev
= exec_node_get_prev(&instr
->node
);
454 if (exec_node_is_head_sentinel(prev
))
457 return exec_node_data(nir_instr
, prev
, node
);
461 nir_instr_is_first(nir_instr
*instr
)
463 return exec_node_is_head_sentinel(exec_node_get_prev(&instr
->node
));
467 nir_instr_is_last(nir_instr
*instr
)
469 return exec_node_is_tail_sentinel(exec_node_get_next(&instr
->node
));
472 typedef struct nir_ssa_def
{
473 /** for debugging only, can be NULL */
476 /** generic SSA definition index. */
479 /** Index into the live_in and live_out bitfields */
482 nir_instr
*parent_instr
;
484 /** set of nir_instr's where this register is used (read from) */
485 struct list_head uses
;
487 /** set of nir_if's where this register is used as a condition */
488 struct list_head if_uses
;
490 uint8_t num_components
;
497 struct nir_src
*indirect
; /** < NULL for no indirect offset */
498 unsigned base_offset
;
500 /* TODO use-def chain goes here */
504 nir_instr
*parent_instr
;
505 struct list_head def_link
;
508 struct nir_src
*indirect
; /** < NULL for no indirect offset */
509 unsigned base_offset
;
511 /* TODO def-use chain goes here */
516 typedef struct nir_src
{
518 nir_instr
*parent_instr
;
519 struct nir_if
*parent_if
;
522 struct list_head use_link
;
532 #define NIR_SRC_INIT (nir_src) { { NULL } }
534 #define nir_foreach_use(reg_or_ssa_def, src) \
535 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
537 #define nir_foreach_use_safe(reg_or_ssa_def, src) \
538 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
540 #define nir_foreach_if_use(reg_or_ssa_def, src) \
541 list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
543 #define nir_foreach_if_use_safe(reg_or_ssa_def, src) \
544 list_for_each_entry_safe(nir_src, src, &(reg_or_ssa_def)->if_uses, use_link)
555 #define NIR_DEST_INIT (nir_dest) { { { NULL } } }
557 #define nir_foreach_def(reg, dest) \
558 list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
560 #define nir_foreach_def_safe(reg, dest) \
561 list_for_each_entry_safe(nir_dest, dest, &(reg)->defs, reg.def_link)
563 static inline nir_src
564 nir_src_for_ssa(nir_ssa_def
*def
)
566 nir_src src
= NIR_SRC_INIT
;
574 static inline nir_src
575 nir_src_for_reg(nir_register
*reg
)
577 nir_src src
= NIR_SRC_INIT
;
581 src
.reg
.indirect
= NULL
;
582 src
.reg
.base_offset
= 0;
587 static inline nir_dest
588 nir_dest_for_reg(nir_register
*reg
)
590 nir_dest dest
= NIR_DEST_INIT
;
597 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *instr_or_if
);
598 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
);
604 * \name input modifiers
608 * For inputs interpreted as floating point, flips the sign bit. For
609 * inputs interpreted as integers, performs the two's complement negation.
614 * Clears the sign bit for floating point values, and computes the integer
615 * absolute value for integers. Note that the negate modifier acts after
616 * the absolute value modifier, therefore if both are set then all inputs
617 * will become negative.
623 * For each input component, says which component of the register it is
624 * chosen from. Note that which elements of the swizzle are used and which
625 * are ignored are based on the write mask for most opcodes - for example,
626 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
627 * a swizzle of {2, x, 1, 0} where x means "don't care."
636 * \name saturate output modifier
638 * Only valid for opcodes that output floating-point numbers. Clamps the
639 * output to between 0.0 and 1.0 inclusive.
644 unsigned write_mask
: 4; /* ignored if dest.is_ssa is true */
648 nir_type_invalid
= 0, /* Not a valid type */
656 NIR_OP_IS_COMMUTATIVE
= (1 << 0),
657 NIR_OP_IS_ASSOCIATIVE
= (1 << 1),
658 } nir_op_algebraic_property
;
666 * The number of components in the output
668 * If non-zero, this is the size of the output and input sizes are
669 * explicitly given; swizzle and writemask are still in effect, but if
670 * the output component is masked out, then the input component may
673 * If zero, the opcode acts in the standard, per-component manner; the
674 * operation is performed on each component (except the ones that are
675 * masked out) with the input being taken from the input swizzle for
678 * The size of some of the inputs may be given (i.e. non-zero) even
679 * though output_size is zero; in that case, the inputs with a zero
680 * size act per-component, while the inputs with non-zero size don't.
682 unsigned output_size
;
685 * The type of vector that the instruction outputs. Note that the
686 * staurate modifier is only allowed on outputs with the float type.
689 nir_alu_type output_type
;
692 * The number of components in each input
694 unsigned input_sizes
[4];
697 * The type of vector that each input takes. Note that negate and
698 * absolute value are only allowed on inputs with int or float type and
699 * behave differently on the two.
701 nir_alu_type input_types
[4];
703 nir_op_algebraic_property algebraic_properties
;
706 extern const nir_op_info nir_op_infos
[nir_num_opcodes
];
708 typedef struct nir_alu_instr
{
715 void nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
716 nir_alu_instr
*instr
);
717 void nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
718 nir_alu_instr
*instr
);
720 /* is this source channel used? */
722 nir_alu_instr_channel_used(nir_alu_instr
*instr
, unsigned src
, unsigned channel
)
724 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
725 return channel
< nir_op_infos
[instr
->op
].input_sizes
[src
];
727 return (instr
->dest
.write_mask
>> channel
) & 1;
731 * For instructions whose destinations are SSA, get the number of channels
734 static inline unsigned
735 nir_ssa_alu_instr_src_components(const nir_alu_instr
*instr
, unsigned src
)
737 assert(instr
->dest
.dest
.is_ssa
);
739 if (nir_op_infos
[instr
->op
].input_sizes
[src
] > 0)
740 return nir_op_infos
[instr
->op
].input_sizes
[src
];
742 return instr
->dest
.dest
.ssa
.num_components
;
747 nir_deref_type_array
,
748 nir_deref_type_struct
751 typedef struct nir_deref
{
752 nir_deref_type deref_type
;
753 struct nir_deref
*child
;
754 const struct glsl_type
*type
;
763 /* This enum describes how the array is referenced. If the deref is
764 * direct then the base_offset is used. If the deref is indirect then then
765 * offset is given by base_offset + indirect. If the deref is a wildcard
766 * then the deref refers to all of the elements of the array at the same
767 * time. Wildcard dereferences are only ever allowed in copy_var
768 * intrinsics and the source and destination derefs must have matching
772 nir_deref_array_type_direct
,
773 nir_deref_array_type_indirect
,
774 nir_deref_array_type_wildcard
,
775 } nir_deref_array_type
;
780 nir_deref_array_type deref_array_type
;
781 unsigned base_offset
;
791 NIR_DEFINE_CAST(nir_deref_as_var
, nir_deref
, nir_deref_var
, deref
)
792 NIR_DEFINE_CAST(nir_deref_as_array
, nir_deref
, nir_deref_array
, deref
)
793 NIR_DEFINE_CAST(nir_deref_as_struct
, nir_deref
, nir_deref_struct
, deref
)
795 /* Returns the last deref in the chain. */
796 static inline nir_deref
*
797 nir_deref_tail(nir_deref
*deref
)
800 deref
= deref
->child
;
808 nir_deref_var
**params
;
809 nir_deref_var
*return_deref
;
811 struct nir_function
*callee
;
814 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
815 num_variables, num_indices, idx0, idx1, idx2, flags) \
816 nir_intrinsic_##name,
818 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
821 #include "nir_intrinsics.h"
822 nir_num_intrinsics
= nir_last_intrinsic
+ 1
826 #undef LAST_INTRINSIC
828 #define NIR_INTRINSIC_MAX_CONST_INDEX 3
830 /** Represents an intrinsic
832 * An intrinsic is an instruction type for handling things that are
833 * more-or-less regular operations but don't just consume and produce SSA
834 * values like ALU operations do. Intrinsics are not for things that have
835 * special semantic meaning such as phi nodes and parallel copies.
836 * Examples of intrinsics include variable load/store operations, system
837 * value loads, and the like. Even though texturing more-or-less falls
838 * under this category, texturing is its own instruction type because
839 * trying to represent texturing with intrinsics would lead to a
840 * combinatorial explosion of intrinsic opcodes.
842 * By having a single instruction type for handling a lot of different
843 * cases, optimization passes can look for intrinsics and, for the most
844 * part, completely ignore them. Each intrinsic type also has a few
845 * possible flags that govern whether or not they can be reordered or
846 * eliminated. That way passes like dead code elimination can still work
847 * on intrisics without understanding the meaning of each.
849 * Each intrinsic has some number of constant indices, some number of
850 * variables, and some number of sources. What these sources, variables,
851 * and indices mean depends on the intrinsic and is documented with the
852 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
853 * instructions are the only types of instruction that can operate on
859 nir_intrinsic_op intrinsic
;
863 /** number of components if this is a vectorized intrinsic
865 * Similarly to ALU operations, some intrinsics are vectorized.
866 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
867 * For vectorized intrinsics, the num_components field specifies the
868 * number of destination components and the number of source components
869 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
871 uint8_t num_components
;
873 int const_index
[NIR_INTRINSIC_MAX_CONST_INDEX
];
875 nir_deref_var
*variables
[2];
878 } nir_intrinsic_instr
;
881 * \name NIR intrinsics semantic flags
883 * information about what the compiler can do with the intrinsics.
885 * \sa nir_intrinsic_info::flags
889 * whether the intrinsic can be safely eliminated if none of its output
890 * value is not being used.
892 NIR_INTRINSIC_CAN_ELIMINATE
= (1 << 0),
895 * Whether the intrinsic can be reordered with respect to any other
896 * intrinsic, i.e. whether the only reordering dependencies of the
897 * intrinsic are due to the register reads/writes.
899 NIR_INTRINSIC_CAN_REORDER
= (1 << 1),
900 } nir_intrinsic_semantic_flag
;
903 * \name NIR intrinsics const-index flag
905 * Indicates the usage of a const_index slot.
907 * \sa nir_intrinsic_info::index_map
911 * Generally instructions that take a offset src argument, can encode
912 * a constant 'base' value which is added to the offset.
914 NIR_INTRINSIC_BASE
= 1,
917 * For store instructions, a writemask for the store.
919 NIR_INTRINSIC_WRMASK
= 2,
922 * The stream-id for GS emit_vertex/end_primitive intrinsics.
924 NIR_INTRINSIC_STREAM_ID
= 3,
927 * The clip-plane id for load_user_clip_plane intrinsic.
929 NIR_INTRINSIC_UCP_ID
= 4,
932 * The range of a load operation. This specifies the maximum amount of
933 * data starting at the base offset (if any) that can be accessed.
935 NIR_INTRINSIC_RANGE
= 5,
938 * The Vulkan descriptor set for vulkan_resource_index intrinsic.
940 NIR_INTRINSIC_DESC_SET
= 6,
943 * The Vulkan descriptor set binding for vulkan_resource_index intrinsic.
945 NIR_INTRINSIC_BINDING
= 7,
947 NIR_INTRINSIC_NUM_INDEX_FLAGS
,
949 } nir_intrinsic_index_flag
;
951 #define NIR_INTRINSIC_MAX_INPUTS 4
956 unsigned num_srcs
; /** < number of register/SSA inputs */
958 /** number of components of each input register
960 * If this value is 0, the number of components is given by the
961 * num_components field of nir_intrinsic_instr.
963 unsigned src_components
[NIR_INTRINSIC_MAX_INPUTS
];
967 /** number of components of the output register
969 * If this value is 0, the number of components is given by the
970 * num_components field of nir_intrinsic_instr.
972 unsigned dest_components
;
974 /** the number of inputs/outputs that are variables */
975 unsigned num_variables
;
977 /** the number of constant indices used by the intrinsic */
978 unsigned num_indices
;
980 /** indicates the usage of intr->const_index[n] */
981 unsigned index_map
[NIR_INTRINSIC_NUM_INDEX_FLAGS
];
983 /** semantic flags for calls to this intrinsic */
984 nir_intrinsic_semantic_flag flags
;
985 } nir_intrinsic_info
;
987 extern const nir_intrinsic_info nir_intrinsic_infos
[nir_num_intrinsics
];
990 #define INTRINSIC_IDX_ACCESSORS(name, flag, type) \
992 nir_intrinsic_##name(nir_intrinsic_instr *instr) \
994 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
995 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
996 return instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1]; \
999 nir_intrinsic_set_##name(nir_intrinsic_instr *instr, type val) \
1001 const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic]; \
1002 assert(info->index_map[NIR_INTRINSIC_##flag] > 0); \
1003 instr->const_index[info->index_map[NIR_INTRINSIC_##flag] - 1] = val; \
1006 INTRINSIC_IDX_ACCESSORS(write_mask
, WRMASK
, unsigned)
1007 INTRINSIC_IDX_ACCESSORS(base
, BASE
, int)
1008 INTRINSIC_IDX_ACCESSORS(stream_id
, STREAM_ID
, unsigned)
1009 INTRINSIC_IDX_ACCESSORS(ucp_id
, UCP_ID
, unsigned)
1010 INTRINSIC_IDX_ACCESSORS(range
, RANGE
, unsigned)
1011 INTRINSIC_IDX_ACCESSORS(desc_set
, DESC_SET
, unsigned)
1012 INTRINSIC_IDX_ACCESSORS(binding
, BINDING
, unsigned)
1015 * \group texture information
1017 * This gives semantic information about textures which is useful to the
1018 * frontend, the backend, and lowering passes, but not the optimizer.
1023 nir_tex_src_projector
,
1024 nir_tex_src_comparitor
, /* shadow comparitor */
1028 nir_tex_src_ms_index
, /* MSAA sample index */
1031 nir_tex_src_texture_offset
, /* < dynamically uniform indirect offset */
1032 nir_tex_src_sampler_offset
, /* < dynamically uniform indirect offset */
1033 nir_num_tex_src_types
1038 nir_tex_src_type src_type
;
1042 nir_texop_tex
, /**< Regular texture look-up */
1043 nir_texop_txb
, /**< Texture look-up with LOD bias */
1044 nir_texop_txl
, /**< Texture look-up with explicit LOD */
1045 nir_texop_txd
, /**< Texture look-up with partial derivatvies */
1046 nir_texop_txf
, /**< Texel fetch with explicit LOD */
1047 nir_texop_txf_ms
, /**< Multisample texture fetch */
1048 nir_texop_txs
, /**< Texture size */
1049 nir_texop_lod
, /**< Texture lod query */
1050 nir_texop_tg4
, /**< Texture gather */
1051 nir_texop_query_levels
, /**< Texture levels query */
1052 nir_texop_texture_samples
, /**< Texture samples query */
1053 nir_texop_samples_identical
, /**< Query whether all samples are definitely
1061 enum glsl_sampler_dim sampler_dim
;
1062 nir_alu_type dest_type
;
1067 unsigned num_srcs
, coord_components
;
1068 bool is_array
, is_shadow
;
1071 * If is_shadow is true, whether this is the old-style shadow that outputs 4
1072 * components or the new-style shadow that outputs 1 component.
1074 bool is_new_style_shadow
;
1076 /* gather component selector */
1077 unsigned component
: 2;
1079 /** The texture index
1081 * If this texture instruction has a nir_tex_src_texture_offset source,
1082 * then the texture index is given by texture_index + texture_offset.
1084 unsigned texture_index
;
1086 /** The size of the texture array or 0 if it's not an array */
1087 unsigned texture_array_size
;
1089 /** The texture deref
1091 * If this is null, use texture_index instead.
1093 nir_deref_var
*texture
;
1095 /** The sampler index
1097 * The following operations do not require a sampler and, as such, this
1098 * field should be ignored:
1100 * - nir_texop_txf_ms
1104 * - nir_texop_query_levels
1105 * - nir_texop_texture_samples
1106 * - nir_texop_samples_identical
1108 * If this texture instruction has a nir_tex_src_sampler_offset source,
1109 * then the sampler index is given by sampler_index + sampler_offset.
1111 unsigned sampler_index
;
1113 /** The sampler deref
1115 * If this is null, use sampler_index instead.
1117 nir_deref_var
*sampler
;
1120 static inline unsigned
1121 nir_tex_instr_dest_size(nir_tex_instr
*instr
)
1123 switch (instr
->op
) {
1124 case nir_texop_txs
: {
1126 switch (instr
->sampler_dim
) {
1127 case GLSL_SAMPLER_DIM_1D
:
1128 case GLSL_SAMPLER_DIM_BUF
:
1131 case GLSL_SAMPLER_DIM_2D
:
1132 case GLSL_SAMPLER_DIM_CUBE
:
1133 case GLSL_SAMPLER_DIM_MS
:
1134 case GLSL_SAMPLER_DIM_RECT
:
1135 case GLSL_SAMPLER_DIM_EXTERNAL
:
1138 case GLSL_SAMPLER_DIM_3D
:
1142 unreachable("not reached");
1144 if (instr
->is_array
)
1152 case nir_texop_texture_samples
:
1153 case nir_texop_query_levels
:
1154 case nir_texop_samples_identical
:
1158 if (instr
->is_shadow
&& instr
->is_new_style_shadow
)
1165 /* Returns true if this texture operation queries something about the texture
1166 * rather than actually sampling it.
1169 nir_tex_instr_is_query(nir_tex_instr
*instr
)
1171 switch (instr
->op
) {
1174 case nir_texop_texture_samples
:
1175 case nir_texop_query_levels
:
1182 case nir_texop_txf_ms
:
1186 unreachable("Invalid texture opcode");
1190 static inline unsigned
1191 nir_tex_instr_src_size(nir_tex_instr
*instr
, unsigned src
)
1193 if (instr
->src
[src
].src_type
== nir_tex_src_coord
)
1194 return instr
->coord_components
;
1197 if (instr
->src
[src
].src_type
== nir_tex_src_offset
||
1198 instr
->src
[src
].src_type
== nir_tex_src_ddx
||
1199 instr
->src
[src
].src_type
== nir_tex_src_ddy
) {
1200 if (instr
->is_array
)
1201 return instr
->coord_components
- 1;
1203 return instr
->coord_components
;
1210 nir_tex_instr_src_index(nir_tex_instr
*instr
, nir_tex_src_type type
)
1212 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1213 if (instr
->src
[i
].src_type
== type
)
1230 nir_const_value value
;
1233 } nir_load_const_instr
;
1246 /* creates a new SSA variable in an undefined state */
1251 } nir_ssa_undef_instr
;
1254 struct exec_node node
;
1256 /* The predecessor block corresponding to this source */
1257 struct nir_block
*pred
;
1262 #define nir_foreach_phi_src(phi, entry) \
1263 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1264 #define nir_foreach_phi_src_safe(phi, entry) \
1265 foreach_list_typed_safe(nir_phi_src, entry, node, &(phi)->srcs)
1270 struct exec_list srcs
; /** < list of nir_phi_src */
1276 struct exec_node node
;
1279 } nir_parallel_copy_entry
;
1281 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1282 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1287 /* A list of nir_parallel_copy_entry's. The sources of all of the
1288 * entries are copied to the corresponding destinations "in parallel".
1289 * In other words, if we have two entries: a -> b and b -> a, the values
1292 struct exec_list entries
;
1293 } nir_parallel_copy_instr
;
1295 NIR_DEFINE_CAST(nir_instr_as_alu
, nir_instr
, nir_alu_instr
, instr
)
1296 NIR_DEFINE_CAST(nir_instr_as_call
, nir_instr
, nir_call_instr
, instr
)
1297 NIR_DEFINE_CAST(nir_instr_as_jump
, nir_instr
, nir_jump_instr
, instr
)
1298 NIR_DEFINE_CAST(nir_instr_as_tex
, nir_instr
, nir_tex_instr
, instr
)
1299 NIR_DEFINE_CAST(nir_instr_as_intrinsic
, nir_instr
, nir_intrinsic_instr
, instr
)
1300 NIR_DEFINE_CAST(nir_instr_as_load_const
, nir_instr
, nir_load_const_instr
, instr
)
1301 NIR_DEFINE_CAST(nir_instr_as_ssa_undef
, nir_instr
, nir_ssa_undef_instr
, instr
)
1302 NIR_DEFINE_CAST(nir_instr_as_phi
, nir_instr
, nir_phi_instr
, instr
)
1303 NIR_DEFINE_CAST(nir_instr_as_parallel_copy
, nir_instr
,
1304 nir_parallel_copy_instr
, instr
)
1309 * Control flow consists of a tree of control flow nodes, which include
1310 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1311 * instructions that always run start-to-finish. Each basic block also keeps
1312 * track of its successors (blocks which may run immediately after the current
1313 * block) and predecessors (blocks which could have run immediately before the
1314 * current block). Each function also has a start block and an end block which
1315 * all return statements point to (which is always empty). Together, all the
1316 * blocks with their predecessors and successors make up the control flow
1317 * graph (CFG) of the function. There are helpers that modify the tree of
1318 * control flow nodes while modifying the CFG appropriately; these should be
1319 * used instead of modifying the tree directly.
1326 nir_cf_node_function
1329 typedef struct nir_cf_node
{
1330 struct exec_node node
;
1331 nir_cf_node_type type
;
1332 struct nir_cf_node
*parent
;
1335 typedef struct nir_block
{
1336 nir_cf_node cf_node
;
1338 struct exec_list instr_list
; /** < list of nir_instr */
1340 /** generic block index; generated by nir_index_blocks */
1344 * Each block can only have up to 2 successors, so we put them in a simple
1345 * array - no need for anything more complicated.
1347 struct nir_block
*successors
[2];
1349 /* Set of nir_block predecessors in the CFG */
1350 struct set
*predecessors
;
1353 * this node's immediate dominator in the dominance tree - set to NULL for
1356 struct nir_block
*imm_dom
;
1358 /* This node's children in the dominance tree */
1359 unsigned num_dom_children
;
1360 struct nir_block
**dom_children
;
1362 /* Set of nir_block's on the dominance frontier of this block */
1363 struct set
*dom_frontier
;
1366 * These two indices have the property that dom_{pre,post}_index for each
1367 * child of this block in the dominance tree will always be between
1368 * dom_pre_index and dom_post_index for this block, which makes testing if
1369 * a given block is dominated by another block an O(1) operation.
1371 unsigned dom_pre_index
, dom_post_index
;
1373 /* live in and out for this block; used for liveness analysis */
1374 BITSET_WORD
*live_in
;
1375 BITSET_WORD
*live_out
;
1378 static inline nir_instr
*
1379 nir_block_first_instr(nir_block
*block
)
1381 struct exec_node
*head
= exec_list_get_head(&block
->instr_list
);
1382 return exec_node_data(nir_instr
, head
, node
);
1385 static inline nir_instr
*
1386 nir_block_last_instr(nir_block
*block
)
1388 struct exec_node
*tail
= exec_list_get_tail(&block
->instr_list
);
1389 return exec_node_data(nir_instr
, tail
, node
);
1392 #define nir_foreach_instr(block, instr) \
1393 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1394 #define nir_foreach_instr_reverse(block, instr) \
1395 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1396 #define nir_foreach_instr_safe(block, instr) \
1397 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1398 #define nir_foreach_instr_reverse_safe(block, instr) \
1399 foreach_list_typed_reverse_safe(nir_instr, instr, node, &(block)->instr_list)
1401 typedef struct nir_if
{
1402 nir_cf_node cf_node
;
1405 struct exec_list then_list
; /** < list of nir_cf_node */
1406 struct exec_list else_list
; /** < list of nir_cf_node */
1409 static inline nir_cf_node
*
1410 nir_if_first_then_node(nir_if
*if_stmt
)
1412 struct exec_node
*head
= exec_list_get_head(&if_stmt
->then_list
);
1413 return exec_node_data(nir_cf_node
, head
, node
);
1416 static inline nir_cf_node
*
1417 nir_if_last_then_node(nir_if
*if_stmt
)
1419 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->then_list
);
1420 return exec_node_data(nir_cf_node
, tail
, node
);
1423 static inline nir_cf_node
*
1424 nir_if_first_else_node(nir_if
*if_stmt
)
1426 struct exec_node
*head
= exec_list_get_head(&if_stmt
->else_list
);
1427 return exec_node_data(nir_cf_node
, head
, node
);
1430 static inline nir_cf_node
*
1431 nir_if_last_else_node(nir_if
*if_stmt
)
1433 struct exec_node
*tail
= exec_list_get_tail(&if_stmt
->else_list
);
1434 return exec_node_data(nir_cf_node
, tail
, node
);
1438 nir_cf_node cf_node
;
1440 struct exec_list body
; /** < list of nir_cf_node */
1443 static inline nir_cf_node
*
1444 nir_loop_first_cf_node(nir_loop
*loop
)
1446 return exec_node_data(nir_cf_node
, exec_list_get_head(&loop
->body
), node
);
1449 static inline nir_cf_node
*
1450 nir_loop_last_cf_node(nir_loop
*loop
)
1452 return exec_node_data(nir_cf_node
, exec_list_get_tail(&loop
->body
), node
);
1456 * Various bits of metadata that can may be created or required by
1457 * optimization and analysis passes
1460 nir_metadata_none
= 0x0,
1461 nir_metadata_block_index
= 0x1,
1462 nir_metadata_dominance
= 0x2,
1463 nir_metadata_live_ssa_defs
= 0x4,
1464 nir_metadata_not_properly_reset
= 0x8,
1468 nir_cf_node cf_node
;
1470 /** pointer to the function of which this is an implementation */
1471 struct nir_function
*function
;
1473 struct exec_list body
; /** < list of nir_cf_node */
1475 nir_block
*end_block
;
1477 /** list for all local variables in the function */
1478 struct exec_list locals
;
1480 /** array of variables used as parameters */
1481 unsigned num_params
;
1482 nir_variable
**params
;
1484 /** variable used to hold the result of the function */
1485 nir_variable
*return_var
;
1487 /** list of local registers in the function */
1488 struct exec_list registers
;
1490 /** next available local register index */
1493 /** next available SSA value index */
1496 /* total number of basic blocks, only valid when block_index_dirty = false */
1497 unsigned num_blocks
;
1499 nir_metadata valid_metadata
;
1500 } nir_function_impl
;
1502 static inline nir_block
*
1503 nir_start_block(nir_function_impl
*impl
)
1505 return (nir_block
*) exec_list_get_head(&impl
->body
);
1508 static inline nir_cf_node
*
1509 nir_cf_node_next(nir_cf_node
*node
)
1511 struct exec_node
*next
= exec_node_get_next(&node
->node
);
1512 if (exec_node_is_tail_sentinel(next
))
1515 return exec_node_data(nir_cf_node
, next
, node
);
1518 static inline nir_cf_node
*
1519 nir_cf_node_prev(nir_cf_node
*node
)
1521 struct exec_node
*prev
= exec_node_get_prev(&node
->node
);
1522 if (exec_node_is_head_sentinel(prev
))
1525 return exec_node_data(nir_cf_node
, prev
, node
);
1529 nir_cf_node_is_first(const nir_cf_node
*node
)
1531 return exec_node_is_head_sentinel(node
->node
.prev
);
1535 nir_cf_node_is_last(const nir_cf_node
*node
)
1537 return exec_node_is_tail_sentinel(node
->node
.next
);
1540 NIR_DEFINE_CAST(nir_cf_node_as_block
, nir_cf_node
, nir_block
, cf_node
)
1541 NIR_DEFINE_CAST(nir_cf_node_as_if
, nir_cf_node
, nir_if
, cf_node
)
1542 NIR_DEFINE_CAST(nir_cf_node_as_loop
, nir_cf_node
, nir_loop
, cf_node
)
1543 NIR_DEFINE_CAST(nir_cf_node_as_function
, nir_cf_node
, nir_function_impl
, cf_node
)
1548 nir_parameter_inout
,
1549 } nir_parameter_type
;
1552 nir_parameter_type param_type
;
1553 const struct glsl_type
*type
;
1556 typedef struct nir_function
{
1557 struct exec_node node
;
1560 struct nir_shader
*shader
;
1562 unsigned num_params
;
1563 nir_parameter
*params
;
1564 const struct glsl_type
*return_type
;
1566 /** The implementation of this function.
1568 * If the function is only declared and not implemented, this is NULL.
1570 nir_function_impl
*impl
;
1573 typedef struct nir_shader_compiler_options
{
1581 bool lower_bitfield_extract
;
1582 bool lower_bitfield_insert
;
1583 bool lower_uadd_carry
;
1584 bool lower_usub_borrow
;
1585 /** lowers fneg and ineg to fsub and isub. */
1587 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1590 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1593 /* Does the native fdot instruction replicate its result for four
1594 * components? If so, then opt_algebraic_late will turn all fdotN
1595 * instructions into fdot_replicatedN instructions.
1597 bool fdot_replicates
;
1599 /** lowers ffract to fsub+ffloor: */
1602 bool lower_pack_half_2x16
;
1603 bool lower_pack_unorm_2x16
;
1604 bool lower_pack_snorm_2x16
;
1605 bool lower_pack_unorm_4x8
;
1606 bool lower_pack_snorm_4x8
;
1607 bool lower_unpack_half_2x16
;
1608 bool lower_unpack_unorm_2x16
;
1609 bool lower_unpack_snorm_2x16
;
1610 bool lower_unpack_unorm_4x8
;
1611 bool lower_unpack_snorm_4x8
;
1613 bool lower_extract_byte
;
1614 bool lower_extract_word
;
1617 * Does the driver support real 32-bit integers? (Otherwise, integers
1618 * are simulated by floats.)
1620 bool native_integers
;
1622 /* Indicates that the driver only has zero-based vertex id */
1623 bool vertex_id_zero_based
;
1624 } nir_shader_compiler_options
;
1626 typedef struct nir_shader_info
{
1629 /* Descriptive name provided by the client; may be NULL */
1632 /* Number of textures used by this shader */
1633 unsigned num_textures
;
1634 /* Number of uniform buffers used by this shader */
1636 /* Number of atomic buffers used by this shader */
1638 /* Number of shader storage buffers used by this shader */
1640 /* Number of images used by this shader */
1641 unsigned num_images
;
1643 /* Which inputs are actually read */
1644 uint64_t inputs_read
;
1645 /* Which outputs are actually written */
1646 uint64_t outputs_written
;
1647 /* Which system values are actually read */
1648 uint64_t system_values_read
;
1650 /* Which patch inputs are actually read */
1651 uint32_t patch_inputs_read
;
1652 /* Which patch outputs are actually written */
1653 uint32_t patch_outputs_written
;
1655 /* Whether or not this shader ever uses textureGather() */
1656 bool uses_texture_gather
;
1658 /* Whether or not this shader uses the gl_ClipDistance output */
1659 bool uses_clip_distance_out
;
1661 /* Whether or not separate shader objects were used */
1662 bool separate_shader
;
1664 /** Was this shader linked with any transform feedback varyings? */
1665 bool has_transform_feedback_varyings
;
1669 /** The number of vertices recieves per input primitive */
1670 unsigned vertices_in
;
1672 /** The output primitive type (GL enum value) */
1673 unsigned output_primitive
;
1675 /** The maximum number of vertices the geometry shader might write. */
1676 unsigned vertices_out
;
1678 /** 1 .. MAX_GEOMETRY_SHADER_INVOCATIONS */
1679 unsigned invocations
;
1681 /** Whether or not this shader uses EndPrimitive */
1682 bool uses_end_primitive
;
1684 /** Whether or not this shader uses non-zero streams */
1692 * Whether early fragment tests are enabled as defined by
1693 * ARB_shader_image_load_store.
1695 bool early_fragment_tests
;
1697 /** gl_FragDepth layout for ARB_conservative_depth. */
1698 enum gl_frag_depth_layout depth_layout
;
1702 unsigned local_size
[3];
1706 /** The number of vertices in the TCS output patch. */
1707 unsigned vertices_out
;
1712 typedef struct nir_shader
{
1713 /** list of uniforms (nir_variable) */
1714 struct exec_list uniforms
;
1716 /** list of inputs (nir_variable) */
1717 struct exec_list inputs
;
1719 /** list of outputs (nir_variable) */
1720 struct exec_list outputs
;
1722 /** list of shared compute variables (nir_variable) */
1723 struct exec_list shared
;
1725 /** Set of driver-specific options for the shader.
1727 * The memory for the options is expected to be kept in a single static
1728 * copy by the driver.
1730 const struct nir_shader_compiler_options
*options
;
1732 /** Various bits of compile-time information about a given shader */
1733 struct nir_shader_info info
;
1735 /** list of global variables in the shader (nir_variable) */
1736 struct exec_list globals
;
1738 /** list of system value variables in the shader (nir_variable) */
1739 struct exec_list system_values
;
1741 struct exec_list functions
; /** < list of nir_function */
1743 /** list of global register in the shader */
1744 struct exec_list registers
;
1746 /** next available global register index */
1750 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1753 unsigned num_inputs
, num_uniforms
, num_outputs
, num_shared
;
1755 /** The shader stage, such as MESA_SHADER_VERTEX. */
1756 gl_shader_stage stage
;
1759 static inline nir_function
*
1760 nir_shader_get_entrypoint(nir_shader
*shader
)
1762 assert(exec_list_length(&shader
->functions
) == 1);
1763 struct exec_node
*func_node
= exec_list_get_head(&shader
->functions
);
1764 nir_function
*func
= exec_node_data(nir_function
, func_node
, node
);
1768 #define nir_foreach_function(shader, func) \
1769 foreach_list_typed(nir_function, func, node, &(shader)->functions)
1771 nir_shader
*nir_shader_create(void *mem_ctx
,
1772 gl_shader_stage stage
,
1773 const nir_shader_compiler_options
*options
);
1775 /** creates a register, including assigning it an index and adding it to the list */
1776 nir_register
*nir_global_reg_create(nir_shader
*shader
);
1778 nir_register
*nir_local_reg_create(nir_function_impl
*impl
);
1780 void nir_reg_remove(nir_register
*reg
);
1782 /** Adds a variable to the appropreate list in nir_shader */
1783 void nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
);
1786 nir_function_impl_add_variable(nir_function_impl
*impl
, nir_variable
*var
)
1788 assert(var
->data
.mode
== nir_var_local
);
1789 exec_list_push_tail(&impl
->locals
, &var
->node
);
1792 /** creates a variable, sets a few defaults, and adds it to the list */
1793 nir_variable
*nir_variable_create(nir_shader
*shader
,
1794 nir_variable_mode mode
,
1795 const struct glsl_type
*type
,
1797 /** creates a local variable and adds it to the list */
1798 nir_variable
*nir_local_variable_create(nir_function_impl
*impl
,
1799 const struct glsl_type
*type
,
1802 /** creates a function and adds it to the shader's list of functions */
1803 nir_function
*nir_function_create(nir_shader
*shader
, const char *name
);
1805 nir_function_impl
*nir_function_impl_create(nir_function
*func
);
1806 /** creates a function_impl that isn't tied to any particular function */
1807 nir_function_impl
*nir_function_impl_create_bare(nir_shader
*shader
);
1809 nir_block
*nir_block_create(nir_shader
*shader
);
1810 nir_if
*nir_if_create(nir_shader
*shader
);
1811 nir_loop
*nir_loop_create(nir_shader
*shader
);
1813 nir_function_impl
*nir_cf_node_get_function(nir_cf_node
*node
);
1815 /** requests that the given pieces of metadata be generated */
1816 void nir_metadata_require(nir_function_impl
*impl
, nir_metadata required
);
1817 /** dirties all but the preserved metadata */
1818 void nir_metadata_preserve(nir_function_impl
*impl
, nir_metadata preserved
);
1820 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1821 nir_alu_instr
*nir_alu_instr_create(nir_shader
*shader
, nir_op op
);
1823 nir_jump_instr
*nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
);
1825 nir_load_const_instr
*nir_load_const_instr_create(nir_shader
*shader
,
1826 unsigned num_components
);
1828 nir_intrinsic_instr
*nir_intrinsic_instr_create(nir_shader
*shader
,
1829 nir_intrinsic_op op
);
1831 nir_call_instr
*nir_call_instr_create(nir_shader
*shader
,
1832 nir_function
*callee
);
1834 nir_tex_instr
*nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
);
1836 nir_phi_instr
*nir_phi_instr_create(nir_shader
*shader
);
1838 nir_parallel_copy_instr
*nir_parallel_copy_instr_create(nir_shader
*shader
);
1840 nir_ssa_undef_instr
*nir_ssa_undef_instr_create(nir_shader
*shader
,
1841 unsigned num_components
);
1843 nir_deref_var
*nir_deref_var_create(void *mem_ctx
, nir_variable
*var
);
1844 nir_deref_array
*nir_deref_array_create(void *mem_ctx
);
1845 nir_deref_struct
*nir_deref_struct_create(void *mem_ctx
, unsigned field_index
);
1847 nir_deref
*nir_copy_deref(void *mem_ctx
, nir_deref
*deref
);
1849 nir_load_const_instr
*
1850 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
);
1853 * NIR Cursors and Instruction Insertion API
1856 * A tiny struct representing a point to insert/extract instructions or
1857 * control flow nodes. Helps reduce the combinatorial explosion of possible
1858 * points to insert/extract.
1860 * \sa nir_control_flow.h
1863 nir_cursor_before_block
,
1864 nir_cursor_after_block
,
1865 nir_cursor_before_instr
,
1866 nir_cursor_after_instr
,
1867 } nir_cursor_option
;
1870 nir_cursor_option option
;
1877 static inline nir_block
*
1878 nir_cursor_current_block(nir_cursor cursor
)
1880 if (cursor
.option
== nir_cursor_before_instr
||
1881 cursor
.option
== nir_cursor_after_instr
) {
1882 return cursor
.instr
->block
;
1884 return cursor
.block
;
1888 bool nir_cursors_equal(nir_cursor a
, nir_cursor b
);
1890 static inline nir_cursor
1891 nir_before_block(nir_block
*block
)
1894 cursor
.option
= nir_cursor_before_block
;
1895 cursor
.block
= block
;
1899 static inline nir_cursor
1900 nir_after_block(nir_block
*block
)
1903 cursor
.option
= nir_cursor_after_block
;
1904 cursor
.block
= block
;
1908 static inline nir_cursor
1909 nir_before_instr(nir_instr
*instr
)
1912 cursor
.option
= nir_cursor_before_instr
;
1913 cursor
.instr
= instr
;
1917 static inline nir_cursor
1918 nir_after_instr(nir_instr
*instr
)
1921 cursor
.option
= nir_cursor_after_instr
;
1922 cursor
.instr
= instr
;
1926 static inline nir_cursor
1927 nir_after_block_before_jump(nir_block
*block
)
1929 nir_instr
*last_instr
= nir_block_last_instr(block
);
1930 if (last_instr
&& last_instr
->type
== nir_instr_type_jump
) {
1931 return nir_before_instr(last_instr
);
1933 return nir_after_block(block
);
1937 static inline nir_cursor
1938 nir_before_cf_node(nir_cf_node
*node
)
1940 if (node
->type
== nir_cf_node_block
)
1941 return nir_before_block(nir_cf_node_as_block(node
));
1943 return nir_after_block(nir_cf_node_as_block(nir_cf_node_prev(node
)));
1946 static inline nir_cursor
1947 nir_after_cf_node(nir_cf_node
*node
)
1949 if (node
->type
== nir_cf_node_block
)
1950 return nir_after_block(nir_cf_node_as_block(node
));
1952 return nir_before_block(nir_cf_node_as_block(nir_cf_node_next(node
)));
1955 static inline nir_cursor
1956 nir_after_cf_node_and_phis(nir_cf_node
*node
)
1958 if (node
->type
== nir_cf_node_block
)
1959 return nir_after_block(nir_cf_node_as_block(node
));
1961 nir_block
*block
= nir_cf_node_as_block(nir_cf_node_next(node
));
1962 assert(block
->cf_node
.type
== nir_cf_node_block
);
1964 nir_foreach_instr(block
, instr
) {
1965 if (instr
->type
!= nir_instr_type_phi
)
1966 return nir_before_instr(instr
);
1968 return nir_after_block(block
);
1971 static inline nir_cursor
1972 nir_before_cf_list(struct exec_list
*cf_list
)
1974 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1975 exec_list_get_head(cf_list
), node
);
1976 return nir_before_cf_node(first_node
);
1979 static inline nir_cursor
1980 nir_after_cf_list(struct exec_list
*cf_list
)
1982 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1983 exec_list_get_tail(cf_list
), node
);
1984 return nir_after_cf_node(last_node
);
1988 * Insert a NIR instruction at the given cursor.
1990 * Note: This does not update the cursor.
1992 void nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
);
1995 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1997 nir_instr_insert(nir_before_instr(instr
), before
);
2001 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
2003 nir_instr_insert(nir_after_instr(instr
), after
);
2007 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
2009 nir_instr_insert(nir_before_block(block
), before
);
2013 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
2015 nir_instr_insert(nir_after_block(block
), after
);
2019 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
2021 nir_instr_insert(nir_before_cf_node(node
), before
);
2025 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
2027 nir_instr_insert(nir_after_cf_node(node
), after
);
2031 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
2033 nir_instr_insert(nir_before_cf_list(list
), before
);
2037 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
2039 nir_instr_insert(nir_after_cf_list(list
), after
);
2042 void nir_instr_remove(nir_instr
*instr
);
2046 typedef bool (*nir_foreach_ssa_def_cb
)(nir_ssa_def
*def
, void *state
);
2047 typedef bool (*nir_foreach_dest_cb
)(nir_dest
*dest
, void *state
);
2048 typedef bool (*nir_foreach_src_cb
)(nir_src
*src
, void *state
);
2049 bool nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
,
2051 bool nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
);
2052 bool nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
);
2054 nir_const_value
*nir_src_as_const_value(nir_src src
);
2055 bool nir_src_is_dynamically_uniform(nir_src src
);
2056 bool nir_srcs_equal(nir_src src1
, nir_src src2
);
2057 void nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
);
2058 void nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
);
2059 void nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
);
2060 void nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
,
2063 void nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
2064 unsigned num_components
, const char *name
);
2065 void nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
2066 unsigned num_components
, const char *name
);
2067 void nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
);
2068 void nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
2069 nir_instr
*after_me
);
2071 /* visits basic blocks in source-code order */
2072 typedef bool (*nir_foreach_block_cb
)(nir_block
*block
, void *state
);
2073 bool nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
2075 bool nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
2077 bool nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
2080 /* If the following CF node is an if, this function returns that if.
2081 * Otherwise, it returns NULL.
2083 nir_if
*nir_block_get_following_if(nir_block
*block
);
2085 nir_loop
*nir_block_get_following_loop(nir_block
*block
);
2087 void nir_index_local_regs(nir_function_impl
*impl
);
2088 void nir_index_global_regs(nir_shader
*shader
);
2089 void nir_index_ssa_defs(nir_function_impl
*impl
);
2090 unsigned nir_index_instrs(nir_function_impl
*impl
);
2092 void nir_index_blocks(nir_function_impl
*impl
);
2094 void nir_print_shader(nir_shader
*shader
, FILE *fp
);
2095 void nir_print_instr(const nir_instr
*instr
, FILE *fp
);
2097 nir_shader
* nir_shader_clone(void *mem_ctx
, const nir_shader
*s
);
2098 nir_function_impl
*nir_function_impl_clone(const nir_function_impl
*fi
);
2099 nir_constant
*nir_constant_clone(const nir_constant
*c
, nir_variable
*var
);
2102 void nir_validate_shader(nir_shader
*shader
);
2103 void nir_metadata_set_validation_flag(nir_shader
*shader
);
2104 void nir_metadata_check_validation_flag(nir_shader
*shader
);
2106 #include "util/debug.h"
2108 should_clone_nir(void)
2110 static int should_clone
= -1;
2111 if (should_clone
< 0)
2112 should_clone
= env_var_as_boolean("NIR_TEST_CLONE", false);
2114 return should_clone
;
2117 static inline void nir_validate_shader(nir_shader
*shader
) { (void) shader
; }
2118 static inline void nir_metadata_set_validation_flag(nir_shader
*shader
) { (void) shader
; }
2119 static inline void nir_metadata_check_validation_flag(nir_shader
*shader
) { (void) shader
; }
2120 static inline bool should_clone_nir(void) { return false; }
2123 #define _PASS(nir, do_pass) do { \
2125 nir_validate_shader(nir); \
2126 if (should_clone_nir()) { \
2127 nir_shader *clone = nir_shader_clone(ralloc_parent(nir), nir); \
2133 #define NIR_PASS(progress, nir, pass, ...) _PASS(nir, \
2134 nir_metadata_set_validation_flag(nir); \
2135 if (pass(nir, ##__VA_ARGS__)) { \
2137 nir_metadata_check_validation_flag(nir); \
2141 #define NIR_PASS_V(nir, pass, ...) _PASS(nir, \
2142 pass(nir, ##__VA_ARGS__); \
2145 void nir_calc_dominance_impl(nir_function_impl
*impl
);
2146 void nir_calc_dominance(nir_shader
*shader
);
2148 nir_block
*nir_dominance_lca(nir_block
*b1
, nir_block
*b2
);
2149 bool nir_block_dominates(nir_block
*parent
, nir_block
*child
);
2151 void nir_dump_dom_tree_impl(nir_function_impl
*impl
, FILE *fp
);
2152 void nir_dump_dom_tree(nir_shader
*shader
, FILE *fp
);
2154 void nir_dump_dom_frontier_impl(nir_function_impl
*impl
, FILE *fp
);
2155 void nir_dump_dom_frontier(nir_shader
*shader
, FILE *fp
);
2157 void nir_dump_cfg_impl(nir_function_impl
*impl
, FILE *fp
);
2158 void nir_dump_cfg(nir_shader
*shader
, FILE *fp
);
2160 int nir_gs_count_vertices(const nir_shader
*shader
);
2162 bool nir_split_var_copies(nir_shader
*shader
);
2164 bool nir_lower_returns_impl(nir_function_impl
*impl
);
2165 bool nir_lower_returns(nir_shader
*shader
);
2167 bool nir_inline_functions(nir_shader
*shader
);
2169 void nir_lower_var_copy_instr(nir_intrinsic_instr
*copy
, void *mem_ctx
);
2170 void nir_lower_var_copies(nir_shader
*shader
);
2172 bool nir_lower_global_vars_to_local(nir_shader
*shader
);
2174 bool nir_lower_indirect_derefs(nir_shader
*shader
, uint32_t mode_mask
);
2176 bool nir_lower_locals_to_regs(nir_shader
*shader
);
2178 void nir_lower_outputs_to_temporaries(nir_shader
*shader
,
2179 nir_function
*entrypoint
);
2181 void nir_shader_gather_info(nir_shader
*shader
, nir_function_impl
*entrypoint
);
2183 void nir_assign_var_locations(struct exec_list
*var_list
,
2185 int (*type_size
)(const struct glsl_type
*));
2187 void nir_lower_io(nir_shader
*shader
,
2188 nir_variable_mode mode
,
2189 int (*type_size
)(const struct glsl_type
*));
2190 nir_src
*nir_get_io_offset_src(nir_intrinsic_instr
*instr
);
2191 nir_src
*nir_get_io_vertex_index_src(nir_intrinsic_instr
*instr
);
2193 void nir_lower_vars_to_ssa(nir_shader
*shader
);
2195 bool nir_remove_dead_variables(nir_shader
*shader
, nir_variable_mode mode
);
2197 void nir_move_vec_src_uses_to_dest(nir_shader
*shader
);
2198 bool nir_lower_vec_to_movs(nir_shader
*shader
);
2199 void nir_lower_alu_to_scalar(nir_shader
*shader
);
2200 void nir_lower_load_const_to_scalar(nir_shader
*shader
);
2202 void nir_lower_phis_to_scalar(nir_shader
*shader
);
2204 void nir_lower_samplers(nir_shader
*shader
,
2205 const struct gl_shader_program
*shader_program
);
2207 bool nir_lower_system_values(nir_shader
*shader
);
2209 typedef struct nir_lower_tex_options
{
2211 * bitmask of (1 << GLSL_SAMPLER_DIM_x) to control for which
2212 * sampler types a texture projector is lowered.
2217 * If true, lower rect textures to 2D, using txs to fetch the
2218 * texture dimensions and dividing the texture coords by the
2219 * texture dims to normalize.
2224 * To emulate certain texture wrap modes, this can be used
2225 * to saturate the specified tex coord to [0.0, 1.0]. The
2226 * bits are according to sampler #, ie. if, for example:
2228 * (conf->saturate_s & (1 << n))
2230 * is true, then the s coord for sampler n is saturated.
2232 * Note that clamping must happen *after* projector lowering
2233 * so any projected texture sample instruction with a clamped
2234 * coordinate gets automatically lowered, regardless of the
2235 * 'lower_txp' setting.
2237 unsigned saturate_s
;
2238 unsigned saturate_t
;
2239 unsigned saturate_r
;
2241 /* Bitmask of textures that need swizzling.
2243 * If (swizzle_result & (1 << texture_index)), then the swizzle in
2244 * swizzles[texture_index] is applied to the result of the texturing
2247 unsigned swizzle_result
;
2249 /* A swizzle for each texture. Values 0-3 represent x, y, z, or w swizzles
2250 * while 4 and 5 represent 0 and 1 respectively.
2252 uint8_t swizzles
[32][4];
2253 } nir_lower_tex_options
;
2255 bool nir_lower_tex(nir_shader
*shader
,
2256 const nir_lower_tex_options
*options
);
2258 void nir_lower_idiv(nir_shader
*shader
);
2260 void nir_lower_clip_vs(nir_shader
*shader
, unsigned ucp_enables
);
2261 void nir_lower_clip_fs(nir_shader
*shader
, unsigned ucp_enables
);
2263 void nir_lower_two_sided_color(nir_shader
*shader
);
2265 void nir_lower_atomics(nir_shader
*shader
,
2266 const struct gl_shader_program
*shader_program
);
2267 void nir_lower_to_source_mods(nir_shader
*shader
);
2269 bool nir_lower_gs_intrinsics(nir_shader
*shader
);
2271 bool nir_normalize_cubemap_coords(nir_shader
*shader
);
2273 void nir_live_ssa_defs_impl(nir_function_impl
*impl
);
2274 bool nir_ssa_defs_interfere(nir_ssa_def
*a
, nir_ssa_def
*b
);
2276 void nir_convert_to_ssa_impl(nir_function_impl
*impl
);
2277 void nir_convert_to_ssa(nir_shader
*shader
);
2279 bool nir_repair_ssa_impl(nir_function_impl
*impl
);
2280 bool nir_repair_ssa(nir_shader
*shader
);
2282 /* If phi_webs_only is true, only convert SSA values involved in phi nodes to
2283 * registers. If false, convert all values (even those not involved in a phi
2284 * node) to registers.
2286 void nir_convert_from_ssa(nir_shader
*shader
, bool phi_webs_only
);
2288 bool nir_opt_algebraic(nir_shader
*shader
);
2289 bool nir_opt_algebraic_late(nir_shader
*shader
);
2290 bool nir_opt_constant_folding(nir_shader
*shader
);
2292 bool nir_opt_global_to_local(nir_shader
*shader
);
2294 bool nir_copy_prop(nir_shader
*shader
);
2296 bool nir_opt_cse(nir_shader
*shader
);
2298 bool nir_opt_dce(nir_shader
*shader
);
2300 bool nir_opt_dead_cf(nir_shader
*shader
);
2302 void nir_opt_gcm(nir_shader
*shader
);
2304 bool nir_opt_peephole_select(nir_shader
*shader
);
2306 bool nir_opt_remove_phis(nir_shader
*shader
);
2308 bool nir_opt_undef(nir_shader
*shader
);
2310 void nir_sweep(nir_shader
*shader
);
2312 nir_intrinsic_op
nir_intrinsic_from_system_value(gl_system_value val
);
2313 gl_system_value
nir_system_value_from_intrinsic(nir_intrinsic_op intrin
);