nir: Store num_direct_uniforms in the nir_shader.
[mesa.git] / src / glsl / nir / nir.h
1 /*
2 * Copyright © 2014 Connor Abbott
3 *
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:
10 *
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
13 * Software.
14 *
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
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 * Connor Abbott (cwabbott0@gmail.com)
25 *
26 */
27
28 #pragma once
29
30 #include "util/hash_table.h"
31 #include "../list.h"
32 #include "GL/gl.h" /* GLenum */
33 #include "util/ralloc.h"
34 #include "util/set.h"
35 #include "util/bitset.h"
36 #include "nir_types.h"
37 #include "glsl/shader_enums.h"
38 #include <stdio.h>
39
40 #include "nir_opcodes.h"
41
42 #ifdef __cplusplus
43 extern "C" {
44 #endif
45
46 struct gl_program;
47 struct gl_shader_program;
48
49 #define NIR_FALSE 0u
50 #define NIR_TRUE (~0u)
51
52 /** Defines a cast function
53 *
54 * This macro defines a cast function from in_type to out_type where
55 * out_type is some structure type that contains a field of type out_type.
56 *
57 * Note that you have to be a bit careful as the generated cast function
58 * destroys constness.
59 */
60 #define NIR_DEFINE_CAST(name, in_type, out_type, field) \
61 static inline out_type * \
62 name(const in_type *parent) \
63 { \
64 return exec_node_data(out_type, parent, field); \
65 }
66
67 struct nir_function_overload;
68 struct nir_function;
69 struct nir_shader;
70 struct nir_instr;
71
72
73 /**
74 * Description of built-in state associated with a uniform
75 *
76 * \sa nir_variable::state_slots
77 */
78 typedef struct {
79 int tokens[5];
80 int swizzle;
81 } nir_state_slot;
82
83 typedef enum {
84 nir_var_shader_in,
85 nir_var_shader_out,
86 nir_var_global,
87 nir_var_local,
88 nir_var_uniform,
89 nir_var_system_value
90 } nir_variable_mode;
91
92 /**
93 * Data stored in an nir_constant
94 */
95 union nir_constant_data {
96 unsigned u[16];
97 int i[16];
98 float f[16];
99 bool b[16];
100 };
101
102 typedef struct nir_constant {
103 /**
104 * Value of the constant.
105 *
106 * The field used to back the values supplied by the constant is determined
107 * by the type associated with the \c nir_variable. Constants may be
108 * scalars, vectors, or matrices.
109 */
110 union nir_constant_data value;
111
112 /* Array elements / Structure Fields */
113 struct nir_constant **elements;
114 } nir_constant;
115
116 /**
117 * \brief Layout qualifiers for gl_FragDepth.
118 *
119 * The AMD/ARB_conservative_depth extensions allow gl_FragDepth to be redeclared
120 * with a layout qualifier.
121 */
122 typedef enum {
123 nir_depth_layout_none, /**< No depth layout is specified. */
124 nir_depth_layout_any,
125 nir_depth_layout_greater,
126 nir_depth_layout_less,
127 nir_depth_layout_unchanged
128 } nir_depth_layout;
129
130 /**
131 * Either a uniform, global variable, shader input, or shader output. Based on
132 * ir_variable - it should be easy to translate between the two.
133 */
134
135 typedef struct {
136 struct exec_node node;
137
138 /**
139 * Declared type of the variable
140 */
141 const struct glsl_type *type;
142
143 /**
144 * Declared name of the variable
145 */
146 char *name;
147
148 /**
149 * For variables which satisfy the is_interface_instance() predicate, this
150 * points to an array of integers such that if the ith member of the
151 * interface block is an array, max_ifc_array_access[i] is the maximum
152 * array element of that member that has been accessed. If the ith member
153 * of the interface block is not an array, max_ifc_array_access[i] is
154 * unused.
155 *
156 * For variables whose type is not an interface block, this pointer is
157 * NULL.
158 */
159 unsigned *max_ifc_array_access;
160
161 struct nir_variable_data {
162
163 /**
164 * Is the variable read-only?
165 *
166 * This is set for variables declared as \c const, shader inputs,
167 * and uniforms.
168 */
169 unsigned read_only:1;
170 unsigned centroid:1;
171 unsigned sample:1;
172 unsigned invariant:1;
173
174 /**
175 * Storage class of the variable.
176 *
177 * \sa nir_variable_mode
178 */
179 nir_variable_mode mode:4;
180
181 /**
182 * Interpolation mode for shader inputs / outputs
183 *
184 * \sa glsl_interp_qualifier
185 */
186 unsigned interpolation:2;
187
188 /**
189 * \name ARB_fragment_coord_conventions
190 * @{
191 */
192 unsigned origin_upper_left:1;
193 unsigned pixel_center_integer:1;
194 /*@}*/
195
196 /**
197 * Was the location explicitly set in the shader?
198 *
199 * If the location is explicitly set in the shader, it \b cannot be changed
200 * by the linker or by the API (e.g., calls to \c glBindAttribLocation have
201 * no effect).
202 */
203 unsigned explicit_location:1;
204 unsigned explicit_index:1;
205
206 /**
207 * Was an initial binding explicitly set in the shader?
208 *
209 * If so, constant_initializer contains an integer nir_constant
210 * representing the initial binding point.
211 */
212 unsigned explicit_binding:1;
213
214 /**
215 * Does this variable have an initializer?
216 *
217 * This is used by the linker to cross-validiate initializers of global
218 * variables.
219 */
220 unsigned has_initializer:1;
221
222 /**
223 * Is this variable a generic output or input that has not yet been matched
224 * up to a variable in another stage of the pipeline?
225 *
226 * This is used by the linker as scratch storage while assigning locations
227 * to generic inputs and outputs.
228 */
229 unsigned is_unmatched_generic_inout:1;
230
231 /**
232 * If non-zero, then this variable may be packed along with other variables
233 * into a single varying slot, so this offset should be applied when
234 * accessing components. For example, an offset of 1 means that the x
235 * component of this variable is actually stored in component y of the
236 * location specified by \c location.
237 */
238 unsigned location_frac:2;
239
240 /**
241 * Non-zero if this variable was created by lowering a named interface
242 * block which was not an array.
243 *
244 * Note that this variable and \c from_named_ifc_block_array will never
245 * both be non-zero.
246 */
247 unsigned from_named_ifc_block_nonarray:1;
248
249 /**
250 * Non-zero if this variable was created by lowering a named interface
251 * block which was an array.
252 *
253 * Note that this variable and \c from_named_ifc_block_nonarray will never
254 * both be non-zero.
255 */
256 unsigned from_named_ifc_block_array:1;
257
258 /**
259 * \brief Layout qualifier for gl_FragDepth.
260 *
261 * This is not equal to \c ir_depth_layout_none if and only if this
262 * variable is \c gl_FragDepth and a layout qualifier is specified.
263 */
264 nir_depth_layout depth_layout;
265
266 /**
267 * Storage location of the base of this variable
268 *
269 * The precise meaning of this field depends on the nature of the variable.
270 *
271 * - Vertex shader input: one of the values from \c gl_vert_attrib.
272 * - Vertex shader output: one of the values from \c gl_varying_slot.
273 * - Geometry shader input: one of the values from \c gl_varying_slot.
274 * - Geometry shader output: one of the values from \c gl_varying_slot.
275 * - Fragment shader input: one of the values from \c gl_varying_slot.
276 * - Fragment shader output: one of the values from \c gl_frag_result.
277 * - Uniforms: Per-stage uniform slot number for default uniform block.
278 * - Uniforms: Index within the uniform block definition for UBO members.
279 * - Other: This field is not currently used.
280 *
281 * If the variable is a uniform, shader input, or shader output, and the
282 * slot has not been assigned, the value will be -1.
283 */
284 int location;
285
286 /**
287 * The actual location of the variable in the IR. Only valid for inputs
288 * and outputs.
289 */
290 unsigned int driver_location;
291
292 /**
293 * output index for dual source blending.
294 */
295 int index;
296
297 /**
298 * Initial binding point for a sampler or UBO.
299 *
300 * For array types, this represents the binding point for the first element.
301 */
302 int binding;
303
304 /**
305 * Location an atomic counter is stored at.
306 */
307 struct {
308 unsigned buffer_index;
309 unsigned offset;
310 } atomic;
311
312 /**
313 * ARB_shader_image_load_store qualifiers.
314 */
315 struct {
316 bool read_only; /**< "readonly" qualifier. */
317 bool write_only; /**< "writeonly" qualifier. */
318 bool coherent;
319 bool _volatile;
320 bool restrict_flag;
321
322 /** Image internal format if specified explicitly, otherwise GL_NONE. */
323 GLenum format;
324 } image;
325
326 /**
327 * Highest element accessed with a constant expression array index
328 *
329 * Not used for non-array variables.
330 */
331 unsigned max_array_access;
332
333 } data;
334
335 /**
336 * Built-in state that backs this uniform
337 *
338 * Once set at variable creation, \c state_slots must remain invariant.
339 * This is because, ideally, this array would be shared by all clones of
340 * this variable in the IR tree. In other words, we'd really like for it
341 * to be a fly-weight.
342 *
343 * If the variable is not a uniform, \c num_state_slots will be zero and
344 * \c state_slots will be \c NULL.
345 */
346 /*@{*/
347 unsigned num_state_slots; /**< Number of state slots used */
348 nir_state_slot *state_slots; /**< State descriptors. */
349 /*@}*/
350
351 /**
352 * Constant expression assigned in the initializer of the variable
353 */
354 nir_constant *constant_initializer;
355
356 /**
357 * For variables that are in an interface block or are an instance of an
358 * interface block, this is the \c GLSL_TYPE_INTERFACE type for that block.
359 *
360 * \sa ir_variable::location
361 */
362 const struct glsl_type *interface_type;
363 } nir_variable;
364
365 typedef struct {
366 struct exec_node node;
367
368 unsigned num_components; /** < number of vector components */
369 unsigned num_array_elems; /** < size of array (0 for no array) */
370
371 /** generic register index. */
372 unsigned index;
373
374 /** only for debug purposes, can be NULL */
375 const char *name;
376
377 /** whether this register is local (per-function) or global (per-shader) */
378 bool is_global;
379
380 /**
381 * If this flag is set to true, then accessing channels >= num_components
382 * is well-defined, and simply spills over to the next array element. This
383 * is useful for backends that can do per-component accessing, in
384 * particular scalar backends. By setting this flag and making
385 * num_components equal to 1, structures can be packed tightly into
386 * registers and then registers can be accessed per-component to get to
387 * each structure member, even if it crosses vec4 boundaries.
388 */
389 bool is_packed;
390
391 /**
392 * If this pointer is non-NULL then this register has exactly one
393 * definition and that definition dominates all of its uses. This is
394 * set by the out-of-SSA pass so that backends can get SSA-like
395 * information even once they have gone out of SSA.
396 */
397 struct nir_instr *parent_instr;
398
399 /** set of nir_instr's where this register is used (read from) */
400 struct set *uses;
401
402 /** set of nir_instr's where this register is defined (written to) */
403 struct set *defs;
404
405 /** set of nir_if's where this register is used as a condition */
406 struct set *if_uses;
407 } nir_register;
408
409 typedef enum {
410 nir_instr_type_alu,
411 nir_instr_type_call,
412 nir_instr_type_tex,
413 nir_instr_type_intrinsic,
414 nir_instr_type_load_const,
415 nir_instr_type_jump,
416 nir_instr_type_ssa_undef,
417 nir_instr_type_phi,
418 nir_instr_type_parallel_copy,
419 } nir_instr_type;
420
421 typedef struct nir_instr {
422 struct exec_node node;
423 nir_instr_type type;
424 struct nir_block *block;
425
426 /* A temporary for optimization and analysis passes to use for storing
427 * flags. For instance, DCE uses this to store the "dead/live" info.
428 */
429 uint8_t pass_flags;
430 } nir_instr;
431
432 static inline nir_instr *
433 nir_instr_next(nir_instr *instr)
434 {
435 struct exec_node *next = exec_node_get_next(&instr->node);
436 if (exec_node_is_tail_sentinel(next))
437 return NULL;
438 else
439 return exec_node_data(nir_instr, next, node);
440 }
441
442 static inline nir_instr *
443 nir_instr_prev(nir_instr *instr)
444 {
445 struct exec_node *prev = exec_node_get_prev(&instr->node);
446 if (exec_node_is_head_sentinel(prev))
447 return NULL;
448 else
449 return exec_node_data(nir_instr, prev, node);
450 }
451
452 typedef struct {
453 /** for debugging only, can be NULL */
454 const char* name;
455
456 /** generic SSA definition index. */
457 unsigned index;
458
459 /** Index into the live_in and live_out bitfields */
460 unsigned live_index;
461
462 nir_instr *parent_instr;
463
464 /** set of nir_instr's where this register is used (read from) */
465 struct set *uses;
466
467 /** set of nir_if's where this register is used as a condition */
468 struct set *if_uses;
469
470 uint8_t num_components;
471 } nir_ssa_def;
472
473 struct nir_src;
474
475 typedef struct {
476 nir_register *reg;
477 struct nir_src *indirect; /** < NULL for no indirect offset */
478 unsigned base_offset;
479
480 /* TODO use-def chain goes here */
481 } nir_reg_src;
482
483 typedef struct {
484 nir_register *reg;
485 struct nir_src *indirect; /** < NULL for no indirect offset */
486 unsigned base_offset;
487
488 /* TODO def-use chain goes here */
489 } nir_reg_dest;
490
491 typedef struct nir_src {
492 union {
493 nir_reg_src reg;
494 nir_ssa_def *ssa;
495 };
496
497 bool is_ssa;
498 } nir_src;
499
500 typedef struct {
501 union {
502 nir_reg_dest reg;
503 nir_ssa_def ssa;
504 };
505
506 bool is_ssa;
507 } nir_dest;
508
509 static inline nir_src
510 nir_src_for_ssa(nir_ssa_def *def)
511 {
512 nir_src src;
513
514 src.is_ssa = true;
515 src.ssa = def;
516
517 return src;
518 }
519
520 static inline nir_src
521 nir_src_for_reg(nir_register *reg)
522 {
523 nir_src src;
524
525 src.is_ssa = false;
526 src.reg.reg = reg;
527 src.reg.indirect = NULL;
528 src.reg.base_offset = 0;
529
530 return src;
531 }
532
533 static inline nir_instr *
534 nir_src_get_parent_instr(const nir_src *src)
535 {
536 if (src->is_ssa) {
537 return src->ssa->parent_instr;
538 } else {
539 return src->reg.reg->parent_instr;
540 }
541 }
542
543 static inline nir_dest
544 nir_dest_for_reg(nir_register *reg)
545 {
546 nir_dest dest;
547
548 dest.is_ssa = false;
549 dest.reg.reg = reg;
550 dest.reg.indirect = NULL;
551 dest.reg.base_offset = 0;
552
553 return dest;
554 }
555
556 void nir_src_copy(nir_src *dest, const nir_src *src, void *mem_ctx);
557 void nir_dest_copy(nir_dest *dest, const nir_dest *src, void *mem_ctx);
558
559 typedef struct {
560 nir_src src;
561
562 /**
563 * \name input modifiers
564 */
565 /*@{*/
566 /**
567 * For inputs interpreted as floating point, flips the sign bit. For
568 * inputs interpreted as integers, performs the two's complement negation.
569 */
570 bool negate;
571
572 /**
573 * Clears the sign bit for floating point values, and computes the integer
574 * absolute value for integers. Note that the negate modifier acts after
575 * the absolute value modifier, therefore if both are set then all inputs
576 * will become negative.
577 */
578 bool abs;
579 /*@}*/
580
581 /**
582 * For each input component, says which component of the register it is
583 * chosen from. Note that which elements of the swizzle are used and which
584 * are ignored are based on the write mask for most opcodes - for example,
585 * a statement like "foo.xzw = bar.zyx" would have a writemask of 1101b and
586 * a swizzle of {2, x, 1, 0} where x means "don't care."
587 */
588 uint8_t swizzle[4];
589 } nir_alu_src;
590
591 typedef struct {
592 nir_dest dest;
593
594 /**
595 * \name saturate output modifier
596 *
597 * Only valid for opcodes that output floating-point numbers. Clamps the
598 * output to between 0.0 and 1.0 inclusive.
599 */
600
601 bool saturate;
602
603 unsigned write_mask : 4; /* ignored if dest.is_ssa is true */
604 } nir_alu_dest;
605
606 void nir_alu_src_copy(nir_alu_src *dest, const nir_alu_src *src, void *mem_ctx);
607 void nir_alu_dest_copy(nir_alu_dest *dest, const nir_alu_dest *src,
608 void *mem_ctx);
609
610 typedef enum {
611 nir_type_invalid = 0, /* Not a valid type */
612 nir_type_float,
613 nir_type_int,
614 nir_type_unsigned,
615 nir_type_bool
616 } nir_alu_type;
617
618 typedef enum {
619 NIR_OP_IS_COMMUTATIVE = (1 << 0),
620 NIR_OP_IS_ASSOCIATIVE = (1 << 1),
621 } nir_op_algebraic_property;
622
623 typedef struct {
624 const char *name;
625
626 unsigned num_inputs;
627
628 /**
629 * The number of components in the output
630 *
631 * If non-zero, this is the size of the output and input sizes are
632 * explicitly given; swizzle and writemask are still in effect, but if
633 * the output component is masked out, then the input component may
634 * still be in use.
635 *
636 * If zero, the opcode acts in the standard, per-component manner; the
637 * operation is performed on each component (except the ones that are
638 * masked out) with the input being taken from the input swizzle for
639 * that component.
640 *
641 * The size of some of the inputs may be given (i.e. non-zero) even
642 * though output_size is zero; in that case, the inputs with a zero
643 * size act per-component, while the inputs with non-zero size don't.
644 */
645 unsigned output_size;
646
647 /**
648 * The type of vector that the instruction outputs. Note that the
649 * staurate modifier is only allowed on outputs with the float type.
650 */
651
652 nir_alu_type output_type;
653
654 /**
655 * The number of components in each input
656 */
657 unsigned input_sizes[4];
658
659 /**
660 * The type of vector that each input takes. Note that negate and
661 * absolute value are only allowed on inputs with int or float type and
662 * behave differently on the two.
663 */
664 nir_alu_type input_types[4];
665
666 nir_op_algebraic_property algebraic_properties;
667 } nir_op_info;
668
669 extern const nir_op_info nir_op_infos[nir_num_opcodes];
670
671 typedef struct nir_alu_instr {
672 nir_instr instr;
673 nir_op op;
674 nir_alu_dest dest;
675 nir_alu_src src[];
676 } nir_alu_instr;
677
678 /* is this source channel used? */
679 static inline bool
680 nir_alu_instr_channel_used(nir_alu_instr *instr, unsigned src, unsigned channel)
681 {
682 if (nir_op_infos[instr->op].input_sizes[src] > 0)
683 return channel < nir_op_infos[instr->op].input_sizes[src];
684
685 return (instr->dest.write_mask >> channel) & 1;
686 }
687
688 /*
689 * For instructions whose destinations are SSA, get the number of channels
690 * used for a source
691 */
692 static inline unsigned
693 nir_ssa_alu_instr_src_components(nir_alu_instr *instr, unsigned src)
694 {
695 assert(instr->dest.dest.is_ssa);
696
697 if (nir_op_infos[instr->op].input_sizes[src] > 0)
698 return nir_op_infos[instr->op].input_sizes[src];
699
700 return instr->dest.dest.ssa.num_components;
701 }
702
703 typedef enum {
704 nir_deref_type_var,
705 nir_deref_type_array,
706 nir_deref_type_struct
707 } nir_deref_type;
708
709 typedef struct nir_deref {
710 nir_deref_type deref_type;
711 struct nir_deref *child;
712 const struct glsl_type *type;
713 } nir_deref;
714
715 typedef struct {
716 nir_deref deref;
717
718 nir_variable *var;
719 } nir_deref_var;
720
721 /* This enum describes how the array is referenced. If the deref is
722 * direct then the base_offset is used. If the deref is indirect then then
723 * offset is given by base_offset + indirect. If the deref is a wildcard
724 * then the deref refers to all of the elements of the array at the same
725 * time. Wildcard dereferences are only ever allowed in copy_var
726 * intrinsics and the source and destination derefs must have matching
727 * wildcards.
728 */
729 typedef enum {
730 nir_deref_array_type_direct,
731 nir_deref_array_type_indirect,
732 nir_deref_array_type_wildcard,
733 } nir_deref_array_type;
734
735 typedef struct {
736 nir_deref deref;
737
738 nir_deref_array_type deref_array_type;
739 unsigned base_offset;
740 nir_src indirect;
741 } nir_deref_array;
742
743 typedef struct {
744 nir_deref deref;
745
746 unsigned index;
747 } nir_deref_struct;
748
749 NIR_DEFINE_CAST(nir_deref_as_var, nir_deref, nir_deref_var, deref)
750 NIR_DEFINE_CAST(nir_deref_as_array, nir_deref, nir_deref_array, deref)
751 NIR_DEFINE_CAST(nir_deref_as_struct, nir_deref, nir_deref_struct, deref)
752
753 typedef struct {
754 nir_instr instr;
755
756 unsigned num_params;
757 nir_deref_var **params;
758 nir_deref_var *return_deref;
759
760 struct nir_function_overload *callee;
761 } nir_call_instr;
762
763 #define INTRINSIC(name, num_srcs, src_components, has_dest, dest_components, \
764 num_variables, num_indices, flags) \
765 nir_intrinsic_##name,
766
767 #define LAST_INTRINSIC(name) nir_last_intrinsic = nir_intrinsic_##name,
768
769 typedef enum {
770 #include "nir_intrinsics.h"
771 nir_num_intrinsics = nir_last_intrinsic + 1
772 } nir_intrinsic_op;
773
774 #undef INTRINSIC
775 #undef LAST_INTRINSIC
776
777 /** Represents an intrinsic
778 *
779 * An intrinsic is an instruction type for handling things that are
780 * more-or-less regular operations but don't just consume and produce SSA
781 * values like ALU operations do. Intrinsics are not for things that have
782 * special semantic meaning such as phi nodes and parallel copies.
783 * Examples of intrinsics include variable load/store operations, system
784 * value loads, and the like. Even though texturing more-or-less falls
785 * under this category, texturing is its own instruction type because
786 * trying to represent texturing with intrinsics would lead to a
787 * combinatorial explosion of intrinsic opcodes.
788 *
789 * By having a single instruction type for handling a lot of different
790 * cases, optimization passes can look for intrinsics and, for the most
791 * part, completely ignore them. Each intrinsic type also has a few
792 * possible flags that govern whether or not they can be reordered or
793 * eliminated. That way passes like dead code elimination can still work
794 * on intrisics without understanding the meaning of each.
795 *
796 * Each intrinsic has some number of constant indices, some number of
797 * variables, and some number of sources. What these sources, variables,
798 * and indices mean depends on the intrinsic and is documented with the
799 * intrinsic declaration in nir_intrinsics.h. Intrinsics and texture
800 * instructions are the only types of instruction that can operate on
801 * variables.
802 */
803 typedef struct {
804 nir_instr instr;
805
806 nir_intrinsic_op intrinsic;
807
808 nir_dest dest;
809
810 /** number of components if this is a vectorized intrinsic
811 *
812 * Similarly to ALU operations, some intrinsics are vectorized.
813 * An intrinsic is vectorized if nir_intrinsic_infos.dest_components == 0.
814 * For vectorized intrinsics, the num_components field specifies the
815 * number of destination components and the number of source components
816 * for all sources with nir_intrinsic_infos.src_components[i] == 0.
817 */
818 uint8_t num_components;
819
820 int const_index[3];
821
822 nir_deref_var *variables[2];
823
824 nir_src src[];
825 } nir_intrinsic_instr;
826
827 /**
828 * \name NIR intrinsics semantic flags
829 *
830 * information about what the compiler can do with the intrinsics.
831 *
832 * \sa nir_intrinsic_info::flags
833 */
834 typedef enum {
835 /**
836 * whether the intrinsic can be safely eliminated if none of its output
837 * value is not being used.
838 */
839 NIR_INTRINSIC_CAN_ELIMINATE = (1 << 0),
840
841 /**
842 * Whether the intrinsic can be reordered with respect to any other
843 * intrinsic, i.e. whether the only reordering dependencies of the
844 * intrinsic are due to the register reads/writes.
845 */
846 NIR_INTRINSIC_CAN_REORDER = (1 << 1),
847 } nir_intrinsic_semantic_flag;
848
849 #define NIR_INTRINSIC_MAX_INPUTS 4
850
851 typedef struct {
852 const char *name;
853
854 unsigned num_srcs; /** < number of register/SSA inputs */
855
856 /** number of components of each input register
857 *
858 * If this value is 0, the number of components is given by the
859 * num_components field of nir_intrinsic_instr.
860 */
861 unsigned src_components[NIR_INTRINSIC_MAX_INPUTS];
862
863 bool has_dest;
864
865 /** number of components of the output register
866 *
867 * If this value is 0, the number of components is given by the
868 * num_components field of nir_intrinsic_instr.
869 */
870 unsigned dest_components;
871
872 /** the number of inputs/outputs that are variables */
873 unsigned num_variables;
874
875 /** the number of constant indices used by the intrinsic */
876 unsigned num_indices;
877
878 /** semantic flags for calls to this intrinsic */
879 nir_intrinsic_semantic_flag flags;
880 } nir_intrinsic_info;
881
882 extern const nir_intrinsic_info nir_intrinsic_infos[nir_num_intrinsics];
883
884 /**
885 * \group texture information
886 *
887 * This gives semantic information about textures which is useful to the
888 * frontend, the backend, and lowering passes, but not the optimizer.
889 */
890
891 typedef enum {
892 nir_tex_src_coord,
893 nir_tex_src_projector,
894 nir_tex_src_comparitor, /* shadow comparitor */
895 nir_tex_src_offset,
896 nir_tex_src_bias,
897 nir_tex_src_lod,
898 nir_tex_src_ms_index, /* MSAA sample index */
899 nir_tex_src_ddx,
900 nir_tex_src_ddy,
901 nir_tex_src_sampler_offset, /* < dynamically uniform indirect offset */
902 nir_num_tex_src_types
903 } nir_tex_src_type;
904
905 typedef struct {
906 nir_src src;
907 nir_tex_src_type src_type;
908 } nir_tex_src;
909
910 typedef enum {
911 nir_texop_tex, /**< Regular texture look-up */
912 nir_texop_txb, /**< Texture look-up with LOD bias */
913 nir_texop_txl, /**< Texture look-up with explicit LOD */
914 nir_texop_txd, /**< Texture look-up with partial derivatvies */
915 nir_texop_txf, /**< Texel fetch with explicit LOD */
916 nir_texop_txf_ms, /**< Multisample texture fetch */
917 nir_texop_txs, /**< Texture size */
918 nir_texop_lod, /**< Texture lod query */
919 nir_texop_tg4, /**< Texture gather */
920 nir_texop_query_levels /**< Texture levels query */
921 } nir_texop;
922
923 typedef struct {
924 nir_instr instr;
925
926 enum glsl_sampler_dim sampler_dim;
927 nir_alu_type dest_type;
928
929 nir_texop op;
930 nir_dest dest;
931 nir_tex_src *src;
932 unsigned num_srcs, coord_components;
933 bool is_array, is_shadow;
934
935 /**
936 * If is_shadow is true, whether this is the old-style shadow that outputs 4
937 * components or the new-style shadow that outputs 1 component.
938 */
939 bool is_new_style_shadow;
940
941 /* constant offset - must be 0 if the offset source is used */
942 int const_offset[4];
943
944 /* gather component selector */
945 unsigned component : 2;
946
947 /** The sampler index
948 *
949 * If this texture instruction has a nir_tex_src_sampler_offset source,
950 * then the sampler index is given by sampler_index + sampler_offset.
951 */
952 unsigned sampler_index;
953
954 /** The size of the sampler array or 0 if it's not an array */
955 unsigned sampler_array_size;
956
957 nir_deref_var *sampler; /* if this is NULL, use sampler_index instead */
958 } nir_tex_instr;
959
960 static inline unsigned
961 nir_tex_instr_dest_size(nir_tex_instr *instr)
962 {
963 if (instr->op == nir_texop_txs) {
964 unsigned ret;
965 switch (instr->sampler_dim) {
966 case GLSL_SAMPLER_DIM_1D:
967 case GLSL_SAMPLER_DIM_BUF:
968 ret = 1;
969 break;
970 case GLSL_SAMPLER_DIM_2D:
971 case GLSL_SAMPLER_DIM_CUBE:
972 case GLSL_SAMPLER_DIM_MS:
973 case GLSL_SAMPLER_DIM_RECT:
974 case GLSL_SAMPLER_DIM_EXTERNAL:
975 ret = 2;
976 break;
977 case GLSL_SAMPLER_DIM_3D:
978 ret = 3;
979 break;
980 default:
981 unreachable("not reached");
982 }
983 if (instr->is_array)
984 ret++;
985 return ret;
986 }
987
988 if (instr->op == nir_texop_query_levels)
989 return 2;
990
991 if (instr->is_shadow && instr->is_new_style_shadow)
992 return 1;
993
994 return 4;
995 }
996
997 static inline unsigned
998 nir_tex_instr_src_size(nir_tex_instr *instr, unsigned src)
999 {
1000 if (instr->src[src].src_type == nir_tex_src_coord)
1001 return instr->coord_components;
1002
1003
1004 if (instr->src[src].src_type == nir_tex_src_offset ||
1005 instr->src[src].src_type == nir_tex_src_ddx ||
1006 instr->src[src].src_type == nir_tex_src_ddy) {
1007 if (instr->is_array)
1008 return instr->coord_components - 1;
1009 else
1010 return instr->coord_components;
1011 }
1012
1013 return 1;
1014 }
1015
1016 static inline int
1017 nir_tex_instr_src_index(nir_tex_instr *instr, nir_tex_src_type type)
1018 {
1019 for (unsigned i = 0; i < instr->num_srcs; i++)
1020 if (instr->src[i].src_type == type)
1021 return (int) i;
1022
1023 return -1;
1024 }
1025
1026 typedef struct {
1027 union {
1028 float f[4];
1029 int32_t i[4];
1030 uint32_t u[4];
1031 };
1032 } nir_const_value;
1033
1034 typedef struct {
1035 nir_instr instr;
1036
1037 nir_const_value value;
1038
1039 nir_ssa_def def;
1040 } nir_load_const_instr;
1041
1042 typedef enum {
1043 nir_jump_return,
1044 nir_jump_break,
1045 nir_jump_continue,
1046 } nir_jump_type;
1047
1048 typedef struct {
1049 nir_instr instr;
1050 nir_jump_type type;
1051 } nir_jump_instr;
1052
1053 /* creates a new SSA variable in an undefined state */
1054
1055 typedef struct {
1056 nir_instr instr;
1057 nir_ssa_def def;
1058 } nir_ssa_undef_instr;
1059
1060 typedef struct {
1061 struct exec_node node;
1062
1063 /* The predecessor block corresponding to this source */
1064 struct nir_block *pred;
1065
1066 nir_src src;
1067 } nir_phi_src;
1068
1069 #define nir_foreach_phi_src(phi, entry) \
1070 foreach_list_typed(nir_phi_src, entry, node, &(phi)->srcs)
1071
1072 typedef struct {
1073 nir_instr instr;
1074
1075 struct exec_list srcs; /** < list of nir_phi_src */
1076
1077 nir_dest dest;
1078 } nir_phi_instr;
1079
1080 typedef struct {
1081 struct exec_node node;
1082 nir_src src;
1083 nir_dest dest;
1084 } nir_parallel_copy_entry;
1085
1086 #define nir_foreach_parallel_copy_entry(pcopy, entry) \
1087 foreach_list_typed(nir_parallel_copy_entry, entry, node, &(pcopy)->entries)
1088
1089 typedef struct {
1090 nir_instr instr;
1091
1092 /* A list of nir_parallel_copy_entry's. The sources of all of the
1093 * entries are copied to the corresponding destinations "in parallel".
1094 * In other words, if we have two entries: a -> b and b -> a, the values
1095 * get swapped.
1096 */
1097 struct exec_list entries;
1098 } nir_parallel_copy_instr;
1099
1100 NIR_DEFINE_CAST(nir_instr_as_alu, nir_instr, nir_alu_instr, instr)
1101 NIR_DEFINE_CAST(nir_instr_as_call, nir_instr, nir_call_instr, instr)
1102 NIR_DEFINE_CAST(nir_instr_as_jump, nir_instr, nir_jump_instr, instr)
1103 NIR_DEFINE_CAST(nir_instr_as_tex, nir_instr, nir_tex_instr, instr)
1104 NIR_DEFINE_CAST(nir_instr_as_intrinsic, nir_instr, nir_intrinsic_instr, instr)
1105 NIR_DEFINE_CAST(nir_instr_as_load_const, nir_instr, nir_load_const_instr, instr)
1106 NIR_DEFINE_CAST(nir_instr_as_ssa_undef, nir_instr, nir_ssa_undef_instr, instr)
1107 NIR_DEFINE_CAST(nir_instr_as_phi, nir_instr, nir_phi_instr, instr)
1108 NIR_DEFINE_CAST(nir_instr_as_parallel_copy, nir_instr,
1109 nir_parallel_copy_instr, instr)
1110
1111 /*
1112 * Control flow
1113 *
1114 * Control flow consists of a tree of control flow nodes, which include
1115 * if-statements and loops. The leaves of the tree are basic blocks, lists of
1116 * instructions that always run start-to-finish. Each basic block also keeps
1117 * track of its successors (blocks which may run immediately after the current
1118 * block) and predecessors (blocks which could have run immediately before the
1119 * current block). Each function also has a start block and an end block which
1120 * all return statements point to (which is always empty). Together, all the
1121 * blocks with their predecessors and successors make up the control flow
1122 * graph (CFG) of the function. There are helpers that modify the tree of
1123 * control flow nodes while modifying the CFG appropriately; these should be
1124 * used instead of modifying the tree directly.
1125 */
1126
1127 typedef enum {
1128 nir_cf_node_block,
1129 nir_cf_node_if,
1130 nir_cf_node_loop,
1131 nir_cf_node_function
1132 } nir_cf_node_type;
1133
1134 typedef struct nir_cf_node {
1135 struct exec_node node;
1136 nir_cf_node_type type;
1137 struct nir_cf_node *parent;
1138 } nir_cf_node;
1139
1140 typedef struct nir_block {
1141 nir_cf_node cf_node;
1142
1143 struct exec_list instr_list; /** < list of nir_instr */
1144
1145 /** generic block index; generated by nir_index_blocks */
1146 unsigned index;
1147
1148 /*
1149 * Each block can only have up to 2 successors, so we put them in a simple
1150 * array - no need for anything more complicated.
1151 */
1152 struct nir_block *successors[2];
1153
1154 /* Set of nir_block predecessors in the CFG */
1155 struct set *predecessors;
1156
1157 /*
1158 * this node's immediate dominator in the dominance tree - set to NULL for
1159 * the start block.
1160 */
1161 struct nir_block *imm_dom;
1162
1163 /* This node's children in the dominance tree */
1164 unsigned num_dom_children;
1165 struct nir_block **dom_children;
1166
1167 /* Set of nir_block's on the dominance frontier of this block */
1168 struct set *dom_frontier;
1169
1170 /*
1171 * These two indices have the property that dom_{pre,post}_index for each
1172 * child of this block in the dominance tree will always be between
1173 * dom_pre_index and dom_post_index for this block, which makes testing if
1174 * a given block is dominated by another block an O(1) operation.
1175 */
1176 unsigned dom_pre_index, dom_post_index;
1177
1178 /* live in and out for this block; used for liveness analysis */
1179 BITSET_WORD *live_in;
1180 BITSET_WORD *live_out;
1181 } nir_block;
1182
1183 static inline nir_instr *
1184 nir_block_first_instr(nir_block *block)
1185 {
1186 struct exec_node *head = exec_list_get_head(&block->instr_list);
1187 return exec_node_data(nir_instr, head, node);
1188 }
1189
1190 static inline nir_instr *
1191 nir_block_last_instr(nir_block *block)
1192 {
1193 struct exec_node *tail = exec_list_get_tail(&block->instr_list);
1194 return exec_node_data(nir_instr, tail, node);
1195 }
1196
1197 #define nir_foreach_instr(block, instr) \
1198 foreach_list_typed(nir_instr, instr, node, &(block)->instr_list)
1199 #define nir_foreach_instr_reverse(block, instr) \
1200 foreach_list_typed_reverse(nir_instr, instr, node, &(block)->instr_list)
1201 #define nir_foreach_instr_safe(block, instr) \
1202 foreach_list_typed_safe(nir_instr, instr, node, &(block)->instr_list)
1203
1204 typedef struct {
1205 nir_cf_node cf_node;
1206 nir_src condition;
1207
1208 struct exec_list then_list; /** < list of nir_cf_node */
1209 struct exec_list else_list; /** < list of nir_cf_node */
1210 } nir_if;
1211
1212 static inline nir_cf_node *
1213 nir_if_first_then_node(nir_if *if_stmt)
1214 {
1215 struct exec_node *head = exec_list_get_head(&if_stmt->then_list);
1216 return exec_node_data(nir_cf_node, head, node);
1217 }
1218
1219 static inline nir_cf_node *
1220 nir_if_last_then_node(nir_if *if_stmt)
1221 {
1222 struct exec_node *tail = exec_list_get_tail(&if_stmt->then_list);
1223 return exec_node_data(nir_cf_node, tail, node);
1224 }
1225
1226 static inline nir_cf_node *
1227 nir_if_first_else_node(nir_if *if_stmt)
1228 {
1229 struct exec_node *head = exec_list_get_head(&if_stmt->else_list);
1230 return exec_node_data(nir_cf_node, head, node);
1231 }
1232
1233 static inline nir_cf_node *
1234 nir_if_last_else_node(nir_if *if_stmt)
1235 {
1236 struct exec_node *tail = exec_list_get_tail(&if_stmt->else_list);
1237 return exec_node_data(nir_cf_node, tail, node);
1238 }
1239
1240 typedef struct {
1241 nir_cf_node cf_node;
1242
1243 struct exec_list body; /** < list of nir_cf_node */
1244 } nir_loop;
1245
1246 static inline nir_cf_node *
1247 nir_loop_first_cf_node(nir_loop *loop)
1248 {
1249 return exec_node_data(nir_cf_node, exec_list_get_head(&loop->body), node);
1250 }
1251
1252 static inline nir_cf_node *
1253 nir_loop_last_cf_node(nir_loop *loop)
1254 {
1255 return exec_node_data(nir_cf_node, exec_list_get_tail(&loop->body), node);
1256 }
1257
1258 /**
1259 * Various bits of metadata that can may be created or required by
1260 * optimization and analysis passes
1261 */
1262 typedef enum {
1263 nir_metadata_none = 0x0,
1264 nir_metadata_block_index = 0x1,
1265 nir_metadata_dominance = 0x2,
1266 nir_metadata_live_variables = 0x4,
1267 } nir_metadata;
1268
1269 typedef struct {
1270 nir_cf_node cf_node;
1271
1272 /** pointer to the overload of which this is an implementation */
1273 struct nir_function_overload *overload;
1274
1275 struct exec_list body; /** < list of nir_cf_node */
1276
1277 nir_block *start_block, *end_block;
1278
1279 /** list for all local variables in the function */
1280 struct exec_list locals;
1281
1282 /** array of variables used as parameters */
1283 unsigned num_params;
1284 nir_variable **params;
1285
1286 /** variable used to hold the result of the function */
1287 nir_variable *return_var;
1288
1289 /** list of local registers in the function */
1290 struct exec_list registers;
1291
1292 /** next available local register index */
1293 unsigned reg_alloc;
1294
1295 /** next available SSA value index */
1296 unsigned ssa_alloc;
1297
1298 /* total number of basic blocks, only valid when block_index_dirty = false */
1299 unsigned num_blocks;
1300
1301 nir_metadata valid_metadata;
1302 } nir_function_impl;
1303
1304 static inline nir_cf_node *
1305 nir_cf_node_next(nir_cf_node *node)
1306 {
1307 struct exec_node *next = exec_node_get_next(&node->node);
1308 if (exec_node_is_tail_sentinel(next))
1309 return NULL;
1310 else
1311 return exec_node_data(nir_cf_node, next, node);
1312 }
1313
1314 static inline nir_cf_node *
1315 nir_cf_node_prev(nir_cf_node *node)
1316 {
1317 struct exec_node *prev = exec_node_get_prev(&node->node);
1318 if (exec_node_is_head_sentinel(prev))
1319 return NULL;
1320 else
1321 return exec_node_data(nir_cf_node, prev, node);
1322 }
1323
1324 static inline bool
1325 nir_cf_node_is_first(const nir_cf_node *node)
1326 {
1327 return exec_node_is_head_sentinel(node->node.prev);
1328 }
1329
1330 static inline bool
1331 nir_cf_node_is_last(const nir_cf_node *node)
1332 {
1333 return exec_node_is_tail_sentinel(node->node.next);
1334 }
1335
1336 NIR_DEFINE_CAST(nir_cf_node_as_block, nir_cf_node, nir_block, cf_node)
1337 NIR_DEFINE_CAST(nir_cf_node_as_if, nir_cf_node, nir_if, cf_node)
1338 NIR_DEFINE_CAST(nir_cf_node_as_loop, nir_cf_node, nir_loop, cf_node)
1339 NIR_DEFINE_CAST(nir_cf_node_as_function, nir_cf_node, nir_function_impl, cf_node)
1340
1341 typedef enum {
1342 nir_parameter_in,
1343 nir_parameter_out,
1344 nir_parameter_inout,
1345 } nir_parameter_type;
1346
1347 typedef struct {
1348 nir_parameter_type param_type;
1349 const struct glsl_type *type;
1350 } nir_parameter;
1351
1352 typedef struct nir_function_overload {
1353 struct exec_node node;
1354
1355 unsigned num_params;
1356 nir_parameter *params;
1357 const struct glsl_type *return_type;
1358
1359 nir_function_impl *impl; /** < NULL if the overload is only declared yet */
1360
1361 /** pointer to the function of which this is an overload */
1362 struct nir_function *function;
1363 } nir_function_overload;
1364
1365 typedef struct nir_function {
1366 struct exec_node node;
1367
1368 struct exec_list overload_list; /** < list of nir_function_overload */
1369 const char *name;
1370 struct nir_shader *shader;
1371 } nir_function;
1372
1373 #define nir_function_first_overload(func) \
1374 exec_node_data(nir_function_overload, \
1375 exec_list_get_head(&(func)->overload_list), node)
1376
1377 typedef struct nir_shader_compiler_options {
1378 bool lower_ffma;
1379 bool lower_flrp;
1380 bool lower_fpow;
1381 bool lower_fsat;
1382 bool lower_fsqrt;
1383 /** lowers fneg and ineg to fsub and isub. */
1384 bool lower_negate;
1385 /** lowers fsub and isub to fadd+fneg and iadd+ineg. */
1386 bool lower_sub;
1387
1388 /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
1389 bool lower_scmp;
1390
1391 /**
1392 * Does the driver support real 32-bit integers? (Otherwise, integers
1393 * are simulated by floats.)
1394 */
1395 bool native_integers;
1396 } nir_shader_compiler_options;
1397
1398 typedef struct nir_shader {
1399 /** hash table of name -> uniform nir_variable */
1400 struct exec_list uniforms;
1401
1402 /** hash table of name -> input nir_variable */
1403 struct exec_list inputs;
1404
1405 /** hash table of name -> output nir_variable */
1406 struct exec_list outputs;
1407
1408 /** Set of driver-specific options for the shader.
1409 *
1410 * The memory for the options is expected to be kept in a single static
1411 * copy by the driver.
1412 */
1413 const struct nir_shader_compiler_options *options;
1414
1415 /** list of global variables in the shader */
1416 struct exec_list globals;
1417
1418 /** list of system value variables in the shader */
1419 struct exec_list system_values;
1420
1421 struct exec_list functions; /** < list of nir_function */
1422
1423 /** list of global register in the shader */
1424 struct exec_list registers;
1425
1426 /** next available global register index */
1427 unsigned reg_alloc;
1428
1429 /**
1430 * the highest index a load_input_*, load_uniform_*, etc. intrinsic can
1431 * access plus one
1432 */
1433 unsigned num_inputs, num_uniforms, num_outputs;
1434
1435 /** the number of uniforms that are only accessed directly */
1436 unsigned num_direct_uniforms;
1437 } nir_shader;
1438
1439 #define nir_foreach_overload(shader, overload) \
1440 foreach_list_typed(nir_function, func, node, &(shader)->functions) \
1441 foreach_list_typed(nir_function_overload, overload, node, \
1442 &(func)->overload_list)
1443
1444 nir_shader *nir_shader_create(void *mem_ctx,
1445 const nir_shader_compiler_options *options);
1446
1447 /** creates a register, including assigning it an index and adding it to the list */
1448 nir_register *nir_global_reg_create(nir_shader *shader);
1449
1450 nir_register *nir_local_reg_create(nir_function_impl *impl);
1451
1452 void nir_reg_remove(nir_register *reg);
1453
1454 /** creates a function and adds it to the shader's list of functions */
1455 nir_function *nir_function_create(nir_shader *shader, const char *name);
1456
1457 /** creates a null function returning null */
1458 nir_function_overload *nir_function_overload_create(nir_function *func);
1459
1460 nir_function_impl *nir_function_impl_create(nir_function_overload *func);
1461
1462 nir_block *nir_block_create(void *mem_ctx);
1463 nir_if *nir_if_create(void *mem_ctx);
1464 nir_loop *nir_loop_create(void *mem_ctx);
1465
1466 nir_function_impl *nir_cf_node_get_function(nir_cf_node *node);
1467
1468 /** puts a control flow node immediately after another control flow node */
1469 void nir_cf_node_insert_after(nir_cf_node *node, nir_cf_node *after);
1470
1471 /** puts a control flow node immediately before another control flow node */
1472 void nir_cf_node_insert_before(nir_cf_node *node, nir_cf_node *before);
1473
1474 /** puts a control flow node at the beginning of a list from an if, loop, or function */
1475 void nir_cf_node_insert_begin(struct exec_list *list, nir_cf_node *node);
1476
1477 /** puts a control flow node at the end of a list from an if, loop, or function */
1478 void nir_cf_node_insert_end(struct exec_list *list, nir_cf_node *node);
1479
1480 /** removes a control flow node, doing any cleanup necessary */
1481 void nir_cf_node_remove(nir_cf_node *node);
1482
1483 /** requests that the given pieces of metadata be generated */
1484 void nir_metadata_require(nir_function_impl *impl, nir_metadata required);
1485 /** dirties all but the preserved metadata */
1486 void nir_metadata_preserve(nir_function_impl *impl, nir_metadata preserved);
1487
1488 /** creates an instruction with default swizzle/writemask/etc. with NULL registers */
1489 nir_alu_instr *nir_alu_instr_create(nir_shader *shader, nir_op op);
1490
1491 nir_jump_instr *nir_jump_instr_create(nir_shader *shader, nir_jump_type type);
1492
1493 nir_load_const_instr *nir_load_const_instr_create(nir_shader *shader,
1494 unsigned num_components);
1495
1496 nir_intrinsic_instr *nir_intrinsic_instr_create(nir_shader *shader,
1497 nir_intrinsic_op op);
1498
1499 nir_call_instr *nir_call_instr_create(nir_shader *shader,
1500 nir_function_overload *callee);
1501
1502 nir_tex_instr *nir_tex_instr_create(nir_shader *shader, unsigned num_srcs);
1503
1504 nir_phi_instr *nir_phi_instr_create(nir_shader *shader);
1505
1506 nir_parallel_copy_instr *nir_parallel_copy_instr_create(nir_shader *shader);
1507
1508 nir_ssa_undef_instr *nir_ssa_undef_instr_create(nir_shader *shader,
1509 unsigned num_components);
1510
1511 nir_deref_var *nir_deref_var_create(void *mem_ctx, nir_variable *var);
1512 nir_deref_array *nir_deref_array_create(void *mem_ctx);
1513 nir_deref_struct *nir_deref_struct_create(void *mem_ctx, unsigned field_index);
1514
1515 nir_deref *nir_copy_deref(void *mem_ctx, nir_deref *deref);
1516
1517 void nir_instr_insert_before(nir_instr *instr, nir_instr *before);
1518 void nir_instr_insert_after(nir_instr *instr, nir_instr *after);
1519
1520 void nir_instr_insert_before_block(nir_block *block, nir_instr *before);
1521 void nir_instr_insert_after_block(nir_block *block, nir_instr *after);
1522
1523 void nir_instr_insert_before_cf(nir_cf_node *node, nir_instr *before);
1524 void nir_instr_insert_after_cf(nir_cf_node *node, nir_instr *after);
1525
1526 void nir_instr_insert_before_cf_list(struct exec_list *list, nir_instr *before);
1527 void nir_instr_insert_after_cf_list(struct exec_list *list, nir_instr *after);
1528
1529 void nir_instr_remove(nir_instr *instr);
1530
1531 typedef bool (*nir_foreach_ssa_def_cb)(nir_ssa_def *def, void *state);
1532 typedef bool (*nir_foreach_dest_cb)(nir_dest *dest, void *state);
1533 typedef bool (*nir_foreach_src_cb)(nir_src *src, void *state);
1534 bool nir_foreach_ssa_def(nir_instr *instr, nir_foreach_ssa_def_cb cb,
1535 void *state);
1536 bool nir_foreach_dest(nir_instr *instr, nir_foreach_dest_cb cb, void *state);
1537 bool nir_foreach_src(nir_instr *instr, nir_foreach_src_cb cb, void *state);
1538
1539 nir_const_value *nir_src_as_const_value(nir_src src);
1540 bool nir_srcs_equal(nir_src src1, nir_src src2);
1541 void nir_instr_rewrite_src(nir_instr *instr, nir_src *src, nir_src new_src);
1542
1543 void nir_ssa_dest_init(nir_instr *instr, nir_dest *dest,
1544 unsigned num_components, const char *name);
1545 void nir_ssa_def_init(nir_instr *instr, nir_ssa_def *def,
1546 unsigned num_components, const char *name);
1547 void nir_ssa_def_rewrite_uses(nir_ssa_def *def, nir_src new_src, void *mem_ctx);
1548
1549 /* visits basic blocks in source-code order */
1550 typedef bool (*nir_foreach_block_cb)(nir_block *block, void *state);
1551 bool nir_foreach_block(nir_function_impl *impl, nir_foreach_block_cb cb,
1552 void *state);
1553 bool nir_foreach_block_reverse(nir_function_impl *impl, nir_foreach_block_cb cb,
1554 void *state);
1555
1556 /* If the following CF node is an if, this function returns that if.
1557 * Otherwise, it returns NULL.
1558 */
1559 nir_if *nir_block_get_following_if(nir_block *block);
1560
1561 void nir_index_local_regs(nir_function_impl *impl);
1562 void nir_index_global_regs(nir_shader *shader);
1563 void nir_index_ssa_defs(nir_function_impl *impl);
1564
1565 void nir_index_blocks(nir_function_impl *impl);
1566
1567 void nir_print_shader(nir_shader *shader, FILE *fp);
1568 void nir_print_instr(const nir_instr *instr, FILE *fp);
1569
1570 #ifdef DEBUG
1571 void nir_validate_shader(nir_shader *shader);
1572 #else
1573 static inline void nir_validate_shader(nir_shader *shader) { }
1574 #endif /* DEBUG */
1575
1576 void nir_calc_dominance_impl(nir_function_impl *impl);
1577 void nir_calc_dominance(nir_shader *shader);
1578
1579 nir_block *nir_dominance_lca(nir_block *b1, nir_block *b2);
1580 bool nir_block_dominates(nir_block *parent, nir_block *child);
1581
1582 void nir_dump_dom_tree_impl(nir_function_impl *impl, FILE *fp);
1583 void nir_dump_dom_tree(nir_shader *shader, FILE *fp);
1584
1585 void nir_dump_dom_frontier_impl(nir_function_impl *impl, FILE *fp);
1586 void nir_dump_dom_frontier(nir_shader *shader, FILE *fp);
1587
1588 void nir_dump_cfg_impl(nir_function_impl *impl, FILE *fp);
1589 void nir_dump_cfg(nir_shader *shader, FILE *fp);
1590
1591 void nir_split_var_copies(nir_shader *shader);
1592
1593 void nir_lower_var_copy_instr(nir_intrinsic_instr *copy, void *mem_ctx);
1594 void nir_lower_var_copies(nir_shader *shader);
1595
1596 void nir_lower_global_vars_to_local(nir_shader *shader);
1597
1598 void nir_lower_locals_to_regs(nir_shader *shader);
1599
1600 void nir_assign_var_locations_scalar(struct exec_list *var_list,
1601 unsigned *size);
1602 void nir_assign_var_locations_scalar_direct_first(nir_shader *shader,
1603 struct exec_list *var_list,
1604 unsigned *direct_size,
1605 unsigned *size);
1606
1607 void nir_lower_io(nir_shader *shader);
1608
1609 void nir_lower_vars_to_ssa(nir_shader *shader);
1610
1611 void nir_remove_dead_variables(nir_shader *shader);
1612
1613 void nir_lower_vec_to_movs(nir_shader *shader);
1614 void nir_lower_alu_to_scalar(nir_shader *shader);
1615
1616 void nir_lower_phis_to_scalar(nir_shader *shader);
1617
1618 void nir_lower_samplers(nir_shader *shader,
1619 const struct gl_shader_program *shader_program,
1620 gl_shader_stage stage);
1621
1622 void nir_lower_system_values(nir_shader *shader);
1623 void nir_lower_tex_projector(nir_shader *shader);
1624 void nir_lower_idiv(nir_shader *shader);
1625
1626 void nir_lower_atomics(nir_shader *shader);
1627 void nir_lower_to_source_mods(nir_shader *shader);
1628
1629 void nir_normalize_cubemap_coords(nir_shader *shader);
1630
1631 void nir_live_variables_impl(nir_function_impl *impl);
1632 bool nir_ssa_defs_interfere(nir_ssa_def *a, nir_ssa_def *b);
1633
1634 void nir_convert_to_ssa_impl(nir_function_impl *impl);
1635 void nir_convert_to_ssa(nir_shader *shader);
1636 void nir_convert_from_ssa(nir_shader *shader);
1637
1638 bool nir_opt_algebraic(nir_shader *shader);
1639 bool nir_opt_algebraic_late(nir_shader *shader);
1640 bool nir_opt_constant_folding(nir_shader *shader);
1641
1642 bool nir_opt_global_to_local(nir_shader *shader);
1643
1644 bool nir_copy_prop_impl(nir_function_impl *impl);
1645 bool nir_copy_prop(nir_shader *shader);
1646
1647 bool nir_opt_cse(nir_shader *shader);
1648
1649 bool nir_opt_dce_impl(nir_function_impl *impl);
1650 bool nir_opt_dce(nir_shader *shader);
1651
1652 void nir_opt_gcm(nir_shader *shader);
1653
1654 bool nir_opt_peephole_select(nir_shader *shader);
1655 bool nir_opt_peephole_ffma(nir_shader *shader);
1656
1657 bool nir_opt_remove_phis(nir_shader *shader);
1658
1659 void nir_sweep(nir_shader *shader);
1660
1661 #ifdef __cplusplus
1662 } /* extern "C" */
1663 #endif