2 * Copyright © 2014 Intel Corporation
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)
32 nir_shader_create(void *mem_ctx
, const nir_shader_compiler_options
*options
)
34 nir_shader
*shader
= ralloc(mem_ctx
, nir_shader
);
36 exec_list_make_empty(&shader
->uniforms
);
37 exec_list_make_empty(&shader
->inputs
);
38 exec_list_make_empty(&shader
->outputs
);
40 shader
->options
= options
;
42 exec_list_make_empty(&shader
->functions
);
43 exec_list_make_empty(&shader
->registers
);
44 exec_list_make_empty(&shader
->globals
);
45 exec_list_make_empty(&shader
->system_values
);
46 shader
->reg_alloc
= 0;
48 shader
->num_inputs
= 0;
49 shader
->num_outputs
= 0;
50 shader
->num_uniforms
= 0;
56 reg_create(void *mem_ctx
, struct exec_list
*list
)
58 nir_register
*reg
= ralloc(mem_ctx
, nir_register
);
60 reg
->parent_instr
= NULL
;
61 list_inithead(®
->uses
);
62 list_inithead(®
->defs
);
63 list_inithead(®
->if_uses
);
65 reg
->num_components
= 0;
66 reg
->num_array_elems
= 0;
67 reg
->is_packed
= false;
70 exec_list_push_tail(list
, ®
->node
);
76 nir_global_reg_create(nir_shader
*shader
)
78 nir_register
*reg
= reg_create(shader
, &shader
->registers
);
79 reg
->index
= shader
->reg_alloc
++;
80 reg
->is_global
= true;
86 nir_local_reg_create(nir_function_impl
*impl
)
88 nir_register
*reg
= reg_create(ralloc_parent(impl
), &impl
->registers
);
89 reg
->index
= impl
->reg_alloc
++;
90 reg
->is_global
= false;
96 nir_reg_remove(nir_register
*reg
)
98 exec_node_remove(®
->node
);
102 nir_function_create(nir_shader
*shader
, const char *name
)
104 nir_function
*func
= ralloc(shader
, nir_function
);
106 exec_list_push_tail(&shader
->functions
, &func
->node
);
107 exec_list_make_empty(&func
->overload_list
);
108 func
->name
= ralloc_strdup(func
, name
);
109 func
->shader
= shader
;
114 nir_function_overload
*
115 nir_function_overload_create(nir_function
*func
)
117 void *mem_ctx
= ralloc_parent(func
);
119 nir_function_overload
*overload
= ralloc(mem_ctx
, nir_function_overload
);
121 overload
->num_params
= 0;
122 overload
->params
= NULL
;
123 overload
->return_type
= glsl_void_type();
124 overload
->impl
= NULL
;
126 exec_list_push_tail(&func
->overload_list
, &overload
->node
);
127 overload
->function
= func
;
132 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *mem_ctx
)
134 dest
->is_ssa
= src
->is_ssa
;
136 dest
->ssa
= src
->ssa
;
138 dest
->reg
.base_offset
= src
->reg
.base_offset
;
139 dest
->reg
.reg
= src
->reg
.reg
;
140 if (src
->reg
.indirect
) {
141 dest
->reg
.indirect
= ralloc(mem_ctx
, nir_src
);
142 nir_src_copy(dest
->reg
.indirect
, src
->reg
.indirect
, mem_ctx
);
144 dest
->reg
.indirect
= NULL
;
149 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, void *mem_ctx
)
151 dest
->is_ssa
= src
->is_ssa
;
153 dest
->ssa
= src
->ssa
;
155 dest
->reg
.base_offset
= src
->reg
.base_offset
;
156 dest
->reg
.reg
= src
->reg
.reg
;
157 if (src
->reg
.indirect
) {
158 dest
->reg
.indirect
= ralloc(mem_ctx
, nir_src
);
159 nir_src_copy(dest
->reg
.indirect
, src
->reg
.indirect
, mem_ctx
);
161 dest
->reg
.indirect
= NULL
;
167 nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
, void *mem_ctx
)
169 nir_src_copy(&dest
->src
, &src
->src
, mem_ctx
);
170 dest
->abs
= src
->abs
;
171 dest
->negate
= src
->negate
;
172 for (unsigned i
= 0; i
< 4; i
++)
173 dest
->swizzle
[i
] = src
->swizzle
[i
];
177 nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
, void *mem_ctx
)
179 nir_dest_copy(&dest
->dest
, &src
->dest
, mem_ctx
);
180 dest
->write_mask
= src
->write_mask
;
181 dest
->saturate
= src
->saturate
;
185 block_add_pred(nir_block
*block
, nir_block
*pred
)
187 _mesa_set_add(block
->predecessors
, pred
);
191 cf_init(nir_cf_node
*node
, nir_cf_node_type type
)
193 exec_node_init(&node
->node
);
199 link_blocks(nir_block
*pred
, nir_block
*succ1
, nir_block
*succ2
)
201 pred
->successors
[0] = succ1
;
202 block_add_pred(succ1
, pred
);
204 pred
->successors
[1] = succ2
;
206 block_add_pred(succ2
, pred
);
210 unlink_blocks(nir_block
*pred
, nir_block
*succ
)
212 if (pred
->successors
[0] == succ
) {
213 pred
->successors
[0] = pred
->successors
[1];
214 pred
->successors
[1] = NULL
;
216 assert(pred
->successors
[1] == succ
);
217 pred
->successors
[1] = NULL
;
220 struct set_entry
*entry
= _mesa_set_search(succ
->predecessors
, pred
);
224 _mesa_set_remove(succ
->predecessors
, entry
);
228 unlink_block_successors(nir_block
*block
)
230 if (block
->successors
[0] != NULL
)
231 unlink_blocks(block
, block
->successors
[0]);
232 if (block
->successors
[1] != NULL
)
233 unlink_blocks(block
, block
->successors
[1]);
238 nir_function_impl_create(nir_function_overload
*overload
)
240 assert(overload
->impl
== NULL
);
242 void *mem_ctx
= ralloc_parent(overload
);
244 nir_function_impl
*impl
= ralloc(mem_ctx
, nir_function_impl
);
246 overload
->impl
= impl
;
247 impl
->overload
= overload
;
249 cf_init(&impl
->cf_node
, nir_cf_node_function
);
251 exec_list_make_empty(&impl
->body
);
252 exec_list_make_empty(&impl
->registers
);
253 exec_list_make_empty(&impl
->locals
);
254 impl
->num_params
= 0;
256 impl
->return_var
= NULL
;
259 impl
->valid_metadata
= nir_metadata_none
;
261 /* create start & end blocks */
262 nir_block
*start_block
= nir_block_create(mem_ctx
);
263 nir_block
*end_block
= nir_block_create(mem_ctx
);
264 start_block
->cf_node
.parent
= &impl
->cf_node
;
265 end_block
->cf_node
.parent
= &impl
->cf_node
;
266 impl
->start_block
= start_block
;
267 impl
->end_block
= end_block
;
269 exec_list_push_tail(&impl
->body
, &start_block
->cf_node
.node
);
271 start_block
->successors
[0] = end_block
;
272 block_add_pred(end_block
, start_block
);
278 nir_block_create(void *mem_ctx
)
280 nir_block
*block
= ralloc(mem_ctx
, nir_block
);
282 cf_init(&block
->cf_node
, nir_cf_node_block
);
284 block
->successors
[0] = block
->successors
[1] = NULL
;
285 block
->predecessors
= _mesa_set_create(block
, _mesa_hash_pointer
,
286 _mesa_key_pointer_equal
);
287 block
->imm_dom
= NULL
;
288 block
->dom_frontier
= _mesa_set_create(block
, _mesa_hash_pointer
,
289 _mesa_key_pointer_equal
);
291 exec_list_make_empty(&block
->instr_list
);
297 src_init(nir_src
*src
)
301 src
->reg
.indirect
= NULL
;
302 src
->reg
.base_offset
= 0;
306 nir_if_create(void *mem_ctx
)
308 nir_if
*if_stmt
= ralloc(mem_ctx
, nir_if
);
310 cf_init(&if_stmt
->cf_node
, nir_cf_node_if
);
311 src_init(&if_stmt
->condition
);
313 nir_block
*then
= nir_block_create(mem_ctx
);
314 exec_list_make_empty(&if_stmt
->then_list
);
315 exec_list_push_tail(&if_stmt
->then_list
, &then
->cf_node
.node
);
316 then
->cf_node
.parent
= &if_stmt
->cf_node
;
318 nir_block
*else_stmt
= nir_block_create(mem_ctx
);
319 exec_list_make_empty(&if_stmt
->else_list
);
320 exec_list_push_tail(&if_stmt
->else_list
, &else_stmt
->cf_node
.node
);
321 else_stmt
->cf_node
.parent
= &if_stmt
->cf_node
;
327 nir_loop_create(void *mem_ctx
)
329 nir_loop
*loop
= ralloc(mem_ctx
, nir_loop
);
331 cf_init(&loop
->cf_node
, nir_cf_node_loop
);
333 nir_block
*body
= nir_block_create(mem_ctx
);
334 exec_list_make_empty(&loop
->body
);
335 exec_list_push_tail(&loop
->body
, &body
->cf_node
.node
);
336 body
->cf_node
.parent
= &loop
->cf_node
;
338 body
->successors
[0] = body
;
339 block_add_pred(body
, body
);
345 instr_init(nir_instr
*instr
, nir_instr_type type
)
349 exec_node_init(&instr
->node
);
353 dest_init(nir_dest
*dest
)
355 dest
->is_ssa
= false;
356 dest
->reg
.reg
= NULL
;
357 dest
->reg
.indirect
= NULL
;
358 dest
->reg
.base_offset
= 0;
362 alu_dest_init(nir_alu_dest
*dest
)
364 dest_init(&dest
->dest
);
365 dest
->saturate
= false;
366 dest
->write_mask
= 0xf;
370 alu_src_init(nir_alu_src
*src
)
373 src
->abs
= src
->negate
= false;
381 nir_alu_instr_create(nir_shader
*shader
, nir_op op
)
383 unsigned num_srcs
= nir_op_infos
[op
].num_inputs
;
384 nir_alu_instr
*instr
=
386 sizeof(nir_alu_instr
) + num_srcs
* sizeof(nir_alu_src
));
388 instr_init(&instr
->instr
, nir_instr_type_alu
);
390 alu_dest_init(&instr
->dest
);
391 for (unsigned i
= 0; i
< num_srcs
; i
++)
392 alu_src_init(&instr
->src
[i
]);
398 nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
)
400 nir_jump_instr
*instr
= ralloc(shader
, nir_jump_instr
);
401 instr_init(&instr
->instr
, nir_instr_type_jump
);
406 nir_load_const_instr
*
407 nir_load_const_instr_create(nir_shader
*shader
, unsigned num_components
)
409 nir_load_const_instr
*instr
= ralloc(shader
, nir_load_const_instr
);
410 instr_init(&instr
->instr
, nir_instr_type_load_const
);
412 nir_ssa_def_init(&instr
->instr
, &instr
->def
, num_components
, NULL
);
417 nir_intrinsic_instr
*
418 nir_intrinsic_instr_create(nir_shader
*shader
, nir_intrinsic_op op
)
420 unsigned num_srcs
= nir_intrinsic_infos
[op
].num_srcs
;
421 nir_intrinsic_instr
*instr
=
423 sizeof(nir_intrinsic_instr
) + num_srcs
* sizeof(nir_src
));
425 instr_init(&instr
->instr
, nir_instr_type_intrinsic
);
426 instr
->intrinsic
= op
;
428 if (nir_intrinsic_infos
[op
].has_dest
)
429 dest_init(&instr
->dest
);
431 for (unsigned i
= 0; i
< num_srcs
; i
++)
432 src_init(&instr
->src
[i
]);
438 nir_call_instr_create(nir_shader
*shader
, nir_function_overload
*callee
)
440 nir_call_instr
*instr
= ralloc(shader
, nir_call_instr
);
441 instr_init(&instr
->instr
, nir_instr_type_call
);
443 instr
->callee
= callee
;
444 instr
->num_params
= callee
->num_params
;
445 instr
->params
= ralloc_array(instr
, nir_deref_var
*, instr
->num_params
);
446 instr
->return_deref
= NULL
;
452 nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
)
454 nir_tex_instr
*instr
= ralloc(shader
, nir_tex_instr
);
455 instr_init(&instr
->instr
, nir_instr_type_tex
);
457 dest_init(&instr
->dest
);
459 instr
->num_srcs
= num_srcs
;
460 instr
->src
= ralloc_array(instr
, nir_tex_src
, num_srcs
);
461 for (unsigned i
= 0; i
< num_srcs
; i
++)
462 src_init(&instr
->src
[i
].src
);
464 instr
->sampler_index
= 0;
465 instr
->sampler_array_size
= 0;
466 instr
->sampler
= NULL
;
472 nir_phi_instr_create(nir_shader
*shader
)
474 nir_phi_instr
*instr
= ralloc(shader
, nir_phi_instr
);
475 instr_init(&instr
->instr
, nir_instr_type_phi
);
477 dest_init(&instr
->dest
);
478 exec_list_make_empty(&instr
->srcs
);
482 nir_parallel_copy_instr
*
483 nir_parallel_copy_instr_create(nir_shader
*shader
)
485 nir_parallel_copy_instr
*instr
= ralloc(shader
, nir_parallel_copy_instr
);
486 instr_init(&instr
->instr
, nir_instr_type_parallel_copy
);
488 exec_list_make_empty(&instr
->entries
);
493 nir_ssa_undef_instr
*
494 nir_ssa_undef_instr_create(nir_shader
*shader
, unsigned num_components
)
496 nir_ssa_undef_instr
*instr
= ralloc(shader
, nir_ssa_undef_instr
);
497 instr_init(&instr
->instr
, nir_instr_type_ssa_undef
);
499 nir_ssa_def_init(&instr
->instr
, &instr
->def
, num_components
, NULL
);
505 nir_deref_var_create(void *mem_ctx
, nir_variable
*var
)
507 nir_deref_var
*deref
= ralloc(mem_ctx
, nir_deref_var
);
508 deref
->deref
.deref_type
= nir_deref_type_var
;
509 deref
->deref
.child
= NULL
;
510 deref
->deref
.type
= var
->type
;
516 nir_deref_array_create(void *mem_ctx
)
518 nir_deref_array
*deref
= ralloc(mem_ctx
, nir_deref_array
);
519 deref
->deref
.deref_type
= nir_deref_type_array
;
520 deref
->deref
.child
= NULL
;
521 deref
->deref_array_type
= nir_deref_array_type_direct
;
522 src_init(&deref
->indirect
);
523 deref
->base_offset
= 0;
528 nir_deref_struct_create(void *mem_ctx
, unsigned field_index
)
530 nir_deref_struct
*deref
= ralloc(mem_ctx
, nir_deref_struct
);
531 deref
->deref
.deref_type
= nir_deref_type_struct
;
532 deref
->deref
.child
= NULL
;
533 deref
->index
= field_index
;
537 static nir_deref_var
*
538 copy_deref_var(void *mem_ctx
, nir_deref_var
*deref
)
540 nir_deref_var
*ret
= nir_deref_var_create(mem_ctx
, deref
->var
);
541 ret
->deref
.type
= deref
->deref
.type
;
542 if (deref
->deref
.child
)
543 ret
->deref
.child
= nir_copy_deref(ret
, deref
->deref
.child
);
547 static nir_deref_array
*
548 copy_deref_array(void *mem_ctx
, nir_deref_array
*deref
)
550 nir_deref_array
*ret
= nir_deref_array_create(mem_ctx
);
551 ret
->base_offset
= deref
->base_offset
;
552 ret
->deref_array_type
= deref
->deref_array_type
;
553 if (deref
->deref_array_type
== nir_deref_array_type_indirect
) {
554 nir_src_copy(&ret
->indirect
, &deref
->indirect
, mem_ctx
);
556 ret
->deref
.type
= deref
->deref
.type
;
557 if (deref
->deref
.child
)
558 ret
->deref
.child
= nir_copy_deref(ret
, deref
->deref
.child
);
562 static nir_deref_struct
*
563 copy_deref_struct(void *mem_ctx
, nir_deref_struct
*deref
)
565 nir_deref_struct
*ret
= nir_deref_struct_create(mem_ctx
, deref
->index
);
566 ret
->deref
.type
= deref
->deref
.type
;
567 if (deref
->deref
.child
)
568 ret
->deref
.child
= nir_copy_deref(ret
, deref
->deref
.child
);
573 nir_copy_deref(void *mem_ctx
, nir_deref
*deref
)
575 switch (deref
->deref_type
) {
576 case nir_deref_type_var
:
577 return ©_deref_var(mem_ctx
, nir_deref_as_var(deref
))->deref
;
578 case nir_deref_type_array
:
579 return ©_deref_array(mem_ctx
, nir_deref_as_array(deref
))->deref
;
580 case nir_deref_type_struct
:
581 return ©_deref_struct(mem_ctx
, nir_deref_as_struct(deref
))->deref
;
583 unreachable("Invalid dereference type");
589 /* Returns a load_const instruction that represents the constant
590 * initializer for the given deref chain. The caller is responsible for
591 * ensuring that there actually is a constant initializer.
593 nir_load_const_instr
*
594 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
)
596 nir_constant
*constant
= deref
->var
->constant_initializer
;
599 const nir_deref
*tail
= &deref
->deref
;
600 unsigned matrix_offset
= 0;
601 while (tail
->child
) {
602 switch (tail
->child
->deref_type
) {
603 case nir_deref_type_array
: {
604 nir_deref_array
*arr
= nir_deref_as_array(tail
->child
);
605 assert(arr
->deref_array_type
== nir_deref_array_type_direct
);
606 if (glsl_type_is_matrix(tail
->type
)) {
607 assert(arr
->deref
.child
== NULL
);
608 matrix_offset
= arr
->base_offset
;
610 constant
= constant
->elements
[arr
->base_offset
];
615 case nir_deref_type_struct
: {
616 constant
= constant
->elements
[nir_deref_as_struct(tail
->child
)->index
];
621 unreachable("Invalid deref child type");
627 nir_load_const_instr
*load
=
628 nir_load_const_instr_create(shader
, glsl_get_vector_elements(tail
->type
));
630 matrix_offset
*= load
->def
.num_components
;
631 for (unsigned i
= 0; i
< load
->def
.num_components
; i
++) {
632 switch (glsl_get_base_type(tail
->type
)) {
633 case GLSL_TYPE_FLOAT
:
636 load
->value
.u
[i
] = constant
->value
.u
[matrix_offset
+ i
];
639 load
->value
.u
[i
] = constant
->value
.b
[matrix_offset
+ i
] ?
640 NIR_TRUE
: NIR_FALSE
;
643 unreachable("Invalid immediate type");
651 * \name Control flow modification
653 * These functions modify the control flow tree while keeping the control flow
654 * graph up-to-date. The invariants respected are:
655 * 1. Each then statement, else statement, or loop body must have at least one
657 * 2. Each if-statement and loop must have one basic block before it and one
659 * 3. Two basic blocks cannot be directly next to each other.
660 * 4. If a basic block has a jump instruction, there must be only one and it
661 * must be at the end of the block.
662 * 5. The CFG must always be connected - this means that we must insert a fake
663 * CFG edge for loops with no break statement.
665 * The purpose of the second one is so that we have places to insert code during
666 * GCM, as well as eliminating the possibility of critical edges.
671 link_non_block_to_block(nir_cf_node
*node
, nir_block
*block
)
673 if (node
->type
== nir_cf_node_if
) {
675 * We're trying to link an if to a block after it; this just means linking
676 * the last block of the then and else branches.
679 nir_if
*if_stmt
= nir_cf_node_as_if(node
);
681 nir_cf_node
*last_then
= nir_if_last_then_node(if_stmt
);
682 assert(last_then
->type
== nir_cf_node_block
);
683 nir_block
*last_then_block
= nir_cf_node_as_block(last_then
);
685 nir_cf_node
*last_else
= nir_if_last_else_node(if_stmt
);
686 assert(last_else
->type
== nir_cf_node_block
);
687 nir_block
*last_else_block
= nir_cf_node_as_block(last_else
);
689 if (exec_list_is_empty(&last_then_block
->instr_list
) ||
690 nir_block_last_instr(last_then_block
)->type
!= nir_instr_type_jump
) {
691 unlink_block_successors(last_then_block
);
692 link_blocks(last_then_block
, block
, NULL
);
695 if (exec_list_is_empty(&last_else_block
->instr_list
) ||
696 nir_block_last_instr(last_else_block
)->type
!= nir_instr_type_jump
) {
697 unlink_block_successors(last_else_block
);
698 link_blocks(last_else_block
, block
, NULL
);
701 assert(node
->type
== nir_cf_node_loop
);
704 * We can only get to this codepath if we're inserting a new loop, or
705 * at least a loop with no break statements; we can't insert break
706 * statements into a loop when we haven't inserted it into the CFG
707 * because we wouldn't know which block comes after the loop
708 * and therefore, which block should be the successor of the block with
709 * the break). Therefore, we need to insert a fake edge (see invariant
713 nir_loop
*loop
= nir_cf_node_as_loop(node
);
715 nir_cf_node
*last
= nir_loop_last_cf_node(loop
);
716 assert(last
->type
== nir_cf_node_block
);
717 nir_block
*last_block
= nir_cf_node_as_block(last
);
719 last_block
->successors
[1] = block
;
720 block_add_pred(block
, last_block
);
725 link_block_to_non_block(nir_block
*block
, nir_cf_node
*node
)
727 if (node
->type
== nir_cf_node_if
) {
729 * We're trying to link a block to an if after it; this just means linking
730 * the block to the first block of the then and else branches.
733 nir_if
*if_stmt
= nir_cf_node_as_if(node
);
735 nir_cf_node
*first_then
= nir_if_first_then_node(if_stmt
);
736 assert(first_then
->type
== nir_cf_node_block
);
737 nir_block
*first_then_block
= nir_cf_node_as_block(first_then
);
739 nir_cf_node
*first_else
= nir_if_first_else_node(if_stmt
);
740 assert(first_else
->type
== nir_cf_node_block
);
741 nir_block
*first_else_block
= nir_cf_node_as_block(first_else
);
743 unlink_block_successors(block
);
744 link_blocks(block
, first_then_block
, first_else_block
);
747 * For similar reasons as the corresponding case in
748 * link_non_block_to_block(), don't worry about if the loop header has
749 * any predecessors that need to be unlinked.
752 assert(node
->type
== nir_cf_node_loop
);
754 nir_loop
*loop
= nir_cf_node_as_loop(node
);
756 nir_cf_node
*loop_header
= nir_loop_first_cf_node(loop
);
757 assert(loop_header
->type
== nir_cf_node_block
);
758 nir_block
*loop_header_block
= nir_cf_node_as_block(loop_header
);
760 unlink_block_successors(block
);
761 link_blocks(block
, loop_header_block
, NULL
);
767 * Takes a basic block and inserts a new empty basic block before it, making its
768 * predecessors point to the new block. This essentially splits the block into
769 * an empty header and a body so that another non-block CF node can be inserted
770 * between the two. Note that this does *not* link the two basic blocks, so
771 * some kind of cleanup *must* be performed after this call.
775 split_block_beginning(nir_block
*block
)
777 nir_block
*new_block
= nir_block_create(ralloc_parent(block
));
778 new_block
->cf_node
.parent
= block
->cf_node
.parent
;
779 exec_node_insert_node_before(&block
->cf_node
.node
, &new_block
->cf_node
.node
);
781 struct set_entry
*entry
;
782 set_foreach(block
->predecessors
, entry
) {
783 nir_block
*pred
= (nir_block
*) entry
->key
;
785 unlink_blocks(pred
, block
);
786 link_blocks(pred
, new_block
, NULL
);
793 rewrite_phi_preds(nir_block
*block
, nir_block
*old_pred
, nir_block
*new_pred
)
795 nir_foreach_instr_safe(block
, instr
) {
796 if (instr
->type
!= nir_instr_type_phi
)
799 nir_phi_instr
*phi
= nir_instr_as_phi(instr
);
800 nir_foreach_phi_src(phi
, src
) {
801 if (src
->pred
== old_pred
) {
802 src
->pred
= new_pred
;
810 * Moves the successors of source to the successors of dest, leaving both
811 * successors of source NULL.
815 move_successors(nir_block
*source
, nir_block
*dest
)
817 nir_block
*succ1
= source
->successors
[0];
818 nir_block
*succ2
= source
->successors
[1];
821 unlink_blocks(source
, succ1
);
822 rewrite_phi_preds(succ1
, source
, dest
);
826 unlink_blocks(source
, succ2
);
827 rewrite_phi_preds(succ2
, source
, dest
);
830 unlink_block_successors(dest
);
831 link_blocks(dest
, succ1
, succ2
);
835 split_block_end(nir_block
*block
)
837 nir_block
*new_block
= nir_block_create(ralloc_parent(block
));
838 new_block
->cf_node
.parent
= block
->cf_node
.parent
;
839 exec_node_insert_after(&block
->cf_node
.node
, &new_block
->cf_node
.node
);
841 move_successors(block
, new_block
);
847 * Inserts a non-basic block between two basic blocks and links them together.
851 insert_non_block(nir_block
*before
, nir_cf_node
*node
, nir_block
*after
)
853 node
->parent
= before
->cf_node
.parent
;
854 exec_node_insert_after(&before
->cf_node
.node
, &node
->node
);
855 link_block_to_non_block(before
, node
);
856 link_non_block_to_block(node
, after
);
860 * Inserts a non-basic block before a basic block.
864 insert_non_block_before_block(nir_cf_node
*node
, nir_block
*block
)
866 /* split off the beginning of block into new_block */
867 nir_block
*new_block
= split_block_beginning(block
);
869 /* insert our node in between new_block and block */
870 insert_non_block(new_block
, node
, block
);
874 insert_non_block_after_block(nir_block
*block
, nir_cf_node
*node
)
876 /* split off the end of block into new_block */
877 nir_block
*new_block
= split_block_end(block
);
879 /* insert our node in between block and new_block */
880 insert_non_block(block
, node
, new_block
);
883 /* walk up the control flow tree to find the innermost enclosed loop */
885 nearest_loop(nir_cf_node
*node
)
887 while (node
->type
!= nir_cf_node_loop
) {
891 return nir_cf_node_as_loop(node
);
895 nir_cf_node_get_function(nir_cf_node
*node
)
897 while (node
->type
!= nir_cf_node_function
) {
901 return nir_cf_node_as_function(node
);
905 * update the CFG after a jump instruction has been added to the end of a block
909 handle_jump(nir_block
*block
)
911 nir_instr
*instr
= nir_block_last_instr(block
);
912 nir_jump_instr
*jump_instr
= nir_instr_as_jump(instr
);
914 unlink_block_successors(block
);
916 nir_function_impl
*impl
= nir_cf_node_get_function(&block
->cf_node
);
917 nir_metadata_preserve(impl
, nir_metadata_none
);
919 if (jump_instr
->type
== nir_jump_break
||
920 jump_instr
->type
== nir_jump_continue
) {
921 nir_loop
*loop
= nearest_loop(&block
->cf_node
);
923 if (jump_instr
->type
== nir_jump_continue
) {
924 nir_cf_node
*first_node
= nir_loop_first_cf_node(loop
);
925 assert(first_node
->type
== nir_cf_node_block
);
926 nir_block
*first_block
= nir_cf_node_as_block(first_node
);
927 link_blocks(block
, first_block
, NULL
);
929 nir_cf_node
*after
= nir_cf_node_next(&loop
->cf_node
);
930 assert(after
->type
== nir_cf_node_block
);
931 nir_block
*after_block
= nir_cf_node_as_block(after
);
932 link_blocks(block
, after_block
, NULL
);
934 /* If we inserted a fake link, remove it */
935 nir_cf_node
*last
= nir_loop_last_cf_node(loop
);
936 assert(last
->type
== nir_cf_node_block
);
937 nir_block
*last_block
= nir_cf_node_as_block(last
);
938 if (last_block
->successors
[1] != NULL
)
939 unlink_blocks(last_block
, after_block
);
942 assert(jump_instr
->type
== nir_jump_return
);
943 link_blocks(block
, impl
->end_block
, NULL
);
948 handle_remove_jump(nir_block
*block
, nir_jump_type type
)
950 unlink_block_successors(block
);
952 if (exec_node_is_tail_sentinel(block
->cf_node
.node
.next
)) {
953 nir_cf_node
*parent
= block
->cf_node
.parent
;
954 if (parent
->type
== nir_cf_node_if
) {
955 nir_cf_node
*next
= nir_cf_node_next(parent
);
956 assert(next
->type
== nir_cf_node_block
);
957 nir_block
*next_block
= nir_cf_node_as_block(next
);
959 link_blocks(block
, next_block
, NULL
);
961 assert(parent
->type
== nir_cf_node_loop
);
962 nir_loop
*loop
= nir_cf_node_as_loop(parent
);
964 nir_cf_node
*head
= nir_loop_first_cf_node(loop
);
965 assert(head
->type
== nir_cf_node_block
);
966 nir_block
*head_block
= nir_cf_node_as_block(head
);
968 link_blocks(block
, head_block
, NULL
);
971 nir_cf_node
*next
= nir_cf_node_next(&block
->cf_node
);
972 if (next
->type
== nir_cf_node_if
) {
973 nir_if
*next_if
= nir_cf_node_as_if(next
);
975 nir_cf_node
*first_then
= nir_if_first_then_node(next_if
);
976 assert(first_then
->type
== nir_cf_node_block
);
977 nir_block
*first_then_block
= nir_cf_node_as_block(first_then
);
979 nir_cf_node
*first_else
= nir_if_first_else_node(next_if
);
980 assert(first_else
->type
== nir_cf_node_block
);
981 nir_block
*first_else_block
= nir_cf_node_as_block(first_else
);
983 link_blocks(block
, first_then_block
, first_else_block
);
985 assert(next
->type
== nir_cf_node_loop
);
986 nir_loop
*next_loop
= nir_cf_node_as_loop(next
);
988 nir_cf_node
*first
= nir_loop_first_cf_node(next_loop
);
989 assert(first
->type
== nir_cf_node_block
);
990 nir_block
*first_block
= nir_cf_node_as_block(first
);
992 link_blocks(block
, first_block
, NULL
);
996 if (type
== nir_jump_break
) {
997 nir_loop
*loop
= nearest_loop(&block
->cf_node
);
999 nir_cf_node
*next
= nir_cf_node_next(&loop
->cf_node
);
1000 assert(next
->type
== nir_cf_node_block
);
1001 nir_block
*next_block
= nir_cf_node_as_block(next
);
1003 if (next_block
->predecessors
->entries
== 0) {
1004 /* insert fake link */
1005 nir_cf_node
*last
= nir_loop_last_cf_node(loop
);
1006 assert(last
->type
== nir_cf_node_block
);
1007 nir_block
*last_block
= nir_cf_node_as_block(last
);
1009 last_block
->successors
[1] = next_block
;
1010 block_add_pred(next_block
, last_block
);
1014 nir_function_impl
*impl
= nir_cf_node_get_function(&block
->cf_node
);
1015 nir_metadata_preserve(impl
, nir_metadata_none
);
1019 * Inserts a basic block before another by merging the instructions.
1021 * @param block the target of the insertion
1022 * @param before the block to be inserted - must not have been inserted before
1023 * @param has_jump whether \before has a jump instruction at the end
1027 insert_block_before_block(nir_block
*block
, nir_block
*before
, bool has_jump
)
1029 assert(!has_jump
|| exec_list_is_empty(&block
->instr_list
));
1031 foreach_list_typed(nir_instr
, instr
, node
, &before
->instr_list
) {
1032 instr
->block
= block
;
1035 exec_list_prepend(&block
->instr_list
, &before
->instr_list
);
1042 * Inserts a basic block after another by merging the instructions.
1044 * @param block the target of the insertion
1045 * @param after the block to be inserted - must not have been inserted before
1046 * @param has_jump whether \after has a jump instruction at the end
1050 insert_block_after_block(nir_block
*block
, nir_block
*after
, bool has_jump
)
1052 foreach_list_typed(nir_instr
, instr
, node
, &after
->instr_list
) {
1053 instr
->block
= block
;
1056 exec_list_append(&block
->instr_list
, &after
->instr_list
);
1063 update_if_uses(nir_cf_node
*node
)
1065 if (node
->type
!= nir_cf_node_if
)
1068 nir_if
*if_stmt
= nir_cf_node_as_if(node
);
1070 if_stmt
->condition
.parent_if
= if_stmt
;
1071 if (if_stmt
->condition
.is_ssa
) {
1072 list_addtail(&if_stmt
->condition
.use_link
,
1073 &if_stmt
->condition
.ssa
->if_uses
);
1075 list_addtail(&if_stmt
->condition
.use_link
,
1076 &if_stmt
->condition
.reg
.reg
->if_uses
);
1081 nir_cf_node_insert_after(nir_cf_node
*node
, nir_cf_node
*after
)
1083 update_if_uses(after
);
1085 if (after
->type
== nir_cf_node_block
) {
1087 * either node or the one after it must be a basic block, by invariant #2;
1088 * in either case, just merge the blocks together.
1090 nir_block
*after_block
= nir_cf_node_as_block(after
);
1092 bool has_jump
= !exec_list_is_empty(&after_block
->instr_list
) &&
1093 nir_block_last_instr(after_block
)->type
== nir_instr_type_jump
;
1095 if (node
->type
== nir_cf_node_block
) {
1096 insert_block_after_block(nir_cf_node_as_block(node
), after_block
,
1099 nir_cf_node
*next
= nir_cf_node_next(node
);
1100 assert(next
->type
== nir_cf_node_block
);
1101 nir_block
*next_block
= nir_cf_node_as_block(next
);
1103 insert_block_before_block(next_block
, after_block
, has_jump
);
1106 if (node
->type
== nir_cf_node_block
) {
1107 insert_non_block_after_block(nir_cf_node_as_block(node
), after
);
1110 * We have to insert a non-basic block after a non-basic block. Since
1111 * every non-basic block has a basic block after it, this is equivalent
1112 * to inserting a non-basic block before a basic block.
1115 nir_cf_node
*next
= nir_cf_node_next(node
);
1116 assert(next
->type
== nir_cf_node_block
);
1117 nir_block
*next_block
= nir_cf_node_as_block(next
);
1119 insert_non_block_before_block(after
, next_block
);
1123 nir_function_impl
*impl
= nir_cf_node_get_function(node
);
1124 nir_metadata_preserve(impl
, nir_metadata_none
);
1128 nir_cf_node_insert_before(nir_cf_node
*node
, nir_cf_node
*before
)
1130 update_if_uses(before
);
1132 if (before
->type
== nir_cf_node_block
) {
1133 nir_block
*before_block
= nir_cf_node_as_block(before
);
1135 bool has_jump
= !exec_list_is_empty(&before_block
->instr_list
) &&
1136 nir_block_last_instr(before_block
)->type
== nir_instr_type_jump
;
1138 if (node
->type
== nir_cf_node_block
) {
1139 insert_block_before_block(nir_cf_node_as_block(node
), before_block
,
1142 nir_cf_node
*prev
= nir_cf_node_prev(node
);
1143 assert(prev
->type
== nir_cf_node_block
);
1144 nir_block
*prev_block
= nir_cf_node_as_block(prev
);
1146 insert_block_after_block(prev_block
, before_block
, has_jump
);
1149 if (node
->type
== nir_cf_node_block
) {
1150 insert_non_block_before_block(before
, nir_cf_node_as_block(node
));
1153 * We have to insert a non-basic block before a non-basic block. This
1154 * is equivalent to inserting a non-basic block after a basic block.
1157 nir_cf_node
*prev_node
= nir_cf_node_prev(node
);
1158 assert(prev_node
->type
== nir_cf_node_block
);
1159 nir_block
*prev_block
= nir_cf_node_as_block(prev_node
);
1161 insert_non_block_after_block(prev_block
, before
);
1165 nir_function_impl
*impl
= nir_cf_node_get_function(node
);
1166 nir_metadata_preserve(impl
, nir_metadata_none
);
1170 nir_cf_node_insert_begin(struct exec_list
*list
, nir_cf_node
*node
)
1172 nir_cf_node
*begin
= exec_node_data(nir_cf_node
, list
->head
, node
);
1173 nir_cf_node_insert_before(begin
, node
);
1177 nir_cf_node_insert_end(struct exec_list
*list
, nir_cf_node
*node
)
1179 nir_cf_node
*end
= exec_node_data(nir_cf_node
, list
->tail_pred
, node
);
1180 nir_cf_node_insert_after(end
, node
);
1184 * Stitch two basic blocks together into one. The aggregate must have the same
1185 * predecessors as the first and the same successors as the second.
1189 stitch_blocks(nir_block
*before
, nir_block
*after
)
1192 * We move after into before, so we have to deal with up to 2 successors vs.
1193 * possibly a large number of predecessors.
1195 * TODO: special case when before is empty and after isn't?
1198 move_successors(after
, before
);
1200 foreach_list_typed(nir_instr
, instr
, node
, &after
->instr_list
) {
1201 instr
->block
= before
;
1204 exec_list_append(&before
->instr_list
, &after
->instr_list
);
1205 exec_node_remove(&after
->cf_node
.node
);
1209 remove_defs_uses(nir_instr
*instr
);
1212 cleanup_cf_node(nir_cf_node
*node
)
1214 switch (node
->type
) {
1215 case nir_cf_node_block
: {
1216 nir_block
*block
= nir_cf_node_as_block(node
);
1217 /* We need to walk the instructions and clean up defs/uses */
1218 nir_foreach_instr(block
, instr
)
1219 remove_defs_uses(instr
);
1223 case nir_cf_node_if
: {
1224 nir_if
*if_stmt
= nir_cf_node_as_if(node
);
1225 foreach_list_typed(nir_cf_node
, child
, node
, &if_stmt
->then_list
)
1226 cleanup_cf_node(child
);
1227 foreach_list_typed(nir_cf_node
, child
, node
, &if_stmt
->else_list
)
1228 cleanup_cf_node(child
);
1230 list_del(&if_stmt
->condition
.use_link
);
1234 case nir_cf_node_loop
: {
1235 nir_loop
*loop
= nir_cf_node_as_loop(node
);
1236 foreach_list_typed(nir_cf_node
, child
, node
, &loop
->body
)
1237 cleanup_cf_node(child
);
1240 case nir_cf_node_function
: {
1241 nir_function_impl
*impl
= nir_cf_node_as_function(node
);
1242 foreach_list_typed(nir_cf_node
, child
, node
, &impl
->body
)
1243 cleanup_cf_node(child
);
1247 unreachable("Invalid CF node type");
1252 nir_cf_node_remove(nir_cf_node
*node
)
1254 nir_function_impl
*impl
= nir_cf_node_get_function(node
);
1255 nir_metadata_preserve(impl
, nir_metadata_none
);
1257 if (node
->type
== nir_cf_node_block
) {
1259 * Basic blocks can't really be removed by themselves, since they act as
1260 * padding between the non-basic blocks. So all we do here is empty the
1261 * block of instructions.
1263 * TODO: could we assert here?
1265 exec_list_make_empty(&nir_cf_node_as_block(node
)->instr_list
);
1267 nir_cf_node
*before
= nir_cf_node_prev(node
);
1268 assert(before
->type
== nir_cf_node_block
);
1269 nir_block
*before_block
= nir_cf_node_as_block(before
);
1271 nir_cf_node
*after
= nir_cf_node_next(node
);
1272 assert(after
->type
== nir_cf_node_block
);
1273 nir_block
*after_block
= nir_cf_node_as_block(after
);
1275 exec_node_remove(&node
->node
);
1276 stitch_blocks(before_block
, after_block
);
1279 cleanup_cf_node(node
);
1283 add_use_cb(nir_src
*src
, void *state
)
1285 nir_instr
*instr
= state
;
1287 src
->parent_instr
= instr
;
1288 list_addtail(&src
->use_link
,
1289 src
->is_ssa
? &src
->ssa
->uses
: &src
->reg
.reg
->uses
);
1295 add_ssa_def_cb(nir_ssa_def
*def
, void *state
)
1297 nir_instr
*instr
= state
;
1299 if (instr
->block
&& def
->index
== UINT_MAX
) {
1300 nir_function_impl
*impl
=
1301 nir_cf_node_get_function(&instr
->block
->cf_node
);
1303 def
->index
= impl
->ssa_alloc
++;
1310 add_reg_def_cb(nir_dest
*dest
, void *state
)
1312 nir_instr
*instr
= state
;
1314 if (!dest
->is_ssa
) {
1315 dest
->reg
.parent_instr
= instr
;
1316 list_addtail(&dest
->reg
.def_link
, &dest
->reg
.reg
->defs
);
1323 add_defs_uses(nir_instr
*instr
)
1325 nir_foreach_src(instr
, add_use_cb
, instr
);
1326 nir_foreach_dest(instr
, add_reg_def_cb
, instr
);
1327 nir_foreach_ssa_def(instr
, add_ssa_def_cb
, instr
);
1331 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1333 assert(before
->type
!= nir_instr_type_jump
);
1334 before
->block
= instr
->block
;
1335 add_defs_uses(before
);
1336 exec_node_insert_node_before(&instr
->node
, &before
->node
);
1340 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
1342 if (after
->type
== nir_instr_type_jump
) {
1343 assert(instr
== nir_block_last_instr(instr
->block
));
1344 assert(instr
->type
!= nir_instr_type_jump
);
1347 after
->block
= instr
->block
;
1348 add_defs_uses(after
);
1349 exec_node_insert_after(&instr
->node
, &after
->node
);
1351 if (after
->type
== nir_instr_type_jump
)
1352 handle_jump(after
->block
);
1356 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
1358 if (before
->type
== nir_instr_type_jump
)
1359 assert(exec_list_is_empty(&block
->instr_list
));
1361 before
->block
= block
;
1362 add_defs_uses(before
);
1363 exec_list_push_head(&block
->instr_list
, &before
->node
);
1365 if (before
->type
== nir_instr_type_jump
)
1370 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
1372 if (after
->type
== nir_instr_type_jump
) {
1373 assert(exec_list_is_empty(&block
->instr_list
) ||
1374 nir_block_last_instr(block
)->type
!= nir_instr_type_jump
);
1377 after
->block
= block
;
1378 add_defs_uses(after
);
1379 exec_list_push_tail(&block
->instr_list
, &after
->node
);
1381 if (after
->type
== nir_instr_type_jump
)
1386 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
1388 if (node
->type
== nir_cf_node_block
) {
1389 nir_instr_insert_before_block(nir_cf_node_as_block(node
), before
);
1391 nir_cf_node
*prev
= nir_cf_node_prev(node
);
1392 assert(prev
->type
== nir_cf_node_block
);
1393 nir_block
*prev_block
= nir_cf_node_as_block(prev
);
1395 nir_instr_insert_before_block(prev_block
, before
);
1400 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
1402 if (node
->type
== nir_cf_node_block
) {
1403 nir_instr_insert_after_block(nir_cf_node_as_block(node
), after
);
1405 nir_cf_node
*next
= nir_cf_node_next(node
);
1406 assert(next
->type
== nir_cf_node_block
);
1407 nir_block
*next_block
= nir_cf_node_as_block(next
);
1409 nir_instr_insert_before_block(next_block
, after
);
1414 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
1416 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1417 exec_list_get_head(list
), node
);
1418 nir_instr_insert_before_cf(first_node
, before
);
1422 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
1424 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1425 exec_list_get_tail(list
), node
);
1426 nir_instr_insert_after_cf(last_node
, after
);
1430 remove_use_cb(nir_src
*src
, void *state
)
1432 list_del(&src
->use_link
);
1438 remove_def_cb(nir_dest
*dest
, void *state
)
1441 list_del(&dest
->reg
.def_link
);
1447 remove_defs_uses(nir_instr
*instr
)
1449 nir_foreach_dest(instr
, remove_def_cb
, instr
);
1450 nir_foreach_src(instr
, remove_use_cb
, instr
);
1453 void nir_instr_remove(nir_instr
*instr
)
1455 remove_defs_uses(instr
);
1456 exec_node_remove(&instr
->node
);
1458 if (instr
->type
== nir_instr_type_jump
) {
1459 nir_jump_instr
*jump_instr
= nir_instr_as_jump(instr
);
1460 handle_remove_jump(instr
->block
, jump_instr
->type
);
1467 nir_index_local_regs(nir_function_impl
*impl
)
1470 foreach_list_typed(nir_register
, reg
, node
, &impl
->registers
) {
1471 reg
->index
= index
++;
1473 impl
->reg_alloc
= index
;
1477 nir_index_global_regs(nir_shader
*shader
)
1480 foreach_list_typed(nir_register
, reg
, node
, &shader
->registers
) {
1481 reg
->index
= index
++;
1483 shader
->reg_alloc
= index
;
1487 visit_alu_dest(nir_alu_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
1489 return cb(&instr
->dest
.dest
, state
);
1493 visit_intrinsic_dest(nir_intrinsic_instr
*instr
, nir_foreach_dest_cb cb
,
1496 if (nir_intrinsic_infos
[instr
->intrinsic
].has_dest
)
1497 return cb(&instr
->dest
, state
);
1503 visit_texture_dest(nir_tex_instr
*instr
, nir_foreach_dest_cb cb
,
1506 return cb(&instr
->dest
, state
);
1510 visit_phi_dest(nir_phi_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
1512 return cb(&instr
->dest
, state
);
1516 visit_parallel_copy_dest(nir_parallel_copy_instr
*instr
,
1517 nir_foreach_dest_cb cb
, void *state
)
1519 nir_foreach_parallel_copy_entry(instr
, entry
) {
1520 if (!cb(&entry
->dest
, state
))
1528 nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
1530 switch (instr
->type
) {
1531 case nir_instr_type_alu
:
1532 return visit_alu_dest(nir_instr_as_alu(instr
), cb
, state
);
1533 case nir_instr_type_intrinsic
:
1534 return visit_intrinsic_dest(nir_instr_as_intrinsic(instr
), cb
, state
);
1535 case nir_instr_type_tex
:
1536 return visit_texture_dest(nir_instr_as_tex(instr
), cb
, state
);
1537 case nir_instr_type_phi
:
1538 return visit_phi_dest(nir_instr_as_phi(instr
), cb
, state
);
1539 case nir_instr_type_parallel_copy
:
1540 return visit_parallel_copy_dest(nir_instr_as_parallel_copy(instr
),
1543 case nir_instr_type_load_const
:
1544 case nir_instr_type_ssa_undef
:
1545 case nir_instr_type_call
:
1546 case nir_instr_type_jump
:
1550 unreachable("Invalid instruction type");
1557 struct foreach_ssa_def_state
{
1558 nir_foreach_ssa_def_cb cb
;
1563 nir_ssa_def_visitor(nir_dest
*dest
, void *void_state
)
1565 struct foreach_ssa_def_state
*state
= void_state
;
1568 return state
->cb(&dest
->ssa
, state
->client_state
);
1574 nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
, void *state
)
1576 switch (instr
->type
) {
1577 case nir_instr_type_alu
:
1578 case nir_instr_type_tex
:
1579 case nir_instr_type_intrinsic
:
1580 case nir_instr_type_phi
:
1581 case nir_instr_type_parallel_copy
: {
1582 struct foreach_ssa_def_state foreach_state
= {cb
, state
};
1583 return nir_foreach_dest(instr
, nir_ssa_def_visitor
, &foreach_state
);
1586 case nir_instr_type_load_const
:
1587 return cb(&nir_instr_as_load_const(instr
)->def
, state
);
1588 case nir_instr_type_ssa_undef
:
1589 return cb(&nir_instr_as_ssa_undef(instr
)->def
, state
);
1590 case nir_instr_type_call
:
1591 case nir_instr_type_jump
:
1594 unreachable("Invalid instruction type");
1599 visit_src(nir_src
*src
, nir_foreach_src_cb cb
, void *state
)
1601 if (!cb(src
, state
))
1603 if (!src
->is_ssa
&& src
->reg
.indirect
)
1604 return cb(src
->reg
.indirect
, state
);
1609 visit_deref_array_src(nir_deref_array
*deref
, nir_foreach_src_cb cb
,
1612 if (deref
->deref_array_type
== nir_deref_array_type_indirect
)
1613 return visit_src(&deref
->indirect
, cb
, state
);
1618 visit_deref_src(nir_deref_var
*deref
, nir_foreach_src_cb cb
, void *state
)
1620 nir_deref
*cur
= &deref
->deref
;
1621 while (cur
!= NULL
) {
1622 if (cur
->deref_type
== nir_deref_type_array
)
1623 if (!visit_deref_array_src(nir_deref_as_array(cur
), cb
, state
))
1633 visit_alu_src(nir_alu_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1635 for (unsigned i
= 0; i
< nir_op_infos
[instr
->op
].num_inputs
; i
++)
1636 if (!visit_src(&instr
->src
[i
].src
, cb
, state
))
1643 visit_tex_src(nir_tex_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1645 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1646 if (!visit_src(&instr
->src
[i
].src
, cb
, state
))
1649 if (instr
->sampler
!= NULL
)
1650 if (!visit_deref_src(instr
->sampler
, cb
, state
))
1657 visit_intrinsic_src(nir_intrinsic_instr
*instr
, nir_foreach_src_cb cb
,
1660 unsigned num_srcs
= nir_intrinsic_infos
[instr
->intrinsic
].num_srcs
;
1661 for (unsigned i
= 0; i
< num_srcs
; i
++)
1662 if (!visit_src(&instr
->src
[i
], cb
, state
))
1666 nir_intrinsic_infos
[instr
->intrinsic
].num_variables
;
1667 for (unsigned i
= 0; i
< num_vars
; i
++)
1668 if (!visit_deref_src(instr
->variables
[i
], cb
, state
))
1675 visit_call_src(nir_call_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1681 visit_load_const_src(nir_load_const_instr
*instr
, nir_foreach_src_cb cb
,
1688 visit_phi_src(nir_phi_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1690 nir_foreach_phi_src(instr
, src
) {
1691 if (!visit_src(&src
->src
, cb
, state
))
1699 visit_parallel_copy_src(nir_parallel_copy_instr
*instr
,
1700 nir_foreach_src_cb cb
, void *state
)
1702 nir_foreach_parallel_copy_entry(instr
, entry
) {
1703 if (!visit_src(&entry
->src
, cb
, state
))
1712 nir_foreach_src_cb cb
;
1713 } visit_dest_indirect_state
;
1716 visit_dest_indirect(nir_dest
*dest
, void *_state
)
1718 visit_dest_indirect_state
*state
= (visit_dest_indirect_state
*) _state
;
1720 if (!dest
->is_ssa
&& dest
->reg
.indirect
)
1721 return state
->cb(dest
->reg
.indirect
, state
->state
);
1727 nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1729 switch (instr
->type
) {
1730 case nir_instr_type_alu
:
1731 if (!visit_alu_src(nir_instr_as_alu(instr
), cb
, state
))
1734 case nir_instr_type_intrinsic
:
1735 if (!visit_intrinsic_src(nir_instr_as_intrinsic(instr
), cb
, state
))
1738 case nir_instr_type_tex
:
1739 if (!visit_tex_src(nir_instr_as_tex(instr
), cb
, state
))
1742 case nir_instr_type_call
:
1743 if (!visit_call_src(nir_instr_as_call(instr
), cb
, state
))
1746 case nir_instr_type_load_const
:
1747 if (!visit_load_const_src(nir_instr_as_load_const(instr
), cb
, state
))
1750 case nir_instr_type_phi
:
1751 if (!visit_phi_src(nir_instr_as_phi(instr
), cb
, state
))
1754 case nir_instr_type_parallel_copy
:
1755 if (!visit_parallel_copy_src(nir_instr_as_parallel_copy(instr
),
1759 case nir_instr_type_jump
:
1760 case nir_instr_type_ssa_undef
:
1764 unreachable("Invalid instruction type");
1768 visit_dest_indirect_state dest_state
;
1769 dest_state
.state
= state
;
1771 return nir_foreach_dest(instr
, visit_dest_indirect
, &dest_state
);
1775 nir_src_as_const_value(nir_src src
)
1780 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_load_const
)
1783 nir_load_const_instr
*load
= nir_instr_as_load_const(src
.ssa
->parent_instr
);
1785 return &load
->value
;
1789 nir_srcs_equal(nir_src src1
, nir_src src2
)
1793 return src1
.ssa
== src2
.ssa
;
1801 if ((src1
.reg
.indirect
== NULL
) != (src2
.reg
.indirect
== NULL
))
1804 if (src1
.reg
.indirect
) {
1805 if (!nir_srcs_equal(*src1
.reg
.indirect
, *src2
.reg
.indirect
))
1809 return src1
.reg
.reg
== src2
.reg
.reg
&&
1810 src1
.reg
.base_offset
== src2
.reg
.base_offset
;
1816 src_is_valid(const nir_src
*src
)
1818 return src
->is_ssa
? (src
->ssa
!= NULL
) : (src
->reg
.reg
!= NULL
);
1822 src_remove_all_uses(nir_src
*src
)
1824 for (; src
; src
= src
->is_ssa
? NULL
: src
->reg
.indirect
) {
1825 if (!src_is_valid(src
))
1828 list_del(&src
->use_link
);
1833 src_add_all_uses(nir_src
*src
, nir_instr
*parent_instr
, nir_if
*parent_if
)
1835 for (; src
; src
= src
->is_ssa
? NULL
: src
->reg
.indirect
) {
1836 if (!src_is_valid(src
))
1840 src
->parent_instr
= parent_instr
;
1842 list_addtail(&src
->use_link
, &src
->ssa
->uses
);
1844 list_addtail(&src
->use_link
, &src
->reg
.reg
->uses
);
1847 src
->parent_if
= parent_if
;
1849 list_addtail(&src
->use_link
, &src
->ssa
->if_uses
);
1851 list_addtail(&src
->use_link
, &src
->reg
.reg
->if_uses
);
1857 nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
)
1859 assert(!src_is_valid(src
) || src
->parent_instr
== instr
);
1861 src_remove_all_uses(src
);
1863 src_add_all_uses(src
, instr
, NULL
);
1867 nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
)
1869 assert(!src_is_valid(dest
) || dest
->parent_instr
== dest_instr
);
1871 src_remove_all_uses(dest
);
1872 src_remove_all_uses(src
);
1874 *src
= NIR_SRC_INIT
;
1875 src_add_all_uses(dest
, dest_instr
, NULL
);
1879 nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
)
1881 nir_src
*src
= &if_stmt
->condition
;
1882 assert(!src_is_valid(src
) || src
->parent_if
== if_stmt
);
1884 src_remove_all_uses(src
);
1886 src_add_all_uses(src
, NULL
, if_stmt
);
1890 nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1891 unsigned num_components
, const char *name
)
1894 def
->parent_instr
= instr
;
1895 list_inithead(&def
->uses
);
1896 list_inithead(&def
->if_uses
);
1897 def
->num_components
= num_components
;
1900 nir_function_impl
*impl
=
1901 nir_cf_node_get_function(&instr
->block
->cf_node
);
1903 def
->index
= impl
->ssa_alloc
++;
1905 def
->index
= UINT_MAX
;
1910 nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1911 unsigned num_components
, const char *name
)
1913 dest
->is_ssa
= true;
1914 nir_ssa_def_init(instr
, &dest
->ssa
, num_components
, name
);
1918 nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
, void *mem_ctx
)
1920 assert(!new_src
.is_ssa
|| def
!= new_src
.ssa
);
1922 nir_foreach_use_safe(def
, use_src
) {
1923 nir_instr
*src_parent_instr
= use_src
->parent_instr
;
1924 list_del(&use_src
->use_link
);
1925 nir_src_copy(use_src
, &new_src
, mem_ctx
);
1926 src_add_all_uses(use_src
, src_parent_instr
, NULL
);
1929 nir_foreach_if_use_safe(def
, use_src
) {
1930 nir_if
*src_parent_if
= use_src
->parent_if
;
1931 list_del(&use_src
->use_link
);
1932 nir_src_copy(use_src
, &new_src
, mem_ctx
);
1933 src_add_all_uses(use_src
, NULL
, src_parent_if
);
1938 static bool foreach_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1939 bool reverse
, void *state
);
1942 foreach_if(nir_if
*if_stmt
, nir_foreach_block_cb cb
, bool reverse
, void *state
)
1945 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
,
1946 &if_stmt
->else_list
) {
1947 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1951 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
,
1952 &if_stmt
->then_list
) {
1953 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1957 foreach_list_typed_safe(nir_cf_node
, node
, node
, &if_stmt
->then_list
) {
1958 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1962 foreach_list_typed_safe(nir_cf_node
, node
, node
, &if_stmt
->else_list
) {
1963 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1972 foreach_loop(nir_loop
*loop
, nir_foreach_block_cb cb
, bool reverse
, void *state
)
1975 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
, &loop
->body
) {
1976 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1980 foreach_list_typed_safe(nir_cf_node
, node
, node
, &loop
->body
) {
1981 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1990 foreach_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1991 bool reverse
, void *state
)
1993 switch (node
->type
) {
1994 case nir_cf_node_block
:
1995 return cb(nir_cf_node_as_block(node
), state
);
1996 case nir_cf_node_if
:
1997 return foreach_if(nir_cf_node_as_if(node
), cb
, reverse
, state
);
1998 case nir_cf_node_loop
:
1999 return foreach_loop(nir_cf_node_as_loop(node
), cb
, reverse
, state
);
2003 unreachable("Invalid CFG node type");
2011 nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
, void *state
)
2013 foreach_list_typed_safe(nir_cf_node
, node
, node
, &impl
->body
) {
2014 if (!foreach_cf_node(node
, cb
, false, state
))
2018 return cb(impl
->end_block
, state
);
2022 nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
2025 if (!cb(impl
->end_block
, state
))
2028 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
, &impl
->body
) {
2029 if (!foreach_cf_node(node
, cb
, true, state
))
2037 nir_block_get_following_if(nir_block
*block
)
2039 if (exec_node_is_tail_sentinel(&block
->cf_node
.node
))
2042 if (nir_cf_node_is_last(&block
->cf_node
))
2045 nir_cf_node
*next_node
= nir_cf_node_next(&block
->cf_node
);
2047 if (next_node
->type
!= nir_cf_node_if
)
2050 return nir_cf_node_as_if(next_node
);
2054 index_block(nir_block
*block
, void *state
)
2056 unsigned *index
= state
;
2057 block
->index
= (*index
)++;
2062 nir_index_blocks(nir_function_impl
*impl
)
2066 if (impl
->valid_metadata
& nir_metadata_block_index
)
2069 nir_foreach_block(impl
, index_block
, &index
);
2071 impl
->num_blocks
= index
;
2075 index_ssa_def_cb(nir_ssa_def
*def
, void *state
)
2077 unsigned *index
= (unsigned *) state
;
2078 def
->index
= (*index
)++;
2084 index_ssa_block(nir_block
*block
, void *state
)
2086 nir_foreach_instr(block
, instr
)
2087 nir_foreach_ssa_def(instr
, index_ssa_def_cb
, state
);
2093 nir_index_ssa_defs(nir_function_impl
*impl
)
2096 nir_foreach_block(impl
, index_ssa_block
, &index
);
2097 impl
->ssa_alloc
= index
;