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 list_inithead(®
->uses
);
61 list_inithead(®
->defs
);
62 list_inithead(®
->if_uses
);
64 reg
->num_components
= 0;
65 reg
->num_array_elems
= 0;
66 reg
->is_packed
= false;
69 exec_list_push_tail(list
, ®
->node
);
75 nir_global_reg_create(nir_shader
*shader
)
77 nir_register
*reg
= reg_create(shader
, &shader
->registers
);
78 reg
->index
= shader
->reg_alloc
++;
79 reg
->is_global
= true;
85 nir_local_reg_create(nir_function_impl
*impl
)
87 nir_register
*reg
= reg_create(ralloc_parent(impl
), &impl
->registers
);
88 reg
->index
= impl
->reg_alloc
++;
89 reg
->is_global
= false;
95 nir_reg_remove(nir_register
*reg
)
97 exec_node_remove(®
->node
);
101 nir_function_create(nir_shader
*shader
, const char *name
)
103 nir_function
*func
= ralloc(shader
, nir_function
);
105 exec_list_push_tail(&shader
->functions
, &func
->node
);
106 exec_list_make_empty(&func
->overload_list
);
107 func
->name
= ralloc_strdup(func
, name
);
108 func
->shader
= shader
;
113 nir_function_overload
*
114 nir_function_overload_create(nir_function
*func
)
116 void *mem_ctx
= ralloc_parent(func
);
118 nir_function_overload
*overload
= ralloc(mem_ctx
, nir_function_overload
);
120 overload
->num_params
= 0;
121 overload
->params
= NULL
;
122 overload
->return_type
= glsl_void_type();
123 overload
->impl
= NULL
;
125 exec_list_push_tail(&func
->overload_list
, &overload
->node
);
126 overload
->function
= func
;
131 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *mem_ctx
)
133 dest
->is_ssa
= src
->is_ssa
;
135 dest
->ssa
= src
->ssa
;
137 dest
->reg
.base_offset
= src
->reg
.base_offset
;
138 dest
->reg
.reg
= src
->reg
.reg
;
139 if (src
->reg
.indirect
) {
140 dest
->reg
.indirect
= ralloc(mem_ctx
, nir_src
);
141 nir_src_copy(dest
->reg
.indirect
, src
->reg
.indirect
, mem_ctx
);
143 dest
->reg
.indirect
= NULL
;
148 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, void *mem_ctx
)
150 /* Copying an SSA definition makes no sense whatsoever. */
151 assert(!src
->is_ssa
);
153 dest
->is_ssa
= false;
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
;
166 nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
, void *mem_ctx
)
168 nir_src_copy(&dest
->src
, &src
->src
, mem_ctx
);
169 dest
->abs
= src
->abs
;
170 dest
->negate
= src
->negate
;
171 for (unsigned i
= 0; i
< 4; i
++)
172 dest
->swizzle
[i
] = src
->swizzle
[i
];
176 nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
, void *mem_ctx
)
178 nir_dest_copy(&dest
->dest
, &src
->dest
, mem_ctx
);
179 dest
->write_mask
= src
->write_mask
;
180 dest
->saturate
= src
->saturate
;
184 block_add_pred(nir_block
*block
, nir_block
*pred
)
186 _mesa_set_add(block
->predecessors
, pred
);
190 cf_init(nir_cf_node
*node
, nir_cf_node_type type
)
192 exec_node_init(&node
->node
);
198 link_blocks(nir_block
*pred
, nir_block
*succ1
, nir_block
*succ2
)
200 pred
->successors
[0] = succ1
;
201 block_add_pred(succ1
, pred
);
203 pred
->successors
[1] = succ2
;
205 block_add_pred(succ2
, pred
);
209 unlink_blocks(nir_block
*pred
, nir_block
*succ
)
211 if (pred
->successors
[0] == succ
) {
212 pred
->successors
[0] = pred
->successors
[1];
213 pred
->successors
[1] = NULL
;
215 assert(pred
->successors
[1] == succ
);
216 pred
->successors
[1] = NULL
;
219 struct set_entry
*entry
= _mesa_set_search(succ
->predecessors
, pred
);
223 _mesa_set_remove(succ
->predecessors
, entry
);
227 unlink_block_successors(nir_block
*block
)
229 if (block
->successors
[0] != NULL
)
230 unlink_blocks(block
, block
->successors
[0]);
231 if (block
->successors
[1] != NULL
)
232 unlink_blocks(block
, block
->successors
[1]);
237 nir_function_impl_create(nir_function_overload
*overload
)
239 assert(overload
->impl
== NULL
);
241 void *mem_ctx
= ralloc_parent(overload
);
243 nir_function_impl
*impl
= ralloc(mem_ctx
, nir_function_impl
);
245 overload
->impl
= impl
;
246 impl
->overload
= overload
;
248 cf_init(&impl
->cf_node
, nir_cf_node_function
);
250 exec_list_make_empty(&impl
->body
);
251 exec_list_make_empty(&impl
->registers
);
252 exec_list_make_empty(&impl
->locals
);
253 impl
->num_params
= 0;
255 impl
->return_var
= NULL
;
258 impl
->valid_metadata
= nir_metadata_none
;
260 /* create start & end blocks */
261 nir_block
*start_block
= nir_block_create(mem_ctx
);
262 nir_block
*end_block
= nir_block_create(mem_ctx
);
263 start_block
->cf_node
.parent
= &impl
->cf_node
;
264 end_block
->cf_node
.parent
= &impl
->cf_node
;
265 impl
->start_block
= start_block
;
266 impl
->end_block
= end_block
;
268 exec_list_push_tail(&impl
->body
, &start_block
->cf_node
.node
);
270 start_block
->successors
[0] = end_block
;
271 block_add_pred(end_block
, start_block
);
277 nir_block_create(void *mem_ctx
)
279 nir_block
*block
= ralloc(mem_ctx
, nir_block
);
281 cf_init(&block
->cf_node
, nir_cf_node_block
);
283 block
->successors
[0] = block
->successors
[1] = NULL
;
284 block
->predecessors
= _mesa_set_create(block
, _mesa_hash_pointer
,
285 _mesa_key_pointer_equal
);
286 block
->imm_dom
= NULL
;
287 block
->dom_frontier
= _mesa_set_create(block
, _mesa_hash_pointer
,
288 _mesa_key_pointer_equal
);
290 exec_list_make_empty(&block
->instr_list
);
296 src_init(nir_src
*src
)
300 src
->reg
.indirect
= NULL
;
301 src
->reg
.base_offset
= 0;
305 nir_if_create(void *mem_ctx
)
307 nir_if
*if_stmt
= ralloc(mem_ctx
, nir_if
);
309 cf_init(&if_stmt
->cf_node
, nir_cf_node_if
);
310 src_init(&if_stmt
->condition
);
312 nir_block
*then
= nir_block_create(mem_ctx
);
313 exec_list_make_empty(&if_stmt
->then_list
);
314 exec_list_push_tail(&if_stmt
->then_list
, &then
->cf_node
.node
);
315 then
->cf_node
.parent
= &if_stmt
->cf_node
;
317 nir_block
*else_stmt
= nir_block_create(mem_ctx
);
318 exec_list_make_empty(&if_stmt
->else_list
);
319 exec_list_push_tail(&if_stmt
->else_list
, &else_stmt
->cf_node
.node
);
320 else_stmt
->cf_node
.parent
= &if_stmt
->cf_node
;
326 nir_loop_create(void *mem_ctx
)
328 nir_loop
*loop
= ralloc(mem_ctx
, nir_loop
);
330 cf_init(&loop
->cf_node
, nir_cf_node_loop
);
332 nir_block
*body
= nir_block_create(mem_ctx
);
333 exec_list_make_empty(&loop
->body
);
334 exec_list_push_tail(&loop
->body
, &body
->cf_node
.node
);
335 body
->cf_node
.parent
= &loop
->cf_node
;
337 body
->successors
[0] = body
;
338 block_add_pred(body
, body
);
344 instr_init(nir_instr
*instr
, nir_instr_type type
)
348 exec_node_init(&instr
->node
);
352 dest_init(nir_dest
*dest
)
354 dest
->is_ssa
= false;
355 dest
->reg
.reg
= NULL
;
356 dest
->reg
.indirect
= NULL
;
357 dest
->reg
.base_offset
= 0;
361 alu_dest_init(nir_alu_dest
*dest
)
363 dest_init(&dest
->dest
);
364 dest
->saturate
= false;
365 dest
->write_mask
= 0xf;
369 alu_src_init(nir_alu_src
*src
)
372 src
->abs
= src
->negate
= false;
380 nir_alu_instr_create(nir_shader
*shader
, nir_op op
)
382 unsigned num_srcs
= nir_op_infos
[op
].num_inputs
;
383 nir_alu_instr
*instr
=
385 sizeof(nir_alu_instr
) + num_srcs
* sizeof(nir_alu_src
));
387 instr_init(&instr
->instr
, nir_instr_type_alu
);
389 alu_dest_init(&instr
->dest
);
390 for (unsigned i
= 0; i
< num_srcs
; i
++)
391 alu_src_init(&instr
->src
[i
]);
397 nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
)
399 nir_jump_instr
*instr
= ralloc(shader
, nir_jump_instr
);
400 instr_init(&instr
->instr
, nir_instr_type_jump
);
405 nir_load_const_instr
*
406 nir_load_const_instr_create(nir_shader
*shader
, unsigned num_components
)
408 nir_load_const_instr
*instr
= ralloc(shader
, nir_load_const_instr
);
409 instr_init(&instr
->instr
, nir_instr_type_load_const
);
411 nir_ssa_def_init(&instr
->instr
, &instr
->def
, num_components
, NULL
);
416 nir_intrinsic_instr
*
417 nir_intrinsic_instr_create(nir_shader
*shader
, nir_intrinsic_op op
)
419 unsigned num_srcs
= nir_intrinsic_infos
[op
].num_srcs
;
420 nir_intrinsic_instr
*instr
=
422 sizeof(nir_intrinsic_instr
) + num_srcs
* sizeof(nir_src
));
424 instr_init(&instr
->instr
, nir_instr_type_intrinsic
);
425 instr
->intrinsic
= op
;
427 if (nir_intrinsic_infos
[op
].has_dest
)
428 dest_init(&instr
->dest
);
430 for (unsigned i
= 0; i
< num_srcs
; i
++)
431 src_init(&instr
->src
[i
]);
437 nir_call_instr_create(nir_shader
*shader
, nir_function_overload
*callee
)
439 nir_call_instr
*instr
= ralloc(shader
, nir_call_instr
);
440 instr_init(&instr
->instr
, nir_instr_type_call
);
442 instr
->callee
= callee
;
443 instr
->num_params
= callee
->num_params
;
444 instr
->params
= ralloc_array(instr
, nir_deref_var
*, instr
->num_params
);
445 instr
->return_deref
= NULL
;
451 nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
)
453 nir_tex_instr
*instr
= rzalloc(shader
, nir_tex_instr
);
454 instr_init(&instr
->instr
, nir_instr_type_tex
);
456 dest_init(&instr
->dest
);
458 instr
->num_srcs
= num_srcs
;
459 instr
->src
= ralloc_array(instr
, nir_tex_src
, num_srcs
);
460 for (unsigned i
= 0; i
< num_srcs
; i
++)
461 src_init(&instr
->src
[i
].src
);
463 instr
->sampler_index
= 0;
464 instr
->sampler_array_size
= 0;
465 instr
->sampler
= NULL
;
471 nir_phi_instr_create(nir_shader
*shader
)
473 nir_phi_instr
*instr
= ralloc(shader
, nir_phi_instr
);
474 instr_init(&instr
->instr
, nir_instr_type_phi
);
476 dest_init(&instr
->dest
);
477 exec_list_make_empty(&instr
->srcs
);
481 nir_parallel_copy_instr
*
482 nir_parallel_copy_instr_create(nir_shader
*shader
)
484 nir_parallel_copy_instr
*instr
= ralloc(shader
, nir_parallel_copy_instr
);
485 instr_init(&instr
->instr
, nir_instr_type_parallel_copy
);
487 exec_list_make_empty(&instr
->entries
);
492 nir_ssa_undef_instr
*
493 nir_ssa_undef_instr_create(nir_shader
*shader
, unsigned num_components
)
495 nir_ssa_undef_instr
*instr
= ralloc(shader
, nir_ssa_undef_instr
);
496 instr_init(&instr
->instr
, nir_instr_type_ssa_undef
);
498 nir_ssa_def_init(&instr
->instr
, &instr
->def
, num_components
, NULL
);
504 nir_deref_var_create(void *mem_ctx
, nir_variable
*var
)
506 nir_deref_var
*deref
= ralloc(mem_ctx
, nir_deref_var
);
507 deref
->deref
.deref_type
= nir_deref_type_var
;
508 deref
->deref
.child
= NULL
;
509 deref
->deref
.type
= var
->type
;
515 nir_deref_array_create(void *mem_ctx
)
517 nir_deref_array
*deref
= ralloc(mem_ctx
, nir_deref_array
);
518 deref
->deref
.deref_type
= nir_deref_type_array
;
519 deref
->deref
.child
= NULL
;
520 deref
->deref_array_type
= nir_deref_array_type_direct
;
521 src_init(&deref
->indirect
);
522 deref
->base_offset
= 0;
527 nir_deref_struct_create(void *mem_ctx
, unsigned field_index
)
529 nir_deref_struct
*deref
= ralloc(mem_ctx
, nir_deref_struct
);
530 deref
->deref
.deref_type
= nir_deref_type_struct
;
531 deref
->deref
.child
= NULL
;
532 deref
->index
= field_index
;
536 static nir_deref_var
*
537 copy_deref_var(void *mem_ctx
, nir_deref_var
*deref
)
539 nir_deref_var
*ret
= nir_deref_var_create(mem_ctx
, deref
->var
);
540 ret
->deref
.type
= deref
->deref
.type
;
541 if (deref
->deref
.child
)
542 ret
->deref
.child
= nir_copy_deref(ret
, deref
->deref
.child
);
546 static nir_deref_array
*
547 copy_deref_array(void *mem_ctx
, nir_deref_array
*deref
)
549 nir_deref_array
*ret
= nir_deref_array_create(mem_ctx
);
550 ret
->base_offset
= deref
->base_offset
;
551 ret
->deref_array_type
= deref
->deref_array_type
;
552 if (deref
->deref_array_type
== nir_deref_array_type_indirect
) {
553 nir_src_copy(&ret
->indirect
, &deref
->indirect
, mem_ctx
);
555 ret
->deref
.type
= deref
->deref
.type
;
556 if (deref
->deref
.child
)
557 ret
->deref
.child
= nir_copy_deref(ret
, deref
->deref
.child
);
561 static nir_deref_struct
*
562 copy_deref_struct(void *mem_ctx
, nir_deref_struct
*deref
)
564 nir_deref_struct
*ret
= nir_deref_struct_create(mem_ctx
, deref
->index
);
565 ret
->deref
.type
= deref
->deref
.type
;
566 if (deref
->deref
.child
)
567 ret
->deref
.child
= nir_copy_deref(ret
, deref
->deref
.child
);
572 nir_copy_deref(void *mem_ctx
, nir_deref
*deref
)
574 switch (deref
->deref_type
) {
575 case nir_deref_type_var
:
576 return ©_deref_var(mem_ctx
, nir_deref_as_var(deref
))->deref
;
577 case nir_deref_type_array
:
578 return ©_deref_array(mem_ctx
, nir_deref_as_array(deref
))->deref
;
579 case nir_deref_type_struct
:
580 return ©_deref_struct(mem_ctx
, nir_deref_as_struct(deref
))->deref
;
582 unreachable("Invalid dereference type");
588 /* Returns a load_const instruction that represents the constant
589 * initializer for the given deref chain. The caller is responsible for
590 * ensuring that there actually is a constant initializer.
592 nir_load_const_instr
*
593 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
)
595 nir_constant
*constant
= deref
->var
->constant_initializer
;
598 const nir_deref
*tail
= &deref
->deref
;
599 unsigned matrix_offset
= 0;
600 while (tail
->child
) {
601 switch (tail
->child
->deref_type
) {
602 case nir_deref_type_array
: {
603 nir_deref_array
*arr
= nir_deref_as_array(tail
->child
);
604 assert(arr
->deref_array_type
== nir_deref_array_type_direct
);
605 if (glsl_type_is_matrix(tail
->type
)) {
606 assert(arr
->deref
.child
== NULL
);
607 matrix_offset
= arr
->base_offset
;
609 constant
= constant
->elements
[arr
->base_offset
];
614 case nir_deref_type_struct
: {
615 constant
= constant
->elements
[nir_deref_as_struct(tail
->child
)->index
];
620 unreachable("Invalid deref child type");
626 nir_load_const_instr
*load
=
627 nir_load_const_instr_create(shader
, glsl_get_vector_elements(tail
->type
));
629 matrix_offset
*= load
->def
.num_components
;
630 for (unsigned i
= 0; i
< load
->def
.num_components
; i
++) {
631 switch (glsl_get_base_type(tail
->type
)) {
632 case GLSL_TYPE_FLOAT
:
635 load
->value
.u
[i
] = constant
->value
.u
[matrix_offset
+ i
];
638 load
->value
.u
[i
] = constant
->value
.b
[matrix_offset
+ i
] ?
639 NIR_TRUE
: NIR_FALSE
;
642 unreachable("Invalid immediate type");
650 * \name Control flow modification
652 * These functions modify the control flow tree while keeping the control flow
653 * graph up-to-date. The invariants respected are:
654 * 1. Each then statement, else statement, or loop body must have at least one
656 * 2. Each if-statement and loop must have one basic block before it and one
658 * 3. Two basic blocks cannot be directly next to each other.
659 * 4. If a basic block has a jump instruction, there must be only one and it
660 * must be at the end of the block.
661 * 5. The CFG must always be connected - this means that we must insert a fake
662 * CFG edge for loops with no break statement.
664 * The purpose of the second one is so that we have places to insert code during
665 * GCM, as well as eliminating the possibility of critical edges.
670 link_non_block_to_block(nir_cf_node
*node
, nir_block
*block
)
672 if (node
->type
== nir_cf_node_if
) {
674 * We're trying to link an if to a block after it; this just means linking
675 * the last block of the then and else branches.
678 nir_if
*if_stmt
= nir_cf_node_as_if(node
);
680 nir_cf_node
*last_then
= nir_if_last_then_node(if_stmt
);
681 assert(last_then
->type
== nir_cf_node_block
);
682 nir_block
*last_then_block
= nir_cf_node_as_block(last_then
);
684 nir_cf_node
*last_else
= nir_if_last_else_node(if_stmt
);
685 assert(last_else
->type
== nir_cf_node_block
);
686 nir_block
*last_else_block
= nir_cf_node_as_block(last_else
);
688 if (exec_list_is_empty(&last_then_block
->instr_list
) ||
689 nir_block_last_instr(last_then_block
)->type
!= nir_instr_type_jump
) {
690 unlink_block_successors(last_then_block
);
691 link_blocks(last_then_block
, block
, NULL
);
694 if (exec_list_is_empty(&last_else_block
->instr_list
) ||
695 nir_block_last_instr(last_else_block
)->type
!= nir_instr_type_jump
) {
696 unlink_block_successors(last_else_block
);
697 link_blocks(last_else_block
, block
, NULL
);
700 assert(node
->type
== nir_cf_node_loop
);
703 * We can only get to this codepath if we're inserting a new loop, or
704 * at least a loop with no break statements; we can't insert break
705 * statements into a loop when we haven't inserted it into the CFG
706 * because we wouldn't know which block comes after the loop
707 * and therefore, which block should be the successor of the block with
708 * the break). Therefore, we need to insert a fake edge (see invariant
712 nir_loop
*loop
= nir_cf_node_as_loop(node
);
714 nir_cf_node
*last
= nir_loop_last_cf_node(loop
);
715 assert(last
->type
== nir_cf_node_block
);
716 nir_block
*last_block
= nir_cf_node_as_block(last
);
718 last_block
->successors
[1] = block
;
719 block_add_pred(block
, last_block
);
724 link_block_to_non_block(nir_block
*block
, nir_cf_node
*node
)
726 if (node
->type
== nir_cf_node_if
) {
728 * We're trying to link a block to an if after it; this just means linking
729 * the block to the first block of the then and else branches.
732 nir_if
*if_stmt
= nir_cf_node_as_if(node
);
734 nir_cf_node
*first_then
= nir_if_first_then_node(if_stmt
);
735 assert(first_then
->type
== nir_cf_node_block
);
736 nir_block
*first_then_block
= nir_cf_node_as_block(first_then
);
738 nir_cf_node
*first_else
= nir_if_first_else_node(if_stmt
);
739 assert(first_else
->type
== nir_cf_node_block
);
740 nir_block
*first_else_block
= nir_cf_node_as_block(first_else
);
742 unlink_block_successors(block
);
743 link_blocks(block
, first_then_block
, first_else_block
);
746 * For similar reasons as the corresponding case in
747 * link_non_block_to_block(), don't worry about if the loop header has
748 * any predecessors that need to be unlinked.
751 assert(node
->type
== nir_cf_node_loop
);
753 nir_loop
*loop
= nir_cf_node_as_loop(node
);
755 nir_cf_node
*loop_header
= nir_loop_first_cf_node(loop
);
756 assert(loop_header
->type
== nir_cf_node_block
);
757 nir_block
*loop_header_block
= nir_cf_node_as_block(loop_header
);
759 unlink_block_successors(block
);
760 link_blocks(block
, loop_header_block
, NULL
);
766 * Takes a basic block and inserts a new empty basic block before it, making its
767 * predecessors point to the new block. This essentially splits the block into
768 * an empty header and a body so that another non-block CF node can be inserted
769 * between the two. Note that this does *not* link the two basic blocks, so
770 * some kind of cleanup *must* be performed after this call.
774 split_block_beginning(nir_block
*block
)
776 nir_block
*new_block
= nir_block_create(ralloc_parent(block
));
777 new_block
->cf_node
.parent
= block
->cf_node
.parent
;
778 exec_node_insert_node_before(&block
->cf_node
.node
, &new_block
->cf_node
.node
);
780 struct set_entry
*entry
;
781 set_foreach(block
->predecessors
, entry
) {
782 nir_block
*pred
= (nir_block
*) entry
->key
;
784 unlink_blocks(pred
, block
);
785 link_blocks(pred
, new_block
, NULL
);
792 rewrite_phi_preds(nir_block
*block
, nir_block
*old_pred
, nir_block
*new_pred
)
794 nir_foreach_instr_safe(block
, instr
) {
795 if (instr
->type
!= nir_instr_type_phi
)
798 nir_phi_instr
*phi
= nir_instr_as_phi(instr
);
799 nir_foreach_phi_src(phi
, src
) {
800 if (src
->pred
== old_pred
) {
801 src
->pred
= new_pred
;
809 * Moves the successors of source to the successors of dest, leaving both
810 * successors of source NULL.
814 move_successors(nir_block
*source
, nir_block
*dest
)
816 nir_block
*succ1
= source
->successors
[0];
817 nir_block
*succ2
= source
->successors
[1];
820 unlink_blocks(source
, succ1
);
821 rewrite_phi_preds(succ1
, source
, dest
);
825 unlink_blocks(source
, succ2
);
826 rewrite_phi_preds(succ2
, source
, dest
);
829 unlink_block_successors(dest
);
830 link_blocks(dest
, succ1
, succ2
);
834 split_block_end(nir_block
*block
)
836 nir_block
*new_block
= nir_block_create(ralloc_parent(block
));
837 new_block
->cf_node
.parent
= block
->cf_node
.parent
;
838 exec_node_insert_after(&block
->cf_node
.node
, &new_block
->cf_node
.node
);
840 move_successors(block
, new_block
);
846 * Inserts a non-basic block between two basic blocks and links them together.
850 insert_non_block(nir_block
*before
, nir_cf_node
*node
, nir_block
*after
)
852 node
->parent
= before
->cf_node
.parent
;
853 exec_node_insert_after(&before
->cf_node
.node
, &node
->node
);
854 link_block_to_non_block(before
, node
);
855 link_non_block_to_block(node
, after
);
859 * Inserts a non-basic block before a basic block.
863 insert_non_block_before_block(nir_cf_node
*node
, nir_block
*block
)
865 /* split off the beginning of block into new_block */
866 nir_block
*new_block
= split_block_beginning(block
);
868 /* insert our node in between new_block and block */
869 insert_non_block(new_block
, node
, block
);
873 insert_non_block_after_block(nir_block
*block
, nir_cf_node
*node
)
875 /* split off the end of block into new_block */
876 nir_block
*new_block
= split_block_end(block
);
878 /* insert our node in between block and new_block */
879 insert_non_block(block
, node
, new_block
);
882 /* walk up the control flow tree to find the innermost enclosed loop */
884 nearest_loop(nir_cf_node
*node
)
886 while (node
->type
!= nir_cf_node_loop
) {
890 return nir_cf_node_as_loop(node
);
894 nir_cf_node_get_function(nir_cf_node
*node
)
896 while (node
->type
!= nir_cf_node_function
) {
900 return nir_cf_node_as_function(node
);
904 * update the CFG after a jump instruction has been added to the end of a block
908 handle_jump(nir_block
*block
)
910 nir_instr
*instr
= nir_block_last_instr(block
);
911 nir_jump_instr
*jump_instr
= nir_instr_as_jump(instr
);
913 unlink_block_successors(block
);
915 nir_function_impl
*impl
= nir_cf_node_get_function(&block
->cf_node
);
916 nir_metadata_preserve(impl
, nir_metadata_none
);
918 if (jump_instr
->type
== nir_jump_break
||
919 jump_instr
->type
== nir_jump_continue
) {
920 nir_loop
*loop
= nearest_loop(&block
->cf_node
);
922 if (jump_instr
->type
== nir_jump_continue
) {
923 nir_cf_node
*first_node
= nir_loop_first_cf_node(loop
);
924 assert(first_node
->type
== nir_cf_node_block
);
925 nir_block
*first_block
= nir_cf_node_as_block(first_node
);
926 link_blocks(block
, first_block
, NULL
);
928 nir_cf_node
*after
= nir_cf_node_next(&loop
->cf_node
);
929 assert(after
->type
== nir_cf_node_block
);
930 nir_block
*after_block
= nir_cf_node_as_block(after
);
931 link_blocks(block
, after_block
, NULL
);
933 /* If we inserted a fake link, remove it */
934 nir_cf_node
*last
= nir_loop_last_cf_node(loop
);
935 assert(last
->type
== nir_cf_node_block
);
936 nir_block
*last_block
= nir_cf_node_as_block(last
);
937 if (last_block
->successors
[1] != NULL
)
938 unlink_blocks(last_block
, after_block
);
941 assert(jump_instr
->type
== nir_jump_return
);
942 link_blocks(block
, impl
->end_block
, NULL
);
947 handle_remove_jump(nir_block
*block
, nir_jump_type type
)
949 unlink_block_successors(block
);
951 if (exec_node_is_tail_sentinel(block
->cf_node
.node
.next
)) {
952 nir_cf_node
*parent
= block
->cf_node
.parent
;
953 if (parent
->type
== nir_cf_node_if
) {
954 nir_cf_node
*next
= nir_cf_node_next(parent
);
955 assert(next
->type
== nir_cf_node_block
);
956 nir_block
*next_block
= nir_cf_node_as_block(next
);
958 link_blocks(block
, next_block
, NULL
);
960 assert(parent
->type
== nir_cf_node_loop
);
961 nir_loop
*loop
= nir_cf_node_as_loop(parent
);
963 nir_cf_node
*head
= nir_loop_first_cf_node(loop
);
964 assert(head
->type
== nir_cf_node_block
);
965 nir_block
*head_block
= nir_cf_node_as_block(head
);
967 link_blocks(block
, head_block
, NULL
);
970 nir_cf_node
*next
= nir_cf_node_next(&block
->cf_node
);
971 if (next
->type
== nir_cf_node_if
) {
972 nir_if
*next_if
= nir_cf_node_as_if(next
);
974 nir_cf_node
*first_then
= nir_if_first_then_node(next_if
);
975 assert(first_then
->type
== nir_cf_node_block
);
976 nir_block
*first_then_block
= nir_cf_node_as_block(first_then
);
978 nir_cf_node
*first_else
= nir_if_first_else_node(next_if
);
979 assert(first_else
->type
== nir_cf_node_block
);
980 nir_block
*first_else_block
= nir_cf_node_as_block(first_else
);
982 link_blocks(block
, first_then_block
, first_else_block
);
984 assert(next
->type
== nir_cf_node_loop
);
985 nir_loop
*next_loop
= nir_cf_node_as_loop(next
);
987 nir_cf_node
*first
= nir_loop_first_cf_node(next_loop
);
988 assert(first
->type
== nir_cf_node_block
);
989 nir_block
*first_block
= nir_cf_node_as_block(first
);
991 link_blocks(block
, first_block
, NULL
);
995 if (type
== nir_jump_break
) {
996 nir_loop
*loop
= nearest_loop(&block
->cf_node
);
998 nir_cf_node
*next
= nir_cf_node_next(&loop
->cf_node
);
999 assert(next
->type
== nir_cf_node_block
);
1000 nir_block
*next_block
= nir_cf_node_as_block(next
);
1002 if (next_block
->predecessors
->entries
== 0) {
1003 /* insert fake link */
1004 nir_cf_node
*last
= nir_loop_last_cf_node(loop
);
1005 assert(last
->type
== nir_cf_node_block
);
1006 nir_block
*last_block
= nir_cf_node_as_block(last
);
1008 last_block
->successors
[1] = next_block
;
1009 block_add_pred(next_block
, last_block
);
1013 nir_function_impl
*impl
= nir_cf_node_get_function(&block
->cf_node
);
1014 nir_metadata_preserve(impl
, nir_metadata_none
);
1018 * Inserts a basic block before another by merging the instructions.
1020 * @param block the target of the insertion
1021 * @param before the block to be inserted - must not have been inserted before
1022 * @param has_jump whether \before has a jump instruction at the end
1026 insert_block_before_block(nir_block
*block
, nir_block
*before
, bool has_jump
)
1028 assert(!has_jump
|| exec_list_is_empty(&block
->instr_list
));
1030 foreach_list_typed(nir_instr
, instr
, node
, &before
->instr_list
) {
1031 instr
->block
= block
;
1034 exec_list_prepend(&block
->instr_list
, &before
->instr_list
);
1041 * Inserts a basic block after another by merging the instructions.
1043 * @param block the target of the insertion
1044 * @param after the block to be inserted - must not have been inserted before
1045 * @param has_jump whether \after has a jump instruction at the end
1049 insert_block_after_block(nir_block
*block
, nir_block
*after
, bool has_jump
)
1051 foreach_list_typed(nir_instr
, instr
, node
, &after
->instr_list
) {
1052 instr
->block
= block
;
1055 exec_list_append(&block
->instr_list
, &after
->instr_list
);
1062 update_if_uses(nir_cf_node
*node
)
1064 if (node
->type
!= nir_cf_node_if
)
1067 nir_if
*if_stmt
= nir_cf_node_as_if(node
);
1069 if_stmt
->condition
.parent_if
= if_stmt
;
1070 if (if_stmt
->condition
.is_ssa
) {
1071 list_addtail(&if_stmt
->condition
.use_link
,
1072 &if_stmt
->condition
.ssa
->if_uses
);
1074 list_addtail(&if_stmt
->condition
.use_link
,
1075 &if_stmt
->condition
.reg
.reg
->if_uses
);
1080 nir_cf_node_insert_after(nir_cf_node
*node
, nir_cf_node
*after
)
1082 update_if_uses(after
);
1084 if (after
->type
== nir_cf_node_block
) {
1086 * either node or the one after it must be a basic block, by invariant #2;
1087 * in either case, just merge the blocks together.
1089 nir_block
*after_block
= nir_cf_node_as_block(after
);
1091 bool has_jump
= !exec_list_is_empty(&after_block
->instr_list
) &&
1092 nir_block_last_instr(after_block
)->type
== nir_instr_type_jump
;
1094 if (node
->type
== nir_cf_node_block
) {
1095 insert_block_after_block(nir_cf_node_as_block(node
), after_block
,
1098 nir_cf_node
*next
= nir_cf_node_next(node
);
1099 assert(next
->type
== nir_cf_node_block
);
1100 nir_block
*next_block
= nir_cf_node_as_block(next
);
1102 insert_block_before_block(next_block
, after_block
, has_jump
);
1105 if (node
->type
== nir_cf_node_block
) {
1106 insert_non_block_after_block(nir_cf_node_as_block(node
), after
);
1109 * We have to insert a non-basic block after a non-basic block. Since
1110 * every non-basic block has a basic block after it, this is equivalent
1111 * to inserting a non-basic block before a basic block.
1114 nir_cf_node
*next
= nir_cf_node_next(node
);
1115 assert(next
->type
== nir_cf_node_block
);
1116 nir_block
*next_block
= nir_cf_node_as_block(next
);
1118 insert_non_block_before_block(after
, next_block
);
1122 nir_function_impl
*impl
= nir_cf_node_get_function(node
);
1123 nir_metadata_preserve(impl
, nir_metadata_none
);
1127 nir_cf_node_insert_before(nir_cf_node
*node
, nir_cf_node
*before
)
1129 update_if_uses(before
);
1131 if (before
->type
== nir_cf_node_block
) {
1132 nir_block
*before_block
= nir_cf_node_as_block(before
);
1134 bool has_jump
= !exec_list_is_empty(&before_block
->instr_list
) &&
1135 nir_block_last_instr(before_block
)->type
== nir_instr_type_jump
;
1137 if (node
->type
== nir_cf_node_block
) {
1138 insert_block_before_block(nir_cf_node_as_block(node
), before_block
,
1141 nir_cf_node
*prev
= nir_cf_node_prev(node
);
1142 assert(prev
->type
== nir_cf_node_block
);
1143 nir_block
*prev_block
= nir_cf_node_as_block(prev
);
1145 insert_block_after_block(prev_block
, before_block
, has_jump
);
1148 if (node
->type
== nir_cf_node_block
) {
1149 insert_non_block_before_block(before
, nir_cf_node_as_block(node
));
1152 * We have to insert a non-basic block before a non-basic block. This
1153 * is equivalent to inserting a non-basic block after a basic block.
1156 nir_cf_node
*prev_node
= nir_cf_node_prev(node
);
1157 assert(prev_node
->type
== nir_cf_node_block
);
1158 nir_block
*prev_block
= nir_cf_node_as_block(prev_node
);
1160 insert_non_block_after_block(prev_block
, before
);
1164 nir_function_impl
*impl
= nir_cf_node_get_function(node
);
1165 nir_metadata_preserve(impl
, nir_metadata_none
);
1169 nir_cf_node_insert_begin(struct exec_list
*list
, nir_cf_node
*node
)
1171 nir_cf_node
*begin
= exec_node_data(nir_cf_node
, list
->head
, node
);
1172 nir_cf_node_insert_before(begin
, node
);
1176 nir_cf_node_insert_end(struct exec_list
*list
, nir_cf_node
*node
)
1178 nir_cf_node
*end
= exec_node_data(nir_cf_node
, list
->tail_pred
, node
);
1179 nir_cf_node_insert_after(end
, node
);
1183 * Stitch two basic blocks together into one. The aggregate must have the same
1184 * predecessors as the first and the same successors as the second.
1188 stitch_blocks(nir_block
*before
, nir_block
*after
)
1191 * We move after into before, so we have to deal with up to 2 successors vs.
1192 * possibly a large number of predecessors.
1194 * TODO: special case when before is empty and after isn't?
1197 move_successors(after
, before
);
1199 foreach_list_typed(nir_instr
, instr
, node
, &after
->instr_list
) {
1200 instr
->block
= before
;
1203 exec_list_append(&before
->instr_list
, &after
->instr_list
);
1204 exec_node_remove(&after
->cf_node
.node
);
1208 remove_defs_uses(nir_instr
*instr
);
1211 cleanup_cf_node(nir_cf_node
*node
)
1213 switch (node
->type
) {
1214 case nir_cf_node_block
: {
1215 nir_block
*block
= nir_cf_node_as_block(node
);
1216 /* We need to walk the instructions and clean up defs/uses */
1217 nir_foreach_instr(block
, instr
)
1218 remove_defs_uses(instr
);
1222 case nir_cf_node_if
: {
1223 nir_if
*if_stmt
= nir_cf_node_as_if(node
);
1224 foreach_list_typed(nir_cf_node
, child
, node
, &if_stmt
->then_list
)
1225 cleanup_cf_node(child
);
1226 foreach_list_typed(nir_cf_node
, child
, node
, &if_stmt
->else_list
)
1227 cleanup_cf_node(child
);
1229 list_del(&if_stmt
->condition
.use_link
);
1233 case nir_cf_node_loop
: {
1234 nir_loop
*loop
= nir_cf_node_as_loop(node
);
1235 foreach_list_typed(nir_cf_node
, child
, node
, &loop
->body
)
1236 cleanup_cf_node(child
);
1239 case nir_cf_node_function
: {
1240 nir_function_impl
*impl
= nir_cf_node_as_function(node
);
1241 foreach_list_typed(nir_cf_node
, child
, node
, &impl
->body
)
1242 cleanup_cf_node(child
);
1246 unreachable("Invalid CF node type");
1251 nir_cf_node_remove(nir_cf_node
*node
)
1253 nir_function_impl
*impl
= nir_cf_node_get_function(node
);
1254 nir_metadata_preserve(impl
, nir_metadata_none
);
1256 if (node
->type
== nir_cf_node_block
) {
1258 * Basic blocks can't really be removed by themselves, since they act as
1259 * padding between the non-basic blocks. So all we do here is empty the
1260 * block of instructions.
1262 * TODO: could we assert here?
1264 exec_list_make_empty(&nir_cf_node_as_block(node
)->instr_list
);
1266 nir_cf_node
*before
= nir_cf_node_prev(node
);
1267 assert(before
->type
== nir_cf_node_block
);
1268 nir_block
*before_block
= nir_cf_node_as_block(before
);
1270 nir_cf_node
*after
= nir_cf_node_next(node
);
1271 assert(after
->type
== nir_cf_node_block
);
1272 nir_block
*after_block
= nir_cf_node_as_block(after
);
1274 exec_node_remove(&node
->node
);
1275 stitch_blocks(before_block
, after_block
);
1278 cleanup_cf_node(node
);
1282 add_use_cb(nir_src
*src
, void *state
)
1284 nir_instr
*instr
= state
;
1286 src
->parent_instr
= instr
;
1287 list_addtail(&src
->use_link
,
1288 src
->is_ssa
? &src
->ssa
->uses
: &src
->reg
.reg
->uses
);
1294 add_ssa_def_cb(nir_ssa_def
*def
, void *state
)
1296 nir_instr
*instr
= state
;
1298 if (instr
->block
&& def
->index
== UINT_MAX
) {
1299 nir_function_impl
*impl
=
1300 nir_cf_node_get_function(&instr
->block
->cf_node
);
1302 def
->index
= impl
->ssa_alloc
++;
1309 add_reg_def_cb(nir_dest
*dest
, void *state
)
1311 nir_instr
*instr
= state
;
1313 if (!dest
->is_ssa
) {
1314 dest
->reg
.parent_instr
= instr
;
1315 list_addtail(&dest
->reg
.def_link
, &dest
->reg
.reg
->defs
);
1322 add_defs_uses(nir_instr
*instr
)
1324 nir_foreach_src(instr
, add_use_cb
, instr
);
1325 nir_foreach_dest(instr
, add_reg_def_cb
, instr
);
1326 nir_foreach_ssa_def(instr
, add_ssa_def_cb
, instr
);
1330 nir_instr_insert_before(nir_instr
*instr
, nir_instr
*before
)
1332 assert(before
->type
!= nir_instr_type_jump
);
1333 before
->block
= instr
->block
;
1334 add_defs_uses(before
);
1335 exec_node_insert_node_before(&instr
->node
, &before
->node
);
1339 nir_instr_insert_after(nir_instr
*instr
, nir_instr
*after
)
1341 if (after
->type
== nir_instr_type_jump
) {
1342 assert(instr
== nir_block_last_instr(instr
->block
));
1343 assert(instr
->type
!= nir_instr_type_jump
);
1346 after
->block
= instr
->block
;
1347 add_defs_uses(after
);
1348 exec_node_insert_after(&instr
->node
, &after
->node
);
1350 if (after
->type
== nir_instr_type_jump
)
1351 handle_jump(after
->block
);
1355 nir_instr_insert_before_block(nir_block
*block
, nir_instr
*before
)
1357 if (before
->type
== nir_instr_type_jump
)
1358 assert(exec_list_is_empty(&block
->instr_list
));
1360 before
->block
= block
;
1361 add_defs_uses(before
);
1362 exec_list_push_head(&block
->instr_list
, &before
->node
);
1364 if (before
->type
== nir_instr_type_jump
)
1369 nir_instr_insert_after_block(nir_block
*block
, nir_instr
*after
)
1371 if (after
->type
== nir_instr_type_jump
) {
1372 assert(exec_list_is_empty(&block
->instr_list
) ||
1373 nir_block_last_instr(block
)->type
!= nir_instr_type_jump
);
1376 after
->block
= block
;
1377 add_defs_uses(after
);
1378 exec_list_push_tail(&block
->instr_list
, &after
->node
);
1380 if (after
->type
== nir_instr_type_jump
)
1385 nir_instr_insert_before_cf(nir_cf_node
*node
, nir_instr
*before
)
1387 if (node
->type
== nir_cf_node_block
) {
1388 nir_instr_insert_before_block(nir_cf_node_as_block(node
), before
);
1390 nir_cf_node
*prev
= nir_cf_node_prev(node
);
1391 assert(prev
->type
== nir_cf_node_block
);
1392 nir_block
*prev_block
= nir_cf_node_as_block(prev
);
1394 nir_instr_insert_before_block(prev_block
, before
);
1399 nir_instr_insert_after_cf(nir_cf_node
*node
, nir_instr
*after
)
1401 if (node
->type
== nir_cf_node_block
) {
1402 nir_instr_insert_after_block(nir_cf_node_as_block(node
), after
);
1404 nir_cf_node
*next
= nir_cf_node_next(node
);
1405 assert(next
->type
== nir_cf_node_block
);
1406 nir_block
*next_block
= nir_cf_node_as_block(next
);
1408 nir_instr_insert_before_block(next_block
, after
);
1413 nir_instr_insert_before_cf_list(struct exec_list
*list
, nir_instr
*before
)
1415 nir_cf_node
*first_node
= exec_node_data(nir_cf_node
,
1416 exec_list_get_head(list
), node
);
1417 nir_instr_insert_before_cf(first_node
, before
);
1421 nir_instr_insert_after_cf_list(struct exec_list
*list
, nir_instr
*after
)
1423 nir_cf_node
*last_node
= exec_node_data(nir_cf_node
,
1424 exec_list_get_tail(list
), node
);
1425 nir_instr_insert_after_cf(last_node
, after
);
1429 remove_use_cb(nir_src
*src
, void *state
)
1431 list_del(&src
->use_link
);
1437 remove_def_cb(nir_dest
*dest
, void *state
)
1440 list_del(&dest
->reg
.def_link
);
1446 remove_defs_uses(nir_instr
*instr
)
1448 nir_foreach_dest(instr
, remove_def_cb
, instr
);
1449 nir_foreach_src(instr
, remove_use_cb
, instr
);
1452 void nir_instr_remove(nir_instr
*instr
)
1454 remove_defs_uses(instr
);
1455 exec_node_remove(&instr
->node
);
1457 if (instr
->type
== nir_instr_type_jump
) {
1458 nir_jump_instr
*jump_instr
= nir_instr_as_jump(instr
);
1459 handle_remove_jump(instr
->block
, jump_instr
->type
);
1466 nir_index_local_regs(nir_function_impl
*impl
)
1469 foreach_list_typed(nir_register
, reg
, node
, &impl
->registers
) {
1470 reg
->index
= index
++;
1472 impl
->reg_alloc
= index
;
1476 nir_index_global_regs(nir_shader
*shader
)
1479 foreach_list_typed(nir_register
, reg
, node
, &shader
->registers
) {
1480 reg
->index
= index
++;
1482 shader
->reg_alloc
= index
;
1486 visit_alu_dest(nir_alu_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
1488 return cb(&instr
->dest
.dest
, state
);
1492 visit_intrinsic_dest(nir_intrinsic_instr
*instr
, nir_foreach_dest_cb cb
,
1495 if (nir_intrinsic_infos
[instr
->intrinsic
].has_dest
)
1496 return cb(&instr
->dest
, state
);
1502 visit_texture_dest(nir_tex_instr
*instr
, nir_foreach_dest_cb cb
,
1505 return cb(&instr
->dest
, state
);
1509 visit_phi_dest(nir_phi_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
1511 return cb(&instr
->dest
, state
);
1515 visit_parallel_copy_dest(nir_parallel_copy_instr
*instr
,
1516 nir_foreach_dest_cb cb
, void *state
)
1518 nir_foreach_parallel_copy_entry(instr
, entry
) {
1519 if (!cb(&entry
->dest
, state
))
1527 nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
1529 switch (instr
->type
) {
1530 case nir_instr_type_alu
:
1531 return visit_alu_dest(nir_instr_as_alu(instr
), cb
, state
);
1532 case nir_instr_type_intrinsic
:
1533 return visit_intrinsic_dest(nir_instr_as_intrinsic(instr
), cb
, state
);
1534 case nir_instr_type_tex
:
1535 return visit_texture_dest(nir_instr_as_tex(instr
), cb
, state
);
1536 case nir_instr_type_phi
:
1537 return visit_phi_dest(nir_instr_as_phi(instr
), cb
, state
);
1538 case nir_instr_type_parallel_copy
:
1539 return visit_parallel_copy_dest(nir_instr_as_parallel_copy(instr
),
1542 case nir_instr_type_load_const
:
1543 case nir_instr_type_ssa_undef
:
1544 case nir_instr_type_call
:
1545 case nir_instr_type_jump
:
1549 unreachable("Invalid instruction type");
1556 struct foreach_ssa_def_state
{
1557 nir_foreach_ssa_def_cb cb
;
1562 nir_ssa_def_visitor(nir_dest
*dest
, void *void_state
)
1564 struct foreach_ssa_def_state
*state
= void_state
;
1567 return state
->cb(&dest
->ssa
, state
->client_state
);
1573 nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
, void *state
)
1575 switch (instr
->type
) {
1576 case nir_instr_type_alu
:
1577 case nir_instr_type_tex
:
1578 case nir_instr_type_intrinsic
:
1579 case nir_instr_type_phi
:
1580 case nir_instr_type_parallel_copy
: {
1581 struct foreach_ssa_def_state foreach_state
= {cb
, state
};
1582 return nir_foreach_dest(instr
, nir_ssa_def_visitor
, &foreach_state
);
1585 case nir_instr_type_load_const
:
1586 return cb(&nir_instr_as_load_const(instr
)->def
, state
);
1587 case nir_instr_type_ssa_undef
:
1588 return cb(&nir_instr_as_ssa_undef(instr
)->def
, state
);
1589 case nir_instr_type_call
:
1590 case nir_instr_type_jump
:
1593 unreachable("Invalid instruction type");
1598 visit_src(nir_src
*src
, nir_foreach_src_cb cb
, void *state
)
1600 if (!cb(src
, state
))
1602 if (!src
->is_ssa
&& src
->reg
.indirect
)
1603 return cb(src
->reg
.indirect
, state
);
1608 visit_deref_array_src(nir_deref_array
*deref
, nir_foreach_src_cb cb
,
1611 if (deref
->deref_array_type
== nir_deref_array_type_indirect
)
1612 return visit_src(&deref
->indirect
, cb
, state
);
1617 visit_deref_src(nir_deref_var
*deref
, nir_foreach_src_cb cb
, void *state
)
1619 nir_deref
*cur
= &deref
->deref
;
1620 while (cur
!= NULL
) {
1621 if (cur
->deref_type
== nir_deref_type_array
)
1622 if (!visit_deref_array_src(nir_deref_as_array(cur
), cb
, state
))
1632 visit_alu_src(nir_alu_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1634 for (unsigned i
= 0; i
< nir_op_infos
[instr
->op
].num_inputs
; i
++)
1635 if (!visit_src(&instr
->src
[i
].src
, cb
, state
))
1642 visit_tex_src(nir_tex_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1644 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
1645 if (!visit_src(&instr
->src
[i
].src
, cb
, state
))
1648 if (instr
->sampler
!= NULL
)
1649 if (!visit_deref_src(instr
->sampler
, cb
, state
))
1656 visit_intrinsic_src(nir_intrinsic_instr
*instr
, nir_foreach_src_cb cb
,
1659 unsigned num_srcs
= nir_intrinsic_infos
[instr
->intrinsic
].num_srcs
;
1660 for (unsigned i
= 0; i
< num_srcs
; i
++)
1661 if (!visit_src(&instr
->src
[i
], cb
, state
))
1665 nir_intrinsic_infos
[instr
->intrinsic
].num_variables
;
1666 for (unsigned i
= 0; i
< num_vars
; i
++)
1667 if (!visit_deref_src(instr
->variables
[i
], cb
, state
))
1674 visit_call_src(nir_call_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1680 visit_load_const_src(nir_load_const_instr
*instr
, nir_foreach_src_cb cb
,
1687 visit_phi_src(nir_phi_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1689 nir_foreach_phi_src(instr
, src
) {
1690 if (!visit_src(&src
->src
, cb
, state
))
1698 visit_parallel_copy_src(nir_parallel_copy_instr
*instr
,
1699 nir_foreach_src_cb cb
, void *state
)
1701 nir_foreach_parallel_copy_entry(instr
, entry
) {
1702 if (!visit_src(&entry
->src
, cb
, state
))
1711 nir_foreach_src_cb cb
;
1712 } visit_dest_indirect_state
;
1715 visit_dest_indirect(nir_dest
*dest
, void *_state
)
1717 visit_dest_indirect_state
*state
= (visit_dest_indirect_state
*) _state
;
1719 if (!dest
->is_ssa
&& dest
->reg
.indirect
)
1720 return state
->cb(dest
->reg
.indirect
, state
->state
);
1726 nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1728 switch (instr
->type
) {
1729 case nir_instr_type_alu
:
1730 if (!visit_alu_src(nir_instr_as_alu(instr
), cb
, state
))
1733 case nir_instr_type_intrinsic
:
1734 if (!visit_intrinsic_src(nir_instr_as_intrinsic(instr
), cb
, state
))
1737 case nir_instr_type_tex
:
1738 if (!visit_tex_src(nir_instr_as_tex(instr
), cb
, state
))
1741 case nir_instr_type_call
:
1742 if (!visit_call_src(nir_instr_as_call(instr
), cb
, state
))
1745 case nir_instr_type_load_const
:
1746 if (!visit_load_const_src(nir_instr_as_load_const(instr
), cb
, state
))
1749 case nir_instr_type_phi
:
1750 if (!visit_phi_src(nir_instr_as_phi(instr
), cb
, state
))
1753 case nir_instr_type_parallel_copy
:
1754 if (!visit_parallel_copy_src(nir_instr_as_parallel_copy(instr
),
1758 case nir_instr_type_jump
:
1759 case nir_instr_type_ssa_undef
:
1763 unreachable("Invalid instruction type");
1767 visit_dest_indirect_state dest_state
;
1768 dest_state
.state
= state
;
1770 return nir_foreach_dest(instr
, visit_dest_indirect
, &dest_state
);
1774 nir_src_as_const_value(nir_src src
)
1779 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_load_const
)
1782 nir_load_const_instr
*load
= nir_instr_as_load_const(src
.ssa
->parent_instr
);
1784 return &load
->value
;
1788 nir_srcs_equal(nir_src src1
, nir_src src2
)
1792 return src1
.ssa
== src2
.ssa
;
1800 if ((src1
.reg
.indirect
== NULL
) != (src2
.reg
.indirect
== NULL
))
1803 if (src1
.reg
.indirect
) {
1804 if (!nir_srcs_equal(*src1
.reg
.indirect
, *src2
.reg
.indirect
))
1808 return src1
.reg
.reg
== src2
.reg
.reg
&&
1809 src1
.reg
.base_offset
== src2
.reg
.base_offset
;
1815 src_is_valid(const nir_src
*src
)
1817 return src
->is_ssa
? (src
->ssa
!= NULL
) : (src
->reg
.reg
!= NULL
);
1821 src_remove_all_uses(nir_src
*src
)
1823 for (; src
; src
= src
->is_ssa
? NULL
: src
->reg
.indirect
) {
1824 if (!src_is_valid(src
))
1827 list_del(&src
->use_link
);
1832 src_add_all_uses(nir_src
*src
, nir_instr
*parent_instr
, nir_if
*parent_if
)
1834 for (; src
; src
= src
->is_ssa
? NULL
: src
->reg
.indirect
) {
1835 if (!src_is_valid(src
))
1839 src
->parent_instr
= parent_instr
;
1841 list_addtail(&src
->use_link
, &src
->ssa
->uses
);
1843 list_addtail(&src
->use_link
, &src
->reg
.reg
->uses
);
1846 src
->parent_if
= parent_if
;
1848 list_addtail(&src
->use_link
, &src
->ssa
->if_uses
);
1850 list_addtail(&src
->use_link
, &src
->reg
.reg
->if_uses
);
1856 nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
)
1858 assert(!src_is_valid(src
) || src
->parent_instr
== instr
);
1860 src_remove_all_uses(src
);
1862 src_add_all_uses(src
, instr
, NULL
);
1866 nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
)
1868 assert(!src_is_valid(dest
) || dest
->parent_instr
== dest_instr
);
1870 src_remove_all_uses(dest
);
1871 src_remove_all_uses(src
);
1873 *src
= NIR_SRC_INIT
;
1874 src_add_all_uses(dest
, dest_instr
, NULL
);
1878 nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
)
1880 nir_src
*src
= &if_stmt
->condition
;
1881 assert(!src_is_valid(src
) || src
->parent_if
== if_stmt
);
1883 src_remove_all_uses(src
);
1885 src_add_all_uses(src
, NULL
, if_stmt
);
1889 nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1890 unsigned num_components
, const char *name
)
1893 def
->parent_instr
= instr
;
1894 list_inithead(&def
->uses
);
1895 list_inithead(&def
->if_uses
);
1896 def
->num_components
= num_components
;
1899 nir_function_impl
*impl
=
1900 nir_cf_node_get_function(&instr
->block
->cf_node
);
1902 def
->index
= impl
->ssa_alloc
++;
1904 def
->index
= UINT_MAX
;
1909 nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1910 unsigned num_components
, const char *name
)
1912 dest
->is_ssa
= true;
1913 nir_ssa_def_init(instr
, &dest
->ssa
, num_components
, name
);
1917 nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
, void *mem_ctx
)
1919 assert(!new_src
.is_ssa
|| def
!= new_src
.ssa
);
1921 nir_foreach_use_safe(def
, use_src
) {
1922 nir_instr
*src_parent_instr
= use_src
->parent_instr
;
1923 list_del(&use_src
->use_link
);
1924 nir_src_copy(use_src
, &new_src
, mem_ctx
);
1925 src_add_all_uses(use_src
, src_parent_instr
, NULL
);
1928 nir_foreach_if_use_safe(def
, use_src
) {
1929 nir_if
*src_parent_if
= use_src
->parent_if
;
1930 list_del(&use_src
->use_link
);
1931 nir_src_copy(use_src
, &new_src
, mem_ctx
);
1932 src_add_all_uses(use_src
, NULL
, src_parent_if
);
1937 static bool foreach_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1938 bool reverse
, void *state
);
1941 foreach_if(nir_if
*if_stmt
, nir_foreach_block_cb cb
, bool reverse
, void *state
)
1944 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
,
1945 &if_stmt
->else_list
) {
1946 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1950 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
,
1951 &if_stmt
->then_list
) {
1952 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1956 foreach_list_typed_safe(nir_cf_node
, node
, node
, &if_stmt
->then_list
) {
1957 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1961 foreach_list_typed_safe(nir_cf_node
, node
, node
, &if_stmt
->else_list
) {
1962 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1971 foreach_loop(nir_loop
*loop
, nir_foreach_block_cb cb
, bool reverse
, void *state
)
1974 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
, &loop
->body
) {
1975 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1979 foreach_list_typed_safe(nir_cf_node
, node
, node
, &loop
->body
) {
1980 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1989 foreach_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1990 bool reverse
, void *state
)
1992 switch (node
->type
) {
1993 case nir_cf_node_block
:
1994 return cb(nir_cf_node_as_block(node
), state
);
1995 case nir_cf_node_if
:
1996 return foreach_if(nir_cf_node_as_if(node
), cb
, reverse
, state
);
1997 case nir_cf_node_loop
:
1998 return foreach_loop(nir_cf_node_as_loop(node
), cb
, reverse
, state
);
2002 unreachable("Invalid CFG node type");
2010 nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
, void *state
)
2012 foreach_list_typed_safe(nir_cf_node
, node
, node
, &impl
->body
) {
2013 if (!foreach_cf_node(node
, cb
, false, state
))
2017 return cb(impl
->end_block
, state
);
2021 nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
2024 if (!cb(impl
->end_block
, state
))
2027 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
, &impl
->body
) {
2028 if (!foreach_cf_node(node
, cb
, true, state
))
2036 nir_block_get_following_if(nir_block
*block
)
2038 if (exec_node_is_tail_sentinel(&block
->cf_node
.node
))
2041 if (nir_cf_node_is_last(&block
->cf_node
))
2044 nir_cf_node
*next_node
= nir_cf_node_next(&block
->cf_node
);
2046 if (next_node
->type
!= nir_cf_node_if
)
2049 return nir_cf_node_as_if(next_node
);
2053 index_block(nir_block
*block
, void *state
)
2055 unsigned *index
= state
;
2056 block
->index
= (*index
)++;
2061 nir_index_blocks(nir_function_impl
*impl
)
2065 if (impl
->valid_metadata
& nir_metadata_block_index
)
2068 nir_foreach_block(impl
, index_block
, &index
);
2070 impl
->num_blocks
= index
;
2074 index_ssa_def_cb(nir_ssa_def
*def
, void *state
)
2076 unsigned *index
= (unsigned *) state
;
2077 def
->index
= (*index
)++;
2083 index_ssa_block(nir_block
*block
, void *state
)
2085 nir_foreach_instr(block
, instr
)
2086 nir_foreach_ssa_def(instr
, index_ssa_def_cb
, state
);
2092 nir_index_ssa_defs(nir_function_impl
*impl
)
2095 nir_foreach_block(impl
, index_ssa_block
, &index
);
2096 impl
->ssa_alloc
= index
;