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)
29 #include "nir_control_flow_private.h"
33 nir_shader_create(void *mem_ctx
,
34 gl_shader_stage stage
,
35 const nir_shader_compiler_options
*options
)
37 nir_shader
*shader
= ralloc(mem_ctx
, nir_shader
);
39 exec_list_make_empty(&shader
->uniforms
);
40 exec_list_make_empty(&shader
->inputs
);
41 exec_list_make_empty(&shader
->outputs
);
43 shader
->options
= options
;
44 memset(&shader
->info
, 0, sizeof(shader
->info
));
46 exec_list_make_empty(&shader
->functions
);
47 exec_list_make_empty(&shader
->registers
);
48 exec_list_make_empty(&shader
->globals
);
49 exec_list_make_empty(&shader
->system_values
);
50 shader
->reg_alloc
= 0;
52 shader
->num_inputs
= 0;
53 shader
->num_outputs
= 0;
54 shader
->num_uniforms
= 0;
56 shader
->stage
= stage
;
62 reg_create(void *mem_ctx
, struct exec_list
*list
)
64 nir_register
*reg
= ralloc(mem_ctx
, nir_register
);
66 list_inithead(®
->uses
);
67 list_inithead(®
->defs
);
68 list_inithead(®
->if_uses
);
70 reg
->num_components
= 0;
71 reg
->num_array_elems
= 0;
72 reg
->is_packed
= false;
75 exec_list_push_tail(list
, ®
->node
);
81 nir_global_reg_create(nir_shader
*shader
)
83 nir_register
*reg
= reg_create(shader
, &shader
->registers
);
84 reg
->index
= shader
->reg_alloc
++;
85 reg
->is_global
= true;
91 nir_local_reg_create(nir_function_impl
*impl
)
93 nir_register
*reg
= reg_create(ralloc_parent(impl
), &impl
->registers
);
94 reg
->index
= impl
->reg_alloc
++;
95 reg
->is_global
= false;
101 nir_reg_remove(nir_register
*reg
)
103 exec_node_remove(®
->node
);
107 nir_function_create(nir_shader
*shader
, const char *name
)
109 nir_function
*func
= ralloc(shader
, nir_function
);
111 exec_list_push_tail(&shader
->functions
, &func
->node
);
112 exec_list_make_empty(&func
->overload_list
);
113 func
->name
= ralloc_strdup(func
, name
);
114 func
->shader
= shader
;
119 nir_function_overload
*
120 nir_function_overload_create(nir_function
*func
)
122 void *mem_ctx
= ralloc_parent(func
);
124 nir_function_overload
*overload
= ralloc(mem_ctx
, nir_function_overload
);
126 overload
->num_params
= 0;
127 overload
->params
= NULL
;
128 overload
->return_type
= glsl_void_type();
129 overload
->impl
= NULL
;
131 exec_list_push_tail(&func
->overload_list
, &overload
->node
);
132 overload
->function
= func
;
137 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *mem_ctx
)
139 dest
->is_ssa
= src
->is_ssa
;
141 dest
->ssa
= src
->ssa
;
143 dest
->reg
.base_offset
= src
->reg
.base_offset
;
144 dest
->reg
.reg
= src
->reg
.reg
;
145 if (src
->reg
.indirect
) {
146 dest
->reg
.indirect
= ralloc(mem_ctx
, nir_src
);
147 nir_src_copy(dest
->reg
.indirect
, src
->reg
.indirect
, mem_ctx
);
149 dest
->reg
.indirect
= NULL
;
154 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
)
156 /* Copying an SSA definition makes no sense whatsoever. */
157 assert(!src
->is_ssa
);
159 dest
->is_ssa
= false;
161 dest
->reg
.base_offset
= src
->reg
.base_offset
;
162 dest
->reg
.reg
= src
->reg
.reg
;
163 if (src
->reg
.indirect
) {
164 dest
->reg
.indirect
= ralloc(instr
, nir_src
);
165 nir_src_copy(dest
->reg
.indirect
, src
->reg
.indirect
, instr
);
167 dest
->reg
.indirect
= NULL
;
172 nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
173 nir_alu_instr
*instr
)
175 nir_src_copy(&dest
->src
, &src
->src
, &instr
->instr
);
176 dest
->abs
= src
->abs
;
177 dest
->negate
= src
->negate
;
178 for (unsigned i
= 0; i
< 4; i
++)
179 dest
->swizzle
[i
] = src
->swizzle
[i
];
183 nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
184 nir_alu_instr
*instr
)
186 nir_dest_copy(&dest
->dest
, &src
->dest
, &instr
->instr
);
187 dest
->write_mask
= src
->write_mask
;
188 dest
->saturate
= src
->saturate
;
193 cf_init(nir_cf_node
*node
, nir_cf_node_type type
)
195 exec_node_init(&node
->node
);
201 nir_function_impl_create(nir_function_overload
*overload
)
203 assert(overload
->impl
== NULL
);
205 void *mem_ctx
= ralloc_parent(overload
);
207 nir_function_impl
*impl
= ralloc(mem_ctx
, nir_function_impl
);
209 overload
->impl
= impl
;
210 impl
->overload
= overload
;
212 cf_init(&impl
->cf_node
, nir_cf_node_function
);
214 exec_list_make_empty(&impl
->body
);
215 exec_list_make_empty(&impl
->registers
);
216 exec_list_make_empty(&impl
->locals
);
217 impl
->num_params
= 0;
219 impl
->return_var
= NULL
;
222 impl
->valid_metadata
= nir_metadata_none
;
224 /* create start & end blocks */
225 nir_block
*start_block
= nir_block_create(mem_ctx
);
226 nir_block
*end_block
= nir_block_create(mem_ctx
);
227 start_block
->cf_node
.parent
= &impl
->cf_node
;
228 end_block
->cf_node
.parent
= &impl
->cf_node
;
229 impl
->end_block
= end_block
;
231 exec_list_push_tail(&impl
->body
, &start_block
->cf_node
.node
);
233 start_block
->successors
[0] = end_block
;
234 _mesa_set_add(end_block
->predecessors
, start_block
);
239 nir_block_create(void *mem_ctx
)
241 nir_block
*block
= ralloc(mem_ctx
, nir_block
);
243 cf_init(&block
->cf_node
, nir_cf_node_block
);
245 block
->successors
[0] = block
->successors
[1] = NULL
;
246 block
->predecessors
= _mesa_set_create(block
, _mesa_hash_pointer
,
247 _mesa_key_pointer_equal
);
248 block
->imm_dom
= NULL
;
249 block
->dom_frontier
= _mesa_set_create(block
, _mesa_hash_pointer
,
250 _mesa_key_pointer_equal
);
252 exec_list_make_empty(&block
->instr_list
);
258 src_init(nir_src
*src
)
262 src
->reg
.indirect
= NULL
;
263 src
->reg
.base_offset
= 0;
267 nir_if_create(void *mem_ctx
)
269 nir_if
*if_stmt
= ralloc(mem_ctx
, nir_if
);
271 cf_init(&if_stmt
->cf_node
, nir_cf_node_if
);
272 src_init(&if_stmt
->condition
);
274 nir_block
*then
= nir_block_create(mem_ctx
);
275 exec_list_make_empty(&if_stmt
->then_list
);
276 exec_list_push_tail(&if_stmt
->then_list
, &then
->cf_node
.node
);
277 then
->cf_node
.parent
= &if_stmt
->cf_node
;
279 nir_block
*else_stmt
= nir_block_create(mem_ctx
);
280 exec_list_make_empty(&if_stmt
->else_list
);
281 exec_list_push_tail(&if_stmt
->else_list
, &else_stmt
->cf_node
.node
);
282 else_stmt
->cf_node
.parent
= &if_stmt
->cf_node
;
288 nir_loop_create(void *mem_ctx
)
290 nir_loop
*loop
= ralloc(mem_ctx
, nir_loop
);
292 cf_init(&loop
->cf_node
, nir_cf_node_loop
);
294 nir_block
*body
= nir_block_create(mem_ctx
);
295 exec_list_make_empty(&loop
->body
);
296 exec_list_push_tail(&loop
->body
, &body
->cf_node
.node
);
297 body
->cf_node
.parent
= &loop
->cf_node
;
299 body
->successors
[0] = body
;
300 _mesa_set_add(body
->predecessors
, body
);
306 instr_init(nir_instr
*instr
, nir_instr_type type
)
310 exec_node_init(&instr
->node
);
314 dest_init(nir_dest
*dest
)
316 dest
->is_ssa
= false;
317 dest
->reg
.reg
= NULL
;
318 dest
->reg
.indirect
= NULL
;
319 dest
->reg
.base_offset
= 0;
323 alu_dest_init(nir_alu_dest
*dest
)
325 dest_init(&dest
->dest
);
326 dest
->saturate
= false;
327 dest
->write_mask
= 0xf;
331 alu_src_init(nir_alu_src
*src
)
334 src
->abs
= src
->negate
= false;
342 nir_alu_instr_create(nir_shader
*shader
, nir_op op
)
344 unsigned num_srcs
= nir_op_infos
[op
].num_inputs
;
345 nir_alu_instr
*instr
=
347 sizeof(nir_alu_instr
) + num_srcs
* sizeof(nir_alu_src
));
349 instr_init(&instr
->instr
, nir_instr_type_alu
);
351 alu_dest_init(&instr
->dest
);
352 for (unsigned i
= 0; i
< num_srcs
; i
++)
353 alu_src_init(&instr
->src
[i
]);
359 nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
)
361 nir_jump_instr
*instr
= ralloc(shader
, nir_jump_instr
);
362 instr_init(&instr
->instr
, nir_instr_type_jump
);
367 nir_load_const_instr
*
368 nir_load_const_instr_create(nir_shader
*shader
, unsigned num_components
)
370 nir_load_const_instr
*instr
= ralloc(shader
, nir_load_const_instr
);
371 instr_init(&instr
->instr
, nir_instr_type_load_const
);
373 nir_ssa_def_init(&instr
->instr
, &instr
->def
, num_components
, NULL
);
378 nir_intrinsic_instr
*
379 nir_intrinsic_instr_create(nir_shader
*shader
, nir_intrinsic_op op
)
381 unsigned num_srcs
= nir_intrinsic_infos
[op
].num_srcs
;
382 nir_intrinsic_instr
*instr
=
384 sizeof(nir_intrinsic_instr
) + num_srcs
* sizeof(nir_src
));
386 instr_init(&instr
->instr
, nir_instr_type_intrinsic
);
387 instr
->intrinsic
= op
;
389 if (nir_intrinsic_infos
[op
].has_dest
)
390 dest_init(&instr
->dest
);
392 for (unsigned i
= 0; i
< num_srcs
; i
++)
393 src_init(&instr
->src
[i
]);
399 nir_call_instr_create(nir_shader
*shader
, nir_function_overload
*callee
)
401 nir_call_instr
*instr
= ralloc(shader
, nir_call_instr
);
402 instr_init(&instr
->instr
, nir_instr_type_call
);
404 instr
->callee
= callee
;
405 instr
->num_params
= callee
->num_params
;
406 instr
->params
= ralloc_array(instr
, nir_deref_var
*, instr
->num_params
);
407 instr
->return_deref
= NULL
;
413 nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
)
415 nir_tex_instr
*instr
= rzalloc(shader
, nir_tex_instr
);
416 instr_init(&instr
->instr
, nir_instr_type_tex
);
418 dest_init(&instr
->dest
);
420 instr
->num_srcs
= num_srcs
;
421 instr
->src
= ralloc_array(instr
, nir_tex_src
, num_srcs
);
422 for (unsigned i
= 0; i
< num_srcs
; i
++)
423 src_init(&instr
->src
[i
].src
);
425 instr
->sampler_index
= 0;
426 instr
->sampler_array_size
= 0;
427 instr
->sampler
= NULL
;
433 nir_phi_instr_create(nir_shader
*shader
)
435 nir_phi_instr
*instr
= ralloc(shader
, nir_phi_instr
);
436 instr_init(&instr
->instr
, nir_instr_type_phi
);
438 dest_init(&instr
->dest
);
439 exec_list_make_empty(&instr
->srcs
);
443 nir_parallel_copy_instr
*
444 nir_parallel_copy_instr_create(nir_shader
*shader
)
446 nir_parallel_copy_instr
*instr
= ralloc(shader
, nir_parallel_copy_instr
);
447 instr_init(&instr
->instr
, nir_instr_type_parallel_copy
);
449 exec_list_make_empty(&instr
->entries
);
454 nir_ssa_undef_instr
*
455 nir_ssa_undef_instr_create(nir_shader
*shader
, unsigned num_components
)
457 nir_ssa_undef_instr
*instr
= ralloc(shader
, nir_ssa_undef_instr
);
458 instr_init(&instr
->instr
, nir_instr_type_ssa_undef
);
460 nir_ssa_def_init(&instr
->instr
, &instr
->def
, num_components
, NULL
);
466 nir_deref_var_create(void *mem_ctx
, nir_variable
*var
)
468 nir_deref_var
*deref
= ralloc(mem_ctx
, nir_deref_var
);
469 deref
->deref
.deref_type
= nir_deref_type_var
;
470 deref
->deref
.child
= NULL
;
471 deref
->deref
.type
= var
->type
;
477 nir_deref_array_create(void *mem_ctx
)
479 nir_deref_array
*deref
= ralloc(mem_ctx
, nir_deref_array
);
480 deref
->deref
.deref_type
= nir_deref_type_array
;
481 deref
->deref
.child
= NULL
;
482 deref
->deref_array_type
= nir_deref_array_type_direct
;
483 src_init(&deref
->indirect
);
484 deref
->base_offset
= 0;
489 nir_deref_struct_create(void *mem_ctx
, unsigned field_index
)
491 nir_deref_struct
*deref
= ralloc(mem_ctx
, nir_deref_struct
);
492 deref
->deref
.deref_type
= nir_deref_type_struct
;
493 deref
->deref
.child
= NULL
;
494 deref
->index
= field_index
;
498 static nir_deref_var
*
499 copy_deref_var(void *mem_ctx
, nir_deref_var
*deref
)
501 nir_deref_var
*ret
= nir_deref_var_create(mem_ctx
, deref
->var
);
502 ret
->deref
.type
= deref
->deref
.type
;
503 if (deref
->deref
.child
)
504 ret
->deref
.child
= nir_copy_deref(ret
, deref
->deref
.child
);
508 static nir_deref_array
*
509 copy_deref_array(void *mem_ctx
, nir_deref_array
*deref
)
511 nir_deref_array
*ret
= nir_deref_array_create(mem_ctx
);
512 ret
->base_offset
= deref
->base_offset
;
513 ret
->deref_array_type
= deref
->deref_array_type
;
514 if (deref
->deref_array_type
== nir_deref_array_type_indirect
) {
515 nir_src_copy(&ret
->indirect
, &deref
->indirect
, mem_ctx
);
517 ret
->deref
.type
= deref
->deref
.type
;
518 if (deref
->deref
.child
)
519 ret
->deref
.child
= nir_copy_deref(ret
, deref
->deref
.child
);
523 static nir_deref_struct
*
524 copy_deref_struct(void *mem_ctx
, nir_deref_struct
*deref
)
526 nir_deref_struct
*ret
= nir_deref_struct_create(mem_ctx
, deref
->index
);
527 ret
->deref
.type
= deref
->deref
.type
;
528 if (deref
->deref
.child
)
529 ret
->deref
.child
= nir_copy_deref(ret
, deref
->deref
.child
);
534 nir_copy_deref(void *mem_ctx
, nir_deref
*deref
)
536 switch (deref
->deref_type
) {
537 case nir_deref_type_var
:
538 return ©_deref_var(mem_ctx
, nir_deref_as_var(deref
))->deref
;
539 case nir_deref_type_array
:
540 return ©_deref_array(mem_ctx
, nir_deref_as_array(deref
))->deref
;
541 case nir_deref_type_struct
:
542 return ©_deref_struct(mem_ctx
, nir_deref_as_struct(deref
))->deref
;
544 unreachable("Invalid dereference type");
550 /* Returns a load_const instruction that represents the constant
551 * initializer for the given deref chain. The caller is responsible for
552 * ensuring that there actually is a constant initializer.
554 nir_load_const_instr
*
555 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
)
557 nir_constant
*constant
= deref
->var
->constant_initializer
;
560 const nir_deref
*tail
= &deref
->deref
;
561 unsigned matrix_offset
= 0;
562 while (tail
->child
) {
563 switch (tail
->child
->deref_type
) {
564 case nir_deref_type_array
: {
565 nir_deref_array
*arr
= nir_deref_as_array(tail
->child
);
566 assert(arr
->deref_array_type
== nir_deref_array_type_direct
);
567 if (glsl_type_is_matrix(tail
->type
)) {
568 assert(arr
->deref
.child
== NULL
);
569 matrix_offset
= arr
->base_offset
;
571 constant
= constant
->elements
[arr
->base_offset
];
576 case nir_deref_type_struct
: {
577 constant
= constant
->elements
[nir_deref_as_struct(tail
->child
)->index
];
582 unreachable("Invalid deref child type");
588 nir_load_const_instr
*load
=
589 nir_load_const_instr_create(shader
, glsl_get_vector_elements(tail
->type
));
591 matrix_offset
*= load
->def
.num_components
;
592 for (unsigned i
= 0; i
< load
->def
.num_components
; i
++) {
593 switch (glsl_get_base_type(tail
->type
)) {
594 case GLSL_TYPE_FLOAT
:
597 load
->value
.u
[i
] = constant
->value
.u
[matrix_offset
+ i
];
600 load
->value
.u
[i
] = constant
->value
.b
[matrix_offset
+ i
] ?
601 NIR_TRUE
: NIR_FALSE
;
604 unreachable("Invalid immediate type");
612 nir_cf_node_get_function(nir_cf_node
*node
)
614 while (node
->type
!= nir_cf_node_function
) {
618 return nir_cf_node_as_function(node
);
622 add_use_cb(nir_src
*src
, void *state
)
624 nir_instr
*instr
= state
;
626 src
->parent_instr
= instr
;
627 list_addtail(&src
->use_link
,
628 src
->is_ssa
? &src
->ssa
->uses
: &src
->reg
.reg
->uses
);
634 add_ssa_def_cb(nir_ssa_def
*def
, void *state
)
636 nir_instr
*instr
= state
;
638 if (instr
->block
&& def
->index
== UINT_MAX
) {
639 nir_function_impl
*impl
=
640 nir_cf_node_get_function(&instr
->block
->cf_node
);
642 def
->index
= impl
->ssa_alloc
++;
649 add_reg_def_cb(nir_dest
*dest
, void *state
)
651 nir_instr
*instr
= state
;
654 dest
->reg
.parent_instr
= instr
;
655 list_addtail(&dest
->reg
.def_link
, &dest
->reg
.reg
->defs
);
662 add_defs_uses(nir_instr
*instr
)
664 nir_foreach_src(instr
, add_use_cb
, instr
);
665 nir_foreach_dest(instr
, add_reg_def_cb
, instr
);
666 nir_foreach_ssa_def(instr
, add_ssa_def_cb
, instr
);
670 nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
)
672 switch (cursor
.option
) {
673 case nir_cursor_before_block
:
674 /* Only allow inserting jumps into empty blocks. */
675 if (instr
->type
== nir_instr_type_jump
)
676 assert(exec_list_is_empty(&cursor
.block
->instr_list
));
678 instr
->block
= cursor
.block
;
679 add_defs_uses(instr
);
680 exec_list_push_head(&cursor
.block
->instr_list
, &instr
->node
);
682 case nir_cursor_after_block
: {
683 /* Inserting instructions after a jump is illegal. */
684 nir_instr
*last
= nir_block_last_instr(cursor
.block
);
685 assert(last
== NULL
|| last
->type
!= nir_instr_type_jump
);
688 instr
->block
= cursor
.block
;
689 add_defs_uses(instr
);
690 exec_list_push_tail(&cursor
.block
->instr_list
, &instr
->node
);
693 case nir_cursor_before_instr
:
694 assert(instr
->type
!= nir_instr_type_jump
);
695 instr
->block
= cursor
.instr
->block
;
696 add_defs_uses(instr
);
697 exec_node_insert_node_before(&cursor
.instr
->node
, &instr
->node
);
699 case nir_cursor_after_instr
:
700 /* Inserting instructions after a jump is illegal. */
701 assert(cursor
.instr
->type
!= nir_instr_type_jump
);
703 /* Only allow inserting jumps at the end of the block. */
704 if (instr
->type
== nir_instr_type_jump
)
705 assert(cursor
.instr
== nir_block_last_instr(cursor
.instr
->block
));
707 instr
->block
= cursor
.instr
->block
;
708 add_defs_uses(instr
);
709 exec_node_insert_after(&cursor
.instr
->node
, &instr
->node
);
713 if (instr
->type
== nir_instr_type_jump
)
714 nir_handle_add_jump(instr
->block
);
718 src_is_valid(const nir_src
*src
)
720 return src
->is_ssa
? (src
->ssa
!= NULL
) : (src
->reg
.reg
!= NULL
);
724 remove_use_cb(nir_src
*src
, void *state
)
726 if (src_is_valid(src
))
727 list_del(&src
->use_link
);
733 remove_def_cb(nir_dest
*dest
, void *state
)
736 list_del(&dest
->reg
.def_link
);
742 remove_defs_uses(nir_instr
*instr
)
744 nir_foreach_dest(instr
, remove_def_cb
, instr
);
745 nir_foreach_src(instr
, remove_use_cb
, instr
);
748 void nir_instr_remove(nir_instr
*instr
)
750 remove_defs_uses(instr
);
751 exec_node_remove(&instr
->node
);
753 if (instr
->type
== nir_instr_type_jump
) {
754 nir_jump_instr
*jump_instr
= nir_instr_as_jump(instr
);
755 nir_handle_remove_jump(instr
->block
, jump_instr
->type
);
762 nir_index_local_regs(nir_function_impl
*impl
)
765 foreach_list_typed(nir_register
, reg
, node
, &impl
->registers
) {
766 reg
->index
= index
++;
768 impl
->reg_alloc
= index
;
772 nir_index_global_regs(nir_shader
*shader
)
775 foreach_list_typed(nir_register
, reg
, node
, &shader
->registers
) {
776 reg
->index
= index
++;
778 shader
->reg_alloc
= index
;
782 visit_alu_dest(nir_alu_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
784 return cb(&instr
->dest
.dest
, state
);
788 visit_intrinsic_dest(nir_intrinsic_instr
*instr
, nir_foreach_dest_cb cb
,
791 if (nir_intrinsic_infos
[instr
->intrinsic
].has_dest
)
792 return cb(&instr
->dest
, state
);
798 visit_texture_dest(nir_tex_instr
*instr
, nir_foreach_dest_cb cb
,
801 return cb(&instr
->dest
, state
);
805 visit_phi_dest(nir_phi_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
807 return cb(&instr
->dest
, state
);
811 visit_parallel_copy_dest(nir_parallel_copy_instr
*instr
,
812 nir_foreach_dest_cb cb
, void *state
)
814 nir_foreach_parallel_copy_entry(instr
, entry
) {
815 if (!cb(&entry
->dest
, state
))
823 nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
825 switch (instr
->type
) {
826 case nir_instr_type_alu
:
827 return visit_alu_dest(nir_instr_as_alu(instr
), cb
, state
);
828 case nir_instr_type_intrinsic
:
829 return visit_intrinsic_dest(nir_instr_as_intrinsic(instr
), cb
, state
);
830 case nir_instr_type_tex
:
831 return visit_texture_dest(nir_instr_as_tex(instr
), cb
, state
);
832 case nir_instr_type_phi
:
833 return visit_phi_dest(nir_instr_as_phi(instr
), cb
, state
);
834 case nir_instr_type_parallel_copy
:
835 return visit_parallel_copy_dest(nir_instr_as_parallel_copy(instr
),
838 case nir_instr_type_load_const
:
839 case nir_instr_type_ssa_undef
:
840 case nir_instr_type_call
:
841 case nir_instr_type_jump
:
845 unreachable("Invalid instruction type");
852 struct foreach_ssa_def_state
{
853 nir_foreach_ssa_def_cb cb
;
858 nir_ssa_def_visitor(nir_dest
*dest
, void *void_state
)
860 struct foreach_ssa_def_state
*state
= void_state
;
863 return state
->cb(&dest
->ssa
, state
->client_state
);
869 nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
, void *state
)
871 switch (instr
->type
) {
872 case nir_instr_type_alu
:
873 case nir_instr_type_tex
:
874 case nir_instr_type_intrinsic
:
875 case nir_instr_type_phi
:
876 case nir_instr_type_parallel_copy
: {
877 struct foreach_ssa_def_state foreach_state
= {cb
, state
};
878 return nir_foreach_dest(instr
, nir_ssa_def_visitor
, &foreach_state
);
881 case nir_instr_type_load_const
:
882 return cb(&nir_instr_as_load_const(instr
)->def
, state
);
883 case nir_instr_type_ssa_undef
:
884 return cb(&nir_instr_as_ssa_undef(instr
)->def
, state
);
885 case nir_instr_type_call
:
886 case nir_instr_type_jump
:
889 unreachable("Invalid instruction type");
894 visit_src(nir_src
*src
, nir_foreach_src_cb cb
, void *state
)
898 if (!src
->is_ssa
&& src
->reg
.indirect
)
899 return cb(src
->reg
.indirect
, state
);
904 visit_deref_array_src(nir_deref_array
*deref
, nir_foreach_src_cb cb
,
907 if (deref
->deref_array_type
== nir_deref_array_type_indirect
)
908 return visit_src(&deref
->indirect
, cb
, state
);
913 visit_deref_src(nir_deref_var
*deref
, nir_foreach_src_cb cb
, void *state
)
915 nir_deref
*cur
= &deref
->deref
;
916 while (cur
!= NULL
) {
917 if (cur
->deref_type
== nir_deref_type_array
)
918 if (!visit_deref_array_src(nir_deref_as_array(cur
), cb
, state
))
928 visit_alu_src(nir_alu_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
930 for (unsigned i
= 0; i
< nir_op_infos
[instr
->op
].num_inputs
; i
++)
931 if (!visit_src(&instr
->src
[i
].src
, cb
, state
))
938 visit_tex_src(nir_tex_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
940 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++)
941 if (!visit_src(&instr
->src
[i
].src
, cb
, state
))
944 if (instr
->sampler
!= NULL
)
945 if (!visit_deref_src(instr
->sampler
, cb
, state
))
952 visit_intrinsic_src(nir_intrinsic_instr
*instr
, nir_foreach_src_cb cb
,
955 unsigned num_srcs
= nir_intrinsic_infos
[instr
->intrinsic
].num_srcs
;
956 for (unsigned i
= 0; i
< num_srcs
; i
++)
957 if (!visit_src(&instr
->src
[i
], cb
, state
))
961 nir_intrinsic_infos
[instr
->intrinsic
].num_variables
;
962 for (unsigned i
= 0; i
< num_vars
; i
++)
963 if (!visit_deref_src(instr
->variables
[i
], cb
, state
))
970 visit_call_src(nir_call_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
976 visit_load_const_src(nir_load_const_instr
*instr
, nir_foreach_src_cb cb
,
983 visit_phi_src(nir_phi_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
985 nir_foreach_phi_src(instr
, src
) {
986 if (!visit_src(&src
->src
, cb
, state
))
994 visit_parallel_copy_src(nir_parallel_copy_instr
*instr
,
995 nir_foreach_src_cb cb
, void *state
)
997 nir_foreach_parallel_copy_entry(instr
, entry
) {
998 if (!visit_src(&entry
->src
, cb
, state
))
1007 nir_foreach_src_cb cb
;
1008 } visit_dest_indirect_state
;
1011 visit_dest_indirect(nir_dest
*dest
, void *_state
)
1013 visit_dest_indirect_state
*state
= (visit_dest_indirect_state
*) _state
;
1015 if (!dest
->is_ssa
&& dest
->reg
.indirect
)
1016 return state
->cb(dest
->reg
.indirect
, state
->state
);
1022 nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1024 switch (instr
->type
) {
1025 case nir_instr_type_alu
:
1026 if (!visit_alu_src(nir_instr_as_alu(instr
), cb
, state
))
1029 case nir_instr_type_intrinsic
:
1030 if (!visit_intrinsic_src(nir_instr_as_intrinsic(instr
), cb
, state
))
1033 case nir_instr_type_tex
:
1034 if (!visit_tex_src(nir_instr_as_tex(instr
), cb
, state
))
1037 case nir_instr_type_call
:
1038 if (!visit_call_src(nir_instr_as_call(instr
), cb
, state
))
1041 case nir_instr_type_load_const
:
1042 if (!visit_load_const_src(nir_instr_as_load_const(instr
), cb
, state
))
1045 case nir_instr_type_phi
:
1046 if (!visit_phi_src(nir_instr_as_phi(instr
), cb
, state
))
1049 case nir_instr_type_parallel_copy
:
1050 if (!visit_parallel_copy_src(nir_instr_as_parallel_copy(instr
),
1054 case nir_instr_type_jump
:
1055 case nir_instr_type_ssa_undef
:
1059 unreachable("Invalid instruction type");
1063 visit_dest_indirect_state dest_state
;
1064 dest_state
.state
= state
;
1066 return nir_foreach_dest(instr
, visit_dest_indirect
, &dest_state
);
1070 nir_src_as_const_value(nir_src src
)
1075 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_load_const
)
1078 nir_load_const_instr
*load
= nir_instr_as_load_const(src
.ssa
->parent_instr
);
1080 return &load
->value
;
1084 * Returns true if the source is known to be dynamically uniform. Otherwise it
1085 * returns false which means it may or may not be dynamically uniform but it
1086 * can't be determined.
1089 nir_src_is_dynamically_uniform(nir_src src
)
1094 /* Constants are trivially dynamically uniform */
1095 if (src
.ssa
->parent_instr
->type
== nir_instr_type_load_const
)
1098 /* As are uniform variables */
1099 if (src
.ssa
->parent_instr
->type
== nir_instr_type_intrinsic
) {
1100 nir_intrinsic_instr
*intr
= nir_instr_as_intrinsic(src
.ssa
->parent_instr
);
1102 if (intr
->intrinsic
== nir_intrinsic_load_uniform
)
1106 /* XXX: this could have many more tests, such as when a sampler function is
1107 * called with dynamically uniform arguments.
1113 src_remove_all_uses(nir_src
*src
)
1115 for (; src
; src
= src
->is_ssa
? NULL
: src
->reg
.indirect
) {
1116 if (!src_is_valid(src
))
1119 list_del(&src
->use_link
);
1124 src_add_all_uses(nir_src
*src
, nir_instr
*parent_instr
, nir_if
*parent_if
)
1126 for (; src
; src
= src
->is_ssa
? NULL
: src
->reg
.indirect
) {
1127 if (!src_is_valid(src
))
1131 src
->parent_instr
= parent_instr
;
1133 list_addtail(&src
->use_link
, &src
->ssa
->uses
);
1135 list_addtail(&src
->use_link
, &src
->reg
.reg
->uses
);
1138 src
->parent_if
= parent_if
;
1140 list_addtail(&src
->use_link
, &src
->ssa
->if_uses
);
1142 list_addtail(&src
->use_link
, &src
->reg
.reg
->if_uses
);
1148 nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
)
1150 assert(!src_is_valid(src
) || src
->parent_instr
== instr
);
1152 src_remove_all_uses(src
);
1154 src_add_all_uses(src
, instr
, NULL
);
1158 nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
)
1160 assert(!src_is_valid(dest
) || dest
->parent_instr
== dest_instr
);
1162 src_remove_all_uses(dest
);
1163 src_remove_all_uses(src
);
1165 *src
= NIR_SRC_INIT
;
1166 src_add_all_uses(dest
, dest_instr
, NULL
);
1170 nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
)
1172 nir_src
*src
= &if_stmt
->condition
;
1173 assert(!src_is_valid(src
) || src
->parent_if
== if_stmt
);
1175 src_remove_all_uses(src
);
1177 src_add_all_uses(src
, NULL
, if_stmt
);
1181 nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
, nir_dest new_dest
)
1184 /* We can only overwrite an SSA destination if it has no uses. */
1185 assert(list_empty(&dest
->ssa
.uses
) && list_empty(&dest
->ssa
.if_uses
));
1187 list_del(&dest
->reg
.def_link
);
1188 if (dest
->reg
.indirect
)
1189 src_remove_all_uses(dest
->reg
.indirect
);
1192 /* We can't re-write with an SSA def */
1193 assert(!new_dest
.is_ssa
);
1195 nir_dest_copy(dest
, &new_dest
, instr
);
1197 dest
->reg
.parent_instr
= instr
;
1198 list_addtail(&dest
->reg
.def_link
, &new_dest
.reg
.reg
->defs
);
1200 if (dest
->reg
.indirect
)
1201 src_add_all_uses(dest
->reg
.indirect
, instr
, NULL
);
1205 nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1206 unsigned num_components
, const char *name
)
1209 def
->parent_instr
= instr
;
1210 list_inithead(&def
->uses
);
1211 list_inithead(&def
->if_uses
);
1212 def
->num_components
= num_components
;
1215 nir_function_impl
*impl
=
1216 nir_cf_node_get_function(&instr
->block
->cf_node
);
1218 def
->index
= impl
->ssa_alloc
++;
1220 def
->index
= UINT_MAX
;
1225 nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1226 unsigned num_components
, const char *name
)
1228 dest
->is_ssa
= true;
1229 nir_ssa_def_init(instr
, &dest
->ssa
, num_components
, name
);
1233 nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
)
1235 assert(!new_src
.is_ssa
|| def
!= new_src
.ssa
);
1237 nir_foreach_use_safe(def
, use_src
) {
1238 nir_instr
*src_parent_instr
= use_src
->parent_instr
;
1239 list_del(&use_src
->use_link
);
1240 nir_src_copy(use_src
, &new_src
, src_parent_instr
);
1241 src_add_all_uses(use_src
, src_parent_instr
, NULL
);
1244 nir_foreach_if_use_safe(def
, use_src
) {
1245 nir_if
*src_parent_if
= use_src
->parent_if
;
1246 list_del(&use_src
->use_link
);
1247 nir_src_copy(use_src
, &new_src
, src_parent_if
);
1248 src_add_all_uses(use_src
, NULL
, src_parent_if
);
1253 static bool foreach_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1254 bool reverse
, void *state
);
1257 foreach_if(nir_if
*if_stmt
, nir_foreach_block_cb cb
, bool reverse
, void *state
)
1260 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
,
1261 &if_stmt
->else_list
) {
1262 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1266 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
,
1267 &if_stmt
->then_list
) {
1268 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1272 foreach_list_typed_safe(nir_cf_node
, node
, node
, &if_stmt
->then_list
) {
1273 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1277 foreach_list_typed_safe(nir_cf_node
, node
, node
, &if_stmt
->else_list
) {
1278 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1287 foreach_loop(nir_loop
*loop
, nir_foreach_block_cb cb
, bool reverse
, void *state
)
1290 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
, &loop
->body
) {
1291 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1295 foreach_list_typed_safe(nir_cf_node
, node
, node
, &loop
->body
) {
1296 if (!foreach_cf_node(node
, cb
, reverse
, state
))
1305 foreach_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1306 bool reverse
, void *state
)
1308 switch (node
->type
) {
1309 case nir_cf_node_block
:
1310 return cb(nir_cf_node_as_block(node
), state
);
1311 case nir_cf_node_if
:
1312 return foreach_if(nir_cf_node_as_if(node
), cb
, reverse
, state
);
1313 case nir_cf_node_loop
:
1314 return foreach_loop(nir_cf_node_as_loop(node
), cb
, reverse
, state
);
1318 unreachable("Invalid CFG node type");
1326 nir_foreach_block_in_cf_node(nir_cf_node
*node
, nir_foreach_block_cb cb
,
1329 return foreach_cf_node(node
, cb
, false, state
);
1333 nir_foreach_block(nir_function_impl
*impl
, nir_foreach_block_cb cb
, void *state
)
1335 foreach_list_typed_safe(nir_cf_node
, node
, node
, &impl
->body
) {
1336 if (!foreach_cf_node(node
, cb
, false, state
))
1340 return cb(impl
->end_block
, state
);
1344 nir_foreach_block_reverse(nir_function_impl
*impl
, nir_foreach_block_cb cb
,
1347 if (!cb(impl
->end_block
, state
))
1350 foreach_list_typed_safe_reverse(nir_cf_node
, node
, node
, &impl
->body
) {
1351 if (!foreach_cf_node(node
, cb
, true, state
))
1359 nir_block_get_following_if(nir_block
*block
)
1361 if (exec_node_is_tail_sentinel(&block
->cf_node
.node
))
1364 if (nir_cf_node_is_last(&block
->cf_node
))
1367 nir_cf_node
*next_node
= nir_cf_node_next(&block
->cf_node
);
1369 if (next_node
->type
!= nir_cf_node_if
)
1372 return nir_cf_node_as_if(next_node
);
1376 nir_block_get_following_loop(nir_block
*block
)
1378 if (exec_node_is_tail_sentinel(&block
->cf_node
.node
))
1381 if (nir_cf_node_is_last(&block
->cf_node
))
1384 nir_cf_node
*next_node
= nir_cf_node_next(&block
->cf_node
);
1386 if (next_node
->type
!= nir_cf_node_loop
)
1389 return nir_cf_node_as_loop(next_node
);
1392 index_block(nir_block
*block
, void *state
)
1394 unsigned *index
= state
;
1395 block
->index
= (*index
)++;
1400 nir_index_blocks(nir_function_impl
*impl
)
1404 if (impl
->valid_metadata
& nir_metadata_block_index
)
1407 nir_foreach_block(impl
, index_block
, &index
);
1409 impl
->num_blocks
= index
;
1413 index_ssa_def_cb(nir_ssa_def
*def
, void *state
)
1415 unsigned *index
= (unsigned *) state
;
1416 def
->index
= (*index
)++;
1422 index_ssa_block(nir_block
*block
, void *state
)
1424 nir_foreach_instr(block
, instr
)
1425 nir_foreach_ssa_def(instr
, index_ssa_def_cb
, state
);
1431 * The indices are applied top-to-bottom which has the very nice property
1432 * that, if A dominates B, then A->index <= B->index.
1435 nir_index_ssa_defs(nir_function_impl
*impl
)
1438 nir_foreach_block(impl
, index_ssa_block
, &index
);
1439 impl
->ssa_alloc
= index
;
1443 index_instrs_block(nir_block
*block
, void *state
)
1445 unsigned *index
= state
;
1446 nir_foreach_instr(block
, instr
)
1447 instr
->index
= (*index
)++;
1453 * The indices are applied top-to-bottom which has the very nice property
1454 * that, if A dominates B, then A->index <= B->index.
1457 nir_index_instrs(nir_function_impl
*impl
)
1460 nir_foreach_block(impl
, index_instrs_block
, &index
);
1465 nir_intrinsic_from_system_value(gl_system_value val
)
1468 case SYSTEM_VALUE_VERTEX_ID
:
1469 return nir_intrinsic_load_vertex_id
;
1470 case SYSTEM_VALUE_INSTANCE_ID
:
1471 return nir_intrinsic_load_instance_id
;
1472 case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
:
1473 return nir_intrinsic_load_vertex_id_zero_base
;
1474 case SYSTEM_VALUE_BASE_VERTEX
:
1475 return nir_intrinsic_load_base_vertex
;
1476 case SYSTEM_VALUE_INVOCATION_ID
:
1477 return nir_intrinsic_load_invocation_id
;
1478 case SYSTEM_VALUE_FRONT_FACE
:
1479 return nir_intrinsic_load_front_face
;
1480 case SYSTEM_VALUE_SAMPLE_ID
:
1481 return nir_intrinsic_load_sample_id
;
1482 case SYSTEM_VALUE_SAMPLE_POS
:
1483 return nir_intrinsic_load_sample_pos
;
1484 case SYSTEM_VALUE_SAMPLE_MASK_IN
:
1485 return nir_intrinsic_load_sample_mask_in
;
1486 case SYSTEM_VALUE_LOCAL_INVOCATION_ID
:
1487 return nir_intrinsic_load_local_invocation_id
;
1488 case SYSTEM_VALUE_WORK_GROUP_ID
:
1489 return nir_intrinsic_load_work_group_id
;
1490 case SYSTEM_VALUE_NUM_WORK_GROUPS
:
1491 return nir_intrinsic_load_num_work_groups
;
1492 case SYSTEM_VALUE_PRIMITIVE_ID
:
1493 return nir_intrinsic_load_primitive_id
;
1494 /* FINISHME: Add tessellation intrinsics.
1495 case SYSTEM_VALUE_TESS_COORD:
1496 case SYSTEM_VALUE_VERTICES_IN:
1497 case SYSTEM_VALUE_TESS_LEVEL_OUTER:
1498 case SYSTEM_VALUE_TESS_LEVEL_INNER:
1501 unreachable("system value does not directly correspond to intrinsic");
1506 nir_system_value_from_intrinsic(nir_intrinsic_op intrin
)
1509 case nir_intrinsic_load_vertex_id
:
1510 return SYSTEM_VALUE_VERTEX_ID
;
1511 case nir_intrinsic_load_instance_id
:
1512 return SYSTEM_VALUE_INSTANCE_ID
;
1513 case nir_intrinsic_load_vertex_id_zero_base
:
1514 return SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
;
1515 case nir_intrinsic_load_base_vertex
:
1516 return SYSTEM_VALUE_BASE_VERTEX
;
1517 case nir_intrinsic_load_invocation_id
:
1518 return SYSTEM_VALUE_INVOCATION_ID
;
1519 case nir_intrinsic_load_front_face
:
1520 return SYSTEM_VALUE_FRONT_FACE
;
1521 case nir_intrinsic_load_sample_id
:
1522 return SYSTEM_VALUE_SAMPLE_ID
;
1523 case nir_intrinsic_load_sample_pos
:
1524 return SYSTEM_VALUE_SAMPLE_POS
;
1525 case nir_intrinsic_load_sample_mask_in
:
1526 return SYSTEM_VALUE_SAMPLE_MASK_IN
;
1527 case nir_intrinsic_load_local_invocation_id
:
1528 return SYSTEM_VALUE_LOCAL_INVOCATION_ID
;
1529 case nir_intrinsic_load_num_work_groups
:
1530 return SYSTEM_VALUE_NUM_WORK_GROUPS
;
1531 case nir_intrinsic_load_work_group_id
:
1532 return SYSTEM_VALUE_WORK_GROUP_ID
;
1533 case nir_intrinsic_load_primitive_id
:
1534 return SYSTEM_VALUE_PRIMITIVE_ID
;
1535 /* FINISHME: Add tessellation intrinsics.
1536 return SYSTEM_VALUE_TESS_COORD;
1537 return SYSTEM_VALUE_VERTICES_IN;
1538 return SYSTEM_VALUE_PRIMITIVE_ID;
1539 return SYSTEM_VALUE_TESS_LEVEL_OUTER;
1540 return SYSTEM_VALUE_TESS_LEVEL_INNER;
1543 unreachable("intrinsic doesn't produce a system value");