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
,
38 nir_shader
*shader
= rzalloc(mem_ctx
, nir_shader
);
40 exec_list_make_empty(&shader
->uniforms
);
41 exec_list_make_empty(&shader
->inputs
);
42 exec_list_make_empty(&shader
->outputs
);
43 exec_list_make_empty(&shader
->shared
);
45 shader
->options
= options
;
50 exec_list_make_empty(&shader
->functions
);
51 exec_list_make_empty(&shader
->registers
);
52 exec_list_make_empty(&shader
->globals
);
53 exec_list_make_empty(&shader
->system_values
);
54 shader
->reg_alloc
= 0;
56 shader
->num_inputs
= 0;
57 shader
->num_outputs
= 0;
58 shader
->num_uniforms
= 0;
59 shader
->num_shared
= 0;
61 shader
->stage
= stage
;
67 reg_create(void *mem_ctx
, struct exec_list
*list
)
69 nir_register
*reg
= ralloc(mem_ctx
, nir_register
);
71 list_inithead(®
->uses
);
72 list_inithead(®
->defs
);
73 list_inithead(®
->if_uses
);
75 reg
->num_components
= 0;
77 reg
->num_array_elems
= 0;
78 reg
->is_packed
= false;
81 exec_list_push_tail(list
, ®
->node
);
87 nir_global_reg_create(nir_shader
*shader
)
89 nir_register
*reg
= reg_create(shader
, &shader
->registers
);
90 reg
->index
= shader
->reg_alloc
++;
91 reg
->is_global
= true;
97 nir_local_reg_create(nir_function_impl
*impl
)
99 nir_register
*reg
= reg_create(ralloc_parent(impl
), &impl
->registers
);
100 reg
->index
= impl
->reg_alloc
++;
101 reg
->is_global
= false;
107 nir_reg_remove(nir_register
*reg
)
109 exec_node_remove(®
->node
);
113 nir_shader_add_variable(nir_shader
*shader
, nir_variable
*var
)
115 switch (var
->data
.mode
) {
117 assert(!"invalid mode");
121 assert(!"nir_shader_add_variable cannot be used for local variables");
125 assert(!"nir_shader_add_variable cannot be used for function parameters");
129 exec_list_push_tail(&shader
->globals
, &var
->node
);
132 case nir_var_shader_in
:
133 exec_list_push_tail(&shader
->inputs
, &var
->node
);
136 case nir_var_shader_out
:
137 exec_list_push_tail(&shader
->outputs
, &var
->node
);
140 case nir_var_uniform
:
141 case nir_var_shader_storage
:
142 exec_list_push_tail(&shader
->uniforms
, &var
->node
);
146 assert(shader
->stage
== MESA_SHADER_COMPUTE
);
147 exec_list_push_tail(&shader
->shared
, &var
->node
);
150 case nir_var_system_value
:
151 exec_list_push_tail(&shader
->system_values
, &var
->node
);
157 nir_variable_create(nir_shader
*shader
, nir_variable_mode mode
,
158 const struct glsl_type
*type
, const char *name
)
160 nir_variable
*var
= rzalloc(shader
, nir_variable
);
161 var
->name
= ralloc_strdup(var
, name
);
163 var
->data
.mode
= mode
;
165 if ((mode
== nir_var_shader_in
&& shader
->stage
!= MESA_SHADER_VERTEX
) ||
166 (mode
== nir_var_shader_out
&& shader
->stage
!= MESA_SHADER_FRAGMENT
))
167 var
->data
.interpolation
= INTERP_MODE_SMOOTH
;
169 if (mode
== nir_var_shader_in
|| mode
== nir_var_uniform
)
170 var
->data
.read_only
= true;
172 nir_shader_add_variable(shader
, var
);
178 nir_local_variable_create(nir_function_impl
*impl
,
179 const struct glsl_type
*type
, const char *name
)
181 nir_variable
*var
= rzalloc(impl
->function
->shader
, nir_variable
);
182 var
->name
= ralloc_strdup(var
, name
);
184 var
->data
.mode
= nir_var_local
;
186 nir_function_impl_add_variable(impl
, var
);
192 nir_function_create(nir_shader
*shader
, const char *name
)
194 nir_function
*func
= ralloc(shader
, nir_function
);
196 exec_list_push_tail(&shader
->functions
, &func
->node
);
198 func
->name
= ralloc_strdup(func
, name
);
199 func
->shader
= shader
;
200 func
->num_params
= 0;
202 func
->return_type
= glsl_void_type();
208 void nir_src_copy(nir_src
*dest
, const nir_src
*src
, void *mem_ctx
)
210 dest
->is_ssa
= src
->is_ssa
;
212 dest
->ssa
= src
->ssa
;
214 dest
->reg
.base_offset
= src
->reg
.base_offset
;
215 dest
->reg
.reg
= src
->reg
.reg
;
216 if (src
->reg
.indirect
) {
217 dest
->reg
.indirect
= ralloc(mem_ctx
, nir_src
);
218 nir_src_copy(dest
->reg
.indirect
, src
->reg
.indirect
, mem_ctx
);
220 dest
->reg
.indirect
= NULL
;
225 void nir_dest_copy(nir_dest
*dest
, const nir_dest
*src
, nir_instr
*instr
)
227 /* Copying an SSA definition makes no sense whatsoever. */
228 assert(!src
->is_ssa
);
230 dest
->is_ssa
= false;
232 dest
->reg
.base_offset
= src
->reg
.base_offset
;
233 dest
->reg
.reg
= src
->reg
.reg
;
234 if (src
->reg
.indirect
) {
235 dest
->reg
.indirect
= ralloc(instr
, nir_src
);
236 nir_src_copy(dest
->reg
.indirect
, src
->reg
.indirect
, instr
);
238 dest
->reg
.indirect
= NULL
;
243 nir_alu_src_copy(nir_alu_src
*dest
, const nir_alu_src
*src
,
244 nir_alu_instr
*instr
)
246 nir_src_copy(&dest
->src
, &src
->src
, &instr
->instr
);
247 dest
->abs
= src
->abs
;
248 dest
->negate
= src
->negate
;
249 for (unsigned i
= 0; i
< 4; i
++)
250 dest
->swizzle
[i
] = src
->swizzle
[i
];
254 nir_alu_dest_copy(nir_alu_dest
*dest
, const nir_alu_dest
*src
,
255 nir_alu_instr
*instr
)
257 nir_dest_copy(&dest
->dest
, &src
->dest
, &instr
->instr
);
258 dest
->write_mask
= src
->write_mask
;
259 dest
->saturate
= src
->saturate
;
264 cf_init(nir_cf_node
*node
, nir_cf_node_type type
)
266 exec_node_init(&node
->node
);
272 nir_function_impl_create_bare(nir_shader
*shader
)
274 nir_function_impl
*impl
= ralloc(shader
, nir_function_impl
);
276 impl
->function
= NULL
;
278 cf_init(&impl
->cf_node
, nir_cf_node_function
);
280 exec_list_make_empty(&impl
->body
);
281 exec_list_make_empty(&impl
->registers
);
282 exec_list_make_empty(&impl
->locals
);
283 impl
->num_params
= 0;
285 impl
->return_var
= NULL
;
288 impl
->valid_metadata
= nir_metadata_none
;
290 /* create start & end blocks */
291 nir_block
*start_block
= nir_block_create(shader
);
292 nir_block
*end_block
= nir_block_create(shader
);
293 start_block
->cf_node
.parent
= &impl
->cf_node
;
294 end_block
->cf_node
.parent
= &impl
->cf_node
;
295 impl
->end_block
= end_block
;
297 exec_list_push_tail(&impl
->body
, &start_block
->cf_node
.node
);
299 start_block
->successors
[0] = end_block
;
300 _mesa_set_add(end_block
->predecessors
, start_block
);
305 nir_function_impl_create(nir_function
*function
)
307 assert(function
->impl
== NULL
);
309 nir_function_impl
*impl
= nir_function_impl_create_bare(function
->shader
);
311 function
->impl
= impl
;
312 impl
->function
= function
;
314 impl
->num_params
= function
->num_params
;
315 impl
->params
= ralloc_array(function
->shader
,
316 nir_variable
*, impl
->num_params
);
318 for (unsigned i
= 0; i
< impl
->num_params
; i
++) {
319 impl
->params
[i
] = rzalloc(function
->shader
, nir_variable
);
320 impl
->params
[i
]->type
= function
->params
[i
].type
;
321 impl
->params
[i
]->data
.mode
= nir_var_param
;
322 impl
->params
[i
]->data
.location
= i
;
325 if (!glsl_type_is_void(function
->return_type
)) {
326 impl
->return_var
= rzalloc(function
->shader
, nir_variable
);
327 impl
->return_var
->type
= function
->return_type
;
328 impl
->return_var
->data
.mode
= nir_var_param
;
329 impl
->return_var
->data
.location
= -1;
331 impl
->return_var
= NULL
;
338 nir_block_create(nir_shader
*shader
)
340 nir_block
*block
= rzalloc(shader
, nir_block
);
342 cf_init(&block
->cf_node
, nir_cf_node_block
);
344 block
->successors
[0] = block
->successors
[1] = NULL
;
345 block
->predecessors
= _mesa_set_create(block
, _mesa_hash_pointer
,
346 _mesa_key_pointer_equal
);
347 block
->imm_dom
= NULL
;
348 /* XXX maybe it would be worth it to defer allocation? This
349 * way it doesn't get allocated for shader refs that never run
350 * nir_calc_dominance? For example, state-tracker creates an
351 * initial IR, clones that, runs appropriate lowering pass, passes
352 * to driver which does common lowering/opt, and then stores ref
353 * which is later used to do state specific lowering and futher
354 * opt. Do any of the references not need dominance metadata?
356 block
->dom_frontier
= _mesa_set_create(block
, _mesa_hash_pointer
,
357 _mesa_key_pointer_equal
);
359 exec_list_make_empty(&block
->instr_list
);
365 src_init(nir_src
*src
)
369 src
->reg
.indirect
= NULL
;
370 src
->reg
.base_offset
= 0;
374 nir_if_create(nir_shader
*shader
)
376 nir_if
*if_stmt
= ralloc(shader
, nir_if
);
378 cf_init(&if_stmt
->cf_node
, nir_cf_node_if
);
379 src_init(&if_stmt
->condition
);
381 nir_block
*then
= nir_block_create(shader
);
382 exec_list_make_empty(&if_stmt
->then_list
);
383 exec_list_push_tail(&if_stmt
->then_list
, &then
->cf_node
.node
);
384 then
->cf_node
.parent
= &if_stmt
->cf_node
;
386 nir_block
*else_stmt
= nir_block_create(shader
);
387 exec_list_make_empty(&if_stmt
->else_list
);
388 exec_list_push_tail(&if_stmt
->else_list
, &else_stmt
->cf_node
.node
);
389 else_stmt
->cf_node
.parent
= &if_stmt
->cf_node
;
395 nir_loop_create(nir_shader
*shader
)
397 nir_loop
*loop
= rzalloc(shader
, nir_loop
);
399 cf_init(&loop
->cf_node
, nir_cf_node_loop
);
401 nir_block
*body
= nir_block_create(shader
);
402 exec_list_make_empty(&loop
->body
);
403 exec_list_push_tail(&loop
->body
, &body
->cf_node
.node
);
404 body
->cf_node
.parent
= &loop
->cf_node
;
406 body
->successors
[0] = body
;
407 _mesa_set_add(body
->predecessors
, body
);
413 instr_init(nir_instr
*instr
, nir_instr_type type
)
417 exec_node_init(&instr
->node
);
421 dest_init(nir_dest
*dest
)
423 dest
->is_ssa
= false;
424 dest
->reg
.reg
= NULL
;
425 dest
->reg
.indirect
= NULL
;
426 dest
->reg
.base_offset
= 0;
430 alu_dest_init(nir_alu_dest
*dest
)
432 dest_init(&dest
->dest
);
433 dest
->saturate
= false;
434 dest
->write_mask
= 0xf;
438 alu_src_init(nir_alu_src
*src
)
441 src
->abs
= src
->negate
= false;
449 nir_alu_instr_create(nir_shader
*shader
, nir_op op
)
451 unsigned num_srcs
= nir_op_infos
[op
].num_inputs
;
452 /* TODO: don't use rzalloc */
453 nir_alu_instr
*instr
=
455 sizeof(nir_alu_instr
) + num_srcs
* sizeof(nir_alu_src
));
457 instr_init(&instr
->instr
, nir_instr_type_alu
);
459 alu_dest_init(&instr
->dest
);
460 for (unsigned i
= 0; i
< num_srcs
; i
++)
461 alu_src_init(&instr
->src
[i
]);
467 nir_jump_instr_create(nir_shader
*shader
, nir_jump_type type
)
469 nir_jump_instr
*instr
= ralloc(shader
, nir_jump_instr
);
470 instr_init(&instr
->instr
, nir_instr_type_jump
);
475 nir_load_const_instr
*
476 nir_load_const_instr_create(nir_shader
*shader
, unsigned num_components
,
479 nir_load_const_instr
*instr
= ralloc(shader
, nir_load_const_instr
);
480 instr_init(&instr
->instr
, nir_instr_type_load_const
);
482 nir_ssa_def_init(&instr
->instr
, &instr
->def
, num_components
, bit_size
, NULL
);
487 nir_intrinsic_instr
*
488 nir_intrinsic_instr_create(nir_shader
*shader
, nir_intrinsic_op op
)
490 unsigned num_srcs
= nir_intrinsic_infos
[op
].num_srcs
;
491 /* TODO: don't use rzalloc */
492 nir_intrinsic_instr
*instr
=
494 sizeof(nir_intrinsic_instr
) + num_srcs
* sizeof(nir_src
));
496 instr_init(&instr
->instr
, nir_instr_type_intrinsic
);
497 instr
->intrinsic
= op
;
499 if (nir_intrinsic_infos
[op
].has_dest
)
500 dest_init(&instr
->dest
);
502 for (unsigned i
= 0; i
< num_srcs
; i
++)
503 src_init(&instr
->src
[i
]);
509 nir_call_instr_create(nir_shader
*shader
, nir_function
*callee
)
511 nir_call_instr
*instr
= ralloc(shader
, nir_call_instr
);
512 instr_init(&instr
->instr
, nir_instr_type_call
);
514 instr
->callee
= callee
;
515 instr
->num_params
= callee
->num_params
;
516 instr
->params
= ralloc_array(instr
, nir_deref_var
*, instr
->num_params
);
517 instr
->return_deref
= NULL
;
523 nir_tex_instr_create(nir_shader
*shader
, unsigned num_srcs
)
525 nir_tex_instr
*instr
= rzalloc(shader
, nir_tex_instr
);
526 instr_init(&instr
->instr
, nir_instr_type_tex
);
528 dest_init(&instr
->dest
);
530 instr
->num_srcs
= num_srcs
;
531 instr
->src
= ralloc_array(instr
, nir_tex_src
, num_srcs
);
532 for (unsigned i
= 0; i
< num_srcs
; i
++)
533 src_init(&instr
->src
[i
].src
);
535 instr
->texture_index
= 0;
536 instr
->texture_array_size
= 0;
537 instr
->texture
= NULL
;
538 instr
->sampler_index
= 0;
539 instr
->sampler
= NULL
;
545 nir_tex_instr_remove_src(nir_tex_instr
*tex
, unsigned src_idx
)
547 assert(src_idx
< tex
->num_srcs
);
549 /* First rewrite the source to NIR_SRC_INIT */
550 nir_instr_rewrite_src(&tex
->instr
, &tex
->src
[src_idx
].src
, NIR_SRC_INIT
);
552 /* Now, move all of the other sources down */
553 for (unsigned i
= src_idx
+ 1; i
< tex
->num_srcs
; i
++) {
554 tex
->src
[i
-1].src_type
= tex
->src
[i
].src_type
;
555 nir_instr_move_src(&tex
->instr
, &tex
->src
[i
-1].src
, &tex
->src
[i
].src
);
561 nir_phi_instr_create(nir_shader
*shader
)
563 nir_phi_instr
*instr
= ralloc(shader
, nir_phi_instr
);
564 instr_init(&instr
->instr
, nir_instr_type_phi
);
566 dest_init(&instr
->dest
);
567 exec_list_make_empty(&instr
->srcs
);
571 nir_parallel_copy_instr
*
572 nir_parallel_copy_instr_create(nir_shader
*shader
)
574 nir_parallel_copy_instr
*instr
= ralloc(shader
, nir_parallel_copy_instr
);
575 instr_init(&instr
->instr
, nir_instr_type_parallel_copy
);
577 exec_list_make_empty(&instr
->entries
);
582 nir_ssa_undef_instr
*
583 nir_ssa_undef_instr_create(nir_shader
*shader
,
584 unsigned num_components
,
587 nir_ssa_undef_instr
*instr
= ralloc(shader
, nir_ssa_undef_instr
);
588 instr_init(&instr
->instr
, nir_instr_type_ssa_undef
);
590 nir_ssa_def_init(&instr
->instr
, &instr
->def
, num_components
, bit_size
, NULL
);
596 nir_deref_var_create(void *mem_ctx
, nir_variable
*var
)
598 nir_deref_var
*deref
= ralloc(mem_ctx
, nir_deref_var
);
599 deref
->deref
.deref_type
= nir_deref_type_var
;
600 deref
->deref
.child
= NULL
;
601 deref
->deref
.type
= var
->type
;
607 nir_deref_array_create(void *mem_ctx
)
609 nir_deref_array
*deref
= ralloc(mem_ctx
, nir_deref_array
);
610 deref
->deref
.deref_type
= nir_deref_type_array
;
611 deref
->deref
.child
= NULL
;
612 deref
->deref_array_type
= nir_deref_array_type_direct
;
613 src_init(&deref
->indirect
);
614 deref
->base_offset
= 0;
619 nir_deref_struct_create(void *mem_ctx
, unsigned field_index
)
621 nir_deref_struct
*deref
= ralloc(mem_ctx
, nir_deref_struct
);
622 deref
->deref
.deref_type
= nir_deref_type_struct
;
623 deref
->deref
.child
= NULL
;
624 deref
->index
= field_index
;
629 nir_deref_var_clone(const nir_deref_var
*deref
, void *mem_ctx
)
634 nir_deref_var
*ret
= nir_deref_var_create(mem_ctx
, deref
->var
);
635 ret
->deref
.type
= deref
->deref
.type
;
636 if (deref
->deref
.child
)
637 ret
->deref
.child
= nir_deref_clone(deref
->deref
.child
, ret
);
641 static nir_deref_array
*
642 deref_array_clone(const nir_deref_array
*deref
, void *mem_ctx
)
644 nir_deref_array
*ret
= nir_deref_array_create(mem_ctx
);
645 ret
->base_offset
= deref
->base_offset
;
646 ret
->deref_array_type
= deref
->deref_array_type
;
647 if (deref
->deref_array_type
== nir_deref_array_type_indirect
) {
648 nir_src_copy(&ret
->indirect
, &deref
->indirect
, mem_ctx
);
650 ret
->deref
.type
= deref
->deref
.type
;
651 if (deref
->deref
.child
)
652 ret
->deref
.child
= nir_deref_clone(deref
->deref
.child
, ret
);
656 static nir_deref_struct
*
657 deref_struct_clone(const nir_deref_struct
*deref
, void *mem_ctx
)
659 nir_deref_struct
*ret
= nir_deref_struct_create(mem_ctx
, deref
->index
);
660 ret
->deref
.type
= deref
->deref
.type
;
661 if (deref
->deref
.child
)
662 ret
->deref
.child
= nir_deref_clone(deref
->deref
.child
, ret
);
667 nir_deref_clone(const nir_deref
*deref
, void *mem_ctx
)
672 switch (deref
->deref_type
) {
673 case nir_deref_type_var
:
674 return &nir_deref_var_clone(nir_deref_as_var(deref
), mem_ctx
)->deref
;
675 case nir_deref_type_array
:
676 return &deref_array_clone(nir_deref_as_array(deref
), mem_ctx
)->deref
;
677 case nir_deref_type_struct
:
678 return &deref_struct_clone(nir_deref_as_struct(deref
), mem_ctx
)->deref
;
680 unreachable("Invalid dereference type");
686 /* This is the second step in the recursion. We've found the tail and made a
687 * copy. Now we need to iterate over all possible leaves and call the
688 * callback on each one.
691 deref_foreach_leaf_build_recur(nir_deref_var
*deref
, nir_deref
*tail
,
692 nir_deref_foreach_leaf_cb cb
, void *state
)
697 nir_deref_struct str
;
700 assert(tail
->child
== NULL
);
701 switch (glsl_get_base_type(tail
->type
)) {
703 case GLSL_TYPE_UINT64
:
705 case GLSL_TYPE_INT64
:
706 case GLSL_TYPE_FLOAT
:
707 case GLSL_TYPE_DOUBLE
:
709 if (glsl_type_is_vector_or_scalar(tail
->type
))
710 return cb(deref
, state
);
713 case GLSL_TYPE_ARRAY
:
714 tmp
.arr
.deref
.deref_type
= nir_deref_type_array
;
715 tmp
.arr
.deref
.type
= glsl_get_array_element(tail
->type
);
716 tmp
.arr
.deref_array_type
= nir_deref_array_type_direct
;
717 tmp
.arr
.indirect
= NIR_SRC_INIT
;
718 tail
->child
= &tmp
.arr
.deref
;
720 length
= glsl_get_length(tail
->type
);
721 for (unsigned i
= 0; i
< length
; i
++) {
722 tmp
.arr
.deref
.child
= NULL
;
723 tmp
.arr
.base_offset
= i
;
724 if (!deref_foreach_leaf_build_recur(deref
, &tmp
.arr
.deref
, cb
, state
))
729 case GLSL_TYPE_STRUCT
:
730 tmp
.str
.deref
.deref_type
= nir_deref_type_struct
;
731 tail
->child
= &tmp
.str
.deref
;
733 length
= glsl_get_length(tail
->type
);
734 for (unsigned i
= 0; i
< length
; i
++) {
735 tmp
.arr
.deref
.child
= NULL
;
736 tmp
.str
.deref
.type
= glsl_get_struct_field(tail
->type
, i
);
738 if (!deref_foreach_leaf_build_recur(deref
, &tmp
.arr
.deref
, cb
, state
))
744 unreachable("Invalid type for dereference");
748 /* This is the first step of the foreach_leaf recursion. In this step we are
749 * walking to the end of the deref chain and making a copy in the stack as we
750 * go. This is because we don't want to mutate the deref chain that was
751 * passed in by the caller. The downside is that this deref chain is on the
752 * stack and , if the caller wants to do anything with it, they will have to
753 * make their own copy because this one will go away.
756 deref_foreach_leaf_copy_recur(nir_deref_var
*deref
, nir_deref
*tail
,
757 nir_deref_foreach_leaf_cb cb
, void *state
)
761 nir_deref_struct str
;
765 switch (tail
->child
->deref_type
) {
766 case nir_deref_type_array
:
767 c
.arr
= *nir_deref_as_array(tail
->child
);
768 tail
->child
= &c
.arr
.deref
;
769 return deref_foreach_leaf_copy_recur(deref
, &c
.arr
.deref
, cb
, state
);
771 case nir_deref_type_struct
:
772 c
.str
= *nir_deref_as_struct(tail
->child
);
773 tail
->child
= &c
.str
.deref
;
774 return deref_foreach_leaf_copy_recur(deref
, &c
.str
.deref
, cb
, state
);
776 case nir_deref_type_var
:
778 unreachable("Invalid deref type for a child");
781 /* We've gotten to the end of the original deref. Time to start
782 * building our own derefs.
784 return deref_foreach_leaf_build_recur(deref
, tail
, cb
, state
);
789 * This function iterates over all of the possible derefs that can be created
790 * with the given deref as the head. It then calls the provided callback with
791 * a full deref for each one.
793 * The deref passed to the callback will be allocated on the stack. You will
794 * need to make a copy if you want it to hang around.
797 nir_deref_foreach_leaf(nir_deref_var
*deref
,
798 nir_deref_foreach_leaf_cb cb
, void *state
)
800 nir_deref_var copy
= *deref
;
801 return deref_foreach_leaf_copy_recur(©
, ©
.deref
, cb
, state
);
804 /* Returns a load_const instruction that represents the constant
805 * initializer for the given deref chain. The caller is responsible for
806 * ensuring that there actually is a constant initializer.
808 nir_load_const_instr
*
809 nir_deref_get_const_initializer_load(nir_shader
*shader
, nir_deref_var
*deref
)
811 nir_constant
*constant
= deref
->var
->constant_initializer
;
814 const nir_deref
*tail
= &deref
->deref
;
815 unsigned matrix_col
= 0;
816 while (tail
->child
) {
817 switch (tail
->child
->deref_type
) {
818 case nir_deref_type_array
: {
819 nir_deref_array
*arr
= nir_deref_as_array(tail
->child
);
820 assert(arr
->deref_array_type
== nir_deref_array_type_direct
);
821 if (glsl_type_is_matrix(tail
->type
)) {
822 assert(arr
->deref
.child
== NULL
);
823 matrix_col
= arr
->base_offset
;
825 constant
= constant
->elements
[arr
->base_offset
];
830 case nir_deref_type_struct
: {
831 constant
= constant
->elements
[nir_deref_as_struct(tail
->child
)->index
];
836 unreachable("Invalid deref child type");
842 unsigned bit_size
= glsl_get_bit_size(tail
->type
);
843 nir_load_const_instr
*load
=
844 nir_load_const_instr_create(shader
, glsl_get_vector_elements(tail
->type
),
847 switch (glsl_get_base_type(tail
->type
)) {
848 case GLSL_TYPE_FLOAT
:
851 case GLSL_TYPE_DOUBLE
:
852 case GLSL_TYPE_UINT64
:
853 case GLSL_TYPE_INT64
:
855 load
->value
= constant
->values
[matrix_col
];
858 unreachable("Invalid immediate type");
865 nir_cf_node_get_function(nir_cf_node
*node
)
867 while (node
->type
!= nir_cf_node_function
) {
871 return nir_cf_node_as_function(node
);
874 /* Reduces a cursor by trying to convert everything to after and trying to
875 * go up to block granularity when possible.
878 reduce_cursor(nir_cursor cursor
)
880 switch (cursor
.option
) {
881 case nir_cursor_before_block
:
882 assert(nir_cf_node_prev(&cursor
.block
->cf_node
) == NULL
||
883 nir_cf_node_prev(&cursor
.block
->cf_node
)->type
!= nir_cf_node_block
);
884 if (exec_list_is_empty(&cursor
.block
->instr_list
)) {
885 /* Empty block. After is as good as before. */
886 cursor
.option
= nir_cursor_after_block
;
890 case nir_cursor_after_block
:
893 case nir_cursor_before_instr
: {
894 nir_instr
*prev_instr
= nir_instr_prev(cursor
.instr
);
896 /* Before this instruction is after the previous */
897 cursor
.instr
= prev_instr
;
898 cursor
.option
= nir_cursor_after_instr
;
900 /* No previous instruction. Switch to before block */
901 cursor
.block
= cursor
.instr
->block
;
902 cursor
.option
= nir_cursor_before_block
;
904 return reduce_cursor(cursor
);
907 case nir_cursor_after_instr
:
908 if (nir_instr_next(cursor
.instr
) == NULL
) {
909 /* This is the last instruction, switch to after block */
910 cursor
.option
= nir_cursor_after_block
;
911 cursor
.block
= cursor
.instr
->block
;
916 unreachable("Inavlid cursor option");
921 nir_cursors_equal(nir_cursor a
, nir_cursor b
)
923 /* Reduced cursors should be unique */
924 a
= reduce_cursor(a
);
925 b
= reduce_cursor(b
);
927 return a
.block
== b
.block
&& a
.option
== b
.option
;
931 add_use_cb(nir_src
*src
, void *state
)
933 nir_instr
*instr
= state
;
935 src
->parent_instr
= instr
;
936 list_addtail(&src
->use_link
,
937 src
->is_ssa
? &src
->ssa
->uses
: &src
->reg
.reg
->uses
);
943 add_ssa_def_cb(nir_ssa_def
*def
, void *state
)
945 nir_instr
*instr
= state
;
947 if (instr
->block
&& def
->index
== UINT_MAX
) {
948 nir_function_impl
*impl
=
949 nir_cf_node_get_function(&instr
->block
->cf_node
);
951 def
->index
= impl
->ssa_alloc
++;
958 add_reg_def_cb(nir_dest
*dest
, void *state
)
960 nir_instr
*instr
= state
;
963 dest
->reg
.parent_instr
= instr
;
964 list_addtail(&dest
->reg
.def_link
, &dest
->reg
.reg
->defs
);
971 add_defs_uses(nir_instr
*instr
)
973 nir_foreach_src(instr
, add_use_cb
, instr
);
974 nir_foreach_dest(instr
, add_reg_def_cb
, instr
);
975 nir_foreach_ssa_def(instr
, add_ssa_def_cb
, instr
);
979 nir_instr_insert(nir_cursor cursor
, nir_instr
*instr
)
981 switch (cursor
.option
) {
982 case nir_cursor_before_block
:
983 /* Only allow inserting jumps into empty blocks. */
984 if (instr
->type
== nir_instr_type_jump
)
985 assert(exec_list_is_empty(&cursor
.block
->instr_list
));
987 instr
->block
= cursor
.block
;
988 add_defs_uses(instr
);
989 exec_list_push_head(&cursor
.block
->instr_list
, &instr
->node
);
991 case nir_cursor_after_block
: {
992 /* Inserting instructions after a jump is illegal. */
993 nir_instr
*last
= nir_block_last_instr(cursor
.block
);
994 assert(last
== NULL
|| last
->type
!= nir_instr_type_jump
);
997 instr
->block
= cursor
.block
;
998 add_defs_uses(instr
);
999 exec_list_push_tail(&cursor
.block
->instr_list
, &instr
->node
);
1002 case nir_cursor_before_instr
:
1003 assert(instr
->type
!= nir_instr_type_jump
);
1004 instr
->block
= cursor
.instr
->block
;
1005 add_defs_uses(instr
);
1006 exec_node_insert_node_before(&cursor
.instr
->node
, &instr
->node
);
1008 case nir_cursor_after_instr
:
1009 /* Inserting instructions after a jump is illegal. */
1010 assert(cursor
.instr
->type
!= nir_instr_type_jump
);
1012 /* Only allow inserting jumps at the end of the block. */
1013 if (instr
->type
== nir_instr_type_jump
)
1014 assert(cursor
.instr
== nir_block_last_instr(cursor
.instr
->block
));
1016 instr
->block
= cursor
.instr
->block
;
1017 add_defs_uses(instr
);
1018 exec_node_insert_after(&cursor
.instr
->node
, &instr
->node
);
1022 if (instr
->type
== nir_instr_type_jump
)
1023 nir_handle_add_jump(instr
->block
);
1027 src_is_valid(const nir_src
*src
)
1029 return src
->is_ssa
? (src
->ssa
!= NULL
) : (src
->reg
.reg
!= NULL
);
1033 remove_use_cb(nir_src
*src
, void *state
)
1037 if (src_is_valid(src
))
1038 list_del(&src
->use_link
);
1044 remove_def_cb(nir_dest
*dest
, void *state
)
1049 list_del(&dest
->reg
.def_link
);
1055 remove_defs_uses(nir_instr
*instr
)
1057 nir_foreach_dest(instr
, remove_def_cb
, instr
);
1058 nir_foreach_src(instr
, remove_use_cb
, instr
);
1061 void nir_instr_remove(nir_instr
*instr
)
1063 remove_defs_uses(instr
);
1064 exec_node_remove(&instr
->node
);
1066 if (instr
->type
== nir_instr_type_jump
) {
1067 nir_jump_instr
*jump_instr
= nir_instr_as_jump(instr
);
1068 nir_handle_remove_jump(instr
->block
, jump_instr
->type
);
1075 nir_index_local_regs(nir_function_impl
*impl
)
1078 foreach_list_typed(nir_register
, reg
, node
, &impl
->registers
) {
1079 reg
->index
= index
++;
1081 impl
->reg_alloc
= index
;
1085 nir_index_global_regs(nir_shader
*shader
)
1088 foreach_list_typed(nir_register
, reg
, node
, &shader
->registers
) {
1089 reg
->index
= index
++;
1091 shader
->reg_alloc
= index
;
1095 visit_alu_dest(nir_alu_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
1097 return cb(&instr
->dest
.dest
, state
);
1101 visit_intrinsic_dest(nir_intrinsic_instr
*instr
, nir_foreach_dest_cb cb
,
1104 if (nir_intrinsic_infos
[instr
->intrinsic
].has_dest
)
1105 return cb(&instr
->dest
, state
);
1111 visit_texture_dest(nir_tex_instr
*instr
, nir_foreach_dest_cb cb
,
1114 return cb(&instr
->dest
, state
);
1118 visit_phi_dest(nir_phi_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
1120 return cb(&instr
->dest
, state
);
1124 visit_parallel_copy_dest(nir_parallel_copy_instr
*instr
,
1125 nir_foreach_dest_cb cb
, void *state
)
1127 nir_foreach_parallel_copy_entry(entry
, instr
) {
1128 if (!cb(&entry
->dest
, state
))
1136 nir_foreach_dest(nir_instr
*instr
, nir_foreach_dest_cb cb
, void *state
)
1138 switch (instr
->type
) {
1139 case nir_instr_type_alu
:
1140 return visit_alu_dest(nir_instr_as_alu(instr
), cb
, state
);
1141 case nir_instr_type_intrinsic
:
1142 return visit_intrinsic_dest(nir_instr_as_intrinsic(instr
), cb
, state
);
1143 case nir_instr_type_tex
:
1144 return visit_texture_dest(nir_instr_as_tex(instr
), cb
, state
);
1145 case nir_instr_type_phi
:
1146 return visit_phi_dest(nir_instr_as_phi(instr
), cb
, state
);
1147 case nir_instr_type_parallel_copy
:
1148 return visit_parallel_copy_dest(nir_instr_as_parallel_copy(instr
),
1151 case nir_instr_type_load_const
:
1152 case nir_instr_type_ssa_undef
:
1153 case nir_instr_type_call
:
1154 case nir_instr_type_jump
:
1158 unreachable("Invalid instruction type");
1165 struct foreach_ssa_def_state
{
1166 nir_foreach_ssa_def_cb cb
;
1171 nir_ssa_def_visitor(nir_dest
*dest
, void *void_state
)
1173 struct foreach_ssa_def_state
*state
= void_state
;
1176 return state
->cb(&dest
->ssa
, state
->client_state
);
1182 nir_foreach_ssa_def(nir_instr
*instr
, nir_foreach_ssa_def_cb cb
, void *state
)
1184 switch (instr
->type
) {
1185 case nir_instr_type_alu
:
1186 case nir_instr_type_tex
:
1187 case nir_instr_type_intrinsic
:
1188 case nir_instr_type_phi
:
1189 case nir_instr_type_parallel_copy
: {
1190 struct foreach_ssa_def_state foreach_state
= {cb
, state
};
1191 return nir_foreach_dest(instr
, nir_ssa_def_visitor
, &foreach_state
);
1194 case nir_instr_type_load_const
:
1195 return cb(&nir_instr_as_load_const(instr
)->def
, state
);
1196 case nir_instr_type_ssa_undef
:
1197 return cb(&nir_instr_as_ssa_undef(instr
)->def
, state
);
1198 case nir_instr_type_call
:
1199 case nir_instr_type_jump
:
1202 unreachable("Invalid instruction type");
1207 visit_src(nir_src
*src
, nir_foreach_src_cb cb
, void *state
)
1209 if (!cb(src
, state
))
1211 if (!src
->is_ssa
&& src
->reg
.indirect
)
1212 return cb(src
->reg
.indirect
, state
);
1217 visit_deref_array_src(nir_deref_array
*deref
, nir_foreach_src_cb cb
,
1220 if (deref
->deref_array_type
== nir_deref_array_type_indirect
)
1221 return visit_src(&deref
->indirect
, cb
, state
);
1226 visit_deref_src(nir_deref_var
*deref
, nir_foreach_src_cb cb
, void *state
)
1228 nir_deref
*cur
= &deref
->deref
;
1229 while (cur
!= NULL
) {
1230 if (cur
->deref_type
== nir_deref_type_array
) {
1231 if (!visit_deref_array_src(nir_deref_as_array(cur
), cb
, state
))
1242 visit_alu_src(nir_alu_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1244 for (unsigned i
= 0; i
< nir_op_infos
[instr
->op
].num_inputs
; i
++)
1245 if (!visit_src(&instr
->src
[i
].src
, cb
, state
))
1252 visit_tex_src(nir_tex_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1254 for (unsigned i
= 0; i
< instr
->num_srcs
; i
++) {
1255 if (!visit_src(&instr
->src
[i
].src
, cb
, state
))
1259 if (instr
->texture
!= NULL
) {
1260 if (!visit_deref_src(instr
->texture
, cb
, state
))
1264 if (instr
->sampler
!= NULL
) {
1265 if (!visit_deref_src(instr
->sampler
, cb
, state
))
1273 visit_intrinsic_src(nir_intrinsic_instr
*instr
, nir_foreach_src_cb cb
,
1276 unsigned num_srcs
= nir_intrinsic_infos
[instr
->intrinsic
].num_srcs
;
1277 for (unsigned i
= 0; i
< num_srcs
; i
++) {
1278 if (!visit_src(&instr
->src
[i
], cb
, state
))
1283 nir_intrinsic_infos
[instr
->intrinsic
].num_variables
;
1284 for (unsigned i
= 0; i
< num_vars
; i
++) {
1285 if (!visit_deref_src(instr
->variables
[i
], cb
, state
))
1293 visit_phi_src(nir_phi_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1295 nir_foreach_phi_src(src
, instr
) {
1296 if (!visit_src(&src
->src
, cb
, state
))
1304 visit_parallel_copy_src(nir_parallel_copy_instr
*instr
,
1305 nir_foreach_src_cb cb
, void *state
)
1307 nir_foreach_parallel_copy_entry(entry
, instr
) {
1308 if (!visit_src(&entry
->src
, cb
, state
))
1317 nir_foreach_src_cb cb
;
1318 } visit_dest_indirect_state
;
1321 visit_dest_indirect(nir_dest
*dest
, void *_state
)
1323 visit_dest_indirect_state
*state
= (visit_dest_indirect_state
*) _state
;
1325 if (!dest
->is_ssa
&& dest
->reg
.indirect
)
1326 return state
->cb(dest
->reg
.indirect
, state
->state
);
1332 nir_foreach_src(nir_instr
*instr
, nir_foreach_src_cb cb
, void *state
)
1334 switch (instr
->type
) {
1335 case nir_instr_type_alu
:
1336 if (!visit_alu_src(nir_instr_as_alu(instr
), cb
, state
))
1339 case nir_instr_type_intrinsic
:
1340 if (!visit_intrinsic_src(nir_instr_as_intrinsic(instr
), cb
, state
))
1343 case nir_instr_type_tex
:
1344 if (!visit_tex_src(nir_instr_as_tex(instr
), cb
, state
))
1347 case nir_instr_type_call
:
1348 /* Call instructions have no regular sources */
1350 case nir_instr_type_load_const
:
1351 /* Constant load instructions have no regular sources */
1353 case nir_instr_type_phi
:
1354 if (!visit_phi_src(nir_instr_as_phi(instr
), cb
, state
))
1357 case nir_instr_type_parallel_copy
:
1358 if (!visit_parallel_copy_src(nir_instr_as_parallel_copy(instr
),
1362 case nir_instr_type_jump
:
1363 case nir_instr_type_ssa_undef
:
1367 unreachable("Invalid instruction type");
1371 visit_dest_indirect_state dest_state
;
1372 dest_state
.state
= state
;
1374 return nir_foreach_dest(instr
, visit_dest_indirect
, &dest_state
);
1378 nir_src_as_const_value(nir_src src
)
1383 if (src
.ssa
->parent_instr
->type
!= nir_instr_type_load_const
)
1386 nir_load_const_instr
*load
= nir_instr_as_load_const(src
.ssa
->parent_instr
);
1388 return &load
->value
;
1392 * Returns true if the source is known to be dynamically uniform. Otherwise it
1393 * returns false which means it may or may not be dynamically uniform but it
1394 * can't be determined.
1397 nir_src_is_dynamically_uniform(nir_src src
)
1402 /* Constants are trivially dynamically uniform */
1403 if (src
.ssa
->parent_instr
->type
== nir_instr_type_load_const
)
1406 /* As are uniform variables */
1407 if (src
.ssa
->parent_instr
->type
== nir_instr_type_intrinsic
) {
1408 nir_intrinsic_instr
*intr
= nir_instr_as_intrinsic(src
.ssa
->parent_instr
);
1410 if (intr
->intrinsic
== nir_intrinsic_load_uniform
)
1414 /* XXX: this could have many more tests, such as when a sampler function is
1415 * called with dynamically uniform arguments.
1421 src_remove_all_uses(nir_src
*src
)
1423 for (; src
; src
= src
->is_ssa
? NULL
: src
->reg
.indirect
) {
1424 if (!src_is_valid(src
))
1427 list_del(&src
->use_link
);
1432 src_add_all_uses(nir_src
*src
, nir_instr
*parent_instr
, nir_if
*parent_if
)
1434 for (; src
; src
= src
->is_ssa
? NULL
: src
->reg
.indirect
) {
1435 if (!src_is_valid(src
))
1439 src
->parent_instr
= parent_instr
;
1441 list_addtail(&src
->use_link
, &src
->ssa
->uses
);
1443 list_addtail(&src
->use_link
, &src
->reg
.reg
->uses
);
1446 src
->parent_if
= parent_if
;
1448 list_addtail(&src
->use_link
, &src
->ssa
->if_uses
);
1450 list_addtail(&src
->use_link
, &src
->reg
.reg
->if_uses
);
1456 nir_instr_rewrite_src(nir_instr
*instr
, nir_src
*src
, nir_src new_src
)
1458 assert(!src_is_valid(src
) || src
->parent_instr
== instr
);
1460 src_remove_all_uses(src
);
1462 src_add_all_uses(src
, instr
, NULL
);
1466 nir_instr_move_src(nir_instr
*dest_instr
, nir_src
*dest
, nir_src
*src
)
1468 assert(!src_is_valid(dest
) || dest
->parent_instr
== dest_instr
);
1470 src_remove_all_uses(dest
);
1471 src_remove_all_uses(src
);
1473 *src
= NIR_SRC_INIT
;
1474 src_add_all_uses(dest
, dest_instr
, NULL
);
1478 nir_if_rewrite_condition(nir_if
*if_stmt
, nir_src new_src
)
1480 nir_src
*src
= &if_stmt
->condition
;
1481 assert(!src_is_valid(src
) || src
->parent_if
== if_stmt
);
1483 src_remove_all_uses(src
);
1485 src_add_all_uses(src
, NULL
, if_stmt
);
1489 nir_instr_rewrite_dest(nir_instr
*instr
, nir_dest
*dest
, nir_dest new_dest
)
1492 /* We can only overwrite an SSA destination if it has no uses. */
1493 assert(list_empty(&dest
->ssa
.uses
) && list_empty(&dest
->ssa
.if_uses
));
1495 list_del(&dest
->reg
.def_link
);
1496 if (dest
->reg
.indirect
)
1497 src_remove_all_uses(dest
->reg
.indirect
);
1500 /* We can't re-write with an SSA def */
1501 assert(!new_dest
.is_ssa
);
1503 nir_dest_copy(dest
, &new_dest
, instr
);
1505 dest
->reg
.parent_instr
= instr
;
1506 list_addtail(&dest
->reg
.def_link
, &new_dest
.reg
.reg
->defs
);
1508 if (dest
->reg
.indirect
)
1509 src_add_all_uses(dest
->reg
.indirect
, instr
, NULL
);
1513 nir_instr_rewrite_deref(nir_instr
*instr
, nir_deref_var
**deref
,
1514 nir_deref_var
*new_deref
)
1517 visit_deref_src(*deref
, remove_use_cb
, NULL
);
1522 visit_deref_src(*deref
, add_use_cb
, instr
);
1525 /* note: does *not* take ownership of 'name' */
1527 nir_ssa_def_init(nir_instr
*instr
, nir_ssa_def
*def
,
1528 unsigned num_components
,
1529 unsigned bit_size
, const char *name
)
1531 def
->name
= ralloc_strdup(instr
, name
);
1532 def
->parent_instr
= instr
;
1533 list_inithead(&def
->uses
);
1534 list_inithead(&def
->if_uses
);
1535 def
->num_components
= num_components
;
1536 def
->bit_size
= bit_size
;
1539 nir_function_impl
*impl
=
1540 nir_cf_node_get_function(&instr
->block
->cf_node
);
1542 def
->index
= impl
->ssa_alloc
++;
1544 def
->index
= UINT_MAX
;
1548 /* note: does *not* take ownership of 'name' */
1550 nir_ssa_dest_init(nir_instr
*instr
, nir_dest
*dest
,
1551 unsigned num_components
, unsigned bit_size
,
1554 dest
->is_ssa
= true;
1555 nir_ssa_def_init(instr
, &dest
->ssa
, num_components
, bit_size
, name
);
1559 nir_ssa_def_rewrite_uses(nir_ssa_def
*def
, nir_src new_src
)
1561 assert(!new_src
.is_ssa
|| def
!= new_src
.ssa
);
1563 nir_foreach_use_safe(use_src
, def
)
1564 nir_instr_rewrite_src(use_src
->parent_instr
, use_src
, new_src
);
1566 nir_foreach_if_use_safe(use_src
, def
)
1567 nir_if_rewrite_condition(use_src
->parent_if
, new_src
);
1571 is_instr_between(nir_instr
*start
, nir_instr
*end
, nir_instr
*between
)
1573 assert(start
->block
== end
->block
);
1575 if (between
->block
!= start
->block
)
1578 /* Search backwards looking for "between" */
1579 while (start
!= end
) {
1583 end
= nir_instr_prev(end
);
1590 /* Replaces all uses of the given SSA def with the given source but only if
1591 * the use comes after the after_me instruction. This can be useful if you
1592 * are emitting code to fix up the result of some instruction: you can freely
1593 * use the result in that code and then call rewrite_uses_after and pass the
1594 * last fixup instruction as after_me and it will replace all of the uses you
1595 * want without touching the fixup code.
1597 * This function assumes that after_me is in the same block as
1598 * def->parent_instr and that after_me comes after def->parent_instr.
1601 nir_ssa_def_rewrite_uses_after(nir_ssa_def
*def
, nir_src new_src
,
1602 nir_instr
*after_me
)
1604 assert(!new_src
.is_ssa
|| def
!= new_src
.ssa
);
1606 nir_foreach_use_safe(use_src
, def
) {
1607 assert(use_src
->parent_instr
!= def
->parent_instr
);
1608 /* Since def already dominates all of its uses, the only way a use can
1609 * not be dominated by after_me is if it is between def and after_me in
1610 * the instruction list.
1612 if (!is_instr_between(def
->parent_instr
, after_me
, use_src
->parent_instr
))
1613 nir_instr_rewrite_src(use_src
->parent_instr
, use_src
, new_src
);
1616 nir_foreach_if_use_safe(use_src
, def
)
1617 nir_if_rewrite_condition(use_src
->parent_if
, new_src
);
1621 nir_ssa_def_components_read(nir_ssa_def
*def
)
1623 uint8_t read_mask
= 0;
1624 nir_foreach_use(use
, def
) {
1625 if (use
->parent_instr
->type
== nir_instr_type_alu
) {
1626 nir_alu_instr
*alu
= nir_instr_as_alu(use
->parent_instr
);
1627 nir_alu_src
*alu_src
= exec_node_data(nir_alu_src
, use
, src
);
1628 int src_idx
= alu_src
- &alu
->src
[0];
1629 assert(src_idx
>= 0 && src_idx
< nir_op_infos
[alu
->op
].num_inputs
);
1631 for (unsigned c
= 0; c
< 4; c
++) {
1632 if (!nir_alu_instr_channel_used(alu
, src_idx
, c
))
1635 read_mask
|= (1 << alu_src
->swizzle
[c
]);
1638 return (1 << def
->num_components
) - 1;
1646 nir_block_cf_tree_next(nir_block
*block
)
1648 if (block
== NULL
) {
1649 /* nir_foreach_block_safe() will call this function on a NULL block
1650 * after the last iteration, but it won't use the result so just return
1656 nir_cf_node
*cf_next
= nir_cf_node_next(&block
->cf_node
);
1658 return nir_cf_node_cf_tree_first(cf_next
);
1660 nir_cf_node
*parent
= block
->cf_node
.parent
;
1662 switch (parent
->type
) {
1663 case nir_cf_node_if
: {
1664 /* Are we at the end of the if? Go to the beginning of the else */
1665 nir_if
*if_stmt
= nir_cf_node_as_if(parent
);
1666 if (block
== nir_if_last_then_block(if_stmt
))
1667 return nir_if_first_else_block(if_stmt
);
1669 assert(block
== nir_if_last_else_block(if_stmt
));
1673 case nir_cf_node_loop
:
1674 return nir_cf_node_as_block(nir_cf_node_next(parent
));
1676 case nir_cf_node_function
:
1680 unreachable("unknown cf node type");
1685 nir_block_cf_tree_prev(nir_block
*block
)
1687 if (block
== NULL
) {
1688 /* do this for consistency with nir_block_cf_tree_next() */
1692 nir_cf_node
*cf_prev
= nir_cf_node_prev(&block
->cf_node
);
1694 return nir_cf_node_cf_tree_last(cf_prev
);
1696 nir_cf_node
*parent
= block
->cf_node
.parent
;
1698 switch (parent
->type
) {
1699 case nir_cf_node_if
: {
1700 /* Are we at the beginning of the else? Go to the end of the if */
1701 nir_if
*if_stmt
= nir_cf_node_as_if(parent
);
1702 if (block
== nir_if_first_else_block(if_stmt
))
1703 return nir_if_last_then_block(if_stmt
);
1705 assert(block
== nir_if_first_then_block(if_stmt
));
1709 case nir_cf_node_loop
:
1710 return nir_cf_node_as_block(nir_cf_node_prev(parent
));
1712 case nir_cf_node_function
:
1716 unreachable("unknown cf node type");
1720 nir_block
*nir_cf_node_cf_tree_first(nir_cf_node
*node
)
1722 switch (node
->type
) {
1723 case nir_cf_node_function
: {
1724 nir_function_impl
*impl
= nir_cf_node_as_function(node
);
1725 return nir_start_block(impl
);
1728 case nir_cf_node_if
: {
1729 nir_if
*if_stmt
= nir_cf_node_as_if(node
);
1730 return nir_if_first_then_block(if_stmt
);
1733 case nir_cf_node_loop
: {
1734 nir_loop
*loop
= nir_cf_node_as_loop(node
);
1735 return nir_loop_first_block(loop
);
1738 case nir_cf_node_block
: {
1739 return nir_cf_node_as_block(node
);
1743 unreachable("unknown node type");
1747 nir_block
*nir_cf_node_cf_tree_last(nir_cf_node
*node
)
1749 switch (node
->type
) {
1750 case nir_cf_node_function
: {
1751 nir_function_impl
*impl
= nir_cf_node_as_function(node
);
1752 return nir_impl_last_block(impl
);
1755 case nir_cf_node_if
: {
1756 nir_if
*if_stmt
= nir_cf_node_as_if(node
);
1757 return nir_if_last_else_block(if_stmt
);
1760 case nir_cf_node_loop
: {
1761 nir_loop
*loop
= nir_cf_node_as_loop(node
);
1762 return nir_loop_last_block(loop
);
1765 case nir_cf_node_block
: {
1766 return nir_cf_node_as_block(node
);
1770 unreachable("unknown node type");
1774 nir_block
*nir_cf_node_cf_tree_next(nir_cf_node
*node
)
1776 if (node
->type
== nir_cf_node_block
)
1777 return nir_block_cf_tree_next(nir_cf_node_as_block(node
));
1778 else if (node
->type
== nir_cf_node_function
)
1781 return nir_cf_node_as_block(nir_cf_node_next(node
));
1785 nir_block_get_following_if(nir_block
*block
)
1787 if (exec_node_is_tail_sentinel(&block
->cf_node
.node
))
1790 if (nir_cf_node_is_last(&block
->cf_node
))
1793 nir_cf_node
*next_node
= nir_cf_node_next(&block
->cf_node
);
1795 if (next_node
->type
!= nir_cf_node_if
)
1798 return nir_cf_node_as_if(next_node
);
1802 nir_block_get_following_loop(nir_block
*block
)
1804 if (exec_node_is_tail_sentinel(&block
->cf_node
.node
))
1807 if (nir_cf_node_is_last(&block
->cf_node
))
1810 nir_cf_node
*next_node
= nir_cf_node_next(&block
->cf_node
);
1812 if (next_node
->type
!= nir_cf_node_loop
)
1815 return nir_cf_node_as_loop(next_node
);
1819 nir_index_blocks(nir_function_impl
*impl
)
1823 if (impl
->valid_metadata
& nir_metadata_block_index
)
1826 nir_foreach_block(block
, impl
) {
1827 block
->index
= index
++;
1830 impl
->num_blocks
= index
;
1834 index_ssa_def_cb(nir_ssa_def
*def
, void *state
)
1836 unsigned *index
= (unsigned *) state
;
1837 def
->index
= (*index
)++;
1843 * The indices are applied top-to-bottom which has the very nice property
1844 * that, if A dominates B, then A->index <= B->index.
1847 nir_index_ssa_defs(nir_function_impl
*impl
)
1851 nir_foreach_block(block
, impl
) {
1852 nir_foreach_instr(instr
, block
)
1853 nir_foreach_ssa_def(instr
, index_ssa_def_cb
, &index
);
1856 impl
->ssa_alloc
= index
;
1860 * The indices are applied top-to-bottom which has the very nice property
1861 * that, if A dominates B, then A->index <= B->index.
1864 nir_index_instrs(nir_function_impl
*impl
)
1868 nir_foreach_block(block
, impl
) {
1869 nir_foreach_instr(instr
, block
)
1870 instr
->index
= index
++;
1877 nir_intrinsic_from_system_value(gl_system_value val
)
1880 case SYSTEM_VALUE_VERTEX_ID
:
1881 return nir_intrinsic_load_vertex_id
;
1882 case SYSTEM_VALUE_INSTANCE_ID
:
1883 return nir_intrinsic_load_instance_id
;
1884 case SYSTEM_VALUE_DRAW_ID
:
1885 return nir_intrinsic_load_draw_id
;
1886 case SYSTEM_VALUE_BASE_INSTANCE
:
1887 return nir_intrinsic_load_base_instance
;
1888 case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
:
1889 return nir_intrinsic_load_vertex_id_zero_base
;
1890 case SYSTEM_VALUE_BASE_VERTEX
:
1891 return nir_intrinsic_load_base_vertex
;
1892 case SYSTEM_VALUE_INVOCATION_ID
:
1893 return nir_intrinsic_load_invocation_id
;
1894 case SYSTEM_VALUE_FRONT_FACE
:
1895 return nir_intrinsic_load_front_face
;
1896 case SYSTEM_VALUE_SAMPLE_ID
:
1897 return nir_intrinsic_load_sample_id
;
1898 case SYSTEM_VALUE_SAMPLE_POS
:
1899 return nir_intrinsic_load_sample_pos
;
1900 case SYSTEM_VALUE_SAMPLE_MASK_IN
:
1901 return nir_intrinsic_load_sample_mask_in
;
1902 case SYSTEM_VALUE_LOCAL_INVOCATION_ID
:
1903 return nir_intrinsic_load_local_invocation_id
;
1904 case SYSTEM_VALUE_LOCAL_INVOCATION_INDEX
:
1905 return nir_intrinsic_load_local_invocation_index
;
1906 case SYSTEM_VALUE_WORK_GROUP_ID
:
1907 return nir_intrinsic_load_work_group_id
;
1908 case SYSTEM_VALUE_NUM_WORK_GROUPS
:
1909 return nir_intrinsic_load_num_work_groups
;
1910 case SYSTEM_VALUE_PRIMITIVE_ID
:
1911 return nir_intrinsic_load_primitive_id
;
1912 case SYSTEM_VALUE_TESS_COORD
:
1913 return nir_intrinsic_load_tess_coord
;
1914 case SYSTEM_VALUE_TESS_LEVEL_OUTER
:
1915 return nir_intrinsic_load_tess_level_outer
;
1916 case SYSTEM_VALUE_TESS_LEVEL_INNER
:
1917 return nir_intrinsic_load_tess_level_inner
;
1918 case SYSTEM_VALUE_VERTICES_IN
:
1919 return nir_intrinsic_load_patch_vertices_in
;
1920 case SYSTEM_VALUE_HELPER_INVOCATION
:
1921 return nir_intrinsic_load_helper_invocation
;
1922 case SYSTEM_VALUE_VIEW_INDEX
:
1923 return nir_intrinsic_load_view_index
;
1924 case SYSTEM_VALUE_SUBGROUP_SIZE
:
1925 return nir_intrinsic_load_subgroup_size
;
1926 case SYSTEM_VALUE_SUBGROUP_INVOCATION
:
1927 return nir_intrinsic_load_subgroup_invocation
;
1928 case SYSTEM_VALUE_SUBGROUP_EQ_MASK
:
1929 return nir_intrinsic_load_subgroup_eq_mask
;
1930 case SYSTEM_VALUE_SUBGROUP_GE_MASK
:
1931 return nir_intrinsic_load_subgroup_ge_mask
;
1932 case SYSTEM_VALUE_SUBGROUP_GT_MASK
:
1933 return nir_intrinsic_load_subgroup_gt_mask
;
1934 case SYSTEM_VALUE_SUBGROUP_LE_MASK
:
1935 return nir_intrinsic_load_subgroup_le_mask
;
1936 case SYSTEM_VALUE_SUBGROUP_LT_MASK
:
1937 return nir_intrinsic_load_subgroup_lt_mask
;
1939 unreachable("system value does not directly correspond to intrinsic");
1944 nir_system_value_from_intrinsic(nir_intrinsic_op intrin
)
1947 case nir_intrinsic_load_vertex_id
:
1948 return SYSTEM_VALUE_VERTEX_ID
;
1949 case nir_intrinsic_load_instance_id
:
1950 return SYSTEM_VALUE_INSTANCE_ID
;
1951 case nir_intrinsic_load_draw_id
:
1952 return SYSTEM_VALUE_DRAW_ID
;
1953 case nir_intrinsic_load_base_instance
:
1954 return SYSTEM_VALUE_BASE_INSTANCE
;
1955 case nir_intrinsic_load_vertex_id_zero_base
:
1956 return SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
;
1957 case nir_intrinsic_load_base_vertex
:
1958 return SYSTEM_VALUE_BASE_VERTEX
;
1959 case nir_intrinsic_load_invocation_id
:
1960 return SYSTEM_VALUE_INVOCATION_ID
;
1961 case nir_intrinsic_load_front_face
:
1962 return SYSTEM_VALUE_FRONT_FACE
;
1963 case nir_intrinsic_load_sample_id
:
1964 return SYSTEM_VALUE_SAMPLE_ID
;
1965 case nir_intrinsic_load_sample_pos
:
1966 return SYSTEM_VALUE_SAMPLE_POS
;
1967 case nir_intrinsic_load_sample_mask_in
:
1968 return SYSTEM_VALUE_SAMPLE_MASK_IN
;
1969 case nir_intrinsic_load_local_invocation_id
:
1970 return SYSTEM_VALUE_LOCAL_INVOCATION_ID
;
1971 case nir_intrinsic_load_local_invocation_index
:
1972 return SYSTEM_VALUE_LOCAL_INVOCATION_INDEX
;
1973 case nir_intrinsic_load_num_work_groups
:
1974 return SYSTEM_VALUE_NUM_WORK_GROUPS
;
1975 case nir_intrinsic_load_work_group_id
:
1976 return SYSTEM_VALUE_WORK_GROUP_ID
;
1977 case nir_intrinsic_load_primitive_id
:
1978 return SYSTEM_VALUE_PRIMITIVE_ID
;
1979 case nir_intrinsic_load_tess_coord
:
1980 return SYSTEM_VALUE_TESS_COORD
;
1981 case nir_intrinsic_load_tess_level_outer
:
1982 return SYSTEM_VALUE_TESS_LEVEL_OUTER
;
1983 case nir_intrinsic_load_tess_level_inner
:
1984 return SYSTEM_VALUE_TESS_LEVEL_INNER
;
1985 case nir_intrinsic_load_patch_vertices_in
:
1986 return SYSTEM_VALUE_VERTICES_IN
;
1987 case nir_intrinsic_load_helper_invocation
:
1988 return SYSTEM_VALUE_HELPER_INVOCATION
;
1989 case nir_intrinsic_load_view_index
:
1990 return SYSTEM_VALUE_VIEW_INDEX
;
1991 case SYSTEM_VALUE_SUBGROUP_SIZE
:
1992 return nir_intrinsic_load_subgroup_size
;
1993 case SYSTEM_VALUE_SUBGROUP_INVOCATION
:
1994 return nir_intrinsic_load_subgroup_invocation
;
1995 case nir_intrinsic_load_subgroup_eq_mask
:
1996 return SYSTEM_VALUE_SUBGROUP_EQ_MASK
;
1997 case nir_intrinsic_load_subgroup_ge_mask
:
1998 return SYSTEM_VALUE_SUBGROUP_GE_MASK
;
1999 case nir_intrinsic_load_subgroup_gt_mask
:
2000 return SYSTEM_VALUE_SUBGROUP_GT_MASK
;
2001 case nir_intrinsic_load_subgroup_le_mask
:
2002 return SYSTEM_VALUE_SUBGROUP_LE_MASK
;
2003 case nir_intrinsic_load_subgroup_lt_mask
:
2004 return SYSTEM_VALUE_SUBGROUP_LT_MASK
;
2006 unreachable("intrinsic doesn't produce a system value");