2 * Copyright (c) 2017 Lima Project
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, sub license,
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
12 * next paragraph) shall be included in all copies or substantial portions
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 NON-INFRINGEMENT. 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
21 * DEALINGS IN THE SOFTWARE.
25 #include "util/u_math.h"
26 #include "util/ralloc.h"
27 #include "util/bitscan.h"
31 const ppir_op_info ppir_op_infos
[] = {
35 PPIR_INSTR_SLOT_ALU_SCL_ADD
, PPIR_INSTR_SLOT_ALU_SCL_MUL
,
36 PPIR_INSTR_SLOT_ALU_VEC_ADD
, PPIR_INSTR_SLOT_ALU_VEC_MUL
,
52 PPIR_INSTR_SLOT_ALU_SCL_MUL
, PPIR_INSTR_SLOT_ALU_VEC_MUL
,
59 PPIR_INSTR_SLOT_ALU_SCL_ADD
, PPIR_INSTR_SLOT_ALU_VEC_ADD
,
66 PPIR_INSTR_SLOT_ALU_VEC_ADD
, PPIR_INSTR_SLOT_END
72 PPIR_INSTR_SLOT_ALU_VEC_ADD
, PPIR_INSTR_SLOT_END
78 PPIR_INSTR_SLOT_ALU_COMBINE
, PPIR_INSTR_SLOT_END
84 PPIR_INSTR_SLOT_ALU_COMBINE
, PPIR_INSTR_SLOT_END
90 PPIR_INSTR_SLOT_ALU_COMBINE
, PPIR_INSTR_SLOT_END
96 PPIR_INSTR_SLOT_ALU_COMBINE
, PPIR_INSTR_SLOT_END
102 PPIR_INSTR_SLOT_ALU_COMBINE
, PPIR_INSTR_SLOT_END
108 PPIR_INSTR_SLOT_ALU_COMBINE
, PPIR_INSTR_SLOT_END
114 PPIR_INSTR_SLOT_ALU_SCL_ADD
, PPIR_INSTR_SLOT_ALU_SCL_MUL
,
115 PPIR_INSTR_SLOT_ALU_VEC_ADD
, PPIR_INSTR_SLOT_ALU_VEC_MUL
,
122 PPIR_INSTR_SLOT_ALU_SCL_ADD
, PPIR_INSTR_SLOT_ALU_SCL_MUL
,
123 PPIR_INSTR_SLOT_ALU_VEC_ADD
, PPIR_INSTR_SLOT_ALU_VEC_MUL
,
130 PPIR_INSTR_SLOT_ALU_SCL_ADD
, PPIR_INSTR_SLOT_ALU_VEC_ADD
,
137 PPIR_INSTR_SLOT_ALU_SCL_ADD
, PPIR_INSTR_SLOT_ALU_VEC_ADD
,
144 PPIR_INSTR_SLOT_ALU_SCL_ADD
, PPIR_INSTR_SLOT_ALU_VEC_ADD
,
151 PPIR_INSTR_SLOT_ALU_SCL_ADD
, PPIR_INSTR_SLOT_ALU_VEC_ADD
,
158 PPIR_INSTR_SLOT_ALU_SCL_ADD
, PPIR_INSTR_SLOT_ALU_VEC_ADD
,
165 PPIR_INSTR_SLOT_ALU_SCL_MUL
, PPIR_INSTR_SLOT_ALU_VEC_MUL
,
172 PPIR_INSTR_SLOT_ALU_SCL_MUL
, PPIR_INSTR_SLOT_ALU_VEC_MUL
,
179 PPIR_INSTR_SLOT_ALU_SCL_MUL
, PPIR_INSTR_SLOT_ALU_VEC_MUL
,
186 PPIR_INSTR_SLOT_ALU_SCL_MUL
, PPIR_INSTR_SLOT_ALU_VEC_MUL
,
199 PPIR_INSTR_SLOT_ALU_SCL_MUL
, PPIR_INSTR_SLOT_ALU_SCL_ADD
,
200 PPIR_INSTR_SLOT_ALU_VEC_MUL
, PPIR_INSTR_SLOT_ALU_VEC_ADD
,
207 PPIR_INSTR_SLOT_ALU_SCL_MUL
, PPIR_INSTR_SLOT_ALU_SCL_ADD
,
208 PPIR_INSTR_SLOT_ALU_VEC_MUL
, PPIR_INSTR_SLOT_ALU_VEC_ADD
,
215 PPIR_INSTR_SLOT_ALU_SCL_MUL
, PPIR_INSTR_SLOT_ALU_SCL_ADD
,
216 PPIR_INSTR_SLOT_ALU_VEC_MUL
, PPIR_INSTR_SLOT_ALU_VEC_ADD
,
223 PPIR_INSTR_SLOT_ALU_SCL_MUL
, PPIR_INSTR_SLOT_ALU_SCL_ADD
,
224 PPIR_INSTR_SLOT_ALU_VEC_MUL
, PPIR_INSTR_SLOT_ALU_VEC_ADD
,
228 [ppir_op_sel_cond
] = {
229 /* effectively mov, but must be scheduled only to
230 * PPIR_INSTR_SLOT_ALU_SCL_MUL */
233 PPIR_INSTR_SLOT_ALU_SCL_MUL
, PPIR_INSTR_SLOT_END
239 PPIR_INSTR_SLOT_ALU_SCL_ADD
, PPIR_INSTR_SLOT_ALU_VEC_ADD
,
246 PPIR_INSTR_SLOT_ALU_COMBINE
, PPIR_INSTR_SLOT_END
249 [ppir_op_load_varying
] = {
251 .type
= ppir_node_type_load
,
253 PPIR_INSTR_SLOT_VARYING
, PPIR_INSTR_SLOT_END
256 [ppir_op_load_coords
] = {
258 .type
= ppir_node_type_load
,
260 PPIR_INSTR_SLOT_VARYING
, PPIR_INSTR_SLOT_END
263 [ppir_op_load_fragcoord
] = {
264 .name
= "ld_fragcoord",
265 .type
= ppir_node_type_load
,
267 PPIR_INSTR_SLOT_VARYING
, PPIR_INSTR_SLOT_END
270 [ppir_op_load_pointcoord
] = {
271 .name
= "ld_pointcoord",
272 .type
= ppir_node_type_load
,
274 PPIR_INSTR_SLOT_VARYING
, PPIR_INSTR_SLOT_END
277 [ppir_op_load_frontface
] = {
278 .name
= "ld_frontface",
279 .type
= ppir_node_type_load
,
281 PPIR_INSTR_SLOT_VARYING
, PPIR_INSTR_SLOT_END
284 [ppir_op_load_uniform
] = {
286 .type
= ppir_node_type_load
,
288 PPIR_INSTR_SLOT_UNIFORM
, PPIR_INSTR_SLOT_END
291 [ppir_op_load_texture
] = {
293 .type
= ppir_node_type_load_texture
,
295 PPIR_INSTR_SLOT_TEXLD
, PPIR_INSTR_SLOT_END
298 [ppir_op_load_temp
] = {
300 .type
= ppir_node_type_load
,
302 PPIR_INSTR_SLOT_UNIFORM
, PPIR_INSTR_SLOT_END
307 .type
= ppir_node_type_const
,
309 [ppir_op_store_color
] = {
311 .type
= ppir_node_type_alu
,
313 PPIR_INSTR_SLOT_ALU_VEC_ADD
, PPIR_INSTR_SLOT_ALU_VEC_MUL
,
317 [ppir_op_store_temp
] = {
319 .type
= ppir_node_type_store
,
321 PPIR_INSTR_SLOT_STORE_TEMP
, PPIR_INSTR_SLOT_END
324 [ppir_op_discard
] = {
326 .type
= ppir_node_type_discard
,
328 PPIR_INSTR_SLOT_BRANCH
, PPIR_INSTR_SLOT_END
333 .type
= ppir_node_type_branch
,
335 PPIR_INSTR_SLOT_BRANCH
, PPIR_INSTR_SLOT_END
340 .type
= ppir_node_type_alu
,
346 void *ppir_node_create(ppir_block
*block
, ppir_op op
, int index
, unsigned mask
)
348 ppir_compiler
*comp
= block
->comp
;
349 static const int node_size
[] = {
350 [ppir_node_type_alu
] = sizeof(ppir_alu_node
),
351 [ppir_node_type_const
] = sizeof(ppir_const_node
),
352 [ppir_node_type_load
] = sizeof(ppir_load_node
),
353 [ppir_node_type_store
] = sizeof(ppir_store_node
),
354 [ppir_node_type_load_texture
] = sizeof(ppir_load_texture_node
),
355 [ppir_node_type_discard
] = sizeof(ppir_discard_node
),
356 [ppir_node_type_branch
] = sizeof(ppir_branch_node
),
359 ppir_node_type type
= ppir_op_infos
[op
].type
;
360 int size
= node_size
[type
];
361 ppir_node
*node
= rzalloc_size(block
, size
);
365 list_inithead(&node
->succ_list
);
366 list_inithead(&node
->pred_list
);
370 /* reg has 4 slots for each componemt write node */
372 comp
->var_nodes
[(index
<< 2) + comp
->reg_base
+ u_bit_scan(&mask
)] = node
;
373 snprintf(node
->name
, sizeof(node
->name
), "reg%d", index
);
375 comp
->var_nodes
[index
] = node
;
376 snprintf(node
->name
, sizeof(node
->name
), "ssa%d", index
);
380 snprintf(node
->name
, sizeof(node
->name
), "new");
384 node
->index
= comp
->cur_index
++;
390 void ppir_node_add_dep(ppir_node
*succ
, ppir_node
*pred
,
393 /* don't add dep for two nodes from different block */
394 if (succ
->block
!= pred
->block
)
397 /* don't add duplicated dep */
398 ppir_node_foreach_pred(succ
, dep
) {
399 if (dep
->pred
== pred
)
403 ppir_dep
*dep
= ralloc(succ
, ppir_dep
);
407 list_addtail(&dep
->pred_link
, &succ
->pred_list
);
408 list_addtail(&dep
->succ_link
, &pred
->succ_list
);
411 void ppir_node_remove_dep(ppir_dep
*dep
)
413 list_del(&dep
->succ_link
);
414 list_del(&dep
->pred_link
);
418 static void _ppir_node_replace_child(ppir_src
*src
, ppir_node
*old_child
, ppir_node
*new_child
)
420 ppir_dest
*od
= ppir_node_get_dest(old_child
);
421 if (ppir_node_target_equal(src
, od
)) {
422 ppir_node_target_assign(src
, new_child
);
426 void ppir_node_replace_child(ppir_node
*parent
, ppir_node
*old_child
, ppir_node
*new_child
)
428 switch (parent
->type
) {
429 case ppir_node_type_alu
:
431 ppir_alu_node
*alu
= ppir_node_to_alu(parent
);
432 for (int i
= 0; i
< alu
->num_src
; i
++)
433 _ppir_node_replace_child(alu
->src
+ i
, old_child
, new_child
);
436 case ppir_node_type_branch
:
438 ppir_branch_node
*branch
= ppir_node_to_branch(parent
);
439 for (int i
= 0; i
< 2; i
++)
440 _ppir_node_replace_child(branch
->src
+ i
, old_child
, new_child
);
443 case ppir_node_type_load
:
445 ppir_load_node
*load
= ppir_node_to_load(parent
);
446 _ppir_node_replace_child(&load
->src
, old_child
, new_child
);
449 case ppir_node_type_load_texture
:
451 ppir_load_texture_node
*load_texture
= ppir_node_to_load_texture(parent
);
452 _ppir_node_replace_child(&load_texture
->src_coords
, old_child
, new_child
);
455 case ppir_node_type_store
:
457 ppir_store_node
*store
= ppir_node_to_store(parent
);
458 _ppir_node_replace_child(&store
->src
, old_child
, new_child
);
462 ppir_debug("unknown node type in %s\n", __func__
);
467 void ppir_node_replace_pred(ppir_dep
*dep
, ppir_node
*new_pred
)
469 list_del(&dep
->succ_link
);
470 dep
->pred
= new_pred
;
471 list_addtail(&dep
->succ_link
, &new_pred
->succ_list
);
474 ppir_dep
*ppir_dep_for_pred(ppir_node
*node
, ppir_node
*pred
)
479 if (node
->block
!= pred
->block
)
482 ppir_node_foreach_pred(node
, dep
) {
483 if (dep
->pred
== pred
)
489 void ppir_node_replace_all_succ(ppir_node
*dst
, ppir_node
*src
)
491 ppir_node_foreach_succ_safe(src
, dep
) {
492 ppir_node_replace_pred(dep
, dst
);
493 ppir_node_replace_child(dep
->succ
, src
, dst
);
497 void ppir_node_delete(ppir_node
*node
)
499 ppir_node_foreach_succ_safe(node
, dep
)
500 ppir_node_remove_dep(dep
);
502 ppir_node_foreach_pred_safe(node
, dep
)
503 ppir_node_remove_dep(dep
);
505 list_del(&node
->list
);
509 static void ppir_node_print_dest(ppir_dest
*dest
)
511 switch (dest
->type
) {
512 case ppir_target_ssa
:
513 printf("ssa%d", dest
->ssa
.index
);
515 case ppir_target_pipeline
:
516 printf("pipeline %d", dest
->pipeline
);
518 case ppir_target_register
:
519 printf("reg %d", dest
->reg
->index
);
524 static void ppir_node_print_src(ppir_src
*src
)
527 case ppir_target_ssa
: {
529 printf("ssa node %d", src
->node
->index
);
531 printf("ssa idx %d", src
->ssa
? src
->ssa
->index
: -1);
534 case ppir_target_pipeline
:
536 printf("pipeline %d node %d", src
->pipeline
, src
->node
->index
);
538 printf("pipeline %d", src
->pipeline
);
540 case ppir_target_register
:
541 printf("reg %d", src
->reg
->index
);
546 static void ppir_node_print_node(ppir_node
*node
, int space
)
548 for (int i
= 0; i
< space
; i
++)
551 printf("%s%d: %s %s: ", node
->printed
&& !ppir_node_is_leaf(node
) ? "+" : "",
552 node
->index
, ppir_op_infos
[node
->op
].name
, node
->name
);
554 ppir_dest
*dest
= ppir_node_get_dest(node
);
557 ppir_node_print_dest(dest
);
560 if (ppir_node_get_src_num(node
) > 0) {
563 for (int i
= 0; i
< ppir_node_get_src_num(node
); i
++) {
564 ppir_node_print_src(ppir_node_get_src(node
, i
));
565 if (i
!= (ppir_node_get_src_num(node
) - 1))
570 if (!node
->printed
) {
571 ppir_node_foreach_pred(node
, dep
) {
572 ppir_node
*pred
= dep
->pred
;
573 ppir_node_print_node(pred
, space
+ 2);
576 node
->printed
= true;
580 void ppir_node_print_prog(ppir_compiler
*comp
)
582 if (!(lima_debug
& LIMA_DEBUG_PP
))
585 list_for_each_entry(ppir_block
, block
, &comp
->block_list
, list
) {
586 list_for_each_entry(ppir_node
, node
, &block
->node_list
, list
) {
587 node
->printed
= false;
591 printf("========prog========\n");
592 list_for_each_entry(ppir_block
, block
, &comp
->block_list
, list
) {
593 printf("-------block %3d-------\n", block
->index
);
594 list_for_each_entry(ppir_node
, node
, &block
->node_list
, list
) {
595 if (ppir_node_is_root(node
))
596 ppir_node_print_node(node
, 0);
599 printf("====================\n");
602 static ppir_node
*ppir_node_clone_const(ppir_block
*block
, ppir_node
*node
)
604 ppir_const_node
*cnode
= ppir_node_to_const(node
);
605 ppir_const_node
*new_cnode
= ppir_node_create(block
, ppir_op_const
, -1, 0);
610 list_addtail(&new_cnode
->node
.list
, &block
->node_list
);
612 new_cnode
->constant
.num
= cnode
->constant
.num
;
613 for (int i
= 0; i
< cnode
->constant
.num
; i
++) {
614 new_cnode
->constant
.value
[i
] = cnode
->constant
.value
[i
];
616 new_cnode
->dest
.type
= ppir_target_ssa
;
617 new_cnode
->dest
.ssa
.num_components
= cnode
->dest
.ssa
.num_components
;
618 new_cnode
->dest
.ssa
.live_in
= INT_MAX
;
619 new_cnode
->dest
.ssa
.live_out
= 0;
620 new_cnode
->dest
.write_mask
= cnode
->dest
.write_mask
;
622 return &new_cnode
->node
;
626 ppir_node_clone_load(ppir_block
*block
, ppir_node
*node
)
628 ppir_load_node
*load_node
= ppir_node_to_load(node
);
629 ppir_load_node
*new_lnode
= ppir_node_create(block
, node
->op
, -1, 0);
634 list_addtail(&new_lnode
->node
.list
, &block
->node_list
);
636 new_lnode
->num_components
= load_node
->num_components
;
637 new_lnode
->index
= load_node
->index
;
639 ppir_dest
*dest
= ppir_node_get_dest(node
);
640 new_lnode
->dest
= *dest
;
642 ppir_src
*src
= ppir_node_get_src(node
, 0);
644 new_lnode
->num_src
= 1;
646 case ppir_target_ssa
:
647 ppir_node_target_assign(&new_lnode
->src
, src
->node
);
648 ppir_node_add_dep(&new_lnode
->node
, src
->node
, ppir_dep_src
);
650 case ppir_target_register
:
651 new_lnode
->src
.type
= src
->type
;
652 new_lnode
->src
.reg
= src
->reg
;
653 new_lnode
->src
.node
= NULL
;
656 /* Load nodes can't consume pipeline registers */
661 return &new_lnode
->node
;
664 ppir_node
*ppir_node_clone(ppir_block
*block
, ppir_node
*node
)
668 return ppir_node_clone_const(block
, node
);
669 case ppir_op_load_uniform
:
670 case ppir_op_load_varying
:
671 case ppir_op_load_temp
:
672 case ppir_op_load_coords
:
673 return ppir_node_clone_load(block
, node
);
679 ppir_node
*ppir_node_insert_mov(ppir_node
*node
)
681 ppir_node
*move
= ppir_node_create(node
->block
, ppir_op_mov
, -1, 0);
685 ppir_dest
*dest
= ppir_node_get_dest(node
);
686 ppir_alu_node
*alu
= ppir_node_to_alu(move
);
689 ppir_node_target_assign(alu
->src
, node
);
691 for (int s
= 0; s
< 4; s
++)
692 alu
->src
->swizzle
[s
] = s
;
694 ppir_node_replace_all_succ(move
, node
);
695 ppir_node_add_dep(move
, node
, ppir_dep_src
);
696 list_addtail(&move
->list
, &node
->list
);
701 bool ppir_node_has_single_src_succ(ppir_node
*node
)
703 if (list_is_singular(&node
->succ_list
) &&
704 list_first_entry(&node
->succ_list
,
705 ppir_dep
, succ_link
)->type
== ppir_dep_src
)
709 ppir_node_foreach_succ(node
, dep
) {
710 if (dep
->type
!= ppir_dep_src
)