2 * Copyright 2010 Christoph Bumiller
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 shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
18 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
19 * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 #define NOUVEAU_DEBUG 1
26 #include "nvc0_program.h"
29 nvc0_insn_can_load(struct nv_instruction
*nvi
, int s
,
30 struct nv_instruction
*ld
)
34 if (ld
->opcode
== NV_OP_MOV
&& ld
->src
[0]->value
->reg
.file
== NV_FILE_IMM
) {
35 if (s
> 1 || !(nvc0_op_info_table
[nvi
->opcode
].immediate
& (1 << s
)))
37 if (!(nvc0_op_info_table
[nvi
->opcode
].immediate
& 4))
38 if (ld
->src
[0]->value
->reg
.imm
.u32
& 0xfff)
41 if (!(nvc0_op_info_table
[nvi
->opcode
].memory
& (1 << s
)))
44 if (ld
->indirect
>= 0)
47 for (i
= 0; i
< 3 && nvi
->src
[i
]; ++i
)
48 if (nvi
->src
[i
]->value
->reg
.file
== NV_FILE_IMM
)
54 /* Return whether this instruction can be executed conditionally. */
56 nvc0_insn_is_predicateable(struct nv_instruction
*nvi
)
60 if (!nv_op_predicateable(nvi
->opcode
))
62 if (nvi
->predicate
>= 0)
64 for (s
= 0; s
< 4 && nvi
->src
[s
]; ++s
)
65 if (nvi
->src
[s
]->value
->reg
.file
== NV_FILE_IMM
)
71 nvc0_insn_refcount(struct nv_instruction
*nvi
)
75 for (i
= 0; i
< 5 && nvi
->def
[i
]; ++i
) {
78 rc
+= nvi
->def
[i
]->refc
;
84 nvc0_pc_replace_value(struct nv_pc
*pc
,
85 struct nv_value
*old_val
,
86 struct nv_value
*new_val
)
90 if (old_val
== new_val
)
93 for (i
= 0, n
= 0; i
< pc
->num_refs
; ++i
) {
94 if (pc
->refs
[i
]->value
== old_val
) {
96 for (s
= 0; s
< 6 && pc
->refs
[i
]->insn
->src
[s
]; ++s
)
97 if (pc
->refs
[i
]->insn
->src
[s
] == pc
->refs
[i
])
100 nv_reference(pc
, pc
->refs
[i
]->insn
, s
, new_val
);
107 nvc0_pc_find_constant(struct nv_ref
*ref
)
109 struct nv_value
*src
;
115 while (src
->insn
&& src
->insn
->opcode
== NV_OP_MOV
) {
116 assert(!src
->insn
->src
[0]->mod
);
117 src
= src
->insn
->src
[0]->value
;
119 if ((src
->reg
.file
== NV_FILE_IMM
) ||
121 src
->insn
->opcode
== NV_OP_LD
&&
122 src
->insn
->src
[0]->value
->reg
.file
>= NV_FILE_MEM_C(0) &&
123 src
->insn
->src
[0]->value
->reg
.file
<= NV_FILE_MEM_C(15)))
129 nvc0_pc_find_immediate(struct nv_ref
*ref
)
131 struct nv_value
*src
= nvc0_pc_find_constant(ref
);
133 return (src
&& src
->reg
.file
== NV_FILE_IMM
) ? src
: NULL
;
137 nv_pc_free_refs(struct nv_pc
*pc
)
140 for (i
= 0; i
< pc
->num_refs
; i
+= 64)
146 edge_name(ubyte type
)
149 case CFG_EDGE_FORWARD
: return "forward";
150 case CFG_EDGE_BACK
: return "back";
151 case CFG_EDGE_LOOP_ENTER
: return "loop";
152 case CFG_EDGE_LOOP_LEAVE
: return "break";
153 case CFG_EDGE_FAKE
: return "fake";
160 nvc0_pc_pass_in_order(struct nv_basic_block
*root
, nv_pc_pass_func f
,
163 struct nv_basic_block
*bb
[64], *bbb
[16], *b
;
174 for (j
= 1; j
>= 0; --j
) {
178 switch (b
->out_kind
[j
]) {
181 case CFG_EDGE_FORWARD
:
183 if (++b
->out
[j
]->priv
== b
->out
[j
]->num_in
)
186 case CFG_EDGE_LOOP_ENTER
:
189 case CFG_EDGE_LOOP_LEAVE
:
190 bbb
[pp
++] = b
->out
[j
];
203 bb
[pp
- 1] = bbb
[pp
- 1];
209 nv_do_print_function(void *priv
, struct nv_basic_block
*b
)
211 struct nv_instruction
*i
;
213 debug_printf("=== BB %i ", b
->id
);
215 debug_printf("[%s -> %i] ", edge_name(b
->out_kind
[0]), b
->out
[0]->id
);
217 debug_printf("[%s -> %i] ", edge_name(b
->out_kind
[1]), b
->out
[1]->id
);
218 debug_printf("===\n");
223 for (; i
; i
= i
->next
)
224 nvc0_print_instruction(i
);
228 nvc0_print_function(struct nv_basic_block
*root
)
230 if (root
->subroutine
)
231 debug_printf("SUBROUTINE %i\n", root
->subroutine
);
233 debug_printf("MAIN\n");
235 nvc0_pc_pass_in_order(root
, nv_do_print_function
, root
);
239 nvc0_print_program(struct nv_pc
*pc
)
242 for (i
= 0; i
< pc
->num_subroutines
+ 1; ++i
)
244 nvc0_print_function(pc
->root
[i
]);
247 #if NOUVEAU_DEBUG > 1
249 nv_do_print_cfgraph(struct nv_pc
*pc
, FILE *f
, struct nv_basic_block
*b
)
253 b
->pass_seq
= pc
->pass_seq
;
255 fprintf(f
, "\t%i [shape=box]\n", b
->id
);
257 for (i
= 0; i
< 2; ++i
) {
260 switch (b
->out_kind
[i
]) {
261 case CFG_EDGE_FORWARD
:
262 fprintf(f
, "\t%i -> %i;\n", b
->id
, b
->out
[i
]->id
);
264 case CFG_EDGE_LOOP_ENTER
:
265 fprintf(f
, "\t%i -> %i [color=green];\n", b
->id
, b
->out
[i
]->id
);
267 case CFG_EDGE_LOOP_LEAVE
:
268 fprintf(f
, "\t%i -> %i [color=red];\n", b
->id
, b
->out
[i
]->id
);
271 fprintf(f
, "\t%i -> %i;\n", b
->id
, b
->out
[i
]->id
);
274 fprintf(f
, "\t%i -> %i [style=dotted];\n", b
->id
, b
->out
[i
]->id
);
280 if (b
->out
[i
]->pass_seq
< pc
->pass_seq
)
281 nv_do_print_cfgraph(pc
, f
, b
->out
[i
]);
285 /* Print the control flow graph of subroutine @subr (0 == MAIN) to a file. */
287 nv_print_cfgraph(struct nv_pc
*pc
, const char *filepath
, int subr
)
291 f
= fopen(filepath
, "a");
295 fprintf(f
, "digraph G {\n");
299 nv_do_print_cfgraph(pc
, f
, pc
->root
[subr
]);
308 nvc0_pc_print_binary(struct nv_pc
*pc
)
312 NOUVEAU_DBG("nvc0_pc_print_binary(%u ops)\n", pc
->emit_size
/ 8);
314 for (i
= 0; i
< pc
->emit_size
/ 4; i
+= 2) {
315 debug_printf("0x%08x ", pc
->emit
[i
+ 0]);
316 debug_printf("0x%08x ", pc
->emit
[i
+ 1]);
324 nvc0_emit_program(struct nv_pc
*pc
)
326 uint32_t *code
= pc
->emit
;
329 NOUVEAU_DBG("emitting program: size = %u\n", pc
->emit_size
);
331 for (n
= 0; n
< pc
->num_blocks
; ++n
) {
332 struct nv_instruction
*i
;
333 struct nv_basic_block
*b
= pc
->bb_list
[n
];
335 for (i
= b
->entry
; i
; i
= i
->next
) {
336 nvc0_emit_instruction(pc
, i
);
341 assert(pc
->emit
== &code
[pc
->emit_size
/ 4]);
343 pc
->emit
[0] = 0x00001de7;
344 pc
->emit
[1] = 0x80000000;
350 nvc0_pc_print_binary(pc
);
352 debug_printf("not printing binary\n");
358 nvc0_generate_code(struct nvc0_translation_info
*ti
)
364 pc
= CALLOC_STRUCT(nv_pc
);
368 pc
->is_fragprog
= ti
->prog
->type
== PIPE_SHADER_FRAGMENT
;
370 pc
->root
= CALLOC(ti
->num_subrs
+ 1, sizeof(pc
->root
[0]));
375 pc
->num_subroutines
= ti
->num_subrs
;
377 ret
= nvc0_tgsi_to_nc(pc
, ti
);
380 #if NOUVEAU_DEBUG > 1
381 nvc0_print_program(pc
);
384 pc
->opt_reload_elim
= ti
->require_stores
? FALSE
: TRUE
;
387 ret
= nvc0_pc_exec_pass0(pc
);
391 nvc0_print_program(pc
);
394 /* register allocation */
395 ret
= nvc0_pc_exec_pass1(pc
);
398 #if NOUVEAU_DEBUG > 1
399 nv_print_program(pc
);
400 nv_print_cfgraph(pc
, "nvc0_shader_cfgraph.dot", 0);
403 /* prepare for emission */
404 ret
= nvc0_pc_exec_pass2(pc
);
407 assert(!(pc
->emit_size
% 8));
409 pc
->emit
= CALLOC(pc
->emit_size
/ 4 + 2, 4);
414 ret
= nvc0_emit_program(pc
);
418 ti
->prog
->code
= pc
->emit
;
419 ti
->prog
->code_base
= 0;
420 ti
->prog
->code_size
= pc
->emit_size
;
421 ti
->prog
->parm_size
= 0;
423 ti
->prog
->max_gpr
= MAX2(4, pc
->max_reg
[NV_FILE_GPR
] + 1);
425 ti
->prog
->relocs
= pc
->reloc_entries
;
426 ti
->prog
->num_relocs
= pc
->num_relocs
;
428 NOUVEAU_DBG("SHADER TRANSLATION - %s\n", ret
? "failure" : "success");
433 for (i
= 0; i
< pc
->num_blocks
; ++i
)
434 FREE(pc
->bb_list
[i
]);
438 /* on success, these will be referenced by struct nvc0_program */
443 if (pc
->reloc_entries
)
444 FREE(pc
->reloc_entries
);
451 nvbb_insert_phi(struct nv_basic_block
*b
, struct nv_instruction
*i
)
458 assert(!b
->entry
->prev
&& b
->exit
);
466 if (b
->entry
->opcode
== NV_OP_PHI
) { /* insert after entry */
467 assert(b
->entry
== b
->exit
);
472 } else { /* insert before entry */
473 assert(b
->entry
->prev
&& b
->exit
);
475 i
->prev
= b
->entry
->prev
;
483 nvc0_insn_append(struct nv_basic_block
*b
, struct nv_instruction
*i
)
485 if (i
->opcode
== NV_OP_PHI
) {
486 nvbb_insert_phi(b
, i
);
495 if (i
->prev
&& i
->prev
->opcode
== NV_OP_PHI
)
500 b
->num_instructions
++;
504 nvc0_insn_insert_after(struct nv_instruction
*at
, struct nv_instruction
*ni
)
507 nvc0_insn_append(at
->bb
, ni
);
517 nvc0_insn_delete(struct nv_instruction
*nvi
)
519 struct nv_basic_block
*b
= nvi
->bb
;
522 /* debug_printf("REM: "); nv_print_instruction(nvi); */
524 for (s
= 0; s
< 6 && nvi
->src
[s
]; ++s
)
525 nv_reference(NULL
, nvi
, s
, NULL
);
528 nvi
->next
->prev
= nvi
->prev
;
530 assert(nvi
== b
->exit
);
535 nvi
->prev
->next
= nvi
->next
;
537 if (nvi
== b
->entry
) {
538 /* PHIs don't get hooked to b->entry */
539 b
->entry
= nvi
->next
;
540 assert(!nvi
->prev
|| nvi
->prev
->opcode
== NV_OP_PHI
);
544 if (nvi
->opcode
!= NV_OP_PHI
)
545 NOUVEAU_DBG("NOTE: b->phi points to non-PHI instruction\n");
548 if (!nvi
->next
|| nvi
->next
->opcode
!= NV_OP_PHI
)
556 nvc0_insns_permute(struct nv_instruction
*i1
, struct nv_instruction
*i2
)
558 struct nv_basic_block
*b
= i1
->bb
;
560 assert(i1
->opcode
!= NV_OP_PHI
&&
561 i2
->opcode
!= NV_OP_PHI
);
562 assert(i1
->next
== i2
);
582 nvc0_bblock_attach(struct nv_basic_block
*parent
,
583 struct nv_basic_block
*b
, ubyte edge_kind
)
585 assert(b
->num_in
< 8);
587 if (parent
->out
[0]) {
588 assert(!parent
->out
[1]);
590 parent
->out_kind
[1] = edge_kind
;
593 parent
->out_kind
[0] = edge_kind
;
596 b
->in
[b
->num_in
] = parent
;
597 b
->in_kind
[b
->num_in
++] = edge_kind
;
600 /* NOTE: all BRKs are treated as conditional, so there are 2 outgoing BBs */
603 nvc0_bblock_dominated_by(struct nv_basic_block
*b
, struct nv_basic_block
*d
)
610 for (j
= 0; j
< b
->num_in
; ++j
)
611 if ((b
->in_kind
[j
] != CFG_EDGE_BACK
) &&
612 !nvc0_bblock_dominated_by(b
->in
[j
], d
))
615 return j
? TRUE
: FALSE
;
618 /* check if @bf (future) can be reached from @bp (past), stop at @bt */
620 nvc0_bblock_reachable_by(struct nv_basic_block
*bf
, struct nv_basic_block
*bp
,
621 struct nv_basic_block
*bt
)
623 struct nv_basic_block
*q
[NV_PC_MAX_BASIC_BLOCKS
], *b
;
637 assert(n
<= (1024 - 2));
639 for (i
= 0; i
< 2; ++i
) {
640 if (b
->out
[i
] && !IS_WALL_EDGE(b
->out_kind
[i
]) && !b
->out
[i
]->priv
) {
646 for (--n
; n
>= 0; --n
)
652 static struct nv_basic_block
*
653 nvbb_find_dom_frontier(struct nv_basic_block
*b
, struct nv_basic_block
*df
)
655 struct nv_basic_block
*out
;
658 if (!nvc0_bblock_dominated_by(df
, b
)) {
659 for (i
= 0; i
< df
->num_in
; ++i
) {
660 if (df
->in_kind
[i
] == CFG_EDGE_BACK
)
662 if (nvc0_bblock_dominated_by(df
->in
[i
], b
))
666 for (i
= 0; i
< 2 && df
->out
[i
]; ++i
) {
667 if (df
->out_kind
[i
] == CFG_EDGE_BACK
)
669 if ((out
= nvbb_find_dom_frontier(b
, df
->out
[i
])))
675 struct nv_basic_block
*
676 nvc0_bblock_dom_frontier(struct nv_basic_block
*b
)
678 struct nv_basic_block
*df
;
681 for (i
= 0; i
< 2 && b
->out
[i
]; ++i
)
682 if ((df
= nvbb_find_dom_frontier(b
, b
->out
[i
])))