2 * Linearize - walk the statement tree (but _not_ the expressions)
3 * to generate a linear version of it and the basic blocks.
5 * NOTE! We're not interested in the actual sub-expressions yet,
6 * even though they can generate conditional branches and
7 * subroutine calls. That's all "local" behaviour.
9 * Copyright (C) 2004 Linus Torvalds
10 * Copyright (C) 2004 Christopher Li
20 #include "expression.h"
21 #include "linearize.h"
25 pseudo_t
linearize_statement(struct entrypoint
*ep
, struct statement
*stmt
);
26 pseudo_t
linearize_expression(struct entrypoint
*ep
, struct expression
*expr
);
28 static pseudo_t
add_binary_op(struct entrypoint
*ep
, struct symbol
*ctype
, int op
, pseudo_t left
, pseudo_t right
);
29 static pseudo_t
add_setval(struct entrypoint
*ep
, struct symbol
*ctype
, struct expression
*val
);
30 static pseudo_t
linearize_one_symbol(struct entrypoint
*ep
, struct symbol
*sym
);
33 static pseudo_t
add_load(struct entrypoint
*ep
, struct access_data
*);
34 static pseudo_t
linearize_initializer(struct entrypoint
*ep
, struct expression
*initializer
, struct access_data
*);
35 static pseudo_t
cast_pseudo(struct entrypoint
*ep
, pseudo_t src
, struct symbol
*from
, struct symbol
*to
);
37 struct pseudo void_pseudo
= {};
39 static struct position current_pos
;
41 ALLOCATOR(pseudo_user
, "pseudo_user");
43 static struct instruction
*alloc_instruction(int opcode
, int size
)
45 struct instruction
* insn
= __alloc_instruction(0);
46 insn
->opcode
= opcode
;
48 insn
->pos
= current_pos
;
52 static inline int type_size(struct symbol
*type
)
54 return type
? type
->bit_size
> 0 ? type
->bit_size
: 0 : 0;
57 static struct instruction
*alloc_typed_instruction(int opcode
, struct symbol
*type
)
59 struct instruction
*insn
= alloc_instruction(opcode
, type_size(type
));
64 static struct entrypoint
*alloc_entrypoint(void)
66 return __alloc_entrypoint(0);
69 static struct basic_block
*alloc_basic_block(struct entrypoint
*ep
, struct position pos
)
72 struct basic_block
*bb
= __alloc_basic_block(0);
80 static struct multijmp
*alloc_multijmp(struct basic_block
*target
, int begin
, int end
)
82 struct multijmp
*multijmp
= __alloc_multijmp(0);
83 multijmp
->target
= target
;
84 multijmp
->begin
= begin
;
89 static inline int regno(pseudo_t n
)
92 if (n
&& n
->type
== PSEUDO_REG
)
97 const char *show_pseudo(pseudo_t pseudo
)
100 static char buffer
[4][64];
108 buf
= buffer
[3 & ++n
];
109 switch(pseudo
->type
) {
111 struct symbol
*sym
= pseudo
->sym
;
112 struct expression
*expr
;
114 if (sym
->bb_target
) {
115 snprintf(buf
, 64, ".L%u", sym
->bb_target
->nr
);
119 snprintf(buf
, 64, "%s", show_ident(sym
->ident
));
122 expr
= sym
->initializer
;
123 snprintf(buf
, 64, "<anon symbol:%p>", sym
);
125 switch (expr
->type
) {
127 snprintf(buf
, 64, "<symbol value: %lld>", expr
->value
);
130 return show_string(expr
->string
);
138 i
= snprintf(buf
, 64, "%%r%d", pseudo
->nr
);
140 sprintf(buf
+i
, "(%s)", show_ident(pseudo
->ident
));
143 long long value
= pseudo
->value
;
144 if (value
> 1000 || value
< -1000)
145 snprintf(buf
, 64, "$%#llx", value
);
147 snprintf(buf
, 64, "$%lld", value
);
151 snprintf(buf
, 64, "%%arg%d", pseudo
->nr
);
154 i
= snprintf(buf
, 64, "%%phi%d", pseudo
->nr
);
156 sprintf(buf
+i
, "(%s)", show_ident(pseudo
->ident
));
159 snprintf(buf
, 64, "<bad pseudo type %d>", pseudo
->type
);
164 static const char *opcodes
[] = {
165 [OP_BADOP
] = "bad_op",
168 [OP_ENTRY
] = "<entry-point>",
174 [OP_SWITCH
] = "switch",
175 [OP_INVOKE
] = "invoke",
176 [OP_COMPUTEDGOTO
] = "jmp *",
177 [OP_UNWIND
] = "unwind",
196 [OP_AND_BOOL
] = "and-bool",
197 [OP_OR_BOOL
] = "or-bool",
199 /* Binary comparison */
200 [OP_SET_EQ
] = "seteq",
201 [OP_SET_NE
] = "setne",
202 [OP_SET_LE
] = "setle",
203 [OP_SET_GE
] = "setge",
204 [OP_SET_LT
] = "setlt",
205 [OP_SET_GT
] = "setgt",
208 [OP_SET_BE
] = "setbe",
209 [OP_SET_AE
] = "setae",
215 /* Special three-input */
219 [OP_MALLOC
] = "malloc",
221 [OP_ALLOCA
] = "alloca",
223 [OP_STORE
] = "store",
225 [OP_SYMADDR
] = "symaddr",
226 [OP_GET_ELEMENT_PTR
] = "getelem",
230 [OP_PHISOURCE
] = "phisrc",
232 [OP_SCAST
] = "scast",
233 [OP_FPCAST
] = "fpcast",
234 [OP_PTRCAST
] = "ptrcast",
235 [OP_INLINED_CALL
] = "# call",
237 [OP_VANEXT
] = "va_next",
238 [OP_VAARG
] = "va_arg",
239 [OP_SLICE
] = "slice",
243 [OP_DEATHNOTE
] = "dead",
246 /* Sparse tagging (line numbers, context, whatever) */
247 [OP_CONTEXT
] = "context",
248 [OP_RANGE
] = "range-check",
253 static char *show_asm_constraints(char *buf
, const char *sep
, struct asm_constraint_list
*list
)
255 struct asm_constraint
*entry
;
257 FOR_EACH_PTR(list
, entry
) {
258 buf
+= sprintf(buf
, "%s\"%s\"", sep
, entry
->constraint
);
260 buf
+= sprintf(buf
, " (%s)", show_pseudo(entry
->pseudo
));
262 buf
+= sprintf(buf
, " [%s]", show_ident(entry
->ident
));
264 } END_FOR_EACH_PTR(entry
);
268 static char *show_asm(char *buf
, struct instruction
*insn
)
270 struct asm_rules
*rules
= insn
->asm_rules
;
272 buf
+= sprintf(buf
, "\"%s\"", insn
->string
);
273 buf
= show_asm_constraints(buf
, "\n\t\tout: ", rules
->outputs
);
274 buf
= show_asm_constraints(buf
, "\n\t\tin: ", rules
->inputs
);
275 buf
= show_asm_constraints(buf
, "\n\t\tclobber: ", rules
->clobbers
);
279 const char *show_instruction(struct instruction
*insn
)
281 int opcode
= insn
->opcode
;
282 static char buffer
[4096];
287 buf
+= sprintf(buf
, "# ");
289 if (opcode
< ARRAY_SIZE(opcodes
)) {
290 const char *op
= opcodes
[opcode
];
292 buf
+= sprintf(buf
, "opcode:%d", opcode
);
294 buf
+= sprintf(buf
, "%s", op
);
296 buf
+= sprintf(buf
, ".%d", insn
->size
);
297 memset(buf
, ' ', 20);
301 if (buf
< buffer
+ 12)
305 if (insn
->src
&& insn
->src
!= VOID
)
306 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->src
));
310 buf
+= sprintf(buf
, "%s, .L%u, .L%u", show_pseudo(insn
->cond
), insn
->bb_true
->nr
, insn
->bb_false
->nr
);
314 buf
+= sprintf(buf
, ".L%u", insn
->bb_true
->nr
);
318 struct symbol
*sym
= insn
->symbol
->sym
;
319 buf
+= sprintf(buf
, "%s <- ", show_pseudo(insn
->target
));
321 if (!insn
->bb
&& !sym
)
323 if (sym
->bb_target
) {
324 buf
+= sprintf(buf
, ".L%u", sym
->bb_target
->nr
);
328 buf
+= sprintf(buf
, "%s", show_ident(sym
->ident
));
331 buf
+= sprintf(buf
, "<anon symbol:%p>", sym
);
336 struct expression
*expr
= insn
->val
;
337 buf
+= sprintf(buf
, "%s <- ", show_pseudo(insn
->target
));
340 buf
+= sprintf(buf
, "%s", "<none>");
344 switch (expr
->type
) {
346 buf
+= sprintf(buf
, "%lld", expr
->value
);
349 buf
+= sprintf(buf
, "%Lf", expr
->fvalue
);
352 buf
+= sprintf(buf
, "%.40s", show_string(expr
->string
));
355 buf
+= sprintf(buf
, "%s", show_ident(expr
->symbol
->ident
));
358 buf
+= sprintf(buf
, ".L%u", expr
->symbol
->bb_target
->nr
);
361 buf
+= sprintf(buf
, "SETVAL EXPR TYPE %d", expr
->type
);
366 struct multijmp
*jmp
;
367 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->cond
));
368 FOR_EACH_PTR(insn
->multijmp_list
, jmp
) {
369 if (jmp
->begin
== jmp
->end
)
370 buf
+= sprintf(buf
, ", %d -> .L%u", jmp
->begin
, jmp
->target
->nr
);
371 else if (jmp
->begin
< jmp
->end
)
372 buf
+= sprintf(buf
, ", %d ... %d -> .L%u", jmp
->begin
, jmp
->end
, jmp
->target
->nr
);
374 buf
+= sprintf(buf
, ", default -> .L%u", jmp
->target
->nr
);
375 } END_FOR_EACH_PTR(jmp
);
378 case OP_COMPUTEDGOTO
: {
379 struct multijmp
*jmp
;
380 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->target
));
381 FOR_EACH_PTR(insn
->multijmp_list
, jmp
) {
382 buf
+= sprintf(buf
, ", .L%u", jmp
->target
->nr
);
383 } END_FOR_EACH_PTR(jmp
);
388 struct instruction
*phi
;
389 buf
+= sprintf(buf
, "%s <- %s ", show_pseudo(insn
->target
), show_pseudo(insn
->phi_src
));
390 FOR_EACH_PTR(insn
->phi_users
, phi
) {
391 buf
+= sprintf(buf
, " (%s)", show_pseudo(phi
->target
));
392 } END_FOR_EACH_PTR(phi
);
398 const char *s
= " <-";
399 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->target
));
400 FOR_EACH_PTR(insn
->phi_list
, phi
) {
401 buf
+= sprintf(buf
, "%s %s", s
, show_pseudo(phi
));
403 } END_FOR_EACH_PTR(phi
);
406 case OP_LOAD
: case OP_LNOP
:
407 buf
+= sprintf(buf
, "%s <- %d[%s]", show_pseudo(insn
->target
), insn
->offset
, show_pseudo(insn
->src
));
409 case OP_STORE
: case OP_SNOP
:
410 buf
+= sprintf(buf
, "%s -> %d[%s]", show_pseudo(insn
->target
), insn
->offset
, show_pseudo(insn
->src
));
412 case OP_INLINED_CALL
:
415 if (insn
->target
&& insn
->target
!= VOID
)
416 buf
+= sprintf(buf
, "%s <- ", show_pseudo(insn
->target
));
417 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->func
));
418 FOR_EACH_PTR(insn
->arguments
, arg
) {
419 buf
+= sprintf(buf
, ", %s", show_pseudo(arg
));
420 } END_FOR_EACH_PTR(arg
);
427 buf
+= sprintf(buf
, "%s <- (%d) %s",
428 show_pseudo(insn
->target
),
429 type_size(insn
->orig_type
),
430 show_pseudo(insn
->src
));
432 case OP_BINARY
... OP_BINARY_END
:
433 case OP_BINCMP
... OP_BINCMP_END
:
434 buf
+= sprintf(buf
, "%s <- %s, %s", show_pseudo(insn
->target
), show_pseudo(insn
->src1
), show_pseudo(insn
->src2
));
438 buf
+= sprintf(buf
, "%s <- %s, %s, %s", show_pseudo(insn
->target
),
439 show_pseudo(insn
->src1
), show_pseudo(insn
->src2
), show_pseudo(insn
->src3
));
443 buf
+= sprintf(buf
, "%s <- %s, %d, %d", show_pseudo(insn
->target
), show_pseudo(insn
->base
), insn
->from
, insn
->len
);
446 case OP_NOT
: case OP_NEG
:
447 buf
+= sprintf(buf
, "%s <- %s", show_pseudo(insn
->target
), show_pseudo(insn
->src1
));
451 buf
+= sprintf(buf
, "%s%d", insn
->check
? "check: " : "", insn
->increment
);
454 buf
+= sprintf(buf
, "%s between %s..%s", show_pseudo(insn
->src1
), show_pseudo(insn
->src2
), show_pseudo(insn
->src3
));
457 buf
+= sprintf(buf
, "%s <- %s", show_pseudo(insn
->target
), show_pseudo(insn
->src1
));
460 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->target
));
463 buf
= show_asm(buf
, insn
);
466 buf
+= sprintf(buf
, "%s <- %s", show_pseudo(insn
->target
), show_pseudo(insn
->src
));
472 if (buf
>= buffer
+ sizeof(buffer
))
473 die("instruction buffer overflowed %td\n", buf
- buffer
);
474 do { --buf
; } while (*buf
== ' ');
479 void show_bb(struct basic_block
*bb
)
481 struct instruction
*insn
;
483 printf(".L%u:\n", bb
->nr
);
485 pseudo_t needs
, defines
;
486 printf("%s:%d\n", stream_name(bb
->pos
.stream
), bb
->pos
.line
);
488 FOR_EACH_PTR(bb
->needs
, needs
) {
489 struct instruction
*def
= needs
->def
;
490 if (def
->opcode
!= OP_PHI
) {
491 printf(" **uses %s (from .L%u)**\n", show_pseudo(needs
), def
->bb
->nr
);
494 const char *sep
= " ";
495 printf(" **uses %s (from", show_pseudo(needs
));
496 FOR_EACH_PTR(def
->phi_list
, phi
) {
499 printf("%s(%s:.L%u)", sep
, show_pseudo(phi
), phi
->def
->bb
->nr
);
501 } END_FOR_EACH_PTR(phi
);
504 } END_FOR_EACH_PTR(needs
);
506 FOR_EACH_PTR(bb
->defines
, defines
) {
507 printf(" **defines %s **\n", show_pseudo(defines
));
508 } END_FOR_EACH_PTR(defines
);
511 struct basic_block
*from
;
512 FOR_EACH_PTR(bb
->parents
, from
) {
513 printf(" **from .L%u (%s:%d:%d)**\n", from
->nr
,
514 stream_name(from
->pos
.stream
), from
->pos
.line
, from
->pos
.pos
);
515 } END_FOR_EACH_PTR(from
);
519 struct basic_block
*to
;
520 FOR_EACH_PTR(bb
->children
, to
) {
521 printf(" **to .L%u (%s:%d:%d)**\n", to
->nr
,
522 stream_name(to
->pos
.stream
), to
->pos
.line
, to
->pos
.pos
);
523 } END_FOR_EACH_PTR(to
);
527 FOR_EACH_PTR(bb
->insns
, insn
) {
528 if (!insn
->bb
&& verbose
< 2)
530 printf("\t%s\n", show_instruction(insn
));
531 } END_FOR_EACH_PTR(insn
);
532 if (!bb_terminated(bb
))
536 static void show_symbol_usage(pseudo_t pseudo
)
538 struct pseudo_user
*pu
;
541 FOR_EACH_PTR(pseudo
->users
, pu
) {
542 printf("\t%s\n", show_instruction(pu
->insn
));
543 } END_FOR_EACH_PTR(pu
);
547 void show_entry(struct entrypoint
*ep
)
550 struct basic_block
*bb
;
552 printf("%s:\n", show_ident(ep
->name
->ident
));
555 printf("ep %p: %s\n", ep
, show_ident(ep
->name
->ident
));
557 FOR_EACH_PTR(ep
->syms
, sym
) {
560 if (!sym
->pseudo
->users
)
562 printf(" sym: %p %s\n", sym
, show_ident(sym
->ident
));
563 if (sym
->ctype
.modifiers
& (MOD_EXTERN
| MOD_STATIC
| MOD_ADDRESSABLE
))
564 printf("\texternal visibility\n");
565 show_symbol_usage(sym
->pseudo
);
566 } END_FOR_EACH_PTR(sym
);
571 FOR_EACH_PTR(ep
->bbs
, bb
) {
574 if (!bb
->parents
&& !bb
->children
&& !bb
->insns
&& verbose
< 2)
578 } END_FOR_EACH_PTR(bb
);
583 static void bind_label(struct symbol
*label
, struct basic_block
*bb
, struct position pos
)
585 if (label
->bb_target
)
586 warning(pos
, "label '%s' already bound", show_ident(label
->ident
));
587 label
->bb_target
= bb
;
590 static struct basic_block
* get_bound_block(struct entrypoint
*ep
, struct symbol
*label
)
592 struct basic_block
*bb
= label
->bb_target
;
595 bb
= alloc_basic_block(ep
, label
->pos
);
596 label
->bb_target
= bb
;
601 static void finish_block(struct entrypoint
*ep
)
603 struct basic_block
*src
= ep
->active
;
604 if (bb_reachable(src
))
608 static void add_goto(struct entrypoint
*ep
, struct basic_block
*dst
)
610 struct basic_block
*src
= ep
->active
;
611 if (bb_reachable(src
)) {
612 struct instruction
*br
= alloc_instruction(OP_BR
, 0);
614 add_bb(&dst
->parents
, src
);
615 add_bb(&src
->children
, dst
);
617 add_instruction(&src
->insns
, br
);
622 static void add_one_insn(struct entrypoint
*ep
, struct instruction
*insn
)
624 struct basic_block
*bb
= ep
->active
;
626 if (bb_reachable(bb
)) {
628 add_instruction(&bb
->insns
, insn
);
632 static void set_activeblock(struct entrypoint
*ep
, struct basic_block
*bb
)
634 if (!bb_terminated(ep
->active
))
638 if (bb_reachable(bb
))
639 add_bb(&ep
->bbs
, bb
);
642 static void remove_parent(struct basic_block
*child
, struct basic_block
*parent
)
644 remove_bb_from_list(&child
->parents
, parent
, 1);
646 repeat_phase
|= REPEAT_CFG_CLEANUP
;
649 /* Change a "switch" into a branch */
650 void insert_branch(struct basic_block
*bb
, struct instruction
*jmp
, struct basic_block
*target
)
652 struct instruction
*br
, *old
;
653 struct basic_block
*child
;
655 /* Remove the switch */
656 old
= delete_last_instruction(&bb
->insns
);
659 br
= alloc_instruction(OP_BR
, 0);
661 br
->bb_true
= target
;
662 add_instruction(&bb
->insns
, br
);
664 FOR_EACH_PTR(bb
->children
, child
) {
665 if (child
== target
) {
666 target
= NULL
; /* Trigger just once */
669 DELETE_CURRENT_PTR(child
);
670 remove_parent(child
, bb
);
671 } END_FOR_EACH_PTR(child
);
672 PACK_PTR_LIST(&bb
->children
);
674 if (repeat_phase
& REPEAT_CFG_CLEANUP
)
675 kill_unreachable_bbs(bb
->ep
);
679 void insert_select(struct basic_block
*bb
, struct instruction
*br
, struct instruction
*phi_node
, pseudo_t if_true
, pseudo_t if_false
)
682 struct instruction
*select
;
684 /* Remove the 'br' */
685 delete_last_instruction(&bb
->insns
);
687 select
= alloc_instruction(OP_SEL
, phi_node
->size
);
691 use_pseudo(select
, br
->cond
, &select
->src1
);
693 target
= phi_node
->target
;
694 assert(target
->def
== phi_node
);
695 select
->target
= target
;
696 target
->def
= select
;
698 use_pseudo(select
, if_true
, &select
->src2
);
699 use_pseudo(select
, if_false
, &select
->src3
);
701 add_instruction(&bb
->insns
, select
);
702 add_instruction(&bb
->insns
, br
);
705 static inline int bb_empty(struct basic_block
*bb
)
710 /* Add a label to the currently active block, return new active block */
711 static struct basic_block
* add_label(struct entrypoint
*ep
, struct symbol
*label
)
713 struct basic_block
*bb
= label
->bb_target
;
716 set_activeblock(ep
, bb
);
720 if (!bb_reachable(bb
) || !bb_empty(bb
)) {
721 bb
= alloc_basic_block(ep
, label
->pos
);
722 set_activeblock(ep
, bb
);
724 label
->bb_target
= bb
;
728 static void add_branch(struct entrypoint
*ep
, struct expression
*expr
, pseudo_t cond
, struct basic_block
*bb_true
, struct basic_block
*bb_false
)
730 struct basic_block
*bb
= ep
->active
;
731 struct instruction
*br
;
733 if (bb_reachable(bb
)) {
734 br
= alloc_instruction(OP_CBR
, 0);
735 use_pseudo(br
, cond
, &br
->cond
);
736 br
->bb_true
= bb_true
;
737 br
->bb_false
= bb_false
;
738 add_bb(&bb_true
->parents
, bb
);
739 add_bb(&bb_false
->parents
, bb
);
740 add_bb(&bb
->children
, bb_true
);
741 add_bb(&bb
->children
, bb_false
);
742 add_one_insn(ep
, br
);
746 /* Dummy pseudo allocator */
747 pseudo_t
alloc_pseudo(struct instruction
*def
)
750 struct pseudo
* pseudo
= __alloc_pseudo(0);
751 pseudo
->type
= PSEUDO_REG
;
757 static void clear_symbol_pseudos(struct entrypoint
*ep
)
761 FOR_EACH_PTR(ep
->accesses
, pseudo
) {
762 pseudo
->sym
->pseudo
= NULL
;
763 } END_FOR_EACH_PTR(pseudo
);
766 static pseudo_t
symbol_pseudo(struct entrypoint
*ep
, struct symbol
*sym
)
773 pseudo
= sym
->pseudo
;
775 pseudo
= __alloc_pseudo(0);
777 pseudo
->type
= PSEUDO_SYM
;
779 pseudo
->ident
= sym
->ident
;
780 sym
->pseudo
= pseudo
;
781 add_pseudo(&ep
->accesses
, pseudo
);
783 /* Symbol pseudos have neither nr, usage nor def */
787 pseudo_t
value_pseudo(long long val
)
789 #define MAX_VAL_HASH 64
790 static struct pseudo_list
*prev
[MAX_VAL_HASH
];
791 int hash
= val
& (MAX_VAL_HASH
-1);
792 struct pseudo_list
**list
= prev
+ hash
;
795 FOR_EACH_PTR(*list
, pseudo
) {
796 if (pseudo
->value
== val
)
798 } END_FOR_EACH_PTR(pseudo
);
800 pseudo
= __alloc_pseudo(0);
801 pseudo
->type
= PSEUDO_VAL
;
803 add_pseudo(list
, pseudo
);
805 /* Value pseudos have neither nr, usage nor def */
809 static pseudo_t
argument_pseudo(struct entrypoint
*ep
, int nr
)
811 pseudo_t pseudo
= __alloc_pseudo(0);
812 struct instruction
*entry
= ep
->entry
;
814 pseudo
->type
= PSEUDO_ARG
;
817 add_pseudo(&entry
->arg_list
, pseudo
);
819 /* Argument pseudos have neither usage nor def */
823 pseudo_t
alloc_phi(struct basic_block
*source
, pseudo_t pseudo
, int size
)
825 struct instruction
*insn
= alloc_instruction(OP_PHISOURCE
, size
);
826 pseudo_t phi
= __alloc_pseudo(0);
829 phi
->type
= PSEUDO_PHI
;
833 use_pseudo(insn
, pseudo
, &insn
->phi_src
);
836 add_instruction(&source
->insns
, insn
);
841 * We carry the "access_data" structure around for any accesses,
842 * which simplifies things a lot. It contains all the access
843 * information in one place.
846 struct symbol
*result_type
; // result ctype
847 struct symbol
*source_type
; // source ctype
848 pseudo_t address
; // pseudo containing address ..
849 pseudo_t origval
; // pseudo for original value ..
850 unsigned int offset
, alignment
; // byte offset
851 unsigned int bit_size
, bit_offset
; // which bits
855 static void finish_address_gen(struct entrypoint
*ep
, struct access_data
*ad
)
859 static int linearize_simple_address(struct entrypoint
*ep
,
860 struct expression
*addr
,
861 struct access_data
*ad
)
863 if (addr
->type
== EXPR_SYMBOL
) {
864 linearize_one_symbol(ep
, addr
->symbol
);
865 ad
->address
= symbol_pseudo(ep
, addr
->symbol
);
868 if (addr
->type
== EXPR_BINOP
) {
869 if (addr
->right
->type
== EXPR_VALUE
) {
870 if (addr
->op
== '+') {
871 ad
->offset
+= get_expression_value(addr
->right
);
872 return linearize_simple_address(ep
, addr
->left
, ad
);
876 ad
->address
= linearize_expression(ep
, addr
);
880 static struct symbol
*base_type(struct symbol
*sym
)
882 struct symbol
*base
= sym
;
885 if (sym
->type
== SYM_NODE
)
886 base
= base
->ctype
.base_type
;
887 if (base
->type
== SYM_BITFIELD
)
888 return base
->ctype
.base_type
;
893 static int linearize_address_gen(struct entrypoint
*ep
,
894 struct expression
*expr
,
895 struct access_data
*ad
)
897 struct symbol
*ctype
= expr
->ctype
;
902 ad
->result_type
= ctype
;
903 ad
->source_type
= base_type(ctype
);
904 ad
->bit_size
= ctype
->bit_size
;
905 ad
->alignment
= ctype
->ctype
.alignment
;
906 ad
->bit_offset
= ctype
->bit_offset
;
907 if (expr
->type
== EXPR_PREOP
&& expr
->op
== '*')
908 return linearize_simple_address(ep
, expr
->unop
, ad
);
910 warning(expr
->pos
, "generating address of non-lvalue (%d)", expr
->type
);
914 static pseudo_t
add_load(struct entrypoint
*ep
, struct access_data
*ad
)
916 struct instruction
*insn
;
923 insn
= alloc_typed_instruction(OP_LOAD
, ad
->source_type
);
924 new = alloc_pseudo(insn
);
928 insn
->offset
= ad
->offset
;
929 use_pseudo(insn
, ad
->address
, &insn
->src
);
930 add_one_insn(ep
, insn
);
934 static void add_store(struct entrypoint
*ep
, struct access_data
*ad
, pseudo_t value
)
936 struct basic_block
*bb
= ep
->active
;
938 if (bb_reachable(bb
)) {
939 struct instruction
*store
= alloc_typed_instruction(OP_STORE
, ad
->source_type
);
940 store
->offset
= ad
->offset
;
941 use_pseudo(store
, value
, &store
->target
);
942 use_pseudo(store
, ad
->address
, &store
->src
);
943 add_one_insn(ep
, store
);
947 static pseudo_t
linearize_store_gen(struct entrypoint
*ep
,
949 struct access_data
*ad
)
951 pseudo_t store
= value
;
953 if (type_size(ad
->source_type
) != type_size(ad
->result_type
)) {
954 pseudo_t orig
= add_load(ep
, ad
);
955 int shift
= ad
->bit_offset
;
956 unsigned long long mask
= (1ULL << ad
->bit_size
)-1;
959 store
= add_binary_op(ep
, ad
->source_type
, OP_SHL
, value
, value_pseudo(shift
));
962 orig
= add_binary_op(ep
, ad
->source_type
, OP_AND
, orig
, value_pseudo(~mask
));
963 store
= add_binary_op(ep
, ad
->source_type
, OP_OR
, orig
, store
);
965 add_store(ep
, ad
, store
);
969 static pseudo_t
add_binary_op(struct entrypoint
*ep
, struct symbol
*ctype
, int op
, pseudo_t left
, pseudo_t right
)
971 struct instruction
*insn
= alloc_typed_instruction(op
, ctype
);
972 pseudo_t target
= alloc_pseudo(insn
);
973 insn
->target
= target
;
974 use_pseudo(insn
, left
, &insn
->src1
);
975 use_pseudo(insn
, right
, &insn
->src2
);
976 add_one_insn(ep
, insn
);
980 static pseudo_t
add_setval(struct entrypoint
*ep
, struct symbol
*ctype
, struct expression
*val
)
982 struct instruction
*insn
= alloc_typed_instruction(OP_SETVAL
, ctype
);
983 pseudo_t target
= alloc_pseudo(insn
);
984 insn
->target
= target
;
986 add_one_insn(ep
, insn
);
990 static pseudo_t
add_symbol_address(struct entrypoint
*ep
, struct symbol
*sym
)
992 struct instruction
*insn
= alloc_instruction(OP_SYMADDR
, bits_in_pointer
);
993 pseudo_t target
= alloc_pseudo(insn
);
995 insn
->target
= target
;
996 use_pseudo(insn
, symbol_pseudo(ep
, sym
), &insn
->symbol
);
997 add_one_insn(ep
, insn
);
1001 static pseudo_t
linearize_load_gen(struct entrypoint
*ep
, struct access_data
*ad
)
1003 pseudo_t
new = add_load(ep
, ad
);
1005 if (ad
->bit_offset
) {
1006 pseudo_t shift
= value_pseudo(ad
->bit_offset
);
1007 pseudo_t newval
= add_binary_op(ep
, ad
->source_type
, OP_LSR
, new, shift
);
1010 if (ad
->bit_size
!= type_size(ad
->source_type
))
1011 new = cast_pseudo(ep
, new, ad
->source_type
, ad
->result_type
);
1015 static pseudo_t
linearize_access(struct entrypoint
*ep
, struct expression
*expr
)
1017 struct access_data ad
= { NULL
, };
1020 if (!linearize_address_gen(ep
, expr
, &ad
))
1022 value
= linearize_load_gen(ep
, &ad
);
1023 finish_address_gen(ep
, &ad
);
1028 static pseudo_t
linearize_inc_dec(struct entrypoint
*ep
, struct expression
*expr
, int postop
)
1030 struct access_data ad
= { NULL
, };
1031 pseudo_t old
, new, one
;
1032 int op
= expr
->op
== SPECIAL_INCREMENT
? OP_ADD
: OP_SUB
;
1034 if (!linearize_address_gen(ep
, expr
->unop
, &ad
))
1037 old
= linearize_load_gen(ep
, &ad
);
1038 one
= value_pseudo(expr
->op_value
);
1039 new = add_binary_op(ep
, expr
->ctype
, op
, old
, one
);
1040 linearize_store_gen(ep
, new, &ad
);
1041 finish_address_gen(ep
, &ad
);
1042 return postop
? old
: new;
1045 static pseudo_t
add_uniop(struct entrypoint
*ep
, struct expression
*expr
, int op
, pseudo_t src
)
1047 struct instruction
*insn
= alloc_typed_instruction(op
, expr
->ctype
);
1048 pseudo_t
new = alloc_pseudo(insn
);
1051 use_pseudo(insn
, src
, &insn
->src1
);
1052 add_one_insn(ep
, insn
);
1056 static pseudo_t
linearize_slice(struct entrypoint
*ep
, struct expression
*expr
)
1058 pseudo_t pre
= linearize_expression(ep
, expr
->base
);
1059 struct instruction
*insn
= alloc_typed_instruction(OP_SLICE
, expr
->ctype
);
1060 pseudo_t
new = alloc_pseudo(insn
);
1063 insn
->from
= expr
->r_bitpos
;
1064 insn
->len
= expr
->r_nrbits
;
1065 use_pseudo(insn
, pre
, &insn
->base
);
1066 add_one_insn(ep
, insn
);
1070 static pseudo_t
linearize_regular_preop(struct entrypoint
*ep
, struct expression
*expr
)
1072 pseudo_t pre
= linearize_expression(ep
, expr
->unop
);
1077 pseudo_t zero
= value_pseudo(0);
1078 return add_binary_op(ep
, expr
->ctype
, OP_SET_EQ
, pre
, zero
);
1081 return add_uniop(ep
, expr
, OP_NOT
, pre
);
1083 return add_uniop(ep
, expr
, OP_NEG
, pre
);
1088 static pseudo_t
linearize_preop(struct entrypoint
*ep
, struct expression
*expr
)
1091 * '*' is an lvalue access, and is fundamentally different
1092 * from an arithmetic operation. Maybe it should have an
1093 * expression type of its own..
1095 if (expr
->op
== '*')
1096 return linearize_access(ep
, expr
);
1097 if (expr
->op
== SPECIAL_INCREMENT
|| expr
->op
== SPECIAL_DECREMENT
)
1098 return linearize_inc_dec(ep
, expr
, 0);
1099 return linearize_regular_preop(ep
, expr
);
1102 static pseudo_t
linearize_postop(struct entrypoint
*ep
, struct expression
*expr
)
1104 return linearize_inc_dec(ep
, expr
, 1);
1108 * Casts to pointers are "less safe" than other casts, since
1109 * they imply type-unsafe accesses. "void *" is a special
1110 * case, since you can't access through it anyway without another
1113 static struct instruction
*alloc_cast_instruction(struct symbol
*src
, struct symbol
*ctype
)
1115 int opcode
= OP_CAST
;
1116 struct symbol
*base
= ctype
;
1118 if (src
->ctype
.modifiers
& MOD_SIGNED
)
1120 if (base
->type
== SYM_NODE
)
1121 base
= base
->ctype
.base_type
;
1122 if (base
->type
== SYM_PTR
) {
1123 base
= base
->ctype
.base_type
;
1124 if (base
!= &void_ctype
)
1125 opcode
= OP_PTRCAST
;
1126 } else if (base
->ctype
.base_type
== &fp_type
)
1128 return alloc_typed_instruction(opcode
, ctype
);
1131 static pseudo_t
cast_pseudo(struct entrypoint
*ep
, pseudo_t src
, struct symbol
*from
, struct symbol
*to
)
1134 struct instruction
*insn
;
1140 if (from
->bit_size
< 0 || to
->bit_size
< 0)
1142 insn
= alloc_cast_instruction(from
, to
);
1143 result
= alloc_pseudo(insn
);
1144 insn
->target
= result
;
1145 insn
->orig_type
= from
;
1146 use_pseudo(insn
, src
, &insn
->src
);
1147 add_one_insn(ep
, insn
);
1151 static int opcode_sign(int opcode
, struct symbol
*ctype
)
1153 if (ctype
&& (ctype
->ctype
.modifiers
& MOD_SIGNED
)) {
1155 case OP_MULU
: case OP_DIVU
: case OP_MODU
: case OP_LSR
:
1162 static pseudo_t
linearize_assignment(struct entrypoint
*ep
, struct expression
*expr
)
1164 struct access_data ad
= { NULL
, };
1165 struct expression
*target
= expr
->left
;
1166 struct expression
*src
= expr
->right
;
1167 struct symbol
*ctype
;
1170 value
= linearize_expression(ep
, src
);
1171 if (!target
|| !linearize_address_gen(ep
, target
, &ad
))
1173 if (expr
->op
!= '=') {
1174 pseudo_t oldvalue
= linearize_load_gen(ep
, &ad
);
1176 static const int op_trans
[] = {
1177 [SPECIAL_ADD_ASSIGN
- SPECIAL_BASE
] = OP_ADD
,
1178 [SPECIAL_SUB_ASSIGN
- SPECIAL_BASE
] = OP_SUB
,
1179 [SPECIAL_MUL_ASSIGN
- SPECIAL_BASE
] = OP_MULU
,
1180 [SPECIAL_DIV_ASSIGN
- SPECIAL_BASE
] = OP_DIVU
,
1181 [SPECIAL_MOD_ASSIGN
- SPECIAL_BASE
] = OP_MODU
,
1182 [SPECIAL_SHL_ASSIGN
- SPECIAL_BASE
] = OP_SHL
,
1183 [SPECIAL_SHR_ASSIGN
- SPECIAL_BASE
] = OP_LSR
,
1184 [SPECIAL_AND_ASSIGN
- SPECIAL_BASE
] = OP_AND
,
1185 [SPECIAL_OR_ASSIGN
- SPECIAL_BASE
] = OP_OR
,
1186 [SPECIAL_XOR_ASSIGN
- SPECIAL_BASE
] = OP_XOR
1194 oldvalue
= cast_pseudo(ep
, oldvalue
, target
->ctype
, ctype
);
1195 opcode
= opcode_sign(op_trans
[expr
->op
- SPECIAL_BASE
], ctype
);
1196 dst
= add_binary_op(ep
, ctype
, opcode
, oldvalue
, value
);
1197 value
= cast_pseudo(ep
, dst
, ctype
, expr
->ctype
);
1199 value
= linearize_store_gen(ep
, value
, &ad
);
1200 finish_address_gen(ep
, &ad
);
1204 static pseudo_t
linearize_call_expression(struct entrypoint
*ep
, struct expression
*expr
)
1206 struct expression
*arg
, *fn
;
1207 struct instruction
*insn
= alloc_typed_instruction(OP_CALL
, expr
->ctype
);
1208 pseudo_t retval
, call
;
1209 struct ctype
*ctype
= NULL
;
1210 struct symbol
*fntype
;
1211 struct context
*context
;
1214 warning(expr
->pos
, "call with no type!");
1218 FOR_EACH_PTR(expr
->args
, arg
) {
1219 pseudo_t
new = linearize_expression(ep
, arg
);
1220 use_pseudo(insn
, new, add_pseudo(&insn
->arguments
, new));
1221 } END_FOR_EACH_PTR(arg
);
1226 ctype
= &fn
->ctype
->ctype
;
1230 if (fntype
->type
== SYM_NODE
)
1231 fntype
= fntype
->ctype
.base_type
;
1233 insn
->fntype
= fntype
;
1235 if (fn
->type
== EXPR_PREOP
) {
1236 if (fn
->unop
->type
== EXPR_SYMBOL
) {
1237 struct symbol
*sym
= fn
->unop
->symbol
;
1238 if (sym
->ctype
.base_type
->type
== SYM_FN
)
1242 if (fn
->type
== EXPR_SYMBOL
) {
1243 call
= symbol_pseudo(ep
, fn
->symbol
);
1245 call
= linearize_expression(ep
, fn
);
1247 use_pseudo(insn
, call
, &insn
->func
);
1249 if (expr
->ctype
!= &void_ctype
)
1250 retval
= alloc_pseudo(insn
);
1251 insn
->target
= retval
;
1252 add_one_insn(ep
, insn
);
1255 FOR_EACH_PTR(ctype
->contexts
, context
) {
1256 int in
= context
->in
;
1257 int out
= context
->out
;
1268 context_diff
= out
- in
;
1269 if (check
|| context_diff
) {
1270 insn
= alloc_instruction(OP_CONTEXT
, 0);
1271 insn
->increment
= context_diff
;
1272 insn
->check
= check
;
1273 insn
->context_expr
= context
->context
;
1274 add_one_insn(ep
, insn
);
1276 } END_FOR_EACH_PTR(context
);
1282 static pseudo_t
linearize_binop(struct entrypoint
*ep
, struct expression
*expr
)
1284 pseudo_t src1
, src2
, dst
;
1285 static const int opcode
[] = {
1286 ['+'] = OP_ADD
, ['-'] = OP_SUB
,
1287 ['*'] = OP_MULU
, ['/'] = OP_DIVU
,
1288 ['%'] = OP_MODU
, ['&'] = OP_AND
,
1289 ['|'] = OP_OR
, ['^'] = OP_XOR
,
1290 [SPECIAL_LEFTSHIFT
] = OP_SHL
,
1291 [SPECIAL_RIGHTSHIFT
] = OP_LSR
,
1292 [SPECIAL_LOGICAL_AND
] = OP_AND_BOOL
,
1293 [SPECIAL_LOGICAL_OR
] = OP_OR_BOOL
,
1297 src1
= linearize_expression(ep
, expr
->left
);
1298 src2
= linearize_expression(ep
, expr
->right
);
1299 op
= opcode_sign(opcode
[expr
->op
], expr
->ctype
);
1300 dst
= add_binary_op(ep
, expr
->ctype
, op
, src1
, src2
);
1304 static pseudo_t
linearize_logical_branch(struct entrypoint
*ep
, struct expression
*expr
, struct basic_block
*bb_true
, struct basic_block
*bb_false
);
1306 pseudo_t
linearize_cond_branch(struct entrypoint
*ep
, struct expression
*expr
, struct basic_block
*bb_true
, struct basic_block
*bb_false
);
1308 static pseudo_t
linearize_select(struct entrypoint
*ep
, struct expression
*expr
)
1310 pseudo_t cond
, true, false, res
;
1311 struct instruction
*insn
;
1313 true = linearize_expression(ep
, expr
->cond_true
);
1314 false = linearize_expression(ep
, expr
->cond_false
);
1315 cond
= linearize_expression(ep
, expr
->conditional
);
1317 insn
= alloc_typed_instruction(OP_SEL
, expr
->ctype
);
1318 if (!expr
->cond_true
)
1320 use_pseudo(insn
, cond
, &insn
->src1
);
1321 use_pseudo(insn
, true, &insn
->src2
);
1322 use_pseudo(insn
, false, &insn
->src3
);
1324 res
= alloc_pseudo(insn
);
1326 add_one_insn(ep
, insn
);
1330 static pseudo_t
add_join_conditional(struct entrypoint
*ep
, struct expression
*expr
,
1331 pseudo_t phi1
, pseudo_t phi2
)
1334 struct instruction
*phi_node
;
1341 phi_node
= alloc_typed_instruction(OP_PHI
, expr
->ctype
);
1342 use_pseudo(phi_node
, phi1
, add_pseudo(&phi_node
->phi_list
, phi1
));
1343 use_pseudo(phi_node
, phi2
, add_pseudo(&phi_node
->phi_list
, phi2
));
1344 phi_node
->target
= target
= alloc_pseudo(phi_node
);
1345 add_one_insn(ep
, phi_node
);
1349 static pseudo_t
linearize_short_conditional(struct entrypoint
*ep
, struct expression
*expr
,
1350 struct expression
*cond
,
1351 struct expression
*expr_false
)
1353 pseudo_t src1
, src2
;
1354 struct basic_block
*bb_false
;
1355 struct basic_block
*merge
= alloc_basic_block(ep
, expr
->pos
);
1356 pseudo_t phi1
, phi2
;
1357 int size
= type_size(expr
->ctype
);
1359 if (!expr_false
|| !ep
->active
)
1362 bb_false
= alloc_basic_block(ep
, expr_false
->pos
);
1363 src1
= linearize_expression(ep
, cond
);
1364 phi1
= alloc_phi(ep
->active
, src1
, size
);
1365 add_branch(ep
, expr
, src1
, merge
, bb_false
);
1367 set_activeblock(ep
, bb_false
);
1368 src2
= linearize_expression(ep
, expr_false
);
1369 phi2
= alloc_phi(ep
->active
, src2
, size
);
1370 set_activeblock(ep
, merge
);
1372 return add_join_conditional(ep
, expr
, phi1
, phi2
);
1375 static pseudo_t
linearize_conditional(struct entrypoint
*ep
, struct expression
*expr
,
1376 struct expression
*cond
,
1377 struct expression
*expr_true
,
1378 struct expression
*expr_false
)
1380 pseudo_t src1
, src2
;
1381 pseudo_t phi1
, phi2
;
1382 struct basic_block
*bb_true
, *bb_false
, *merge
;
1383 int size
= type_size(expr
->ctype
);
1385 if (!cond
|| !expr_true
|| !expr_false
|| !ep
->active
)
1387 bb_true
= alloc_basic_block(ep
, expr_true
->pos
);
1388 bb_false
= alloc_basic_block(ep
, expr_false
->pos
);
1389 merge
= alloc_basic_block(ep
, expr
->pos
);
1391 linearize_cond_branch(ep
, cond
, bb_true
, bb_false
);
1393 set_activeblock(ep
, bb_true
);
1394 src1
= linearize_expression(ep
, expr_true
);
1395 phi1
= alloc_phi(ep
->active
, src1
, size
);
1396 add_goto(ep
, merge
);
1398 set_activeblock(ep
, bb_false
);
1399 src2
= linearize_expression(ep
, expr_false
);
1400 phi2
= alloc_phi(ep
->active
, src2
, size
);
1401 set_activeblock(ep
, merge
);
1403 return add_join_conditional(ep
, expr
, phi1
, phi2
);
1406 static pseudo_t
linearize_logical(struct entrypoint
*ep
, struct expression
*expr
)
1408 struct expression
*shortcut
;
1410 shortcut
= alloc_const_expression(expr
->pos
, expr
->op
== SPECIAL_LOGICAL_OR
);
1411 shortcut
->ctype
= expr
->ctype
;
1412 if (expr
->op
== SPECIAL_LOGICAL_OR
)
1413 return linearize_conditional(ep
, expr
, expr
->left
, shortcut
, expr
->right
);
1414 return linearize_conditional(ep
, expr
, expr
->left
, expr
->right
, shortcut
);
1417 static pseudo_t
linearize_compare(struct entrypoint
*ep
, struct expression
*expr
)
1419 static const int cmpop
[] = {
1420 ['>'] = OP_SET_GT
, ['<'] = OP_SET_LT
,
1421 [SPECIAL_EQUAL
] = OP_SET_EQ
,
1422 [SPECIAL_NOTEQUAL
] = OP_SET_NE
,
1423 [SPECIAL_GTE
] = OP_SET_GE
,
1424 [SPECIAL_LTE
] = OP_SET_LE
,
1425 [SPECIAL_UNSIGNED_LT
] = OP_SET_B
,
1426 [SPECIAL_UNSIGNED_GT
] = OP_SET_A
,
1427 [SPECIAL_UNSIGNED_LTE
] = OP_SET_BE
,
1428 [SPECIAL_UNSIGNED_GTE
] = OP_SET_AE
,
1431 pseudo_t src1
= linearize_expression(ep
, expr
->left
);
1432 pseudo_t src2
= linearize_expression(ep
, expr
->right
);
1433 pseudo_t dst
= add_binary_op(ep
, expr
->ctype
, cmpop
[expr
->op
], src1
, src2
);
1438 pseudo_t
linearize_cond_branch(struct entrypoint
*ep
, struct expression
*expr
, struct basic_block
*bb_true
, struct basic_block
*bb_false
)
1442 if (!expr
|| !bb_reachable(ep
->active
))
1445 switch (expr
->type
) {
1449 add_goto(ep
, expr
->value
? bb_true
: bb_false
);
1453 add_goto(ep
, expr
->fvalue
? bb_true
: bb_false
);
1457 linearize_logical_branch(ep
, expr
, bb_true
, bb_false
);
1461 cond
= linearize_compare(ep
, expr
);
1462 add_branch(ep
, expr
, cond
, bb_true
, bb_false
);
1466 if (expr
->op
== '!')
1467 return linearize_cond_branch(ep
, expr
->unop
, bb_false
, bb_true
);
1470 cond
= linearize_expression(ep
, expr
);
1471 add_branch(ep
, expr
, cond
, bb_true
, bb_false
);
1481 static pseudo_t
linearize_logical_branch(struct entrypoint
*ep
, struct expression
*expr
, struct basic_block
*bb_true
, struct basic_block
*bb_false
)
1483 struct basic_block
*next
= alloc_basic_block(ep
, expr
->pos
);
1485 if (expr
->op
== SPECIAL_LOGICAL_OR
)
1486 linearize_cond_branch(ep
, expr
->left
, bb_true
, next
);
1488 linearize_cond_branch(ep
, expr
->left
, next
, bb_false
);
1489 set_activeblock(ep
, next
);
1490 linearize_cond_branch(ep
, expr
->right
, bb_true
, bb_false
);
1494 static pseudo_t
linearize_cast(struct entrypoint
*ep
, struct expression
*expr
)
1497 struct expression
*orig
= expr
->cast_expression
;
1502 src
= linearize_expression(ep
, orig
);
1503 return cast_pseudo(ep
, src
, orig
->ctype
, expr
->ctype
);
1506 static pseudo_t
linearize_position(struct entrypoint
*ep
, struct expression
*pos
, struct access_data
*ad
)
1508 struct expression
*init_expr
= pos
->init_expr
;
1510 ad
->offset
= pos
->init_offset
;
1511 ad
->source_type
= base_type(init_expr
->ctype
);
1512 ad
->result_type
= init_expr
->ctype
;
1513 return linearize_initializer(ep
, init_expr
, ad
);
1516 static pseudo_t
linearize_initializer(struct entrypoint
*ep
, struct expression
*initializer
, struct access_data
*ad
)
1518 switch (initializer
->type
) {
1519 case EXPR_INITIALIZER
: {
1520 struct expression
*expr
;
1521 FOR_EACH_PTR(initializer
->expr_list
, expr
) {
1522 linearize_initializer(ep
, expr
, ad
);
1523 } END_FOR_EACH_PTR(expr
);
1527 linearize_position(ep
, initializer
, ad
);
1530 pseudo_t value
= linearize_expression(ep
, initializer
);
1531 ad
->source_type
= base_type(initializer
->ctype
);
1532 ad
->result_type
= initializer
->ctype
;
1533 linearize_store_gen(ep
, value
, ad
);
1541 static void linearize_argument(struct entrypoint
*ep
, struct symbol
*arg
, int nr
)
1543 struct access_data ad
= { NULL
, };
1545 ad
.source_type
= arg
;
1546 ad
.result_type
= arg
;
1547 ad
.address
= symbol_pseudo(ep
, arg
);
1548 linearize_store_gen(ep
, argument_pseudo(ep
, nr
), &ad
);
1549 finish_address_gen(ep
, &ad
);
1552 pseudo_t
linearize_expression(struct entrypoint
*ep
, struct expression
*expr
)
1557 current_pos
= expr
->pos
;
1558 switch (expr
->type
) {
1560 linearize_one_symbol(ep
, expr
->symbol
);
1561 return add_symbol_address(ep
, expr
->symbol
);
1564 return value_pseudo(expr
->value
);
1566 case EXPR_STRING
: case EXPR_FVALUE
: case EXPR_LABEL
:
1567 return add_setval(ep
, expr
->ctype
, expr
);
1569 case EXPR_STATEMENT
:
1570 return linearize_statement(ep
, expr
->statement
);
1573 return linearize_call_expression(ep
, expr
);
1576 return linearize_binop(ep
, expr
);
1579 return linearize_logical(ep
, expr
);
1582 return linearize_compare(ep
, expr
);
1585 return linearize_select(ep
, expr
);
1587 case EXPR_CONDITIONAL
:
1588 if (!expr
->cond_true
)
1589 return linearize_short_conditional(ep
, expr
, expr
->conditional
, expr
->cond_false
);
1591 return linearize_conditional(ep
, expr
, expr
->conditional
,
1592 expr
->cond_true
, expr
->cond_false
);
1595 linearize_expression(ep
, expr
->left
);
1596 return linearize_expression(ep
, expr
->right
);
1598 case EXPR_ASSIGNMENT
:
1599 return linearize_assignment(ep
, expr
);
1602 return linearize_preop(ep
, expr
);
1605 return linearize_postop(ep
, expr
);
1608 case EXPR_FORCE_CAST
:
1609 case EXPR_IMPLIED_CAST
:
1610 return linearize_cast(ep
, expr
);
1613 return linearize_slice(ep
, expr
);
1615 case EXPR_INITIALIZER
:
1617 warning(expr
->pos
, "unexpected initializer expression (%d %d)", expr
->type
, expr
->op
);
1620 warning(expr
->pos
, "unknown expression (%d %d)", expr
->type
, expr
->op
);
1626 static pseudo_t
linearize_one_symbol(struct entrypoint
*ep
, struct symbol
*sym
)
1628 struct access_data ad
= { NULL
, };
1631 if (!sym
|| !sym
->initializer
|| sym
->initialized
)
1634 /* We need to output these puppies some day too.. */
1635 if (sym
->ctype
.modifiers
& (MOD_STATIC
| MOD_TOPLEVEL
))
1638 sym
->initialized
= 1;
1639 ad
.address
= symbol_pseudo(ep
, sym
);
1640 value
= linearize_initializer(ep
, sym
->initializer
, &ad
);
1641 finish_address_gen(ep
, &ad
);
1645 static pseudo_t
linearize_compound_statement(struct entrypoint
*ep
, struct statement
*stmt
)
1648 struct statement
*s
;
1649 struct symbol
*ret
= stmt
->ret
;
1652 FOR_EACH_PTR(stmt
->stmts
, s
) {
1653 pseudo
= linearize_statement(ep
, s
);
1654 } END_FOR_EACH_PTR(s
);
1657 struct basic_block
*bb
= add_label(ep
, ret
);
1658 struct instruction
*phi_node
= first_instruction(bb
->insns
);
1663 if (pseudo_list_size(phi_node
->phi_list
)==1) {
1664 pseudo
= first_pseudo(phi_node
->phi_list
);
1665 assert(pseudo
->type
== PSEUDO_PHI
);
1666 return pseudo
->def
->src1
;
1668 return phi_node
->target
;
1674 static pseudo_t
linearize_inlined_call(struct entrypoint
*ep
, struct statement
*stmt
)
1676 struct instruction
*insn
= alloc_instruction(OP_INLINED_CALL
, 0);
1677 struct statement
*args
= stmt
->args
;
1678 struct basic_block
*bb
;
1684 concat_symbol_list(args
->declaration
, &ep
->syms
);
1685 FOR_EACH_PTR(args
->declaration
, sym
) {
1686 pseudo_t value
= linearize_one_symbol(ep
, sym
);
1687 use_pseudo(insn
, value
, add_pseudo(&insn
->arguments
, value
));
1688 } END_FOR_EACH_PTR(sym
);
1691 insn
->target
= pseudo
= linearize_compound_statement(ep
, stmt
);
1692 use_pseudo(insn
, symbol_pseudo(ep
, stmt
->inline_fn
), &insn
->func
);
1694 if (bb
&& !bb
->insns
)
1695 bb
->pos
= stmt
->pos
;
1696 add_one_insn(ep
, insn
);
1700 static pseudo_t
linearize_context(struct entrypoint
*ep
, struct statement
*stmt
)
1702 struct instruction
*insn
= alloc_instruction(OP_CONTEXT
, 0);
1703 struct expression
*expr
= stmt
->expression
;
1706 if (expr
->type
== EXPR_VALUE
)
1707 value
= expr
->value
;
1709 insn
->increment
= value
;
1710 insn
->context_expr
= stmt
->context
;
1711 add_one_insn(ep
, insn
);
1715 static pseudo_t
linearize_range(struct entrypoint
*ep
, struct statement
*stmt
)
1717 struct instruction
*insn
= alloc_instruction(OP_RANGE
, 0);
1719 use_pseudo(insn
, linearize_expression(ep
, stmt
->range_expression
), &insn
->src1
);
1720 use_pseudo(insn
, linearize_expression(ep
, stmt
->range_low
), &insn
->src2
);
1721 use_pseudo(insn
, linearize_expression(ep
, stmt
->range_high
), &insn
->src3
);
1722 add_one_insn(ep
, insn
);
1726 ALLOCATOR(asm_rules
, "asm rules");
1727 ALLOCATOR(asm_constraint
, "asm constraints");
1729 static void add_asm_input(struct entrypoint
*ep
, struct instruction
*insn
, struct expression
*expr
,
1730 const char *constraint
, const struct ident
*ident
)
1732 pseudo_t pseudo
= linearize_expression(ep
, expr
);
1733 struct asm_constraint
*rule
= __alloc_asm_constraint(0);
1735 rule
->ident
= ident
;
1736 rule
->constraint
= constraint
;
1737 use_pseudo(insn
, pseudo
, &rule
->pseudo
);
1738 add_ptr_list(&insn
->asm_rules
->inputs
, rule
);
1741 static void add_asm_output(struct entrypoint
*ep
, struct instruction
*insn
, struct expression
*expr
,
1742 const char *constraint
, const struct ident
*ident
)
1744 struct access_data ad
= { NULL
, };
1745 pseudo_t pseudo
= alloc_pseudo(insn
);
1746 struct asm_constraint
*rule
;
1748 if (!expr
|| !linearize_address_gen(ep
, expr
, &ad
))
1750 linearize_store_gen(ep
, pseudo
, &ad
);
1751 finish_address_gen(ep
, &ad
);
1752 rule
= __alloc_asm_constraint(0);
1753 rule
->ident
= ident
;
1754 rule
->constraint
= constraint
;
1755 use_pseudo(insn
, pseudo
, &rule
->pseudo
);
1756 add_ptr_list(&insn
->asm_rules
->outputs
, rule
);
1759 static pseudo_t
linearize_asm_statement(struct entrypoint
*ep
, struct statement
*stmt
)
1762 struct expression
*expr
;
1763 struct instruction
*insn
;
1764 struct asm_rules
*rules
;
1765 const char *constraint
;
1766 struct ident
*ident
;
1768 insn
= alloc_instruction(OP_ASM
, 0);
1769 expr
= stmt
->asm_string
;
1770 if (!expr
|| expr
->type
!= EXPR_STRING
) {
1771 warning(stmt
->pos
, "expected string in inline asm");
1774 insn
->string
= expr
->string
->data
;
1776 rules
= __alloc_asm_rules(0);
1777 insn
->asm_rules
= rules
;
1779 /* Gather the inputs.. */
1783 FOR_EACH_PTR(stmt
->asm_inputs
, expr
) {
1785 case 0: /* Identifier */
1787 ident
= (struct ident
*)expr
;
1790 case 1: /* Constraint */
1792 constraint
= expr
? expr
->string
->data
: "";
1795 case 2: /* Expression */
1797 add_asm_input(ep
, insn
, expr
, constraint
, ident
);
1799 } END_FOR_EACH_PTR(expr
);
1801 add_one_insn(ep
, insn
);
1803 /* Assign the outputs */
1807 FOR_EACH_PTR(stmt
->asm_outputs
, expr
) {
1809 case 0: /* Identifier */
1811 ident
= (struct ident
*)expr
;
1814 case 1: /* Constraint */
1816 constraint
= expr
? expr
->string
->data
: "";
1821 add_asm_output(ep
, insn
, expr
, constraint
, ident
);
1823 } END_FOR_EACH_PTR(expr
);
1828 static int multijmp_cmp(const void *_a
, const void *_b
)
1830 const struct multijmp
*a
= _a
;
1831 const struct multijmp
*b
= _b
;
1834 if (a
->begin
> a
->end
) {
1835 if (b
->begin
> b
->end
)
1839 if (b
->begin
> b
->end
)
1841 if (a
->begin
== b
->begin
) {
1842 if (a
->end
== b
->end
)
1844 return (a
->end
< b
->end
) ? -1 : 1;
1846 return a
->begin
< b
->begin
? -1 : 1;
1849 static void sort_switch_cases(struct instruction
*insn
)
1851 sort_list((struct ptr_list
**)&insn
->multijmp_list
, multijmp_cmp
);
1854 static pseudo_t
linearize_declaration(struct entrypoint
*ep
, struct statement
*stmt
)
1858 concat_symbol_list(stmt
->declaration
, &ep
->syms
);
1860 FOR_EACH_PTR(stmt
->declaration
, sym
) {
1861 linearize_one_symbol(ep
, sym
);
1862 } END_FOR_EACH_PTR(sym
);
1866 static pseudo_t
linearize_return(struct entrypoint
*ep
, struct statement
*stmt
)
1868 struct expression
*expr
= stmt
->expression
;
1869 struct basic_block
*bb_return
= get_bound_block(ep
, stmt
->ret_target
);
1870 struct basic_block
*active
;
1871 pseudo_t src
= linearize_expression(ep
, expr
);
1872 active
= ep
->active
;
1873 if (active
&& src
!= VOID
) {
1874 struct instruction
*phi_node
= first_instruction(bb_return
->insns
);
1877 phi_node
= alloc_typed_instruction(OP_PHI
, expr
->ctype
);
1878 phi_node
->target
= alloc_pseudo(phi_node
);
1879 phi_node
->bb
= bb_return
;
1880 add_instruction(&bb_return
->insns
, phi_node
);
1882 phi
= alloc_phi(active
, src
, type_size(expr
->ctype
));
1883 phi
->ident
= &return_ident
;
1884 use_pseudo(phi_node
, phi
, add_pseudo(&phi_node
->phi_list
, phi
));
1886 add_goto(ep
, bb_return
);
1890 static pseudo_t
linearize_switch(struct entrypoint
*ep
, struct statement
*stmt
)
1893 struct instruction
*switch_ins
;
1894 struct basic_block
*switch_end
= alloc_basic_block(ep
, stmt
->pos
);
1895 struct basic_block
*active
, *default_case
;
1896 struct multijmp
*jmp
;
1899 pseudo
= linearize_expression(ep
, stmt
->switch_expression
);
1901 active
= ep
->active
;
1902 if (!bb_reachable(active
))
1905 switch_ins
= alloc_instruction(OP_SWITCH
, 0);
1906 use_pseudo(switch_ins
, pseudo
, &switch_ins
->cond
);
1907 add_one_insn(ep
, switch_ins
);
1910 default_case
= NULL
;
1911 FOR_EACH_PTR(stmt
->switch_case
->symbol_list
, sym
) {
1912 struct statement
*case_stmt
= sym
->stmt
;
1913 struct basic_block
*bb_case
= get_bound_block(ep
, sym
);
1915 if (!case_stmt
->case_expression
) {
1916 default_case
= bb_case
;
1921 begin
= end
= case_stmt
->case_expression
->value
;
1922 if (case_stmt
->case_to
)
1923 end
= case_stmt
->case_to
->value
;
1925 jmp
= alloc_multijmp(bb_case
, end
, begin
);
1927 jmp
= alloc_multijmp(bb_case
, begin
, end
);
1930 add_multijmp(&switch_ins
->multijmp_list
, jmp
);
1931 add_bb(&bb_case
->parents
, active
);
1932 add_bb(&active
->children
, bb_case
);
1933 } END_FOR_EACH_PTR(sym
);
1935 bind_label(stmt
->switch_break
, switch_end
, stmt
->pos
);
1937 /* And linearize the actual statement */
1938 linearize_statement(ep
, stmt
->switch_statement
);
1939 set_activeblock(ep
, switch_end
);
1942 default_case
= switch_end
;
1944 jmp
= alloc_multijmp(default_case
, 1, 0);
1945 add_multijmp(&switch_ins
->multijmp_list
, jmp
);
1946 add_bb(&default_case
->parents
, active
);
1947 add_bb(&active
->children
, default_case
);
1948 sort_switch_cases(switch_ins
);
1953 static pseudo_t
linearize_iterator(struct entrypoint
*ep
, struct statement
*stmt
)
1955 struct statement
*pre_statement
= stmt
->iterator_pre_statement
;
1956 struct expression
*pre_condition
= stmt
->iterator_pre_condition
;
1957 struct statement
*statement
= stmt
->iterator_statement
;
1958 struct statement
*post_statement
= stmt
->iterator_post_statement
;
1959 struct expression
*post_condition
= stmt
->iterator_post_condition
;
1960 struct basic_block
*loop_top
, *loop_body
, *loop_continue
, *loop_end
;
1963 FOR_EACH_PTR(stmt
->iterator_syms
, sym
) {
1964 linearize_one_symbol(ep
, sym
);
1965 } END_FOR_EACH_PTR(sym
);
1966 concat_symbol_list(stmt
->iterator_syms
, &ep
->syms
);
1967 linearize_statement(ep
, pre_statement
);
1969 loop_body
= loop_top
= alloc_basic_block(ep
, stmt
->pos
);
1970 loop_continue
= alloc_basic_block(ep
, stmt
->pos
);
1971 loop_end
= alloc_basic_block(ep
, stmt
->pos
);
1973 /* An empty post-condition means that it's the same as the pre-condition */
1974 if (!post_condition
) {
1975 loop_top
= alloc_basic_block(ep
, stmt
->pos
);
1976 set_activeblock(ep
, loop_top
);
1980 linearize_cond_branch(ep
, pre_condition
, loop_body
, loop_end
);
1982 bind_label(stmt
->iterator_continue
, loop_continue
, stmt
->pos
);
1983 bind_label(stmt
->iterator_break
, loop_end
, stmt
->pos
);
1985 set_activeblock(ep
, loop_body
);
1986 linearize_statement(ep
, statement
);
1987 add_goto(ep
, loop_continue
);
1989 set_activeblock(ep
, loop_continue
);
1990 linearize_statement(ep
, post_statement
);
1991 if (!post_condition
)
1992 add_goto(ep
, loop_top
);
1994 linearize_cond_branch(ep
, post_condition
, loop_top
, loop_end
);
1995 set_activeblock(ep
, loop_end
);
2000 pseudo_t
linearize_statement(struct entrypoint
*ep
, struct statement
*stmt
)
2002 struct basic_block
*bb
;
2008 if (bb
&& !bb
->insns
)
2009 bb
->pos
= stmt
->pos
;
2010 current_pos
= stmt
->pos
;
2012 switch (stmt
->type
) {
2016 case STMT_DECLARATION
:
2017 return linearize_declaration(ep
, stmt
);
2020 return linearize_context(ep
, stmt
);
2023 return linearize_range(ep
, stmt
);
2025 case STMT_EXPRESSION
:
2026 return linearize_expression(ep
, stmt
->expression
);
2029 return linearize_asm_statement(ep
, stmt
);
2032 return linearize_return(ep
, stmt
);
2035 add_label(ep
, stmt
->case_label
);
2036 linearize_statement(ep
, stmt
->case_statement
);
2041 struct symbol
*label
= stmt
->label_identifier
;
2044 add_label(ep
, label
);
2046 return linearize_statement(ep
, stmt
->label_statement
);
2051 struct expression
*expr
;
2052 struct instruction
*goto_ins
;
2053 struct basic_block
*active
;
2056 active
= ep
->active
;
2057 if (!bb_reachable(active
))
2060 if (stmt
->goto_label
) {
2061 add_goto(ep
, get_bound_block(ep
, stmt
->goto_label
));
2065 expr
= stmt
->goto_expression
;
2069 /* This can happen as part of simplification */
2070 if (expr
->type
== EXPR_LABEL
) {
2071 add_goto(ep
, get_bound_block(ep
, expr
->label_symbol
));
2075 pseudo
= linearize_expression(ep
, expr
);
2076 goto_ins
= alloc_instruction(OP_COMPUTEDGOTO
, 0);
2077 use_pseudo(goto_ins
, pseudo
, &goto_ins
->target
);
2078 add_one_insn(ep
, goto_ins
);
2080 FOR_EACH_PTR(stmt
->target_list
, sym
) {
2081 struct basic_block
*bb_computed
= get_bound_block(ep
, sym
);
2082 struct multijmp
*jmp
= alloc_multijmp(bb_computed
, 1, 0);
2083 add_multijmp(&goto_ins
->multijmp_list
, jmp
);
2084 add_bb(&bb_computed
->parents
, ep
->active
);
2085 add_bb(&active
->children
, bb_computed
);
2086 } END_FOR_EACH_PTR(sym
);
2093 if (stmt
->inline_fn
)
2094 return linearize_inlined_call(ep
, stmt
);
2095 return linearize_compound_statement(ep
, stmt
);
2098 * This could take 'likely/unlikely' into account, and
2099 * switch the arms around appropriately..
2102 struct basic_block
*bb_true
, *bb_false
, *endif
;
2103 struct expression
*cond
= stmt
->if_conditional
;
2105 bb_true
= alloc_basic_block(ep
, stmt
->pos
);
2106 bb_false
= endif
= alloc_basic_block(ep
, stmt
->pos
);
2108 linearize_cond_branch(ep
, cond
, bb_true
, bb_false
);
2110 set_activeblock(ep
, bb_true
);
2111 linearize_statement(ep
, stmt
->if_true
);
2113 if (stmt
->if_false
) {
2114 endif
= alloc_basic_block(ep
, stmt
->pos
);
2115 add_goto(ep
, endif
);
2116 set_activeblock(ep
, bb_false
);
2117 linearize_statement(ep
, stmt
->if_false
);
2119 set_activeblock(ep
, endif
);
2124 return linearize_switch(ep
, stmt
);
2127 return linearize_iterator(ep
, stmt
);
2135 static struct entrypoint
*linearize_fn(struct symbol
*sym
, struct symbol
*base_type
)
2137 struct entrypoint
*ep
;
2138 struct basic_block
*bb
;
2140 struct instruction
*entry
;
2144 if (!base_type
->stmt
)
2147 ep
= alloc_entrypoint();
2148 bb
= alloc_basic_block(ep
, sym
->pos
);
2152 set_activeblock(ep
, bb
);
2154 entry
= alloc_instruction(OP_ENTRY
, 0);
2155 add_one_insn(ep
, entry
);
2158 concat_symbol_list(base_type
->arguments
, &ep
->syms
);
2160 /* FIXME!! We should do something else about varargs.. */
2162 FOR_EACH_PTR(base_type
->arguments
, arg
) {
2163 linearize_argument(ep
, arg
, ++i
);
2164 } END_FOR_EACH_PTR(arg
);
2166 result
= linearize_statement(ep
, base_type
->stmt
);
2167 if (bb_reachable(ep
->active
) && !bb_terminated(ep
->active
)) {
2168 struct symbol
*ret_type
= base_type
->ctype
.base_type
;
2169 struct instruction
*insn
= alloc_typed_instruction(OP_RET
, ret_type
);
2171 if (type_size(ret_type
) > 0)
2172 use_pseudo(insn
, result
, &insn
->src
);
2173 add_one_insn(ep
, insn
);
2177 * Do trivial flow simplification - branches to
2178 * branches, kill dead basicblocks etc
2180 kill_unreachable_bbs(ep
);
2183 * Turn symbols into pseudos
2185 simplify_symbol_usage(ep
);
2189 * Remove trivial instructions, and try to CSE
2193 cleanup_and_cse(ep
);
2194 pack_basic_blocks(ep
);
2195 } while (repeat_phase
& REPEAT_CSE
);
2197 kill_unreachable_bbs(ep
);
2201 clear_symbol_pseudos(ep
);
2203 /* And track pseudo register usage */
2204 track_pseudo_liveness(ep
);
2207 * Some flow optimizations can only effectively
2208 * be done when we've done liveness analysis. But
2209 * if they trigger, we need to start all over
2212 if (simplify_flow(ep
)) {
2217 /* Finally, add deathnotes to pseudos now that we have them */
2219 track_pseudo_death(ep
);
2224 struct entrypoint
*linearize_symbol(struct symbol
*sym
)
2226 struct symbol
*base_type
;
2230 current_pos
= sym
->pos
;
2231 base_type
= sym
->ctype
.base_type
;
2234 if (base_type
->type
== SYM_FN
)
2235 return linearize_fn(sym
, base_type
);