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 void linearize_one_symbol(struct entrypoint
*ep
, struct symbol
*sym
);
33 static pseudo_t
add_load(struct entrypoint
*ep
, struct access_data
*);
34 pseudo_t
linearize_initializer(struct entrypoint
*ep
, struct expression
*initializer
, struct access_data
*);
36 struct pseudo void_pseudo
= {};
38 static struct instruction
*alloc_instruction(int opcode
, int size
)
40 struct instruction
* insn
= __alloc_instruction(0);
41 insn
->opcode
= opcode
;
46 static inline int type_size(struct symbol
*type
)
48 return type
? type
->bit_size
> 0 ? type
->bit_size
: 0 : 0;
51 static struct instruction
*alloc_typed_instruction(int opcode
, struct symbol
*type
)
53 return alloc_instruction(opcode
, type_size(type
));
56 static struct entrypoint
*alloc_entrypoint(void)
58 return __alloc_entrypoint(0);
61 static struct basic_block
*alloc_basic_block(struct entrypoint
*ep
, struct position pos
)
63 struct basic_block
*bb
= __alloc_basic_block(0);
70 static struct multijmp
* alloc_multijmp(struct basic_block
*target
, int begin
, int end
)
72 struct multijmp
*multijmp
= __alloc_multijmp(0);
73 multijmp
->target
= target
;
74 multijmp
->begin
= begin
;
79 static inline int regno(pseudo_t n
)
82 if (n
&& n
->type
== PSEUDO_REG
)
87 const char *show_pseudo(pseudo_t pseudo
)
90 static char buffer
[4][64];
98 buf
= buffer
[3 & ++n
];
99 switch(pseudo
->type
) {
101 struct symbol
*sym
= pseudo
->sym
;
102 struct expression
*expr
;
104 if (sym
->bb_target
) {
105 snprintf(buf
, 64, ".L%p", sym
->bb_target
);
109 snprintf(buf
, 64, "%s", show_ident(sym
->ident
));
112 expr
= sym
->initializer
;
113 snprintf(buf
, 64, "<anon symbol:%p>", sym
);
114 switch (expr
->type
) {
116 snprintf(buf
, 64, "<symbol value: %lld>", expr
->value
);
119 return show_string(expr
->string
);
126 i
= snprintf(buf
, 64, "%%r%d", pseudo
->nr
);
128 sprintf(buf
+i
, "(%s)", show_ident(pseudo
->ident
));
131 long long value
= pseudo
->value
;
132 if (value
> 1000 || value
< -1000)
133 snprintf(buf
, 64, "$%#llx", value
);
135 snprintf(buf
, 64, "$%lld", value
);
139 snprintf(buf
, 64, "%%arg%d", pseudo
->nr
);
142 i
= snprintf(buf
, 64, "%%phi%d", pseudo
->nr
);
144 sprintf(buf
+i
, "(%s)", show_ident(pseudo
->ident
));
147 snprintf(buf
, 64, "<bad pseudo type %d>", pseudo
->type
);
152 static const char* opcodes
[] = {
153 [OP_BADOP
] = "bad_op",
156 [OP_ENTRY
] = "<entry-point>",
161 [OP_SWITCH
] = "switch",
162 [OP_INVOKE
] = "invoke",
163 [OP_COMPUTEDGOTO
] = "jmp *",
164 [OP_UNWIND
] = "unwind",
179 [OP_AND_BOOL
] = "and-bool",
180 [OP_OR_BOOL
] = "or-bool",
182 /* Binary comparison */
183 [OP_SET_EQ
] = "seteq",
184 [OP_SET_NE
] = "setne",
185 [OP_SET_LE
] = "setle",
186 [OP_SET_GE
] = "setge",
187 [OP_SET_LT
] = "setlt",
188 [OP_SET_GT
] = "setgt",
191 [OP_SET_BE
] = "setbe",
192 [OP_SET_AE
] = "setae",
198 /* Special three-input */
202 [OP_MALLOC
] = "malloc",
204 [OP_ALLOCA
] = "alloca",
206 [OP_STORE
] = "store",
208 [OP_GET_ELEMENT_PTR
] = "getelem",
212 [OP_PHISOURCE
] = "phisrc",
214 [OP_PTRCAST
] = "ptrcast",
216 [OP_VANEXT
] = "va_next",
217 [OP_VAARG
] = "va_arg",
218 [OP_SLICE
] = "slice",
222 [OP_DEATHNOTE
] = "dead",
225 /* Sparse tagging (line numbers, context, whatever) */
226 [OP_CONTEXT
] = "context",
229 void show_instruction(struct instruction
*insn
)
231 int opcode
= insn
->opcode
;
232 static char buffer
[1024] = "\t";
239 buf
+= sprintf(buf
, "# ");
242 if (opcode
< sizeof(opcodes
)/sizeof(char *)) {
243 const char *op
= opcodes
[opcode
];
245 buf
+= sprintf(buf
, "opcode:%d", opcode
);
247 buf
+= sprintf(buf
, "%s", op
);
249 buf
+= sprintf(buf
, ".%d", insn
->size
);
250 memset(buf
, ' ', 20);
254 if (buf
< buffer
+ 12)
258 if (insn
->src
&& insn
->src
!= VOID
)
259 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->src
));
262 if (insn
->bb_true
&& insn
->bb_false
) {
263 buf
+= sprintf(buf
, "%s, .L%p, .L%p", show_pseudo(insn
->cond
), insn
->bb_true
, insn
->bb_false
);
266 buf
+= sprintf(buf
, ".L%p", insn
->bb_true
? insn
->bb_true
: insn
->bb_false
);
270 struct expression
*expr
= insn
->val
;
271 pseudo_t pseudo
= insn
->symbol
;
272 buf
+= sprintf(buf
, "%s <- ", show_pseudo(insn
->target
));
274 struct symbol
*sym
= pseudo
->sym
;
276 buf
+= sprintf(buf
, "%s", show_pseudo(pseudo
));
279 if (sym
->bb_target
) {
280 buf
+= sprintf(buf
, ".L%p", sym
->bb_target
);
284 buf
+= sprintf(buf
, "%s", show_ident(sym
->ident
));
287 buf
+= sprintf(buf
, "<anon symbol:%p>", sym
);
292 buf
+= sprintf(buf
, "%s", "<none>");
296 switch (expr
->type
) {
298 buf
+= sprintf(buf
, "%lld", expr
->value
);
301 buf
+= sprintf(buf
, "%Lf", expr
->fvalue
);
304 buf
+= sprintf(buf
, "%.40s", show_string(expr
->string
));
307 buf
+= sprintf(buf
, "%s", show_ident(expr
->symbol
->ident
));
310 buf
+= sprintf(buf
, ".L%p", expr
->symbol
->bb_target
);
313 buf
+= sprintf(buf
, "SETVAL EXPR TYPE %d", expr
->type
);
318 struct multijmp
*jmp
;
319 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->target
));
320 FOR_EACH_PTR(insn
->multijmp_list
, jmp
) {
321 if (jmp
->begin
== jmp
->end
)
322 buf
+= sprintf(buf
, ", %d -> .L%p", jmp
->begin
, jmp
->target
);
323 else if (jmp
->begin
< jmp
->end
)
324 buf
+= sprintf(buf
, ", %d ... %d -> .L%p", jmp
->begin
, jmp
->end
, jmp
->target
);
326 buf
+= sprintf(buf
, ", default -> .L%p", jmp
->target
);
327 } END_FOR_EACH_PTR(jmp
);
330 case OP_COMPUTEDGOTO
: {
331 struct multijmp
*jmp
;
332 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->target
));
333 FOR_EACH_PTR(insn
->multijmp_list
, jmp
) {
334 buf
+= sprintf(buf
, ", .L%p", jmp
->target
);
335 } END_FOR_EACH_PTR(jmp
);
340 struct instruction
*phi
;
341 buf
+= sprintf(buf
, "%s <- %s ", show_pseudo(insn
->target
), show_pseudo(insn
->phi_src
));
342 FOR_EACH_PTR(insn
->phi_users
, phi
) {
343 buf
+= sprintf(buf
, " (%s)", show_pseudo(phi
->target
));
344 } END_FOR_EACH_PTR(phi
);
350 const char *s
= " <-";
351 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->target
));
352 FOR_EACH_PTR(insn
->phi_list
, phi
) {
353 buf
+= sprintf(buf
, "%s %s", s
, show_pseudo(phi
));
355 } END_FOR_EACH_PTR(phi
);
358 case OP_LOAD
: case OP_LNOP
:
359 buf
+= sprintf(buf
, "%s <- %d[%s]", show_pseudo(insn
->target
), insn
->offset
, show_pseudo(insn
->src
));
361 case OP_STORE
: case OP_SNOP
:
362 buf
+= sprintf(buf
, "%s -> %d[%s]", show_pseudo(insn
->target
), insn
->offset
, show_pseudo(insn
->src
));
366 if (insn
->target
&& insn
->target
!= VOID
)
367 buf
+= sprintf(buf
, "%s <- ", show_pseudo(insn
->target
));
368 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->func
));
369 FOR_EACH_PTR(insn
->arguments
, arg
) {
370 buf
+= sprintf(buf
, ", %s", show_pseudo(arg
));
371 } END_FOR_EACH_PTR(arg
);
376 buf
+= sprintf(buf
, "%s <- (%d) %s",
377 show_pseudo(insn
->target
),
378 type_size(insn
->orig_type
),
379 show_pseudo(insn
->src
));
381 case OP_BINARY
... OP_BINARY_END
:
382 case OP_BINCMP
... OP_BINCMP_END
:
383 buf
+= sprintf(buf
, "%s <- %s, %s", show_pseudo(insn
->target
), show_pseudo(insn
->src1
), show_pseudo(insn
->src2
));
387 buf
+= sprintf(buf
, "%s <- %s, %s, %s", show_pseudo(insn
->target
),
388 show_pseudo(insn
->src1
), show_pseudo(insn
->src2
), show_pseudo(insn
->src3
));
392 buf
+= sprintf(buf
, "%s <- %s, %d, %d", show_pseudo(insn
->target
), show_pseudo(insn
->base
), insn
->from
, insn
->len
);
395 case OP_NOT
: case OP_NEG
:
396 buf
+= sprintf(buf
, "%s <- %s", show_pseudo(insn
->target
), show_pseudo(insn
->src1
));
400 buf
+= sprintf(buf
, "%d", insn
->increment
);
403 buf
+= sprintf(buf
, "%s <- %s", show_pseudo(insn
->target
), show_pseudo(insn
->src1
));
406 buf
+= sprintf(buf
, "%s", show_pseudo(insn
->target
));
409 buf
+= sprintf(buf
, "\"%s\"", insn
->string
);
412 buf
+= sprintf(buf
, " (");
413 FOR_EACH_PTR(insn
->outputs
, pseudo
) {
414 buf
+= sprintf(buf
, " %s", show_pseudo(pseudo
));
415 } END_FOR_EACH_PTR(pseudo
);
416 buf
+= sprintf(buf
, " ) <-");
420 buf
+= sprintf(buf
, " (");
421 FOR_EACH_PTR(insn
->inputs
, pseudo
) {
422 buf
+= sprintf(buf
, " %s", show_pseudo(pseudo
));
423 } END_FOR_EACH_PTR(pseudo
);
424 buf
+= sprintf(buf
, " )");
430 do { --buf
; } while (*buf
== ' ');
432 printf("%s\n", buffer
);
435 void show_bb(struct basic_block
*bb
)
437 struct instruction
*insn
;
439 printf(".L%p:\n", bb
);
441 pseudo_t needs
, defines
;
442 printf("%s:%d\n", input_streams
[bb
->pos
.stream
].name
, bb
->pos
.line
);
444 FOR_EACH_PTR(bb
->needs
, needs
) {
445 struct instruction
*def
= needs
->def
;
446 if (def
->opcode
!= OP_PHI
) {
447 printf(" **uses %s (from .L%p)**\n", show_pseudo(needs
), def
->bb
);
450 const char *sep
= " ";
451 printf(" **uses %s (from", show_pseudo(needs
));
452 FOR_EACH_PTR(def
->phi_list
, phi
) {
455 printf("%s(%s:.L%p)", sep
, show_pseudo(phi
), phi
->def
->bb
);
457 } END_FOR_EACH_PTR(phi
);
460 } END_FOR_EACH_PTR(needs
);
462 FOR_EACH_PTR(bb
->defines
, defines
) {
463 printf(" **defines %s **\n", show_pseudo(defines
));
464 } END_FOR_EACH_PTR(defines
);
467 struct basic_block
*from
;
468 FOR_EACH_PTR(bb
->parents
, from
) {
469 printf(" **from %p (%s:%d:%d)**\n", from
,
470 input_streams
[from
->pos
.stream
].name
, from
->pos
.line
, from
->pos
.pos
);
471 } END_FOR_EACH_PTR(from
);
475 struct basic_block
*to
;
476 FOR_EACH_PTR(bb
->children
, to
) {
477 printf(" **to %p (%s:%d:%d)**\n", to
,
478 input_streams
[to
->pos
.stream
].name
, to
->pos
.line
, to
->pos
.pos
);
479 } END_FOR_EACH_PTR(to
);
483 FOR_EACH_PTR(bb
->insns
, insn
) {
484 show_instruction(insn
);
485 } END_FOR_EACH_PTR(insn
);
486 if (!bb_terminated(bb
))
490 static void show_symbol_usage(pseudo_t pseudo
)
494 FOR_EACH_PTR(pseudo
->users
, pp
) {
495 struct instruction
*insn
= container(pp
, struct instruction
, src
);
496 show_instruction(insn
);
497 } END_FOR_EACH_PTR(pp
);
501 void show_entry(struct entrypoint
*ep
)
504 struct basic_block
*bb
;
506 printf("%s:\n", show_ident(ep
->name
->ident
));
509 printf("ep %p: %s\n", ep
, show_ident(ep
->name
->ident
));
511 FOR_EACH_PTR(ep
->syms
, sym
) {
514 if (!sym
->pseudo
->users
)
516 printf(" sym: %p %s\n", sym
, show_ident(sym
->ident
));
517 if (sym
->ctype
.modifiers
& (MOD_EXTERN
| MOD_STATIC
| MOD_ADDRESSABLE
))
518 printf("\texternal visibility\n");
519 show_symbol_usage(sym
->pseudo
);
520 } END_FOR_EACH_PTR(sym
);
525 FOR_EACH_PTR(ep
->bbs
, bb
) {
528 if (!bb
->parents
&& !bb
->children
&& !bb
->insns
&& verbose
< 2)
532 } END_FOR_EACH_PTR(bb
);
537 static void bind_label(struct symbol
*label
, struct basic_block
*bb
, struct position pos
)
539 if (label
->bb_target
)
540 warning(pos
, "label '%s' already bound", show_ident(label
->ident
));
541 label
->bb_target
= bb
;
544 static struct basic_block
* get_bound_block(struct entrypoint
*ep
, struct symbol
*label
)
546 struct basic_block
*bb
= label
->bb_target
;
549 bb
= alloc_basic_block(ep
, label
->pos
);
550 label
->bb_target
= bb
;
555 static void finish_block(struct entrypoint
*ep
)
557 struct basic_block
*src
= ep
->active
;
558 if (bb_reachable(src
))
562 static void add_goto(struct entrypoint
*ep
, struct basic_block
*dst
)
564 struct basic_block
*src
= ep
->active
;
565 if (bb_reachable(src
)) {
566 struct instruction
*br
= alloc_instruction(OP_BR
, 0);
568 add_bb(&dst
->parents
, src
);
569 add_bb(&src
->children
, dst
);
571 add_instruction(&src
->insns
, br
);
576 static void add_one_insn(struct entrypoint
*ep
, struct instruction
*insn
)
578 struct basic_block
*bb
= ep
->active
;
580 if (bb_reachable(bb
)) {
582 add_instruction(&bb
->insns
, insn
);
586 static void set_activeblock(struct entrypoint
*ep
, struct basic_block
*bb
)
588 if (!bb_terminated(ep
->active
))
592 if (bb_reachable(bb
))
593 add_bb(&ep
->bbs
, bb
);
596 static void remove_parent(struct basic_block
*child
, struct basic_block
*parent
)
598 remove_bb_from_list(&child
->parents
, parent
, 1);
603 /* Change a "switch" into a branch */
604 void insert_branch(struct basic_block
*bb
, struct instruction
*jmp
, struct basic_block
*target
)
606 struct instruction
*br
, *old
;
607 struct basic_block
*child
;
609 /* Remove the switch */
610 old
= delete_last_instruction(&bb
->insns
);
613 br
= alloc_instruction(OP_BR
, 0);
615 br
->bb_true
= target
;
616 add_instruction(&bb
->insns
, br
);
618 FOR_EACH_PTR(bb
->children
, child
) {
619 if (child
== target
) {
620 target
= NULL
; /* Trigger just once */
623 DELETE_CURRENT_PTR(child
);
624 remove_parent(child
, bb
);
625 } END_FOR_EACH_PTR(child
);
626 PACK_PTR_LIST(&bb
->children
);
630 void insert_select(struct basic_block
*bb
, struct instruction
*br
, struct instruction
*phi_node
, pseudo_t
true, pseudo_t
false)
633 struct instruction
*select
;
635 /* Remove the 'br' */
636 delete_last_instruction(&bb
->insns
);
638 select
= alloc_instruction(OP_SEL
, phi_node
->size
);
642 use_pseudo(br
->cond
, &select
->src1
);
644 target
= phi_node
->target
;
645 assert(target
->def
== phi_node
);
646 select
->target
= target
;
647 target
->def
= select
;
649 use_pseudo(true, &select
->src2
);
650 use_pseudo(false, &select
->src3
);
652 add_instruction(&bb
->insns
, select
);
653 add_instruction(&bb
->insns
, br
);
656 static inline int bb_empty(struct basic_block
*bb
)
661 /* Add a label to the currently active block, return new active block */
662 static struct basic_block
* add_label(struct entrypoint
*ep
, struct symbol
*label
)
664 struct basic_block
*bb
= label
->bb_target
;
667 set_activeblock(ep
, bb
);
671 if (!bb_reachable(bb
) || !bb_empty(bb
)) {
672 bb
= alloc_basic_block(ep
, label
->pos
);
673 set_activeblock(ep
, bb
);
675 label
->bb_target
= bb
;
679 static void add_branch(struct entrypoint
*ep
, struct expression
*expr
, pseudo_t cond
, struct basic_block
*bb_true
, struct basic_block
*bb_false
)
681 struct basic_block
*bb
= ep
->active
;
682 struct instruction
*br
;
684 if (bb_reachable(bb
)) {
685 br
= alloc_instruction(OP_BR
, 0);
686 use_pseudo(cond
, &br
->cond
);
687 br
->bb_true
= bb_true
;
688 br
->bb_false
= bb_false
;
689 add_bb(&bb_true
->parents
, bb
);
690 add_bb(&bb_false
->parents
, bb
);
691 add_bb(&bb
->children
, bb_true
);
692 add_bb(&bb
->children
, bb_false
);
693 add_one_insn(ep
, br
);
697 /* Dummy pseudo allocator */
698 pseudo_t
alloc_pseudo(struct instruction
*def
)
701 struct pseudo
* pseudo
= __alloc_pseudo(0);
702 pseudo
->type
= PSEUDO_REG
;
708 static void clear_symbol_pseudos(struct entrypoint
*ep
)
712 FOR_EACH_PTR(ep
->accesses
, sym
) {
714 } END_FOR_EACH_PTR(sym
);
717 static pseudo_t
symbol_pseudo(struct entrypoint
*ep
, struct symbol
*sym
)
724 pseudo
= sym
->pseudo
;
726 pseudo
= __alloc_pseudo(0);
727 pseudo
->type
= PSEUDO_SYM
;
729 pseudo
->ident
= sym
->ident
;
730 sym
->pseudo
= pseudo
;
731 add_symbol(&ep
->accesses
, sym
);
733 /* Symbol pseudos have neither nr, usage nor def */
737 pseudo_t
value_pseudo(long long val
)
739 #define MAX_VAL_HASH 64
740 static struct pseudo_list
*prev
[MAX_VAL_HASH
];
741 int hash
= val
& (MAX_VAL_HASH
-1);
742 struct pseudo_list
**list
= prev
+ hash
;
745 FOR_EACH_PTR(*list
, pseudo
) {
746 if (pseudo
->value
== val
)
748 } END_FOR_EACH_PTR(pseudo
);
750 pseudo
= __alloc_pseudo(0);
751 pseudo
->type
= PSEUDO_VAL
;
753 add_pseudo(list
, pseudo
);
755 /* Value pseudos have neither nr, usage nor def */
759 static pseudo_t
argument_pseudo(struct entrypoint
*ep
, int nr
)
761 pseudo_t pseudo
= __alloc_pseudo(0);
762 pseudo
->type
= PSEUDO_ARG
;
764 pseudo
->def
= ep
->entry
;
765 /* Argument pseudos have neither usage nor def */
769 pseudo_t
alloc_phi(struct basic_block
*source
, pseudo_t pseudo
, int size
)
771 struct instruction
*insn
= alloc_instruction(OP_PHISOURCE
, size
);
772 pseudo_t phi
= __alloc_pseudo(0);
775 phi
->type
= PSEUDO_PHI
;
779 use_pseudo(pseudo
, &insn
->phi_src
);
782 add_instruction(&source
->insns
, insn
);
787 * We carry the "access_data" structure around for any accesses,
788 * which simplifies things a lot. It contains all the access
789 * information in one place.
792 struct symbol
*result_type
; // result ctype
793 struct symbol
*source_type
; // source ctype
794 pseudo_t address
; // pseudo containing address ..
795 pseudo_t origval
; // pseudo for original value ..
796 unsigned int offset
, alignment
; // byte offset
797 unsigned int bit_size
, bit_offset
; // which bits
801 static void finish_address_gen(struct entrypoint
*ep
, struct access_data
*ad
)
805 static int linearize_simple_address(struct entrypoint
*ep
,
806 struct expression
*addr
,
807 struct access_data
*ad
)
809 if (addr
->type
== EXPR_SYMBOL
) {
810 linearize_one_symbol(ep
, addr
->symbol
);
811 ad
->address
= symbol_pseudo(ep
, addr
->symbol
);
814 if (addr
->type
== EXPR_BINOP
) {
815 if (addr
->right
->type
== EXPR_VALUE
) {
816 if (addr
->op
== '+') {
817 ad
->offset
+= get_expression_value(addr
->right
);
818 return linearize_simple_address(ep
, addr
->left
, ad
);
822 ad
->address
= linearize_expression(ep
, addr
);
826 static struct symbol
*base_type(struct symbol
*sym
)
828 struct symbol
*base
= sym
;
831 if (sym
->type
== SYM_NODE
)
832 base
= base
->ctype
.base_type
;
833 if (base
->type
== SYM_BITFIELD
)
834 return base
->ctype
.base_type
;
839 static int linearize_address_gen(struct entrypoint
*ep
,
840 struct expression
*expr
,
841 struct access_data
*ad
)
843 struct symbol
*ctype
= expr
->ctype
;
848 ad
->result_type
= ctype
;
849 ad
->source_type
= base_type(ctype
);
850 ad
->bit_size
= ctype
->bit_size
;
851 ad
->alignment
= ctype
->ctype
.alignment
;
852 ad
->bit_offset
= ctype
->bit_offset
;
853 if (expr
->type
== EXPR_PREOP
&& expr
->op
== '*')
854 return linearize_simple_address(ep
, expr
->unop
, ad
);
856 warning(expr
->pos
, "generating address of non-lvalue (%d)", expr
->type
);
860 static pseudo_t
add_load(struct entrypoint
*ep
, struct access_data
*ad
)
862 struct instruction
*insn
;
869 insn
= alloc_typed_instruction(OP_LOAD
, ad
->source_type
);
870 new = alloc_pseudo(insn
);
874 insn
->offset
= ad
->offset
;
875 use_pseudo(ad
->address
, &insn
->src
);
876 add_one_insn(ep
, insn
);
880 static void add_store(struct entrypoint
*ep
, struct access_data
*ad
, pseudo_t value
)
882 struct basic_block
*bb
= ep
->active
;
884 if (bb_reachable(bb
)) {
885 struct instruction
*store
= alloc_typed_instruction(OP_STORE
, ad
->source_type
);
886 store
->offset
= ad
->offset
;
887 use_pseudo(value
, &store
->target
);
888 use_pseudo(ad
->address
, &store
->src
);
889 add_one_insn(ep
, store
);
893 static pseudo_t
linearize_store_gen(struct entrypoint
*ep
,
895 struct access_data
*ad
)
897 pseudo_t store
= value
;
899 if (type_size(ad
->source_type
) != type_size(ad
->result_type
)) {
900 pseudo_t orig
= add_load(ep
, ad
);
901 int shift
= ad
->bit_offset
;
902 unsigned long long mask
= (1ULL << ad
->bit_size
)-1;
905 store
= add_binary_op(ep
, ad
->source_type
, OP_SHL
, value
, value_pseudo(shift
));
908 orig
= add_binary_op(ep
, ad
->source_type
, OP_AND
, orig
, value_pseudo(~mask
));
909 store
= add_binary_op(ep
, ad
->source_type
, OP_OR
, orig
, store
);
911 add_store(ep
, ad
, store
);
915 static pseudo_t
add_binary_op(struct entrypoint
*ep
, struct symbol
*ctype
, int op
, pseudo_t left
, pseudo_t right
)
917 struct instruction
*insn
= alloc_typed_instruction(op
, ctype
);
918 pseudo_t target
= alloc_pseudo(insn
);
919 insn
->target
= target
;
920 use_pseudo(left
, &insn
->src1
);
921 use_pseudo(right
, &insn
->src2
);
922 add_one_insn(ep
, insn
);
926 static pseudo_t
add_setval(struct entrypoint
*ep
, struct symbol
*ctype
, struct expression
*val
)
928 struct instruction
*insn
= alloc_typed_instruction(OP_SETVAL
, ctype
);
929 pseudo_t target
= alloc_pseudo(insn
);
930 insn
->target
= target
;
933 pseudo_t addr
= symbol_pseudo(ep
, ctype
);
934 use_pseudo(addr
, &insn
->symbol
);
935 insn
->size
= bits_in_pointer
;
937 add_one_insn(ep
, insn
);
941 static pseudo_t
linearize_load_gen(struct entrypoint
*ep
, struct access_data
*ad
)
943 pseudo_t
new = add_load(ep
, ad
);
945 if (ad
->bit_offset
) {
946 pseudo_t shift
= value_pseudo(ad
->bit_offset
);
947 pseudo_t newval
= add_binary_op(ep
, ad
->source_type
, OP_SHR
, new, shift
);
954 static pseudo_t
linearize_access(struct entrypoint
*ep
, struct expression
*expr
)
956 struct access_data ad
= { NULL
, };
959 if (!linearize_address_gen(ep
, expr
, &ad
))
961 value
= linearize_load_gen(ep
, &ad
);
962 finish_address_gen(ep
, &ad
);
967 static pseudo_t
linearize_inc_dec(struct entrypoint
*ep
, struct expression
*expr
, int postop
)
969 struct access_data ad
= { NULL
, };
970 pseudo_t old
, new, one
;
971 int op
= expr
->op
== SPECIAL_INCREMENT
? OP_ADD
: OP_SUB
;
973 if (!linearize_address_gen(ep
, expr
->unop
, &ad
))
976 old
= linearize_load_gen(ep
, &ad
);
977 one
= value_pseudo(expr
->op_value
);
978 new = add_binary_op(ep
, expr
->ctype
, op
, old
, one
);
979 linearize_store_gen(ep
, new, &ad
);
980 finish_address_gen(ep
, &ad
);
981 return postop
? old
: new;
984 static pseudo_t
add_uniop(struct entrypoint
*ep
, struct expression
*expr
, int op
, pseudo_t src
)
986 struct instruction
*insn
= alloc_typed_instruction(op
, expr
->ctype
);
987 pseudo_t
new = alloc_pseudo(insn
);
990 use_pseudo(src
, &insn
->src1
);
991 add_one_insn(ep
, insn
);
995 static pseudo_t
linearize_slice(struct entrypoint
*ep
, struct expression
*expr
)
997 pseudo_t pre
= linearize_expression(ep
, expr
->base
);
998 struct instruction
*insn
= alloc_typed_instruction(OP_SLICE
, expr
->ctype
);
999 pseudo_t
new = alloc_pseudo(insn
);
1002 insn
->from
= expr
->r_bitpos
;
1003 insn
->len
= expr
->r_nrbits
;
1004 use_pseudo(pre
, &insn
->base
);
1005 add_one_insn(ep
, insn
);
1009 static pseudo_t
linearize_regular_preop(struct entrypoint
*ep
, struct expression
*expr
)
1011 pseudo_t pre
= linearize_expression(ep
, expr
->unop
);
1016 pseudo_t zero
= value_pseudo(0);
1017 return add_binary_op(ep
, expr
->ctype
, OP_SET_EQ
, pre
, zero
);
1020 return add_uniop(ep
, expr
, OP_NOT
, pre
);
1022 return add_uniop(ep
, expr
, OP_NEG
, pre
);
1027 static pseudo_t
linearize_preop(struct entrypoint
*ep
, struct expression
*expr
)
1030 * '*' is an lvalue access, and is fundamentally different
1031 * from an arithmetic operation. Maybe it should have an
1032 * expression type of its own..
1034 if (expr
->op
== '*')
1035 return linearize_access(ep
, expr
);
1036 if (expr
->op
== SPECIAL_INCREMENT
|| expr
->op
== SPECIAL_DECREMENT
)
1037 return linearize_inc_dec(ep
, expr
, 0);
1038 return linearize_regular_preop(ep
, expr
);
1041 static pseudo_t
linearize_postop(struct entrypoint
*ep
, struct expression
*expr
)
1043 return linearize_inc_dec(ep
, expr
, 1);
1046 static pseudo_t
linearize_assignment(struct entrypoint
*ep
, struct expression
*expr
)
1048 struct access_data ad
= { NULL
, };
1049 struct expression
*target
= expr
->left
;
1052 value
= linearize_expression(ep
, expr
->right
);
1053 if (!linearize_address_gen(ep
, target
, &ad
))
1055 if (expr
->op
!= '=') {
1056 pseudo_t oldvalue
= linearize_load_gen(ep
, &ad
);
1058 static const int op_trans
[] = {
1059 [SPECIAL_ADD_ASSIGN
- SPECIAL_BASE
] = OP_ADD
,
1060 [SPECIAL_SUB_ASSIGN
- SPECIAL_BASE
] = OP_SUB
,
1061 [SPECIAL_MUL_ASSIGN
- SPECIAL_BASE
] = OP_MUL
,
1062 [SPECIAL_DIV_ASSIGN
- SPECIAL_BASE
] = OP_DIV
,
1063 [SPECIAL_MOD_ASSIGN
- SPECIAL_BASE
] = OP_MOD
,
1064 [SPECIAL_SHL_ASSIGN
- SPECIAL_BASE
] = OP_SHL
,
1065 [SPECIAL_SHR_ASSIGN
- SPECIAL_BASE
] = OP_SHR
,
1066 [SPECIAL_AND_ASSIGN
- SPECIAL_BASE
] = OP_AND
,
1067 [SPECIAL_OR_ASSIGN
- SPECIAL_BASE
] = OP_OR
,
1068 [SPECIAL_XOR_ASSIGN
- SPECIAL_BASE
] = OP_XOR
1070 dst
= add_binary_op(ep
, expr
->ctype
, op_trans
[expr
->op
- SPECIAL_BASE
], oldvalue
, value
);
1073 value
= linearize_store_gen(ep
, value
, &ad
);
1074 finish_address_gen(ep
, &ad
);
1078 static pseudo_t
linearize_call_expression(struct entrypoint
*ep
, struct expression
*expr
)
1080 struct expression
*arg
, *fn
;
1081 struct instruction
*insn
= alloc_typed_instruction(OP_CALL
, expr
->ctype
);
1082 pseudo_t retval
, call
;
1086 warning(expr
->pos
, "call with no type!");
1090 FOR_EACH_PTR(expr
->args
, arg
) {
1091 pseudo_t
new = linearize_expression(ep
, arg
);
1092 use_pseudo(new, add_pseudo(&insn
->arguments
, new));
1093 } END_FOR_EACH_PTR(arg
);
1099 int in
= fn
->ctype
->ctype
.in_context
;
1100 int out
= fn
->ctype
->ctype
.out_context
;
1101 if (in
< 0 || out
< 0)
1103 context_diff
= out
- in
;
1106 if (fn
->type
== EXPR_PREOP
) {
1107 if (fn
->unop
->type
== EXPR_SYMBOL
) {
1108 struct symbol
*sym
= fn
->unop
->symbol
;
1109 if (sym
->ctype
.base_type
->type
== SYM_FN
)
1113 if (fn
->type
== EXPR_SYMBOL
) {
1114 call
= symbol_pseudo(ep
, fn
->symbol
);
1116 call
= linearize_expression(ep
, fn
);
1118 use_pseudo(call
, &insn
->func
);
1120 if (expr
->ctype
!= &void_ctype
)
1121 retval
= alloc_pseudo(insn
);
1122 insn
->target
= retval
;
1123 add_one_insn(ep
, insn
);
1126 insn
= alloc_instruction(OP_CONTEXT
, 0);
1127 insn
->increment
= context_diff
;
1128 add_one_insn(ep
, insn
);
1134 static pseudo_t
linearize_binop(struct entrypoint
*ep
, struct expression
*expr
)
1136 pseudo_t src1
, src2
, dst
;
1137 static const int opcode
[] = {
1138 ['+'] = OP_ADD
, ['-'] = OP_SUB
,
1139 ['*'] = OP_MUL
, ['/'] = OP_DIV
,
1140 ['%'] = OP_MOD
, ['&'] = OP_AND
,
1141 ['|'] = OP_OR
, ['^'] = OP_XOR
,
1142 [SPECIAL_LEFTSHIFT
] = OP_SHL
,
1143 [SPECIAL_RIGHTSHIFT
] = OP_SHR
,
1144 [SPECIAL_LOGICAL_AND
] = OP_AND_BOOL
,
1145 [SPECIAL_LOGICAL_OR
] = OP_OR_BOOL
,
1148 src1
= linearize_expression(ep
, expr
->left
);
1149 src2
= linearize_expression(ep
, expr
->right
);
1150 dst
= add_binary_op(ep
, expr
->ctype
, opcode
[expr
->op
], src1
, src2
);
1154 static pseudo_t
linearize_logical_branch(struct entrypoint
*ep
, struct expression
*expr
, struct basic_block
*bb_true
, struct basic_block
*bb_false
);
1156 pseudo_t
linearize_cond_branch(struct entrypoint
*ep
, struct expression
*expr
, struct basic_block
*bb_true
, struct basic_block
*bb_false
);
1158 static pseudo_t
linearize_select(struct entrypoint
*ep
, struct expression
*expr
)
1160 pseudo_t cond
, true, false, res
;
1161 struct instruction
*insn
;
1163 true = linearize_expression(ep
, expr
->cond_true
);
1164 false = linearize_expression(ep
, expr
->cond_false
);
1165 cond
= linearize_expression(ep
, expr
->conditional
);
1167 insn
= alloc_typed_instruction(OP_SEL
, expr
->ctype
);
1168 if (!expr
->cond_true
)
1170 use_pseudo(cond
, &insn
->src1
);
1171 use_pseudo(true, &insn
->src2
);
1172 use_pseudo(false, &insn
->src3
);
1174 res
= alloc_pseudo(insn
);
1176 add_one_insn(ep
, insn
);
1180 static pseudo_t
add_join_conditional(struct entrypoint
*ep
, struct expression
*expr
,
1181 pseudo_t phi1
, pseudo_t phi2
)
1184 struct instruction
*phi_node
;
1191 phi_node
= alloc_typed_instruction(OP_PHI
, expr
->ctype
);
1192 use_pseudo(phi1
, add_pseudo(&phi_node
->phi_list
, phi1
));
1193 use_pseudo(phi2
, add_pseudo(&phi_node
->phi_list
, phi2
));
1194 phi_node
->target
= target
= alloc_pseudo(phi_node
);
1195 add_one_insn(ep
, phi_node
);
1199 static pseudo_t
linearize_short_conditional(struct entrypoint
*ep
, struct expression
*expr
,
1200 struct expression
*cond
,
1201 struct expression
*expr_false
)
1203 pseudo_t src1
, src2
;
1204 struct basic_block
*bb_false
= alloc_basic_block(ep
, expr_false
->pos
);
1205 struct basic_block
*merge
= alloc_basic_block(ep
, expr
->pos
);
1206 pseudo_t phi1
, phi2
;
1207 int size
= type_size(expr
->ctype
);
1209 src1
= linearize_expression(ep
, cond
);
1210 phi1
= alloc_phi(ep
->active
, src1
, size
);
1211 add_branch(ep
, expr
, src1
, merge
, bb_false
);
1213 set_activeblock(ep
, bb_false
);
1214 src2
= linearize_expression(ep
, expr_false
);
1215 phi2
= alloc_phi(ep
->active
, src2
, size
);
1216 set_activeblock(ep
, merge
);
1218 return add_join_conditional(ep
, expr
, phi1
, phi2
);
1221 static pseudo_t
linearize_conditional(struct entrypoint
*ep
, struct expression
*expr
,
1222 struct expression
*cond
,
1223 struct expression
*expr_true
,
1224 struct expression
*expr_false
)
1226 pseudo_t src1
, src2
;
1227 pseudo_t phi1
, phi2
;
1228 struct basic_block
*bb_true
= alloc_basic_block(ep
, expr_true
->pos
);
1229 struct basic_block
*bb_false
= alloc_basic_block(ep
, expr_false
->pos
);
1230 struct basic_block
*merge
= alloc_basic_block(ep
, expr
->pos
);
1231 int size
= type_size(expr
->ctype
);
1233 linearize_cond_branch(ep
, cond
, bb_true
, bb_false
);
1235 set_activeblock(ep
, bb_true
);
1236 src1
= linearize_expression(ep
, expr_true
);
1237 phi1
= alloc_phi(ep
->active
, src1
, size
);
1238 add_goto(ep
, merge
);
1240 set_activeblock(ep
, bb_false
);
1241 src2
= linearize_expression(ep
, expr_false
);
1242 phi2
= alloc_phi(ep
->active
, src2
, size
);
1243 set_activeblock(ep
, merge
);
1245 return add_join_conditional(ep
, expr
, phi1
, phi2
);
1248 static pseudo_t
linearize_logical(struct entrypoint
*ep
, struct expression
*expr
)
1250 struct expression
*shortcut
;
1252 shortcut
= alloc_const_expression(expr
->pos
, expr
->op
== SPECIAL_LOGICAL_OR
);
1253 shortcut
->ctype
= expr
->ctype
;
1254 return linearize_conditional(ep
, expr
, expr
->left
, shortcut
, expr
->right
);
1257 static pseudo_t
linearize_compare(struct entrypoint
*ep
, struct expression
*expr
)
1259 static const int cmpop
[] = {
1260 ['>'] = OP_SET_GT
, ['<'] = OP_SET_LT
,
1261 [SPECIAL_EQUAL
] = OP_SET_EQ
,
1262 [SPECIAL_NOTEQUAL
] = OP_SET_NE
,
1263 [SPECIAL_GTE
] = OP_SET_GE
,
1264 [SPECIAL_LTE
] = OP_SET_LE
,
1265 [SPECIAL_UNSIGNED_LT
] = OP_SET_B
,
1266 [SPECIAL_UNSIGNED_GT
] = OP_SET_A
,
1267 [SPECIAL_UNSIGNED_LTE
] = OP_SET_BE
,
1268 [SPECIAL_UNSIGNED_GTE
] = OP_SET_AE
,
1271 pseudo_t src1
= linearize_expression(ep
, expr
->left
);
1272 pseudo_t src2
= linearize_expression(ep
, expr
->right
);
1273 pseudo_t dst
= add_binary_op(ep
, expr
->ctype
, cmpop
[expr
->op
], src1
, src2
);
1278 pseudo_t
linearize_cond_branch(struct entrypoint
*ep
, struct expression
*expr
, struct basic_block
*bb_true
, struct basic_block
*bb_false
)
1282 if (!expr
|| !bb_reachable(ep
->active
))
1285 switch (expr
->type
) {
1289 add_goto(ep
, expr
->value
? bb_true
: bb_false
);
1293 add_goto(ep
, expr
->fvalue
? bb_true
: bb_false
);
1297 linearize_logical_branch(ep
, expr
, bb_true
, bb_false
);
1301 cond
= linearize_compare(ep
, expr
);
1302 add_branch(ep
, expr
, cond
, bb_true
, bb_false
);
1306 if (expr
->op
== '!')
1307 return linearize_cond_branch(ep
, expr
->unop
, bb_false
, bb_true
);
1310 cond
= linearize_expression(ep
, expr
);
1311 add_branch(ep
, expr
, cond
, bb_true
, bb_false
);
1321 static pseudo_t
linearize_logical_branch(struct entrypoint
*ep
, struct expression
*expr
, struct basic_block
*bb_true
, struct basic_block
*bb_false
)
1323 struct basic_block
*next
= alloc_basic_block(ep
, expr
->pos
);
1325 if (expr
->op
== SPECIAL_LOGICAL_OR
)
1326 linearize_cond_branch(ep
, expr
->left
, bb_true
, next
);
1328 linearize_cond_branch(ep
, expr
->left
, next
, bb_false
);
1329 set_activeblock(ep
, next
);
1330 linearize_cond_branch(ep
, expr
->right
, bb_true
, bb_false
);
1335 * Casts to pointers are "less safe" than other casts, since
1336 * they imply type-unsafe accesses. "void *" is a special
1337 * case, since you can't access through it anyway without another
1340 static struct instruction
*alloc_cast_instruction(struct symbol
*ctype
)
1342 int opcode
= OP_CAST
;
1343 struct symbol
*base
= ctype
;
1345 if (base
->type
== SYM_NODE
)
1346 base
= base
->ctype
.base_type
;
1347 if (base
->type
== SYM_PTR
) {
1348 base
= base
->ctype
.base_type
;
1349 if (base
!= &void_ctype
)
1350 opcode
= OP_PTRCAST
;
1352 return alloc_typed_instruction(opcode
, ctype
);
1355 pseudo_t
linearize_cast(struct entrypoint
*ep
, struct expression
*expr
)
1357 pseudo_t src
, result
;
1358 struct instruction
*insn
;
1360 src
= linearize_expression(ep
, expr
->cast_expression
);
1365 if (expr
->ctype
->bit_size
< 0)
1368 insn
= alloc_cast_instruction(expr
->ctype
);
1369 result
= alloc_pseudo(insn
);
1370 insn
->target
= result
;
1371 insn
->orig_type
= expr
->cast_expression
->ctype
;
1372 use_pseudo(src
, &insn
->src
);
1373 add_one_insn(ep
, insn
);
1377 pseudo_t
linearize_position(struct entrypoint
*ep
, struct expression
*pos
, struct access_data
*ad
)
1379 struct expression
*init_expr
= pos
->init_expr
;
1380 pseudo_t value
= linearize_expression(ep
, init_expr
);
1382 ad
->offset
= pos
->init_offset
;
1383 ad
->source_type
= base_type(init_expr
->ctype
);
1384 ad
->result_type
= init_expr
->ctype
;
1385 linearize_store_gen(ep
, value
, ad
);
1389 pseudo_t
linearize_initializer(struct entrypoint
*ep
, struct expression
*initializer
, struct access_data
*ad
)
1391 switch (initializer
->type
) {
1392 case EXPR_INITIALIZER
: {
1393 struct expression
*expr
;
1394 FOR_EACH_PTR(initializer
->expr_list
, expr
) {
1395 linearize_initializer(ep
, expr
, ad
);
1396 } END_FOR_EACH_PTR(expr
);
1400 linearize_position(ep
, initializer
, ad
);
1403 pseudo_t value
= linearize_expression(ep
, initializer
);
1404 ad
->source_type
= base_type(initializer
->ctype
);
1405 ad
->result_type
= initializer
->ctype
;
1406 linearize_store_gen(ep
, value
, ad
);
1413 void linearize_argument(struct entrypoint
*ep
, struct symbol
*arg
, int nr
)
1415 struct access_data ad
= { NULL
, };
1417 ad
.source_type
= arg
;
1418 ad
.result_type
= arg
;
1419 ad
.address
= symbol_pseudo(ep
, arg
);
1420 linearize_store_gen(ep
, argument_pseudo(ep
, nr
), &ad
);
1421 finish_address_gen(ep
, &ad
);
1424 pseudo_t
linearize_expression(struct entrypoint
*ep
, struct expression
*expr
)
1429 switch (expr
->type
) {
1431 linearize_one_symbol(ep
, expr
->symbol
);
1432 return add_setval(ep
, expr
->symbol
, NULL
);
1435 return value_pseudo(expr
->value
);
1437 case EXPR_STRING
: case EXPR_FVALUE
: case EXPR_LABEL
:
1438 return add_setval(ep
, expr
->ctype
, expr
);
1440 case EXPR_STATEMENT
:
1441 return linearize_statement(ep
, expr
->statement
);
1444 return linearize_call_expression(ep
, expr
);
1447 return linearize_binop(ep
, expr
);
1450 return linearize_logical(ep
, expr
);
1453 return linearize_compare(ep
, expr
);
1456 return linearize_select(ep
, expr
);
1458 case EXPR_CONDITIONAL
:
1459 if (!expr
->cond_true
)
1460 return linearize_short_conditional(ep
, expr
, expr
->conditional
, expr
->cond_false
);
1462 return linearize_conditional(ep
, expr
, expr
->conditional
,
1463 expr
->cond_true
, expr
->cond_false
);
1466 linearize_expression(ep
, expr
->left
);
1467 return linearize_expression(ep
, expr
->right
);
1469 case EXPR_ASSIGNMENT
:
1470 return linearize_assignment(ep
, expr
);
1473 return linearize_preop(ep
, expr
);
1476 return linearize_postop(ep
, expr
);
1479 case EXPR_IMPLIED_CAST
:
1480 return linearize_cast(ep
, expr
);
1483 return linearize_slice(ep
, expr
);
1485 case EXPR_INITIALIZER
:
1487 warning(expr
->pos
, "unexpected initializer expression (%d %d)", expr
->type
, expr
->op
);
1490 warning(expr
->pos
, "unknown expression (%d %d)", expr
->type
, expr
->op
);
1496 static void linearize_one_symbol(struct entrypoint
*ep
, struct symbol
*sym
)
1498 struct access_data ad
= { NULL
, };
1500 if (!sym
|| !sym
->initializer
|| sym
->initialized
)
1503 /* We need to output these puppies some day too.. */
1504 if (sym
->ctype
.modifiers
& (MOD_STATIC
| MOD_TOPLEVEL
))
1507 sym
->initialized
= 1;
1508 ad
.address
= symbol_pseudo(ep
, sym
);
1509 linearize_initializer(ep
, sym
->initializer
, &ad
);
1510 finish_address_gen(ep
, &ad
);
1513 static pseudo_t
linearize_compound_statement(struct entrypoint
*ep
, struct statement
*stmt
)
1516 struct statement
*s
;
1518 struct symbol
*ret
= stmt
->ret
;
1520 concat_symbol_list(stmt
->syms
, &ep
->syms
);
1522 FOR_EACH_PTR(stmt
->syms
, sym
) {
1523 linearize_one_symbol(ep
, sym
);
1524 } END_FOR_EACH_PTR(sym
);
1527 FOR_EACH_PTR(stmt
->stmts
, s
) {
1528 pseudo
= linearize_statement(ep
, s
);
1529 } END_FOR_EACH_PTR(s
);
1532 struct basic_block
*bb
= add_label(ep
, ret
);
1533 struct instruction
*phi_node
= first_instruction(bb
->insns
);
1538 if (pseudo_list_size(phi_node
->phi_list
)==1) {
1539 pseudo
= first_pseudo(phi_node
->phi_list
);
1540 assert(pseudo
->type
== PSEUDO_PHI
);
1541 return pseudo
->def
->src1
;
1543 return phi_node
->target
;
1548 pseudo_t
linearize_internal(struct entrypoint
*ep
, struct statement
*stmt
)
1550 struct instruction
*insn
= alloc_instruction(OP_CONTEXT
, 0);
1551 struct expression
*expr
= stmt
->expression
;
1554 if (expr
->type
== EXPR_VALUE
)
1555 value
= expr
->value
;
1557 insn
->increment
= value
;
1558 add_one_insn(ep
, insn
);
1562 static void add_asm_input(struct entrypoint
*ep
, struct instruction
*insn
, struct expression
*expr
)
1564 pseudo_t pseudo
= linearize_expression(ep
, expr
);
1566 use_pseudo(pseudo
, add_pseudo(&insn
->inputs
, pseudo
));
1569 static void add_asm_output(struct entrypoint
*ep
, struct instruction
*insn
, struct expression
*expr
)
1571 struct access_data ad
= { NULL
, };
1572 pseudo_t pseudo
= alloc_pseudo(insn
);
1574 if (!linearize_address_gen(ep
, expr
, &ad
))
1576 linearize_store_gen(ep
, pseudo
, &ad
);
1577 finish_address_gen(ep
, &ad
);
1578 add_pseudo(&insn
->outputs
, pseudo
);
1581 pseudo_t
linearize_asm_statement(struct entrypoint
*ep
, struct statement
*stmt
)
1584 struct expression
*expr
;
1585 struct instruction
*insn
;
1587 insn
= alloc_instruction(OP_ASM
, 0);
1588 expr
= stmt
->asm_string
;
1589 if (!expr
|| expr
->type
!= EXPR_STRING
) {
1590 warning(stmt
->pos
, "expected string in inline asm");
1593 insn
->string
= expr
->string
->data
;
1595 /* Gather the inputs.. */
1597 FOR_EACH_PTR(stmt
->asm_inputs
, expr
) {
1598 even_odd
= 1 - even_odd
;
1600 /* FIXME! We ignore the constraints for now.. */
1603 add_asm_input(ep
, insn
, expr
);
1604 } END_FOR_EACH_PTR(expr
);
1606 add_one_insn(ep
, insn
);
1608 /* Assign the outputs */
1610 FOR_EACH_PTR(stmt
->asm_outputs
, expr
) {
1611 even_odd
= 1 - even_odd
;
1613 /* FIXME! We ignore the constraints for now.. */
1616 add_asm_output(ep
, insn
, expr
);
1617 } END_FOR_EACH_PTR(expr
);
1622 pseudo_t
linearize_statement(struct entrypoint
*ep
, struct statement
*stmt
)
1624 struct basic_block
*bb
;
1630 if (bb
&& !bb
->insns
)
1631 bb
->pos
= stmt
->pos
;
1633 switch (stmt
->type
) {
1638 return linearize_internal(ep
, stmt
);
1640 case STMT_EXPRESSION
:
1641 return linearize_expression(ep
, stmt
->expression
);
1644 return linearize_asm_statement(ep
, stmt
);
1647 struct expression
*expr
= stmt
->expression
;
1648 struct basic_block
*bb_return
= get_bound_block(ep
, stmt
->ret_target
);
1649 struct basic_block
*active
;
1650 pseudo_t src
= linearize_expression(ep
, expr
);
1651 active
= ep
->active
;
1652 if (active
&& src
!= &void_pseudo
) {
1653 struct instruction
*phi_node
= first_instruction(bb_return
->insns
);
1656 phi_node
= alloc_typed_instruction(OP_PHI
, expr
->ctype
);
1657 phi_node
->target
= alloc_pseudo(phi_node
);
1658 phi_node
->bb
= bb_return
;
1659 add_instruction(&bb_return
->insns
, phi_node
);
1661 phi
= alloc_phi(active
, src
, type_size(expr
->ctype
));
1662 phi
->ident
= &return_ident
;
1663 use_pseudo(phi
, add_pseudo(&phi_node
->phi_list
, phi
));
1665 add_goto(ep
, bb_return
);
1670 add_label(ep
, stmt
->case_label
);
1671 linearize_statement(ep
, stmt
->case_statement
);
1676 struct symbol
*label
= stmt
->label_identifier
;
1679 add_label(ep
, label
);
1680 linearize_statement(ep
, stmt
->label_statement
);
1687 struct expression
*expr
;
1688 struct instruction
*goto_ins
;
1689 struct basic_block
*active
;
1692 active
= ep
->active
;
1693 if (!bb_reachable(active
))
1696 if (stmt
->goto_label
) {
1697 add_goto(ep
, get_bound_block(ep
, stmt
->goto_label
));
1701 expr
= stmt
->goto_expression
;
1705 /* This can happen as part of simplification */
1706 if (expr
->type
== EXPR_LABEL
) {
1707 add_goto(ep
, get_bound_block(ep
, expr
->label_symbol
));
1711 pseudo
= linearize_expression(ep
, expr
);
1712 goto_ins
= alloc_instruction(OP_COMPUTEDGOTO
, 0);
1713 use_pseudo(pseudo
, &goto_ins
->target
);
1714 add_one_insn(ep
, goto_ins
);
1716 FOR_EACH_PTR(stmt
->target_list
, sym
) {
1717 struct basic_block
*bb_computed
= get_bound_block(ep
, sym
);
1718 struct multijmp
*jmp
= alloc_multijmp(bb_computed
, 1, 0);
1719 add_multijmp(&goto_ins
->multijmp_list
, jmp
);
1720 add_bb(&bb_computed
->parents
, ep
->active
);
1721 add_bb(&active
->children
, bb_computed
);
1722 } END_FOR_EACH_PTR(sym
);
1729 return linearize_compound_statement(ep
, stmt
);
1732 * This could take 'likely/unlikely' into account, and
1733 * switch the arms around appropriately..
1736 struct basic_block
*bb_true
, *bb_false
, *endif
;
1737 struct expression
*cond
= stmt
->if_conditional
;
1739 bb_true
= alloc_basic_block(ep
, stmt
->pos
);
1740 bb_false
= endif
= alloc_basic_block(ep
, stmt
->pos
);
1742 linearize_cond_branch(ep
, cond
, bb_true
, bb_false
);
1744 set_activeblock(ep
, bb_true
);
1745 linearize_statement(ep
, stmt
->if_true
);
1747 if (stmt
->if_false
) {
1748 endif
= alloc_basic_block(ep
, stmt
->pos
);
1749 add_goto(ep
, endif
);
1750 set_activeblock(ep
, bb_false
);
1751 linearize_statement(ep
, stmt
->if_false
);
1753 set_activeblock(ep
, endif
);
1759 struct instruction
*switch_ins
;
1760 struct basic_block
*switch_end
= alloc_basic_block(ep
, stmt
->pos
);
1761 struct basic_block
*active
, *default_case
;
1762 struct multijmp
*jmp
;
1765 pseudo
= linearize_expression(ep
, stmt
->switch_expression
);
1767 active
= ep
->active
;
1768 if (!bb_reachable(active
))
1771 switch_ins
= alloc_instruction(OP_SWITCH
, 0);
1772 use_pseudo(pseudo
, &switch_ins
->cond
);
1773 add_one_insn(ep
, switch_ins
);
1776 default_case
= NULL
;
1777 FOR_EACH_PTR(stmt
->switch_case
->symbol_list
, sym
) {
1778 struct statement
*case_stmt
= sym
->stmt
;
1779 struct basic_block
*bb_case
= get_bound_block(ep
, sym
);
1781 if (!case_stmt
->case_expression
) {
1782 default_case
= bb_case
;
1787 begin
= end
= case_stmt
->case_expression
->value
;
1788 if (case_stmt
->case_to
)
1789 end
= case_stmt
->case_to
->value
;
1791 jmp
= alloc_multijmp(bb_case
, end
, begin
);
1793 jmp
= alloc_multijmp(bb_case
, begin
, end
);
1796 add_multijmp(&switch_ins
->multijmp_list
, jmp
);
1797 add_bb(&bb_case
->parents
, active
);
1798 add_bb(&active
->children
, bb_case
);
1799 } END_FOR_EACH_PTR(sym
);
1801 bind_label(stmt
->switch_break
, switch_end
, stmt
->pos
);
1803 /* And linearize the actual statement */
1804 linearize_statement(ep
, stmt
->switch_statement
);
1805 set_activeblock(ep
, switch_end
);
1808 default_case
= switch_end
;
1810 jmp
= alloc_multijmp(default_case
, 1, 0);
1811 add_multijmp(&switch_ins
->multijmp_list
, jmp
);
1812 add_bb(&default_case
->parents
, active
);
1813 add_bb(&active
->children
, default_case
);
1818 case STMT_ITERATOR
: {
1819 struct statement
*pre_statement
= stmt
->iterator_pre_statement
;
1820 struct expression
*pre_condition
= stmt
->iterator_pre_condition
;
1821 struct statement
*statement
= stmt
->iterator_statement
;
1822 struct statement
*post_statement
= stmt
->iterator_post_statement
;
1823 struct expression
*post_condition
= stmt
->iterator_post_condition
;
1824 struct basic_block
*loop_top
, *loop_body
, *loop_continue
, *loop_end
;
1826 concat_symbol_list(stmt
->iterator_syms
, &ep
->syms
);
1827 linearize_statement(ep
, pre_statement
);
1829 loop_body
= loop_top
= alloc_basic_block(ep
, stmt
->pos
);
1830 loop_continue
= alloc_basic_block(ep
, stmt
->pos
);
1831 loop_end
= alloc_basic_block(ep
, stmt
->pos
);
1833 if (pre_condition
== post_condition
) {
1834 loop_top
= alloc_basic_block(ep
, stmt
->pos
);
1835 set_activeblock(ep
, loop_top
);
1839 linearize_cond_branch(ep
, pre_condition
, loop_body
, loop_end
);
1841 bind_label(stmt
->iterator_continue
, loop_continue
, stmt
->pos
);
1842 bind_label(stmt
->iterator_break
, loop_end
, stmt
->pos
);
1844 set_activeblock(ep
, loop_body
);
1845 linearize_statement(ep
, statement
);
1846 add_goto(ep
, loop_continue
);
1848 set_activeblock(ep
, loop_continue
);
1849 linearize_statement(ep
, post_statement
);
1850 if (!post_condition
|| pre_condition
== post_condition
)
1851 add_goto(ep
, loop_top
);
1853 linearize_cond_branch(ep
, post_condition
, loop_top
, loop_end
);
1854 set_activeblock(ep
, loop_end
);
1864 static struct entrypoint
*linearize_fn(struct symbol
*sym
, struct symbol
*base_type
)
1866 struct entrypoint
*ep
;
1867 struct basic_block
*bb
;
1869 struct instruction
*entry
;
1873 if (!base_type
->stmt
)
1876 ep
= alloc_entrypoint();
1877 bb
= alloc_basic_block(ep
, sym
->pos
);
1880 set_activeblock(ep
, bb
);
1882 entry
= alloc_instruction(OP_ENTRY
, 0);
1883 add_one_insn(ep
, entry
);
1886 concat_symbol_list(base_type
->arguments
, &ep
->syms
);
1888 /* FIXME!! We should do something else about varargs.. */
1890 FOR_EACH_PTR(base_type
->arguments
, arg
) {
1891 linearize_argument(ep
, arg
, ++i
);
1892 } END_FOR_EACH_PTR(arg
);
1894 result
= linearize_statement(ep
, base_type
->stmt
);
1895 if (bb_reachable(ep
->active
) && !bb_terminated(ep
->active
)) {
1896 struct symbol
*ret_type
= base_type
->ctype
.base_type
;
1897 struct instruction
*insn
= alloc_typed_instruction(OP_RET
, ret_type
);
1899 if (type_size(ret_type
) > 0)
1900 use_pseudo(result
, &insn
->src
);
1901 add_one_insn(ep
, insn
);
1904 merge_phi_sources
= 1;
1907 * Do trivial flow simplification - branches to
1908 * branches, kill dead basicblocks etc
1910 kill_unreachable_bbs(ep
);
1913 * Turn symbols into pseudos
1915 simplify_symbol_usage(ep
);
1919 * Remove trivial instructions, and try to CSE
1923 cleanup_and_cse(ep
);
1924 pack_basic_blocks(ep
);
1925 } while (repeat_phase
& REPEAT_CSE
);
1927 kill_unreachable_bbs(ep
);
1931 clear_symbol_pseudos(ep
);
1933 /* And track pseudo register usage */
1934 track_pseudo_liveness(ep
);
1937 * Some flow optimizations can only effectively
1938 * be done when we've done liveness analysis. But
1939 * if they trigger, we need to start all over
1942 if (simplify_flow(ep
)) {
1947 /* Finally, add deathnotes to pseudos now that we have them */
1948 track_pseudo_death(ep
);
1953 struct entrypoint
*linearize_symbol(struct symbol
*sym
)
1955 struct symbol
*base_type
;
1959 base_type
= sym
->ctype
.base_type
;
1962 if (base_type
->type
== SYM_FN
)
1963 return linearize_fn(sym
, base_type
);