/* { dg-do run } */
/* { dg-options "-O2 -march=i586 -dA -g -fomit-frame-pointer" { target { {
i?86-*-* x86_64-*-* } && ilp32 } } } */
void __attribute__((noinline))
bar (int *p)
{
(*p)++;
asm volatile ("" : "=r" (*p) : "0" (*p) : "memory");
}
void __attribute__((noinline))
baz (int a, int b, int c, int d, int e, int f)
{
asm volatile ("" : : "r" (a + b + c + d + e + f) : "memory");
}
int __attribute__((noinline))
foo (int x)
{
int y;
y = 5;
bar (&y);
baz (0, 1, 2, 3, 4, 5);
bar (&x);
y++;
baz (1, 2, 3, 4, 5, 6);
bar (&y);
return x + y;
}
int
main ()
{
if (foo (17) != 8 + 18)
__builtin_abort ();
return 0;
}
has wrong debuginfo for variable y - (mem (sp + offset)) live across several sp
adjustments. In the debugger inside foo when using nexti and printing y after
each nexti, it first starts as <optimized out> (expected), then y = 5 is
executed and it prints 5, but then sp is adjusted and it prints garbage because
sp changed.
I believe we should be de-eliminating stack_pointer_rtx resp.
hard_frame_pointer_rtx based MEMs in var-tracking generated notes back to
arg_pointer_rtx / frame_pointer_rtx based notes when they aren't meant to
change whenever sp resp. fp changes.
On the http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43051#c1 testcase this
doesn't show up as wrong debug, but just inefficient location list (LVL1
through LVL2 the argument is fbreg 0, while from LVL2 to LVL12 it is breg5 8
when both
locations are actually the same memory. There is also the additional problem
that we leave out the epilogue ret insn, as if the parameter wasn't defined at
that spot.
--
Summary: [4.5 Regression] Wrong debuginfo with VTA and -fomit-
frame-pointer/-mno-accumulate-outgoing-args
Product: gcc
Version: 4.5.0
Status: UNCONFIRMED
Keywords: wrong-debug
Severity: normal
Priority: P3
Component: debug
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: jakub at gcc dot gnu dot org
GCC target triplet: i686-linux
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43092
