------- Comment #16 from dave at hiauly1 dot hia dot nrc dot ca 2008-10-25
01:11 -------
Subject: Re: [4.4 Regression] Small structs are not passed correctly on
hppa64-*-*
> Can you please look at the testcases why they fail (what is passed
> differently)
> and try to minimize them as much as possible? Finding a bug in RTL or
> assembly
> without knowing what you are looking for is certainly harder than when you
> have
> a debugger and can see what is passed differently...
The following is a simplification of the first failure in
gcc.dg/compat/scalar-by-value-4.
_Complex char g01cc;
void
initcc (_Complex char *p, _Complex char v)
{
*p = v + (0 + 1 * __extension__ 1i);
}
void
checkcc (_Complex char x, _Complex char v)
{
if (x != v + (0 + 1 * __extension__ 1i)) abort ();
}
void
checkgcc (void)
{
checkcc (g01cc, 1);
}
int
main ()
{
initcc (&g01cc, 1);
checkgcc ();
return 0;
}
The first argument passed to checkcc is wrong. The value passed in
r26 (the first argument register) is 1. 4.3.2 passed 0x101. The value
passed for v is 0x100 in both versions.
Now, as I understand the 64-bit runtime specification for hppa64-hpux,
neither GCC version is correct. However, I don't believe HP cc or aCC
supports integer complex values, so I can't check for consistency.
My understanding is that small structs and aggregates should be left
justified. This unfortunately differs from the right justification
used on most big endian targets. I believe tha correct value for
x and v should be 0x0101000000000000 and 0x010000000000000, respectively.
I have checked that the struct { byte x; byte y } is correctly left
justified.
Dave
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37316
