This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: operand_subword question
- To: David Edelsohn <dje at watson dot ibm dot com>
- Subject: Re: operand_subword question
- From: Jeffrey A Law <law at cygnus dot com>
- Date: Tue, 07 Mar 2000 15:43:19 -0700
- cc: gcc at gcc dot gnu dot org
- Reply-To: law at cygnus dot com
In message <200003020421.XAA05112@mal-ach.watson.ibm.com>you write:
> I have not added long double support to the PowerPC port, but it
> certainly could encounter the situation you describe in 64-bit PowerPC
> mode.
We're probably going to have to solve it soon.
I'm using this right now to deal with the problem, but I haven't really
thought about whether or not it's correct in regards to endianness
issues.
#if HOST_BITS_PER_WIDE_INT == 64
else if (BITS_PER_WORD == 64)
{
val = k[i * 2 + ! WORDS_BIG_ENDIAN];
val = (((val & 0xffffffff) ^ 0x80000000) - 0x80000000) << 32;
val |= (HOST_WIDE_INT) k[i * 2 + WORDS_BIG_ENDIAN] & 0xffffffff;
return GEN_INT (val);
}
#endif
It's worth noting that REAL_VALUE_TO_TARGET_LONG_DOUBLE fills in an array
of 32bit wide hunks, not HOST_WIDE_INT wide hunks.
jeff