This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: 1.0.2/m68k/exception/longlong problem


>>>>> "Kamil" == Kamil Iskra <kamil@dwd.interkom.pl> writes:

Kamil> On Sat, 6 Jun 1998, Jes Sorensen wrote:
>> It is being build with bootstrap-lean and I have patched it to
>> compile enquire.c with -O2 as it otherwise crashes with spilled
>> register problems. This is also a problem with all versions of gcc
>> that I know of so I do not think it is related.

Kamil> I wouldn't be so sure... I have built egcs 1.0.2 on my
Kamil> Linux/m68k (with your RH 5.0 installed) and it compiled just
Kamil> fine. All I did was override CFLAGS in order not to compile
Kamil> with -g, if I remember correctly (just for space reasons). It
Kamil> compiled enquire just fine, as well as libgcc2.a.

I believe this is caused by the nono optimized compile of enquire.c
triggers a code-sequence bug in the 68060.

Kamil> I'd advice that you download the official egcs 1.0.2 and try to
Kamil> build this one. Perhaps some local RedHat patches screw things
Kamil> up?

[snip]

Kamil> Oh, I see a difference when compared to my build now. You set
Kamil> CFLAGS to "-O2 -m68020-40 -fomit-frame-pointer", right? Perhaps
Kamil> this triggers the bug.

I tried removing those, didn't make a difference.

Kamil> Also, enabling Haifa (I suppose that's where -DHAIFA comes
Kamil> from?) doesn't make much sense on m68k, since it doesn't do
Kamil> instruction scheduling at all.

I tried compiling 1.0.3a with the same configuration parameters and it
seemed to work - dunno what the difference is, but I am of course
curious.

Jes


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]