"David E. Cross" <crossd@xxxxxxxxxx> writes:
> I am compiling gcc on irix 6.5. I have 4.2.4 successfully build and
> installed on an r4400 iris 6.5.22 indy. from there I tried to build
> 4.5.0, which generated an ICE on a instruction not satisfying prereqs,
> *deep* into the 2nd stage build (it had already built all of binutils
> (I am doing the integrated binutils build), and the gen* programs, and
> run them)... 4.5.1 cam out shortly after.. and I ran it.. same, then I
> backed down to 4.4.4.. same, and 4.3.5 same. the exact instruction
> that generates the ICE is different, and they are in different places
> in each of the 4.x.y builds, but for a GIVEN 4.x.y build it is always
> the same, and overall in the GCC build itself it is always in roughtly
> the same spot (2nd stage after the gen* programs are run, typically in
> optab.c).
>
> I am relatively confident of my 4.2.4 build it passed its own 3 stage
> build test, additionally all of the software that I have built with
> it, including Perl, GMP, MPC, MPFR, PPL, Cloog-ppl have all passed
> their internal test-suites; it would seem if there was anything MAJOR
> wrong, it would have tripped there.
>
> Ideas?
That's a tough one. There is a clearly a bug somewhere. If the bug
happens to be in your installed gcc, then one approach that can work is
make BOOT_CFLAGS="-g"
Use that to bootstrap your new gcc, then install that, and use the newly
installed gcc to do a new bootstrap. This will take longer since the
compiler will not be optimized, but the final result will be optimized.
If that doesn't work, then the bug is pretty certainly in current gcc.
You will have to either debug it or submit a bug report. I doubt
anybody is doing much testing on Irix these days.
Ian
