Re: [Bug target/12371] [3.4 regression] [m68k-linux] bootstrap errorin make compare

[Matthias Klose <doko at cs dot tu-berlin dot de>]

Updated PR12371. I hope I identified two patches (one by Jan, the
other by Roger).

Bernardo Innocenti writes:
> Matthias Klose wrote:
> > bernie at develer dot com writes:
> > 
> >>I've re-read both patches and they seem rather safe, but...
> >>is it possible that the objects produced during stage3 aren't
> >>really ELF files?
> > 
> > 
> > I attach two files. maybe you could get something out of it? I further
> > try to tighten the period, where I see the failure occur.
> 
> The two objects actually _do_ contain different assembly
> code. Both have been generated by the very same version
> of GCC.
> 
> The first few functions are identical, then a very long
> one (gt_pch_pa_reg_base_value) has very different
> register allocation.
> 
> Is it possible that different optimization options were
> used in stage2?

rechecked. no.

> alias.s (GCC: (GNU) 3.4 20031030 (experimental))
>  [...]
>  0000013a <gt_pch_pa_reg_base_value>:
>      13a:       4e56 0000       linkw %fp,#0
>      13e:       48e7 3c20       moveml %d2-%d5/%a2,%sp@-
>      142:       41ee 0008       lea %fp@(8),%a0          <-!!!
>      146:       2810            movel %a0@,%d4
>      148:       2468 0008       moveal %a0@(8),%a2
>      14c:       2a2e 0014       movel %fp@(20),%d5
>  [...]
> 
> stage2/alias.s (GCC: (GNU) 3.4 20031030 (experimental))
>  [...]
>  0000013a <gt_pch_pa_reg_base_value>:
>      13a:       4e56 0000       linkw %fp,#0
>      13e:       48e7 3c20       moveml %d2-%d5/%a2,%sp@-
>      142:       282e 0008       movel %fp@(8),%d4        <-!!!
>      146:       246e 0010       moveal %fp@(16),%a2
>      14a:       2a2e 0014       movel %fp@(20),%d5
>  [...]