Re: SH: abort in reg_overlap_mentioned_for_reload_p

[Hans-Peter Nilsson <hp at bitrange dot com>]

On Sun, 2 Sep 2001, kaz Kojima wrote:
> When I tried to run C testsuits for SH, some tests aborted in
> reg_overlap_mentioned_for_reload_p(X, IN).

FWIW, Honza's patch at
<URL:http://gcc.gnu.org/ml/gcc-patches/2001-08/msg00920.html>
fixes a failure that matches your description.  It lets sh-elf
build newlib (see bootstrap/4196).  Comparing the results of
that patch to the improvements you reported, I don't see these
failures with Honza's patch:

>   gcc.c-torture/compile/920501-4.c
>   gcc.c-torture/compile/950612-1.c
>   gcc.dg/sequence-pt-1.c

Though this failure I see (with -O2 and -O3 -g):

>   gcc.c-torture/compile/990517-1.c

I'll try your patch and see for myself the differences.  I'm
using <URL:http://gcc.gnu.org/simtest-howto.html>, with target
sh-elf and simulator sh-hms-sim.

Thanks.

An interesting observation is that mcore-elf fails on the same
file and function with the same ICE (see bootstrap/4194).
Unfortunately Honza's patch does not fix that one.  I'll test
yours and report.

I'm cheesily staying away from doing any deeper analysis on
these simulator targets; I can fire off a few simulator builds
without much mental involvement, though.

brgds, H-P