REGRESSION in 3.2.1-pre on mips-irix6 (and elsewhere) compiling execute/20011219-1.c

[Franz Sirl]

At 08:11 13.11.2002, Kaveh R. Ghazi wrote:
>Based on the dates in the ChangeLog and the descriptions of the
>patches, I suspect its one of these below.  However it's 2am here so
>I'll have to defer further investigation until tomorrow.
>
>                 --Kaveh
>
>
>2002-11-11  Franz Sirl  <Franz.Sirl-kernel@lauterbach.com>
>
>         PR c/8467
>         * stmt.c (tail_recursion_args): Handle DECL_MODE differing
>         from the mode of DECL_RTL case.
>
>2002-11-11  Graham Stott  <graham.stott@btinternet.com>
>             Jan Hubicka  <jh@suse.cz>
>
>         PR 8502
>         * cfgrtl.c (try_redirect_by_replacing_jump): Do not kill
>         computed jumps post reload.

Jan, Graham,

since powerpc-linux-gnu was a testplatform for the stmt.c patch, your patch 
seemed like the more likely culprit. In fact reverting it fixed it for 
powerpc-linux-gnu.

The failing testcases:

gcc.dg/compile/20020430-1.c at -O2
gcc.c-torture/execute/20011219-1.c at -O3

The failure mode in both cases is:
[fsirl@entropy:~/obj/gcc32/gcc]$ gcc
/home/fsirl/cvsx/gcc32/gcc/testsuite/gcc.dg/20020430-1.c -B./ -c -O2
/home/fsirl/cvsx/gcc32/gcc/testsuite/gcc.dg/20020430-1.c: In function
GetEventEntry':
/home/fsirl/cvsx/gcc32/gcc/testsuite/gcc.dg/20020430-1.c:190: Internal
compiler error in merge_blocks_move_predecessor_nojumps, at cfgcleanup.c:650

Jan, since you have access to entropy and my homedir on it, feel free to 
start a quick debugging session there (source in ~fsirl/cvsx/gcc32/, build 
in ~fsirl/obj/gcc32/). I'm on IRC at irc.openprojects.net, channel #GCC if 
you have any questions.

Franz.