This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Patch to fix powerpc-ibm-aix4.3.3.0 bootstrap
- To: gcc-patches at gcc dot gnu dot org, john at feith dot com
- Subject: Re: Patch to fix powerpc-ibm-aix4.3.3.0 bootstrap
- From: mike stump <mrs at windriver dot com>
- Date: Sat, 6 Oct 2001 10:42:55 -0700 (PDT)
- Cc: dje at watson dot ibm dot com
> Date: Sat, 6 Oct 2001 10:06:49 -0400 (EDT)
> From: John Wehle <john@feith.com>
> To: gcc-patches@gcc.gnu.org
> Cc: mrs@windriver.com, dje@watson.ibm.com
> hash_rtx assumes that identical symbols have the same hash
> which is unfortunately not true on the rs6000 due to the
> backend's use of ggc_strdup. This results in reload_cse_regs_1
> producing varying results depending on how the toc label symbol
> reference hash (which depends on its address) works out.
Yup, that sounds like the type of problem I expected. However, would
be nice to document in bold letters somewhere that things like strdup
cannot be used on label names, ever. Or, that if someone wants to,
they must fix reload_cse_regs_1 to use something more interesting than
the address. That, or we should fix reload_cse_regs_1 to not care
about addresses.