This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Small update to reversed_comparison_code
- To: Jeffrey Oldham <oldham at codesourcery dot com>
- Subject: Re: Small update to reversed_comparison_code
- From: Jan Hubicka <jh at suse dot cz>
- Date: Fri, 2 Mar 2001 11:41:30 +0100
- Cc: jh at suse dot cz, gcc-patches at gcc dot gnu dot org
- References: <20010301112031.A32309@codesourcery.com>
> Your 2001-02-28 patch to gcc/jump.c:reversed_comparison_code_parts()
> breaks gcc bootstrapping on mips-sgi-irix6.5 for "-O1 -mabi=64". To
> reproduce the problem on a i686-pc-linux-gnu or similar configuration,
> build a cross-compiler for mips-sgi-irix6.5. Then compile the
> attached .i file using these commands: "cc1 -O1 wrtfmt.i -mabi=64".
> wrtfmt.i is preprocessed code from libf2c/libI77/wrtfmt.c.
OK, I am going to investigate this later today.
What can help me is the crash or whatever the compiler produces on the
testcase...
>
> Debugging revealed that gcc/optabs.c:prepare_float_lib_cmp() is sent a
> UNGE code with a DFmode mode. The UNGE is produced from reversing an
> LT in gcc/loop.c:canonicalize_condition(). See a call to
> reversed_comparison_code().
This looks correct - we definitly have right to do so with -ffast-math.
Honza