This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: ix86 HImode immediates patch
- To: Richard Henderson <rth at cygnus dot com>
- Subject: Re: ix86 HImode immediates patch
- From: Jan Hubicka <hubicka at atrey dot karlin dot mff dot cuni dot cz>
- Date: Tue, 29 Feb 2000 11:28:50 +0100
- Cc: egcs-patches at egcs dot cygnus dot com
- References: <20000228155853.F18800@atrey.karlin.mff.cuni.cz> <20000228100633.F4483@cygnus.com>
> On Mon, Feb 28, 2000 at 03:58:53PM +0100, Jan Hubicka wrote:
> > Avoiding them brings notable speedups on the PPro on byte benchmark
> > (9.8->10.9).
>
> Please do not over-tune the port to the byte benchmarks. That
> package is an abomination -- run 3 instructions 100 million times
> and find out how long it took.
>
> Please run some more sane benchmarks instead. SuSE ought to be
> able to get you a copy of Spec95 or Spec2000, and there are a few
> good ones in cvs.
Of course I was running other benchmarks too. I don't have Spec available
yet (I will in few months),
but it speeded up the XaoS benchmarks for Hicolor as well as caused only
speedups in the benchmark suite from cvs.
I used byte benchmark only as an example how does the gains look like,
since this is still hard to believe for me.
Honza
>
>
> r~