Re: Announcement: SPEC95 daily testing for GCC

[Diego Novillo <dnovillo at redhat dot com>]

On Sun, 04 Mar 2001, Andreas Jaeger wrote:

> Diego Novillo <dnovillo@redhat.com> writes:
> 
> > On Sun, 04 Mar 2001, Geert Bosch wrote:
> > 
> > > Thanks a lot for setting this up, very useful!
> > > One question, it seems that it would actually be very useful as well,
> > > to have compilation times for compilig the spec benchmarks. That lets
> > > us at least see the effect of trading off compilation speed for execution
> > > speeds. It would be great if that could be added without too much trouble.
> > > 
> > Sure. The data is there in the runspec log file, so collecting it
> > would be trivial. I need to think about what would be the best
> > way of presenting it
> > 
> > We should have a separate chart for compilation times. I don't
> > think we want to put compilation times in the SPEC chart because
> > the scales are so different. Maybe having it side by side with
> > the performance chart?
> > 
> > We should also have cutoff dates for historic data. Having all
> > that daily info becomes meaningless after a while. Maybe take the
> > average of the last 3-4 months and summarize them in one data
> > point.
> 
> It might also be usefull to add "historic" version, e.g. 2.95.2, 2.7.2
> and 2.8.1 to see how the current versions compare to those old
> version.
> 
Yes. We could use historic versions as watermarks in the chart.
We will nead to split up the charts into different views to avoid
making them a blot of ink. Other things we could add:

- A chart for the overall SPECint mark, not just individual
  tests.
  
- Include SPECfp.