Tim,
On Wed, 31 May 2006, Tim Jenness wrote:
> On Thu, 1 Jun 2006, Malcolm J. Currie wrote:
>
> >> So do we really need all the fancy coordinate handling that LINPLOT
> >> provides?
> >
> > There's already a SPECX routine. I thought that part of the motivation
>
> That's the routine people will compare us with.
>
> > for CLINPLOT was to be rigorous about the WCS.
>
> being able to use CURSOR is a definite plus.
Yes, I'm not suggesting dropping all WCS support. I just meant that things
like handling error bars, options for making the X axis linear in pixel,
freq, or log(freq), including lots of different Frames in the stored
AGI Plot, etc, are probably not needed. Also, is is there any point in
providing options for annotating the bottom left spectrum, since the text
will presumably almost always be too small to read.
> > According to Tim's Mac
> > profiling the bulk of the time was going in AST.
>
> I've just re run the profiler:
>
> 34% - agi_slab (via kps1_clipg) with 24% (of the total not the 36%) in
> dat_get0c.
> 34% - kpg1_asset mainly in grp1_grapp via grp_grpex
> 22% in kpg1_plots but that is mainly the ast_write via kpg1_wwrt
>
> So, 1/3 in dat_get0c, 1/3 in grp and the rest in ast_write.
My test was to comment out the calls to kpg1_wwrt. This saved 25% in the
run time. The kpg1_asset routine is the one that gets a style (using GRP)
and applies it to a Plot. This is done for every single spectrum. A design
that does away completely with the drawing of grids round each individual
spectrum would be much better I think. After all, the very nature of this
sort of display is going to mean that annotation,tick marks, etc, around a
single spectrum is likely to be too small to see properly.
David
|