Page 1 of 1

Speeding up surface creation. Hardware support?

Posted: Fri Aug 27, 2004 9:38 am
by mikeb
Possible for microprocessor-specific compiled executable to speed things up? How about external DSP cards or such? I'm using one of the latest, fastest Intel CPUs with 1GB RAM. When sorting nodes (reading .csv files in Random), can sort be sped up? I know I'm stretching, but it doesn't hurt to ask!

Mike

Posted: Fri Aug 27, 2004 1:35 pm
by DPlotAdmin
Nope, never hurts to ask at all :)

It's been a while since I looked at the sort routine and while I do recall that it is now considerably faster than the initial version, I'm sure there are improvements that could be made. Just so we can compare notes if need be - how many points are you working with?

A lot!

Posted: Fri Aug 27, 2004 1:45 pm
by mikeb
.CSV ASCII file (uncompressed) with (X,Y,Z) columns around 5-6 MBytes. I am currently sampling (X,Y,Z) at ~ 100Hz, but I need to change to sample on change in X or Y (change sampling from time domain to displacement domain). I know this would reduce data set size and data would be ready for digital filtering. Would this "fixed spaced" data also sort faster than random?

mikeb

Posted: Fri Aug 27, 2004 1:52 pm
by DPlotAdmin
There currently are no supported file formats that take into account evenly-spaced points (other than DPlot files, but that's no help in this case). If there was such a format, then no sorting would be necessary. Plus the plot would be drawn faster. I'll still take a look at improving sorting efficiency, but if it's no trouble for you to rearrange your data then a better long-term solution is to add a new file format. I'll try to put something together for the next release, most likely in the form of a file import plugin.

Posted: Tue Aug 31, 2004 12:19 am
by DPlotAdmin
:oops: Whoops. A while back I introduced a terribly inefficient routine that eliminates points with overlaying X and Y values, and I'm sure this was the major source of your request for improved sorting. This is fixed in tonight's (8/30/04) update. The problem wasn't really noticeable with a few thousand points, but with 10's of thousands DPlot turned into a real slug. By way of example, a plot with 100,000 data points takes roughly 4 seconds now on a fairly new system, where yesterday it took several minutes.