X-Authentication-Warning: delorie.com: mail set sender to geda-user-bounces using -f
X-Recipient: geda-user AT delorie DOT com
Date: Tue, 28 Jul 2015 05:41:47 +0200 (CEST)
X-X-Sender: igor2 AT igor2priv
To: "Gabriel Paubert (paubert AT iram DOT es) [via geda-user AT delorie DOT com]" <geda-user AT delorie DOT com>
X-Debug: to=geda-user AT delorie DOT com from="gedau AT igor2 DOT repo DOT hu"
From: gedau AT igor2 DOT repo DOT hu
Subject: Re: [geda-user] pcb-rnd feature poll: please vote
In-Reply-To: <20150727105941.GB31438@visitor2.iram.es>
Message-ID: <alpine.DEB.2.00.1507280520260.6924@igor2priv>
References: <alpine DOT DEB DOT 2 DOT 00 DOT 1507251201460 DOT 6924 AT igor2priv> <mp08rt$85h$1 AT ger DOT gmane DOT org> <alpine DOT DEB DOT 2 DOT 00 DOT 1507270806340 DOT 6924 AT igor2priv> <20150727081159 DOT GA31594 AT visitor2 DOT iram DOT es> <alpine DOT DEB DOT 2 DOT 00 DOT 1507271023590 DOT 6924 AT igor2priv>
 <20150727105941 DOT GB31438 AT visitor2 DOT iram DOT es>
User-Agent: Alpine 2.00 (DEB 1167 2008-08-23)
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed
Reply-To: geda-user AT delorie DOT com
Errors-To: nobody AT delorie DOT com
X-Mailing-List: geda-user AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk



On Mon, 27 Jul 2015, Gabriel Paubert (paubert AT iram DOT es) [via geda-user AT delorie DOT com] wrote:

>
> Very likely, for me the two fastest HID are GL followed by lesstif.
> But lesstif is much better on remote displays.
>
> However, I've not tested the non-gl gtk in a long time.
>
>> Will publish my results.
>

*DISCLAIMER* for those who didn't follow the history of this thread: I am 
_not_ trying to fix an opengl driver issue. I merely tried to benchmark 
the HIDs.


Didn't work out as nicely as I planned. Results:

http://igor2.repo.hu/tmp/pcb_bm/

First, I have a set of test files, the same pattern (4 layers) copied in 
more and more times (bm*.pcb.gz). Then I have this hackish bm.sh that 
generates 10 action pairs wrapped in priting the time in high res. The 
output is then the two time stamps. An awk script calculates the delta and 
divides them by 10.

In theory, this shows an objective measure on how fast the whole pattern 
is redrawn, averaged over 10 samples to decrease the errors introduced by 
time quiries and the time sharing OS.

In practice most results are false: gtk with and without gl don't flush 
often enough. Or in other words, executing the next action doesn't really 
wait until the previous has finished drawing. This is a proper 
optimization, but makes my measurements invalid. I didn't find a 
trivial way to force redraws so I eventually gave up.


Instead of objective numbers, I made a set of videos of how crosshair 
moves around on bm32.pcb zoomed so that all objects are visible:

gl*.avi - pcb 200140316, gtk+gl, 2d and 3d view
gtk.avi - pcb-rnd r473 with the gtk hid, sw render
lesstif.avi - pcb-rnd r473 with the lesstif hid

The videos are in real-time, theres's no capture problem. It's like 1 
fps with gl. Things are less laggy and differences are smaller between 
the HIDS on the smaller boards. This is the fastest machine with the most 
advanced GPU I use PCB on.

I did some "select all and move around" tests with similar results: 
lesstif was faster than gtk by only a tiny factor. Although not at 
speed-of-light,  both were pretty much usable. gl was a magnitude 
slower than them.

Regards,

Igor2