[Insight-developers] TestDriver Test

Bill Lorensen wlorens1@nycap.rr.com
Sat, 02 Mar 2002 08:27:18 -0500


Right now, cmake generates several configurations. I think that most users only use one and the others often get in the way since sometimes the wrong configuration gets picked.

It would be nice if the user could have cmake just generate one of these. Would this have any performance improvemement?

Bill

At 10:55 PM 3/1/02 -0500, William A. Hoffman wrote:
>I would not say that.   .NET is quite a bit faster at loading projects, but I would think that most people
>are not seeing 15 second load times.  I just did a quick test with Visual Studio 6, and the load time
>was 40 seconds the old way, and 25 seconds the new way.   This is also a rather new machine, on
>last years PC's load times can be over 1 minute.  So an almost 50% faster load time,  is pretty good,
>and that was only one of the directories.  There are still 160 projects in the workspace, if we did all of them,
>I would expect to see greater speed improvements.
>
>Also, the disk space is quite a good improvement as well, it is 3.5 times less disk space.
>I have not done the timings, but I would think that if the compiler is writing out 3.5 times less data, it
>should run faster.    For perception of ITK, I think anything we can do to make it more manageable
>is a good idea.  I don't think turning off the tests by default is the answer.
>
>
>-Bill
>
>At 06:07 PM 3/1/2002 -0500, Bill Lorensen wrote:
>>So it looks like little improvement except for disk space,
>>
>>Bill
>>
>>At 05:14 PM 3/1/02 -0500, William A. Hoffman wrote:
>>>Hi Folks-
>>>
>>>Just to see if the testing process we discussed on the TCON was
>>>worth doing at all, I went ahead and converted
>>>just the tests in BasicFilters to my approach.  Here are the results
>>>for this one directory:
>>>
>>>The project load time in .NET went from 15 sec. to 8 sec. (1.2 ghz processor).
>>>The number of projects went from 240 to 160.
>>>The binary directory went from 240 megs to 70 megs.
>>>The TestDriver executable was only 2.8 megs.
>>>Changing a test and re-linking/compiling showed no noticeable difference in time.
>>>
>