This is strange because I'm actually working on a single threaded ICENode and don't have any memory leaks. The output port is a single structure of type Vec3 in 0D context. I can run it on a mesh with 3 millions points for 10 000 frames without memory issue.
I'm using 2013 QFE7 btw. On Tue, Oct 1, 2013 at 2:57 PM, Mathias N <[email protected]> wrote: > Thanks for the reply. > > The trial version I'm using is 2014 SP2. > Testing is being done in an empty scene with nothing but the newly created > pointcloud and a simple ice tree. There are no attribute display properties > present. > > To rule out any other possible culprits I created two identical nodes that > simply output the float input (0D component), one single-threaded and the > other multi-threaded. Both created via the wizard with the only manual > change being setting each to pass the value instead of printing it as the > sample code does. > > Switching between the two, the multi-threaded node barely affected memory > usage, while the single-threaded node resulted in a steady increase in > memory usage that was never released. > > It does not appear to have been fixed in SP2 :( > > Given that I am only passing a relatively modest number of components > through it, I can imagine this would be quite a showstopper for larger sets > of data. > > > > On Tue, Oct 1, 2013 at 7:25 PM, Guillaume Laforge < > [email protected]> wrote: > >> Hi, >> >> Indeed, there has been some memory leaks issues in the past. If you were >> testing in 2014, you should try 2014 SP2 as it fixed (some parts) of the >> ICE related memory issue. >> If 2014 SP2 doesn't fix the pb you can also check if you've got some >> attribute display properties on your objects. They could be the culprit. >> >> Guillaume >> >> >> On Tue, Oct 1, 2013 at 12:24 PM, Mathias N <[email protected]> wrote: >> >>> Please excuse any errors; mailing lists are confusing. >>> >>> While wrapping up work on a system that includes 5 custom icenodes, I >>> noticed that memory usage was increasing by about 1mb per second during >>> playback. >>> >>> I eventually narrowed it down to a per-point-array node, the only >>> single-threaded node in the setup. In a test scene I have it reading about >>> 30 turbulized mesh point positions and spitting them into a single >>> per-object array, and after leaving it for 15-30 minutes a while memory >>> usage had increased from 400mb to 1.5gb. It is not released until the >>> program is closed. >>> >>> Being at loss as to what might be causing it I removed everything but >>> the absolute necessary pieces of the code (port declarations etc), only to >>> find that the problem persisted. >>> >>> Next I created a new icenode using the wizard, adding one input and one >>> output port with the default settings, and setting it to single threaded. >>> This also slowly increases memory consumption when executed. Using an 8x8x8 >>> grid as input it increases memory usage by about 150kb per second. >>> >>> If the input is singular (object context) there is no perceivable >>> increase in memory usage. The rate of increase appears to be relative to >>> the number of input components. >>> >>> I primarily use Softimage 2011, but running it in the 2014 trial >>> produces the same result. Were it a legitimate bug it would seem unlikely >>> that it would persist for so long. >>> >>> Could anyone shed some light on what is going on here? Is this normal? >>> >>> Thanks >>> >>> >>> -------------------------- >>> To unsubscribe: mail [email protected] with >>> subject "unsubscribe" and reply to the confirmation email. >>> >> >> >> -------------------------- >> To unsubscribe: mail [email protected] with >> subject "unsubscribe" and reply to the confirmation email. >> > > > -------------------------- > To unsubscribe: mail [email protected] with subject > "unsubscribe" and reply to the confirmation email. >
-------------------------- To unsubscribe: mail [email protected] with subject "unsubscribe" and reply to the confirmation email.
