I made a post a few weeks ago about the same black pixel problem (random pixel read). I use the same nuke version and try to read a single pixel with at() after a request()... This work with most pixels except on some random rows...
Nicolas Envoyé de mon iPhone Le 25 mars 2011 à 18:42, Moritz Moeller <[email protected]> a écrit : > We have a problem where we randomly get black pixels from upstream nodes > we validate() and request() 'somewhere' deep in our engine(). > > Basically we cannot know if we need to validate() (and subsequently > request()) them by the time our own validate() method is ran by Nuke. > > We did not see this in the past, the resp. code has not changed in half > a year. > I'd dare say it feels like it may be related to us building/testing > against 6.2v2 now. > It doesn't matter what nodes are upstream. However, we only query Iops > and Read nodes this way. > > Another thing is that is seems to be #-of-threads-related. I.e. on boxes > with 8 cores it shows more than on boxes with 6 cores and on boxes with > 2 cores we cannot replicate it at all (everything works as expected). > > Our code does only request rows serially though and doesn't use > Row::start()/Row::wait() or anything like that; so the threading stuff > is not on our side. > > > A typical code fragment looks like this: > > _node->validate( true ); > > unsigned short width( _node->info().width() ); > unsigned short height( _node->info().height() ); > > ChannelSet channelMask( Mask_RGBA ); > channelMask &= _node->info().channels(); > unsigned channels( channelMask.size() ); > > _node->request( 0, 0, width, height, channelMask, 1 ); > > Row row( 0, width ); > > for ( unsigned scanLine( 0 ); scanLine < height; ++scanLine ) { > > _node->get( scanLine, 0, width, channelMask, row ); > > foreach( channel, channelMask ) { > // read row[ channel ] and use somehow > } > } > > > > What am I missing here? > > > .mm > > _______________________________________________ > Nuke-dev mailing list > [email protected] > http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev _______________________________________________ Nuke-dev mailing list [email protected] http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-dev
