From: Gerrit V. <vo...@vo...> - 2010-08-16 18:02:57
|
Hi, On Mon, 2010-08-16 at 12:31 -0500, Carsten Neumann wrote: > Hello Gerrit, > > Gerrit Voß wrote: > > On Fri, 2010-08-13 at 17:27 -0500, Carsten Neumann wrote: > >> Should the 'backend' parts of the new shader stuff (ShaderExeChunks, > >> etc.) actually be transmitted in the cluster? > > > > they should not, I checked and they seem to be, so that is wrong I'll > > change that. Let me look through it a little, I have a guess where the > > assert failure might come from. > > your change mostly fixed the problems with our app, I committed a follow > up that marks some additional fields as FClusterLocal as they reference > ShaderExecutable{,Var}Chunks: > > ShaderProgram::_mfParents > ShaderVariable::_mfParents > ShaderValueVariable::_mfVariableIdx > > and: > ShaderProgramVariables::_mfParents > > this last one is a bit iffy as the parents field is actually the one in > the Attachment base class. Is the reuse of the parents field there > intentional [1]? yes, there is some magic with the parents happening, I'm looking into this, as this is where the assert violation comes from. I just wanted to collect all side effects before committing, at least I expect these changes to break the compat SHL version of the shaders. But I'm still verifying this though. > [1] a related question is why StateChunk derives from Attachment in the > first place? Does anyone remember? not sure anymore, I don't know if we ever used them as attachments. kind regards gerrit |