Hi all, I see less problems with OpenCL than with CUDA. Things are getting better on the CUDA side, since CUSP and thrust have reached a certain level of maturity. Anyway, I agree that it's better (in terms of providing good support for GPU-related functionality) to direct people to petsc-dev (master) rather than trying to keep maint somehow up-to-date.
Are there any plans to go for shorter release cycles? If (and only if) so, the burden on 'maint' is smaller and we could consider GPUs for the release. Best regards, Karli On 04/01/2013 08:22 AM, Barry Smith wrote: > > Are we prepared to support the GPU use in PETSc (with CUDA and OpenCL) > and multi-core threaded support in the next release or only a subset or none? > That is will we continue with requesting users to use petsc-dev for some > (all) of the functionality? If so, we should turn off the support in the > release and print a nice helpful error message when configure is run > requested an unsupported mode telling people to use petsc-dev. So we don't > get tons of email from people who didn't read the docs carefully. > > Barry >
