Graphite phone call from 2009_02_04
Attendees: Razya, Kenny, David, Konrad, Tobias, Jan, Christophe, Sebastian.
PPL cleanups: use the PPL constraint interface in graphite.c for building iteration domains and data dependences. The cleanup will be merged to the graphite branch and the branch will then depend on new code that will be integrated in the CLooG-PPL branch.
Tobias reported on the status of the graphite tester, that is for the moment internal to AMD. The plan is to have a similar infrastructure testing both GCC trunk and graphite branch installed on the CompileFarm.
- We also discussed about the loop transform interface versus computing the scattering functions for an objective function that could be "max of data reuse" or "max parallelism".
Razya is interested in auto-parallelization and vectorization as code generation from Graphite, the plan will be defined in the coming weeks.
- Kenny will help with cleaning the SCoP detection to work on Single Entry Single Exit (SESE) regions: the code generation of graphite already works on SESE regions.