Attendees: Razya, Antonio, David, Ramakrishna, Sebastian, Li, Konrad, Michael, Tobias
- Reductions (Sebastian)
- fold-const.i miscompiles just with SCoP detection.
- Almost every benchmark passes, but problems with dependency checks
- (memory usage)
- 3 test cases still fail
- Dependency check (Konrad)
- Remove caching of dependency polyhedron
- cray_pointers2 runtime to long. (380000 polyhedrons)
- Alias analysis (Li, Konrad, Tobi)
- NP hard
- another structure
- Representation as polyhedron but conservative
- Heuristics
- call in gimple alias oracle
- CLooG
- Michael (joined us)
- Vectorizer / Casting Problem (Ramakrishna)
- 2007 bugreport
- not just signed/unsigned but general type problem
- (Modulo semantics)
- number of iterations improvements using PPL (Sebastian)
- Integer polytopes
- Problem of testing for emptyness
- Plans to implement PIP in PPL (by INRIA)
- Current problem in PPL is that it finds to minimize the integral polyhedron,
- but rational points are allowed to.
- Merge
- Autopar patch of Razya
ScalarEvolution patch (Sebastian)
- Reductions patch
- Blocking blocked by interchange heuristics
- Strip mining
- We do no atomic strip mining of the loop
- Atomic strip mining has to be made