[polly] r252713 - ScopInfo: Pass domain constraints through error blocks

Tobias Grosser via llvm-commits llvm-commits at lists.llvm.org
Wed Nov 11 12:37:05 PST 2015


On 11/11/2015 06:36 PM, Johannes Doerfert wrote:
> Did you check if we still need the HasErrorBlock flag? I was under the
> impression we do not because we do not use error block conditions in the
> domain anymore. Thus, there should be no harm in simplifying the
> assumptions with the domain again, even if there are error blocks in the
> SCoP.

I looked at it earlier, but could not convince myself this flag is not 
needed any more. The comment says it prevents a simplification that is
problematic in case certain basic blocks have been removed from the 
scop. Even though we propagate the domains through error blocks, we
still do not generate ScopStmts for them. So the problem described in
the comment seems to remain.

Best,
Tobias



More information about the llvm-commits mailing list