[PATCH] D18670: LoopUnroll: some small fixes/tweaks to make it more useful for partial unrolling

[Evgeny Stupachenko via llvm-commits]

evstupac added a comment.

The only target modifying MaxCount is AMDGPU, but it modifies it to UINT_MAX.
So all checks for exceeding MaxCount are not supposed to hit.

Personally I like Threshold bounding more than Count. So I'd better separate full and partial unroll thresholds in the patch. However there could be something target specific that I'm missing.

So is it possible to reuse MaxCount to bound full unrolling? That way we can introduce an option only for MaxCount?


Repository:
  rL LLVM

http://reviews.llvm.org/D18670