[PATCH] D30128: [LazyMachineBFI] Reimplement with getAnalysisIfAvailable

Adam Nemet via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Fri Feb 17 23:29:11 PST 2017


anemet added a comment.

In https://reviews.llvm.org/D30128#680739, @davide wrote:

> What's the compile time savings you're seeing from not computing these analyses unnecessarily?


This would be hard to measure for the machine passes since GreedyRegAlloc already uses MachineLoopInfo.  This was more of a precaution for the machine passes at this point.  But I can try to measure this when I look at the IR passes.

> Side rant: I think some of the names in LLVM are really poorly chosen, `DominatorTree` in fact is not an analysis but the result of an analysis.

Agreed and then in both MachineLoopInfo and MachineDT, we go out of our way to pretend that the pass has the same APIs as the result.  Very confusing.


https://reviews.llvm.org/D30128





More information about the llvm-commits mailing list