[PATCH] D16616: Avoid inlining call sites in unreachable-terminated block
Jun Bum Lim via llvm-commits
llvm-commits at lists.llvm.org
Fri Jan 29 10:59:03 PST 2016
junbuml added a comment.
Thanks Easwaran for the comment.
I can see that the case you mention could happen with this change. Although I tried to come up with a workaround without BPI, nothing seems to show desirable results in my test environment. To cover this case, I may want to wait until BPI is hooked in inliner and revisit this to elaborate it by checking the actual coldness of the block. So, in future, we may only suppress inlining a call site in a block that is leading to an unreachable and dominated by a block of which all edges from predecessors have zero branch probability.
More information about the llvm-commits