Reduce differences between successor iterators #93445
Conversation
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsThe NumSucc/GetSucc/Succs iterators have two variants. It is hard to understand the distinction. The goal of this change is to reduce the differences, so the only difference is the SWITCH case (iterating either all successors or only unique successors). So:
|
|
@AndyAyersMS PTAL |
src/coreclr/jit/importer.cpp
Outdated
There was a problem hiding this comment.
I couldn't figure out where the "above" was. Would be nice to avoid this special case.
Seems like we should never get into the spill clique case for filters since the exit stack will be emtpy?
There was a problem hiding this comment.
Ah, I can remove this from the spill clique case if you think it's unnecessary. (Just assert block is not a filterret?) I was just being defensive, since I hit an assert without this in the "normal" case.
The NumSucc/GetSucc/Succs iterators have two variants. It is hard to understand the distinction. The goal of this change is to reduce the differences, so the only difference is the SWITCH case (iterating either all successors or only unique successors). So: 1. BBJ_EHFILTERRET: always return the single successor, which is the first block of the filter handler. 2. BBJ_EHFINALLYRET: Move to standard successor iterator
2ae9a3a to
cf4e02b
Compare
|
@AndyAyersMS updated |
The NumSucc/GetSucc/Succs iterators have two variants. It is hard to understand the distinction. The goal of this change is to reduce the differences, so the only difference is the SWITCH case (iterating either all successors or only unique successors).
So: