[PATCH] D76961: [BranchFolder] don't remove MBB's that have their address taken
Nick Desaulniers via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Thu Apr 2 15:44:16 PDT 2020
nickdesaulniers added a comment.
Ok, I think I have a fix in hand; it's still regressing the existing test, but I think it's more along the lines of the above changes regarding `Address of block that was removed by CodeGen` vs `Block address taken` and I don't have the below diff in this new branch I'm working out of yet, but need to double check. It is now producing valid predecessors and successors, and verifies that.
I need to:
1. clean up my spaghetti
2. fix up the existing test case
3. retest kernel builds with this
4. write up an extra detailed commit message, while I still briefly understand a glimpse/sliver of selection dag, instruction selection, instruction scheduling, phis, and virtual registers.
Hopefully I can have that out by EOD, but no promises.
**EDIT**:
The fix wasn't too bad, more of a PITA to debug everything and a firehose to drink from.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D76961/new/
https://reviews.llvm.org/D76961
More information about the llvm-commits
mailing list