When an insertion point is specified in insertMBBIntoMaps the
index range of existing instructions in the block needs to be
considered.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Unit Tests
Time | Test | |
---|---|---|
400 ms | linux > HWAddressSanitizer-x86_64.TestCases::sizes.cpp |
Event Timeline
llvm/include/llvm/CodeGen/SlotIndexes.h | ||
---|---|---|
621–625 | I am not sure I understand the patch. Admittedly, that behavior is weird (because that means the users of this method must make sure to split/join the blocks accordingly), but this is what the comment says! Side-comment: Is that code reachable? A quick grep tells me that InsertionPoint is never passed around. I.e., we should just delete it. |
llvm/include/llvm/CodeGen/SlotIndexes.h | ||
---|---|---|
621–625 | This fixes the case where an inserted block contains multiple instructions. InsertionPoint is used by MachineBasicBlock::splitAt, which I am trying to fix in D91064. Note D91064 contains tests for behaviour fixed by this patch. A couple of other changes are dependent on splitAt working correctly. |
clang-tidy: warning: invalid case style for variable 'isLastMBB' [readability-identifier-naming]
not useful