This is preparatory work to unable storing alignment for AtomicCmpXchgInst.
See D83136 for context and bug: https://bugs.llvm.org/show_bug.cgi?id=27168
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
This patch splits the two implementations, moves a few lines and simplifies the code by inferring constant values.
The semantic is unchanged.
llvm/include/llvm/Bitcode/LLVMBitCodes.h | ||
---|---|---|
539–541 | The documentation here was wrong. | |
llvm/lib/Bitcode/Reader/BitcodeReader.cpp | ||
4991 | Each function taking Slot (previously OpNum) will increase it if successful. |
llvm/include/llvm/Bitcode/LLVMBitCodes.h | ||
---|---|---|
539–541 | It used to only have "ordering", and didn't separate success / failure (so it wasn't optional as much as not there). |
llvm/include/llvm/Bitcode/LLVMBitCodes.h | ||
---|---|---|
539–541 | Thx for the explanation :) |
To connect the dots for anyone following along: this was reverted in cc28058c13e89ecc85dac7e1bd5d13a2ce1bb620.
llvm/lib/Bitcode/Reader/BitcodeReader.cpp | ||
---|---|---|
4991 | I don't think there is any guarantee that they increase it by one. This could break silently in the future... |
llvm/lib/Bitcode/Reader/BitcodeReader.cpp | ||
---|---|---|
4991 | Nevermind, I'm commenting on the old patch, I though I was looking at the new one... |
The documentation here was wrong.
alignment was never stored for FUNC_CODE_INST_CMPXCHG_OLD and failure_ordering and weak were optional.