PPC as AArch64 doesn't have address-spaces so we can drop them in the backend
Details
Diff Detail
- Repository
- rL LLVM
- Build Status
Buildable 15443 Build 15443: arc lint + arc unit
Event Timeline
This seems like a perfectly reasonable thing to do. However, I'd like to understand the implications of this change - does anything change behaviour as a result? If so, can we have a test case? If not, does this just save compile time (i.e. we don't insert casts that will just be cleaned up anyway)? Even if the change is only internal to the SDAG processing and the final code doesn't change, would it be possible to have an MIR test case that shows the difference?
@nemanjai I added two test cases the first taken from ARM.addrspacecast.ll and the second from a bug that the Julia frontend hit.
It should be a non-functional change for the backend itself.
LGTM.
test/CodeGen/PowerPC/addrspacecast.ll | ||
---|---|---|
1 ↗ | (On Diff #138926) | I'm not sure how /dev/null works for targets I'm not familiar with (like Windows). Is this a common thing to do? But in either case, I assume that lit checks for the return code from the llc invocation, so I'm fine either way. |
Thanks! Can you land this for me I don't have the commit bit.
test/CodeGen/PowerPC/addrspacecast.ll | ||
---|---|---|
1 ↗ | (On Diff #138926) | Yeah this seems fairly standard and was taken from the ARM/addrspacecast.ll |