I tested this with "ninja check-llvm" on a Release build with all architectures
enabled, and on a Debug build (for just x86).
Found with llvm-cov.
Paths
| Differential D24435
Delete more dead code in SelectionDAG (NFC) ClosedPublic Authored by vsk on Sep 9 2016, 6:07 PM.
Details Summary I tested this with "ninja check-llvm" on a Release build with all architectures Found with llvm-cov.
Diff Detail Event Timelinevsk updated this object. Comment Actions The SelectNodeTo / getMachineNode helpers can probably go as they're very easy to re-add if we ever get a use case.
Comment Actions As per Mehdi's suggestion - please ask on the llvm-dev list for somebody downstream to take ownership of TargetIndexSDNode/getTargetIndex and improve testing. Otherwise it shouldn't be left to rot in trunk (but that should be a separate patch). The rest of the patch looks alright.
Comment Actions It looks like this patch is OK to land, minus changes related to SelectionDAG:: getTargetIndex. I filed http://llvm.org/PR30625 about adding test coverage for that function. I'll commit this later today if there are no more issues. RKSimon edited edge metadata. Comment Actions
Agreed. LGTM This revision is now accepted and ready to land.Oct 6 2016, 11:41 AM Closed by commit rL283505: Delete some dead code in SelectionDAG (NFC) (authored by vedantk). · Explain WhyOct 6 2016, 4:02 PM This revision was automatically updated to reflect the committed changes.
Revision Contents
Diff 70936 include/llvm/CodeGen/SelectionDAG.h
include/llvm/Target/TargetLowering.h
lib/CodeGen/SelectionDAG/DAGCombiner.cpp
lib/CodeGen/SelectionDAG/SelectionDAG.cpp
lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
|
This should stay while TargetIndexSDNode exists.
However, TargetIndexSDNode doesn't appear to be used so maybe start a conversion on llvm-dev about removing it and if successful then propose a separate patch?