Both WhileLoopStart and LoopDec may get turned into a sub and cmp pair, so add an implicit def to these pseudo instructions in that WLS and LE aren't generated. This has resulted in all the mir test changes.
Details
Details
Diff Detail
Diff Detail
- Repository
- rL LLVM
Event Timeline
Comment Actions
Looks like a sensible fix to me, just one question/nit inline.
lib/Target/ARM/ARMInstrThumb2.td | ||
---|---|---|
5226 ↗ | (On Diff #211722) | Instruction sizes: is that something we can, or actually want to test? I don't think I have seen tests for thhat, but was just wondering it because you're changing it here. If there's no precedent for it, perhaps just a comment that we assume that this expands to 2 instructions? |
lib/Target/ARM/ARMInstrThumb2.td | ||
---|---|---|
5226 ↗ | (On Diff #211722) | I'll add a comment! |
Comment Actions
I made some changes before committing - after noticing I should have attached the def to LoopEnd and not LoopDec.