Prevent folding of operations with memory loads when one of the sources has undefined register update.
Details
Details
Diff Detail
Diff Detail
- Repository
- rL LLVM
Event Timeline
Comment Actions
I'm going to accept this because AVX and SSE are consistent on cvtsi2ss and similar instructions now. But we should file a bug to fix both.
test/CodeGen/X86/fast-isel-int-float-conversion-x86-64.ll | ||
---|---|---|
80 ↗ | (On Diff #149407) | This did nothing to prevent an undef register update. The load goes to a GPR so it didn't do any good. But it is consistent with what was done for the SSE intruction. |