On Dec 14, 2006, at 7:48 PM, Reid Spencer wrote: >> >> The interpreter is broken: all unsigned comparisons with signed >> operands will hit the unhandled type case. > > Yup. Fixed the macro to take two arguments. The first is to match the > type of the operand. > The second is the type to use to pull the values out of the > GenericValue > union, effecitvely doing the type cast by field selection.
Ok >> --- lib/Target/CBackend/Writer.cpp 7 Dec 2006 23:41:45 -0000 1.294 >> +++ lib/Target/CBackend/Writer.cpp 10 Dec 2006 23:26:54 -0000 >> @@ -712,10 +717,44 @@ void CWriter::printConstant(Constant *CP >> case Instruction::SetGT: Out << " > "; break; >> >> This apparently doesn't cast the operands of icmp constant exprs to >> the right sign. > > It does, you just didn't see it in the patch because that part didn't > change. Ok. >> ... There ya go again, trying to inflate your LOC count. > > Just a blunder. Uh huh, sure... right... :) >> >> This (when it becomes active) isn't sufficient. When >> FiniteOnlyFPMath >> is enabled, all the 'o's and 'u's should get dropped. > > Right, missed that. One question: > > For the FCMP_ORD and FCMP_UNO cases, what should the FininteOnlyFPMath > condition be? > There isn't really a mapping for it, right? Just make it the same > as the > regular case? Yeah, please just make them do the 'correct' thing, even though they strictly don't have to. I'm still behind in patch review, I'll try to respond to your other email tomorrow, -Chris _______________________________________________ llvm-commits mailing list llvm-commits@cs.uiuc.edu http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
