[PATCH] Emit minnum / maxnum for __builtin_fmin/fmax

[hfinkel at anl.gov]

This might seem like a silly question, but what's the motivation for doing this? These __builtin__* functions are emitted as readnone which should enable the backend to optimize them using TargetLibraryInfo just as is can for the intrinsics. The intrinsics are important for vectorization, but I'd hope that for scalar optimization, they're essentially the same as the libcalls. Is this not true?

http://reviews.llvm.org/D5896