On 06/25/2018 10:20 AM, Peter Maydell wrote: >> +#define DO_FCMGE(TYPE, X, Y, ST) TYPE##_compare(Y, X, ST) <= 0 > > I was expecting the RHS of this to be TYPE##_le(Y, X, ST). > This prompted me to notice that softfloat has as well as > the generic 'compare' routines also specialized _le/lt/unordered > etc functions for float64 and float32 but not float16, which is > a bit inconsistent...
Yes. I have a mild preference to using floatN_compare everywhere and removing the other functions entirely. But that's a medium sized job across multiple targets. I also agree that in the meantime the inconsistency is surprising. r~
