https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115683
Richard Biener <rguenth at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Last reconfirmed| |2024-07-01 Status|UNCONFIRMED |NEW Ever confirmed|0 |1 --- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> --- For the a > b ? a : b and a < b ? a : b (or is that a <= b ? a : b for min{ps,pd}?) I wonder if we want new optabs like cond_fmax (that name is already taken though - maybe fmax_gt and fmin_le?). I also wonder if there are other archs with similar instructions. The unsigned compare looks like a general trick we could use in vectorizer pattern recognition or alternatively in vectorizable_comparison which needs adjustments anyway in case vec_cmp expanders start to reject some compare operators. Of course the testcase use generic vectors so the same applies to vector lowering (code sharing between vectorization and lowering as far as "tricks" go would be nice though this one looks difficult to generalize).