Not pressing, but if anyone has thoughts on the strict float PR, especially suggestions for testing, I’d love to hear them. Perhaps the first design decision to evaluate is the AllowStrictFloat and ForceStrictFloat target flags. Using per-instruction fast-math flags requires turning off the global fast-math options. (If those options are on globally, all instructions get them implicitly. There’s no way for an instruction to turn off the global option.) I felt the difference between having the global flags on and trying to put the flags on all the instructions instead was a big enough difference to require turning it on with a target flag. I anticipate a lot of people using AllowStrictFloat in their base, always on, flag set.