> BTW how did we end up with the regression differences? Presumably you've
> tried that on your machine and it passed. So if we adjust the expected
> file, won't it fail on some other machines?
I had another patch to check for -0 inside float{4,8}_{div,mul}(). I
dropped it on the last set of patches, so the tests were broken. I
get -0 as a result of -x * 0 both on Mac and Linux.