When looking into #232 I reverted expr.cc back to the upstream gcc 12.2.0. This removed the HACK but also made another slight change regarding PUSH_ROUNDING.
Good point. Still trying to figure this out some more.
It looks like the code in question is not being hit any more, but on the other hand, I do think it should take PUSH_ROUNDING into account if it is ever invoked.
When looking into #232 I reverted expr.cc back to the upstream gcc 12.2.0. This removed the
HACK
but also made another slight change regardingPUSH_ROUNDING
.Now that the hack is gone, you might want to consider this change as well. See https://github.com/gcc-mirror/gcc/blob/releases/gcc-12/gcc/expr.cc#L4430-L4435 for context.