http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46186
--- Comment #24 from joseph at codesourcery dot com <joseph at codesourcery dot com> 2010-10-29 21:58:46 UTC --- On Fri, 29 Oct 2010, sebpop at gmail dot com wrote: > here is a preliminary patch (not tested yet other that the PR testcase). How does this patch deal with needing to compute in a wider type to avoid losing high bits before the division - is that covered by some existing code? It could certainly do with execution testcases that verify cases where computing everything naively in the original type would yield an incorrect result (and those were computing in a type of double the width would still have overflow, etc.).