[Bug tree-optimization/83580] [6/7/8 Regression] Wrong code caused by vectorization

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83580

--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
OK, so this is all a bit fishy.  We're basically looking at

 (ssizetype)exp

splitting exp recursively into X + CST where we keep the overall CST
in ssizetype.  For plus/minus we associate for mult we distribute.
Given we do all this in effective ssizetype we may not cross a
conversion boundary to unsigned, but STRIP_NOPS just does that.

It might be the very outermost STRIP_NOPS is ok (not 100% sure), but
at least I'm testing the fallout of removing the STRIP_NOPS from
split_constant_offset.