https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117476
--- Comment #30 from Jeffrey A. Law <law at gcc dot gnu.org> --- I've reverted the patch and given Alexey some guidance on how to fix the testcase from Zdenek. Alexey, I'd recommend including Zdenek's testcase as well. Note that it uses 128bit objects, so in the test you'll need a line like this: /* { dg-require-effective-target int128 } */ So that Zdenek's test is only run on targets supporting 128bit wide integers.