https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78796
--- Comment #6 from wilco at gcc dot gnu.org --- (In reply to James Greenhalgh from comment #4) > This looks sensible, but why not also drop the: > > if (nopcrelative_literal_loads > > As was done in r237607? > > Wilco, can you comment on why you made that change and whether it would be > appropriate to keep it in the backport? The reason is to remove the 1MB per function limit. If you enable the Cortex-A53 workaround then there is an unexpected 1MB limit for functions in the large model by default, and there is no way to disable this (-mno-pc-relative-literal-loads does not work, which is what my patch added). So yes I agree removing the nopcrelative_literal_loads in the patch is a good idea. It's probably even better to take my whole patch so the undefined reads of aarch64_nopcrelative_literal_loads are fixed too. Yes that simple patch fixed that many issues...
