[Bug rtl-optimization/64317] [5 Regression] Ineffective allocation of PIC base register

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64317

--- Comment #23 from Jeffrey A. Law <law at redhat dot com> ---
So with a functional prototype in place, I can get good code for this test.

However, deeper testing of that prototype is proving difficult simply because
the postreload-gcse.c code very rarely does anything useful, at least on x86 or
x86-64.

I put in some dumping in gcse_after_reload_main to always dump its statistics
if its expression table has > 0 elements.  Believe it or not, this *never*
triggers during a bootstrap on x86_64.  I was so amazed by this I fired up a VM
and bootstraped i686 as well, nope, never fires there either.  Concerned that
maybe my other changes were somehow getting in the way, I put in just the
dumping code, still doesn't fire.  Yet it fires semi regularly in the
testsuite.

The original author of the code in postreload-gcse.c was doing their
development on PPC boxes, so I'm going to grab one and see if I can get some
useful testing from on it.