[Bug rtl-optimization/122215] Nondeterminism in ra on riscv32

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122215

--- Comment #10 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Encouraging:

==497119== Conditional jump or move depends on uninitialised value(s)
==497119==    at 0x8C858B: improve_allocation() (ira-color.cc:3388)
==497119==    by 0x8CE92E: color_allocnos (ira-color.cc:3616)
==497119==    by 0x8CE92E: color_pass(ira_loop_tree_node*) (ira-color.cc:3725)
==497119==    by 0x8B5DA5: ira_traverse_loop_tree(bool, ira_loop_tree_node*,
void (*)(ira_loop_tree_node*), void (*)(ira_loop_tree_node*))
(ira-build.cc:1784)
==497119==    by 0x8CAD27: do_coloring (ira-color.cc:3861)

We're doing a cost based test, so goofing this up won't cause incorrect code,
but could change meaningfully.  Now time to dive into whether or not it's been
fixed or just gone latent.