[Bug rtl-optimization/126914] [17 regressions] noce_try_shifted_store_flag pessimizes "if (c) x = x OP 2^n" on targets with cheap conditional moves

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126914

--- Comment #2 from Jeffrey A. Law <law at gcc dot gnu.org> ---
I wonder if we're going to have to bite the bullet and cost multiple
sequeneces.  ifcvt isn't really set up for that, but it's bubbled up a couple
times already (this case for loongarch being one recent example and conditional
negation being another).

Essentially priority of one sequence over another is statically encoded in the
order in which we try the different approaches rather than building several
sequences, costing them all and selecting the cheapest.