[Bug target/115910] [15 Regression] ((unsigned)x)/3 with a range for (unsigned)x that does not have the sign bit set seems to produce much worse code

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115910

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Note, for 0.0 in 387 stack we use still cost 1, similarly for all ones vectors,
and for various other 387 stack standard constants cost 2, so those again will
misbehave in pattern_cost and DTRT in current SET handling of rtx_cost hook.

Note, the testcases could be simplified to just return x / 3U;