[Bug tree-optimization/123958] [16 regression] ice in useless_type_conversion_p, at gimple-expr.cc:85 since r16-7264-ga70bdd86984c0b6cd8520e30ec713e3101c5d45f

[roger at nextmovesoftware dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123958

--- Comment #7 from Roger Sayle <roger at nextmovesoftware dot com> ---
Can someone re-run the bisect with -fdisable-tree-pow?  The code in FMA to
lower pow(x,2.0) to x*x has been broken since before my errno change.  Indeed
the FMA pass has been relying on x*x to have already been lowered by the
expand_pow pass for sometime.  The bit-rot that now generates bad PHIs could
have been introduced, but been latent, for quite a while [POW conversion to FMA
was introduced in 2019, but has been "vestigial"/unused since the expand_pow
pass was added in 2022].