[Bug target/121266] [16 regression] (x86) Missed an '-Os' optimization on setting a register value to -1

[Explorer09 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121266

--- Comment #4 from Kang-Che Sung <Explorer09 at gmail dot com> ---
Sounds like a microcode issue in processors, no?

I remember that `xor eax, eax`, a common pattern of setting eax value to 0,
doesn't create a dependency on eax as a special case during decode. It's a pity
that this little `or -1, eax` optimization is gone (except in '-Oz' mode).

Note that gcc 15.1 uses two instructions `push -1  pop rax` for this. Same in
Clang 20.1.0 according to my testing.