This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Optimize andl $0xffffffff, %eax
Jakub Jelinek <firstname.lastname@example.org> writes:
> I fear that's too risky, on most targets zero_extend?i?i2 is implemented
> using and, so if simplify-rtx canonicalizes that to zero-extend, we
> risk oscillating forever between the two.
As I said this is already done by make_compound_operation; you're probably not
combining two instructions to get to this pattern on x86 so this never kicks
in. But anyway you wouldn't oscillate because the fact that zero_extend is
implemented with an and is not exposed to the backend but it's hidden in the