https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122904
--- Comment #1 from Jeffrey A. Law <law at gcc dot gnu.org> ---
For rv64 I think the first and second sequences are probably optimal.
The 3rd can be better implemented with a logical right shift and comparison
against a simplified constant:
srli a0, a0, 32
sltiu a0, a0, 3
The 4th can be improved using similar ideas to:
srli a0, a0, 32
sltiu a0, a0, 3
xori a0, a0, 1
But what would be even better (the li has no incoming data dependency):
srli a0, a0, 32
li temp, 2
sgtu a0,a0,temp
In both the func3 and func4 cases we probably need to fix the mvconst_internal
issues for risc-v.
