https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97417
--- Comment #5 from Jim Wilson <wilson at gcc dot gnu.org> --- Yes, the volatile is the problem. We need to disable some optimizations like the combiner to avoid breaking the semantics of volatile. However, if you try looking at other ports, like arm, you can see that they don't have this problem because they generate different RTL at the start and hence do not need the combiner to generate the sign-extended load. So the proposal here is that we modify the RISC-V gcc port to also emit the sign-extended load at RTL generation time, which solves this problem. And then we need to do some testing to make sure that this actually generates good code in every case, as we don't want to accidentally introduce a code size or performance regression while fixing this volatile optimization problem. If you are curious about the combiner issue, see the init_recog_no_volatile call in combine.c. If you comment that out, the andi will be optimized away. But we can't remove that call, because that would break programs using volatile.
