https://bugs.kde.org/show_bug.cgi?id=385409
--- Comment #31 from Julian Seward <jsew...@acm.org> --- (In reply to Florian Weimer from comment #29) > Created attachment 114302 [details] > Use guarded loads in the guest > > This patch changes the guest code to use guarded loads. > > However, this insufficient because the isel implementation does not handle > Ist_LoadG. I don't know how to fix this. Look at how the amd64 back end does it. It basically is going to be a completely normal unconditional load that has a conditional branch in front of it. > s390_insn_cond_move does not perform the U8 → U32 zero extension, so it is > not a good fit here (and even Ist_LoadG with ILGop_8Uto32 is a bit off in > this context, we would need ILGop_Ident8). This doesn't matter very much. Presumably s390 has an instruction that loads 8 bits from memory and zero extends the loaded value out to 64 bits. Then you can use the ILGop_8Uto64 descriptor in your conditional load. The narrowing operations that you will have to use afterwards (64to8, or 32to8) become no-ops in the generated code, so there's no performance disadvantage to using an 8bits-to-64bits-and-back-to-8bits scheme. -- You are receiving this mail because: You are watching all bug changes.