Xi Ruoyao <[email protected]> writes: > On Mon, 2026-03-09 at 14:41 +0000, Richard Sandiford wrote: >> As Andrew said in one of the PRs, although the older MIPS ISA specs >> required inputs to 32-bit instructions to be sign-extended, no MIPS >> hardware that I know of actually trapped on invalid register inputs. >> And almost all MIPS hardware does the obvious thing. (IIRC, SB-1 >> didn't sign-extend the result if the inputs weren't sign-extended, but >> I don't remember any others caring either way.) > > For 32-bit division with non-sign-extended register input, Loongson > 3A4000 really produces trashes (for some cases happen to be the non- > extended result, but completely nonsense for other cases).
Ah, interesting, thanks. I suppose it makes sense that, if one operation was going to be affected, it would be division. Some gates could be saved by assuming that the inputs are correct... Richard
