On Sun, 2018-01-28 at 15:25 +0100, Sébastien Michelland wrote: > > > I thought of forcefully placing my source pointer in a certain > register and issuing an instruction with a hardcoded source operand; > this will probably be my "fallback method". Segher has provided a great answer to this. I think it will solve your problem. > > Would link-time optimization be able to inline a "movua function" for > this purpose? If you put it in a small non-inline function in some translation unit ... according to my experience, with LTO the function will get inlined. Cheers, Oleg
