On Tue, Dec 08, 2020 at 03:18:54PM -0600, Richard Henderson wrote: > As for modifying the fast path cases, the code is quite delicate, > and you run into problems with live registers. Which could be > worked around in each backend, but... why?
Perhaps thinking that working at IR level would prevent these live registers issues, seconded by the fact that vCPU TLBs handling seem to be tcg-target agnostic. But I do understand your position, have no patch suggesting a viable alternative implementation, and most of all don't want to take more of your time. I appreciated the discussion and your help Richard. Thanks again.
