I've not seen SLR, but my comment based on 1990s knowledge is that it's best to keep it simple. Don't access storage unnecessarily (such as L Rn,=F'0') and any of the more obvious methods (such as "LA Rn,0" "SR Rn,Rn" or "XR Rn,Rn") is likely to be optimised as a special case. XR and SR use less memory for the instruction, but LA was the preferred way where I learned most of what I know--perhaps that was because it doesn't set the condition code.
I suspect (I mean guess!) that optimisation for the special case of zeroing a register may mean the advantage of not setting CC is lost in that case. Rupert On Mon, 10 Aug 2020 at 23:51, Paul Gilmartin < [email protected]> wrote: > On 2020-08-10, at 16:28:28, Steve Smith wrote: > > > > The only difference between SR & SLR is how they set the condition code. > > If that could produce a measurable difference in a trillion executions, > I'd > > love to see it. > > > However, the newer z models have a whole family of new > instructions which differ from their older counterparts > only by *not* setting CC. Changing CC can be a pipeline > bottleneck. > > -- gil >
