Yes, your interpretation is what I meant. I was thinking the "branch prediction" would sort that out, but, probably unreasonably, that thought assumes that most programmers make their definitions large enough (and not overly large) to not have overflow, although on the other hand even doing it badly probably doesn't lead to lots of overflow.
I suppose if someone could poke the code into V5, it may be revealing. If anyone wants to try, define all the fields in the LINKAGE SECTION and establish addressing for them. On Sunday, 7 February 2016 02:20:20 UTC, Charles Mills wrote: > Just commenting on one thing: > > > It is again there for the case of an overflow which has resulted in a > > negative zero > > but I can't think why it is not preceded by a branch on condition for the > > overflow > > value in the Condition Code. > > Do I interpret what you are saying as "I get the need for the ZAP, but > couldn't they have saved the ZAP if no overflow by branching around it if no > overflow?" > > If so, I think the answer is because branches are expensive in the brave new > world of caches and so forth. It's probably generally cheaper to do the zap > -- seeing as how you already have the data in write cache -- than to screw up > the pipeline with a branch. Just my slightly informed guess. > > Charles > ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
