Binyamin and Ed, Certainly, if operands were pre-fetched based on 31-bit addresses, the addresses of operands could change once you were in 64-bit mode, so the actual operands could be changed. If operands were changed, since the processor makes pre-fetching decisions based on preliminary compares of possibly changed operands, wouldn't a pipeline flush be in order? It all depends on the millicode and I'm not sure what they do. Perhaps someone with more knowledge will supply us with an answer.
I don't have any hard knowledge on this situation, I'm just speculating. I do know that switch from primary to armode using SAC caused a synchronization, and eventually the SACF instruction was introduced which made the same switch but without causing a synchronization. Just the same, in critical path code, I make a point to stay in the same mode without switching if at all possible. For now, that means running in ARMODE AMODE(31) and soon it will mean running in PRIMMODE AMODE(64). Sometime later (days or weeks), I may have hard data from an analyzer that will reveal, at least on our machine (a z/9) what results we experience. Tom Harper IMS Utilities Development Team NEON Enterprise Software -----Original Message----- From: IBM Mainframe Discussion List [mailto:[EMAIL PROTECTED] On Behalf Of Binyamin Dissen Sent: Wednesday, May 30, 2007 1:33 PM To: [email protected] Subject: Re: Is there a cost in switching AMODEs? On Wed, 30 May 2007 10:39:48 -0700 Edward Jaffe <[EMAIL PROTECTED]> wrote: :>Binyamin Dissen wrote: :>> :>In my experience, excessive switching of addressing mode is quite :>> :>expensive. I suspect it's because the entire pipeline must be flushed. :>> I didn't see that in the POPs. :>AFAIK, the Principles of Operation does not discuss processor pipeline :>considerations for any instruction. There is some cursory information :>here and there regarding instruction prefetch. That's about it. Why would an AMODE change require a pipline flush? It doesn't branch anywhere. The current pipline should be valid. -- Binyamin Dissen <[EMAIL PROTECTED]> http://www.dissensoftware.com Director, Dissen Software, Bar & Grill - Israel ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
