[Info-vax] Moving WASD from one box to another...

[Jan-Erik Söderholm]

John Reagan wrote:
> "Mark Daniel" <mark.daniel at vsm.com.au> wrote in message 
> news:0174f553$0$20652$c3e8da3 at news.astraweb.com...
>> Jan-Erik Söderholm wrote:
>>
>>> The CC compiler has the same writing in the HELP texts, so I guess
>>> this emulation of higher-EV-code is not tied to the compiler
>>> itself.
>> I sure others here can comment authoritatively on compiler construction.
>>
> 
> /ARCH lets the compiler generate new instructions.  EV56 added the byte/word 
> instructions.  They are emulated on pre-EV56 CPUs by the operating system. 
> Instructions added after EV56 have little overall benefit for most 
> applications.  It helps a few applications (mostly SPECmarks at the time) 
> with SQRT and FTOI, but in general, not much for others.  [I'm not sure that 
> the emulator does the right thing for such instructions on pre-EV6 systems. 
> I've never tried it.]
> 
> /OPT=TUNE controls the instruction scheduling for a particular chip variant. 
> Different schedulings will never produce the wrong answer, but might cause a 
> cache miss or a stall.  I've seen folks play with /ARCH=EV56/OPT=TUNE=EV6 
> and such, but again, only playing with SPECmarks.
> 
> Personally, I'd just say /ARCH=EV56 (which will make EV56 the default for 
> /OPT=TUNE) and be done with it if you think you won't see many 
> EV4/EV45/EV5s.
> 
> John Reagan
> 
> 
> 

OK, good info !
So just building with /ARCH=HOST (or whatever it was) on our EV56
will get us pretty as-good-as-possible code on our EV67 system.
Fine...

Jan-Erik.