>From PofOp:
The set-address-space-control-fast facility consists of the SET ADDRESS SPACE 
CONTROL FAST (SACF) instruction, which possibly can be used instead of the 
previously existing SET ADDRESS SPACE CONTROL (SAC) instruction, depending on 
whether all of the SAC functions are required. SACF, unlike SAC, does not 
perform the serialization and checkpoint-synchronization functions, nor does it 
cause copies of prefetched instructions to be discarded. SACF provides improved 
performance on some models.

OK, so ... "on some models". Is this one of those "It used to matter, but 
doesn't any more" deals, or should we still think about using SACF when 
appropriate? I did find this in an old post to ASM370-L:
As I understand it SACF was only implemented (as different from SAC) on a few 
models. The basic difference in how the instruction prefetch queue is handled. 
If you are not modifying the instruction stream or switching from home to 
primary (which causes instruction fetch to load from a different memory) there 
should not be a difference.

And do the serialization caveats mean that in cases where we have absolutely no 
expectation of data change in the target address space (as in, "If they do that 
after calling us, all bets are off anyway"), SACF would be appropriate?

Thanks for any insights!
--

...phsiii


----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to