[Info-vax] NaT consumption faults with COBOL?

[John Reagan]

"IanMiller" <gxys at uk2.net> wrote in message 
news:86368d83-374f-4e2c-bba0-eab236fa0718 at j4g2000yqe.googlegroups.com...
On 20 Nov, 02:04, "John Reagan" <johnrrea... at earthlink.net> wrote:

>
> And there are instructions to load/save the entire set of NaT bits. They
> live in their own application register with special instructions to access
> them. They need to be saved/restore at context swaps, interrupts, etc. The
> problem is that they are pretty slow instructions since they interrupt the
> pipeline/parallel execution nature of the chip.
>
> Same issue with the setf.sig/getf.sig instructions which move integer
> registers to/from the 64-bit mantissa of the floating registers. They are
> used quite a bit, but they are slow since it forces the integer unit on 
> the
> chip to start talking to the floating unit on the chip. They have 
> different
> number of pipeline stages.
>
> John


>All of which demonstrates things are even weirder than they first
>appear, even for the Itainium architecture which is very strange in
>places.

Like what?  Having the ability to attempt to prefetch memory and hoist 
operations out of a loop on the rare chance that the memory doesn't exist 
but if it did, you save lots of cycles?  Both control and data speculation 
on Itanium are one of the few things that really do make a potential 
difference.

Like what?  Slowing the machine down when having two internal functional 
units talk to each other?  The Alpha FTOI and ITOF instructions are just as 
bad.

John