[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: MCL incompatible with Radius accelerator: solutions?

> Hi,
>    I'm starting to get annoyed by the fact that MCL is incompatible with
> a Radious accelerator that we have on one of the SEs in our lab, a
> machine onto which I am sometimes bumped. Is there any known way to make
> MCL work with the accelerator? I'm using MCL 1.3.2.
>              Thanks for any help.
>                          Chris

When it dies, does it die with a "System Error #11 (miscellaneous hardware
exception)" ?

My recollection is that there was a problem that manifested itself
on Radius accelerators that weren't equipped with floating-point coprocessors:
for reasons that aren't entirely clear to me, attempts to execute floating-
point instructions yielded "coprocessor protocol violation" exceptions
instead of the "F-line" exceptions that similar platforms would generate.
(The 1.3.2 kernel tried to detect the presence of an FPU by executing an
"FNOP" instruction and catching any resulting "F-line" exception; coprocessor
protocol violations went undetected.)

My even vaguer recollection is that this had something to do with the way
in which Radius boards were normally configured: a jumper was set to indicate
that an FPU was installed regardless of whether or not this was the case, in
order to simplify a subsequent FPU upgrade.  (I heard this second- or third-
hand a few years ago; although the explanation makes sense to me and seems
to explain the behavior users have reported, I'm not absolutely sure that
it's correct.)

If I'm correct in recalling that a jumper setting was to blame, that would
certainly suggest a solution.  I tried at one point to get a definitive
answer from someone at Radius, but wound up just getting annoyed ...