11 Apr
2026
11 Apr
'26
1:03 p.m.
The VAX vector process was also a great deal of fun for an implementation of Algol 68 we
were working on.
We had access to a VAX 9000 for some testing but from what I remember, we used DEC's
vector emulation software mostly.
We had "great fun" with the attribution for implicit vectorisation of array
processing and collateral clauses and we almost actually got into trouble for including
the RS extensions for loops, the FOR identifier IN (unit delivering [] of some mode) DO
... OD.
Trouble = exceeding our remit.
Anyway, it's all effectively lost. Nobody remembers where any of the project materials
actually went. I hate that.
Also just paged into my memory, we wrote prediction code - part of the code generator
which would output a profile. Along with the runtime profiling, it could (and did
accurately) predict how fast the code would run on the real hardware. We also got into
trouble for that too - disassembling the vector processor emulator code. Did we use Xdelta
for parts of that?
K
-----Original Message-----
From: Dave McGuire [mailto:mcguire@neurotica.com]
Sent: 10 April 2026 21:54
To: The Hobbyist DECnet mailing list <hecnet(a)lists.dfupdate.se>
Subject: [HECnet] Re: quick help on RSX FORTRAN problem?
On 4/10/26 16:42, Paul Koning wrote:
Very much so, yes. It takes two people to move *the power cable* for that machine.
We at LSSM
have a 9000 in the warehouse. I doubt we will try to get it running, but you never know.
One interesting thing about the 9000 is that it's one of only two implementations
of the VAXvector that I'm aware of. We have the other, which is an add-in board for a
VAX-6400. I got that up and running about two years ago; it's a great deal of fun,
but only for FORTRAN.
Does a 9000 need 3-phase power? I did not know there were VAX vector instructions.
Yes, but only those two implementations.
The XMI board, called FV64A, will only work with 6000-400 and -500 processors, though a
box can take one VAXvector XMI board per processor.
It has 16 64-bit- vector registers, vectors can be up to 64 elements long. It crunches
about 90 MFLOPS peak.
I wonder if GCC knows about them (for the VAX back
end). It has really extensive vector support, so if the back end handles that you'd
end up with vectors probably for any language GCC knows about -- or minimally for C and
C++ as well as FORTRAN.
That would be a nice extension, but sadly we can count the number of places that can
take advantage of it on one hand. And the number of machines that implement it that are
actually functional on one finger.
The VAX FORTRAN compiler will emit the vector instructions, and that works well.
Actually it works amazingly well; it autovectorizes code every bit as well as Cray's
compiler, which is really the gold standard.
-Dave
--
Dave McGuire, AK4HZ
New Kensington, PA
_______________________________________________
HECnet mailing list -- hecnet(a)lists.dfupdate.se To unsubscribe send an email to
hecnet-leave(a)lists.dfupdate.se