9 Nov
1999
9 Nov
'99
12:53 p.m.
On 1999/11/09 at 7:18pm +0000, Tony Duell wrote:
And I've over-simplified the PERQ. Believe me,
it's _not_ that simple
when you really start looking at it.
That's OK; I over-simplified the Daybreak :-) Some of the operations are
weird, and only a fraction of encodable operations actually work, due to
timing constraints. I don't fully understand it yet; I'll have to print
the documents first.
[....]
The worst mistake they ever made was to use the 2910. Oh, it's a fine
chip _but_ you can't extend the address width easily. So on the 16K
board, where they needed a 14 bit control store address, they had a '2
bit kludge'.
Hmm. The Daybreak has 8K control store, but only 12 address bits;
everything I've read so far only mentions 12 bit addresses.
Then there's the problem of loading the control
store. The _only_ device
that can address the control store is the 2910. So there's a problem in
actually loading it -- how on earth do you specify the address.
This reminds of a device I once worked on, made by a company whose name
you'd recognize. The part was *almost* a complete processor; it had a
sequencer, ALU, registers, etc. but didn't quite nail down the
instruction set. The manufacturer had designed an evaluation board; I was
evidently the first person to try to program it. They had taken an
extreme RISC approach, providing only one addressing mode: indirect
through a register. The only way to get an arbitrary address into a
register was, of course, to load it from an address in a register....
They changed the design.
BTW, does the Daybreak have any kind of grpahics
processor? I'd wondered
if it was one of the gate arrays on the memory board.
I haven't really looked at it yet. I think two of the gate arrays are
involved, but I think it's just a dumb frame buffer.
Looking again, my machine seems to have mostly standard
TTL number chips
on the CPU and memory boards and mostly 733W numbers on the IOP board.
Very few chips have both numbers :-(
OK. Mine have about 10% 733 only, 60% standard numbers only, and 30%
both, but naturally the same parts appear repeatedly. It seems that
"leading" zeroes (after the W) don't matter. Here are the matching
numbers from the Daybreak boards:
733W 0098 75189
733W 0318 74S00
733W 0319 74S04
733W 0321 74S260
733W 0339 7414
733W 0341 74279
733W 0351 74S257
733W 1523 74S189
733W 1550 AM27S07DC
733W 1606 74S10
733W 1611 74S08
733W 1616 74S138
733W 1619 74S20
733W 1620 74S64
733W 1621 74S51
733W 1624 74S138
733W 1625 74LS240
733W 1626 74LS244
733W 1630 74S175
733W 1633 74S240
733W 1634 74S241
733W 1638 74S280
733W 1640 74S374
733W 1643 74S02
733W 1644 74S11
733W 1646 74S32
733W 1647 74S38
733W 1648 74S86
733W 1652 74S157
733W 1663 74LS393
733W 1675 74LS74
733W 1676 74LS191
733W 1682 10124
733W 1698 74LS374
733W 1699 74S373
733W 1747 74LS259B
733W 1770 74LS163
733W 1771 74S74
733W 2136 74S37
733W 2137 74S151
733W 2369 MCA1300PKG/E5
733W 2380 MCA1300PLM/E4
733W 2538 CY7C122-25PC
733W 2550 C67402J
733W 2552 MK4501N-12
733W 3042 AM2901CPC
I might have missed some, there might be errors here, and I haven't
looked at the Dandelion boards.
--
Kevin Schoedel
schoedel(a)kw.igs.net