cheap analog square function?

Ship diesel engines are much more efficient than steam engines

big low speed "diesel" ship engines also run on bunker C where they are allowed to

I heard that the engine room pics in the Titanic movie were filmed on the

It's old and creaky so they run something crazy like 150 PSI steam pressure.

I think the Titanic was a turbine.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

It had both! 3 propellers, the port and starboard powered by triple expansion steam engines, and then the exhaust from those fed a turbine to power the center propeller at higher speeds. Port and starboard engines only for maneuvering as the turbine couldn't go in reverse.

I remember we had this stuff regularly washing ashore on the Dutch beaches. It's better now, fortunately, even though it still happens now and again. It's vile indeed.

Jeroen Belleman

Focal wasn't up to much. I programmed my machine in assembler - Macro-8 - and used interrupts to get tolerably precise timing.

The chemistry department wrote me off as gadget-happy. which wasn't entirely fair. My gadgets did work

My father got an Argus mini-computer from Ferranti to run his continuous counter-current digestor a bit later. It had enough spare capacity to run a couple of paper machines as well, where it was lot more valuable - the programmed transition from one weight of paper to another went a whole lot faster under computer control.

Eventually the digestor went back to manual control - the control crew had learned how to do it right from the computer program, and didn't need the extra help, and the computer took over a few more paper machine..

Focal-8 was fabulous, a nice recursive language with floats and math functions that ran useful stuff on a 4K-word 12-bit machine. It used all sorts of hashing tricks to save core.

The internal stack-oriented architecture of Focal-8 inspired the cool PDP-11 architecture which in turn inspired the 68K. Pity that IBM picked the Intel dog and Microsoft.

Focal-11 was even better. But it had a terrible random-number function, the silly modulo thing. I patched it to use a PRBS thing, Rick Merrill liked it, and he sent me the program listing.

We keypunched it and spun our own versions. We did pipeline control systems with that and wrote a version for the first satellite navigation experiments that led to GPS.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

how did stack-oriented inspire architectures with lots of orthogonal registers, by showing how it should not be done?

If high precision is not needed and only LF response is enough then I have used PWM techniques:

Circuit A is the basic squaring concept, relies on modern open-drain output comparators being pretty good switches to ground. Assumes you already have a source of sawtooth or triangle waves with defined zero and peak values in the system.

Circuit C shows the triwave source I used, with switching circa 100kHz the squarer/rooter is good to perhaps 1kHz? Steeper filtering could improve that maybe.

Circuit B shows the squarer coerced by opamp feedback into rooting.

I have also implemented this with a self modulating PWM oscillator (one comparator) driving an analog switch to chop Vx by its own ratio. I had a section of a 74HC4053 available.

Have fun!

piglet

Stack-oriented inspired stack-oriented. In the PDP-11, any of the registers could be used as a stack pointer.

Focal-8 faked the stack operations with klunky PDP-8 opcodes. The 8's subroutine call machanism was barbaric: save the return addess at the first word of the destination page. Rick invented PUSHJ, save the RA on the stack and jump anywhere in the 4K space. He made the PDP-8 into a sort of RPN machine.

DEC was going to build yet another clumsy 16-bit computer, like the NOVA or the HP 2114 thing, but Rick talked them into the elegant PDP-11 architecture, a stack-oriented machine with eight registers.

It could do fun things like

NEG PC

negate the program counter.

There was the land mine opcode

MOV -(PC), -(PC)

(014747 octal)

which copied itself one location below itself and re-executed that.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

so it had a bunch of orthogonal registers that could be used for anything you like, including stack pointer(s) though I don't see why you wouldn't always use R6

that is register oriented, stack oriented is things like Forth and Java

R6 was used as the sp for JSR, jump subroutine, calls. And R7 was the program counter. But all opcodes worked on all registers, and any reg could be used as a stack pointer or an index register.

The PDP-11 architecture inspired c, for better or for worse.

DEC did a lot of ugly hairball async logic back when, delay lines and RCs and such. The first 11 was two boards full of TTL nastiness. There was a short paper-tape reader program that would tease a metastability and hang up the CPU logic in a few seconds.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Am 12.02.26 um 23:12 schrieb john larkin:

That does not make it stack-oriented. It had registers and nice addressing modes with pre-decrement and post increment etc. Nice for string handling, too.

The PDP11/40e in our group was the 1st Unix machine on this side of the pond.

Stack-oriented was Burroughs(sp?) B1700 or the p-code machines like UCSD or Andrew Tanenbaum's EM. I wrote a Z8000 version of the EM.

We had to do a 1 semester group project for VLSI design and I persuaded the group to do a somewhat dumbed-down version of Tanenbaums Experimental Machine in HP's full custom dynamic n-mos process.

Unluckily, another working group donated us a large metal square across all of the ALU, so testing was not needed. :-( The design rule checker had its limitations for the multi project wafer.

OpCodes are precious!!!

cheers, Gerhard

I called it stack oriented!

A couple of people made real stack machines, hyper-CISC things with advanced instructions. HP and Intel as I recall. They were pig slow and didn't last.

On the PDP-11 there was a short program to fill every location in memory with zero, the halt opcode.

And another program would fill all of memory with the NOP opcode, and it would cycle on that forever.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Stack-oriented doesn't require use of THE stack but, rather, *A* stack. So, all you need is the ability to address memory through a pointer to implement one.

It is not uncommon to have different stacks for different uses.

An implicit advantage is typically greater locality of reference making caching potentially more effective. And, instruction sets tend to be simpler (cleaner?).

CPUs with register files are more likely to romp around memory indiscriminately as you can easily switch between many "pointers" to many different objects. They typically have much more processor state making them annoying for use with real-time activities (unless they have "tricks" to making preserving state easier/faster)

The B5000 (et al.) were probably the earliest commercially successful "native" stack machines -- and, in general, much more interesting than the more banal designs that have followed.

Thankfully, processors are fast enough that you can *emulate* many of these features instead of being stuck with the more mundane mechanisms present.

Am 13.02.26 um 01:09 schrieb john larkin:

That was the iAPX432. HP's brainchild, Intel prepared the mass production, which never really happened. We had a Multibus board with one 432 in our processor zoo at the Tech Univ Berlin, but nobody adopted it to the point that it did useful work. The entry cost was just too high. With today's cache technology it might make more sense. It had nothing to do with stack machines.

Western Digital had a stack machine for UCSD Pascal; that was essentially the chip set of the LSI 11/23 with custom microprograms.

Our above-mentioned PDP11/40e was also microprogrammable (therefore the e suffix). There were only 5 of it's kind in the world; we had 2 of them in the same unibus. Someone at the institute wrote a microprogrammed stack machine for Per Brinch Hansen's Pascal compiler that ran like a scalded dog.

Luckily for Intel, they did not drop the X86. Internally, a Pentiyummy has nothing in common with x86. There are 100s of registers (called renaming registers) that can hold multiple incarnations of eax, ebx and so on, so that multiple sequential instructions can be executed at the same time, with hardware that checks for dependencies over time.

When the 200 MHz Pentium appeared, that was the EOL for most Riscs, including Mips, Fairchild Clipper, DEC Alpha, IBM Power PC, 32032 and and and.

That filled some badly needed voids.

Gerhard

Could you please explain better how the PWM squarer circuit works? It look like it would be easy to build, but I don't grasp what is going on.

A fairly simple single-opamp or comparator circuit can convert input V to a duty cycle n. A transistor or analog switch or even a diode can multiply V * n. Lowpass filter that and you have V^2.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Memory used to be slow and expensive, so it sort of made sense to have every instruction do lots of complex stuff, like evaluate poynomials. Tons of microcode.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics