Propellurino

Hi all,

here: [www.hobby-roboter.de] you can find a 'Propeller'-shield for the arduino with the capabilities of video- and audio-interfacing and some more capacity - maybe it's interesting for the Arduino-RepRap too?

Viktor

Hurro Viktor

Interesting.

It does beg the question "Why ?" though.

Is'nt the propeller a PIC with a minimalist closed source multitasking OS/monitor preprogrammed on it.....

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

... the propeller ( [en.wikipedia.org] ) is a multicore-chip with 8 32-bit RISC-CPU's inside capable of realtime-video-generation and much much more

In fact it's much more cpable than the arduino but it's harder to code too and lacks enough IO's.

In the propellurino it's an interfacing part for realtime video and audio and could be much more in respect of numberchrunching or higher level logics.

Maybe this will evolve into a link to parallel computing and Neural-Networks cpable of 'learning on doing' how to drive the Arduino-RepRap, optimize the extruder/build or solve other problems ...

Viktor

I stand corrected.

I must confess to being a bit of a parallel junkie having done some professional work with Inmos Transputers some years ago.

8 Risc Cores with Round Robin access to shared memory. ALU with no divide

More processing looks good, lack of a hardware/microcode divide is a touch limiting and the Round Robin Memory access is a bottleneck.

DSP processors go to great lengths to be able to do multiplys and divides in a deterministic and timely fashion.

Given that a divide operation takes the most processor cycles (how many depends on what your are dividing by what) the round robin shared memory access would make more sense with the divide, than without it.

If a core is occupied dividing it would not be contending for round robin access to shared memory. Under a software divide the operating core is contending with the others for shared memory.

I think I agree with you that the Propeller has capabilities that the Atmega8 does'nt. 32 bits versus 8 bits is an obvious starter.

Personaly I would rather head the direction of AVR16 or AVR32 and multiple processors with their own memory (MIMD, no contention) simplistically networked.

Or maybe even an AVR32 coprocessor to number crunch for an Arduino. (Again asks the question Why ?, Just use the 32 instead of the 8)

On video I have been reading an interesting article in an electronics magazine where they get a low end 8 bit PIC to do video using the PWM and SPI ports.

PWM for the video timing and the SPI port to stream the video pixels for each line.

Definitely worth a look at if just for the novelty.

Cheers

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

aka47 Wrote:
-------------------------------------------------------
> ...
> I must confess to being a bit of a parallel junkie
> having done some professional work with Inmos
> Transputers some years ago.
> ...

... me too - in my diploma-thesis (around 1988) i had to research the possibilities of massive parallel networks for numberchrunching and simulation of physical problems in the accellerator-development.

In fact i simulated free scalable and online reconfigurable arrays of crosslinked Transputers (all with own memory and 4xLinker-Chip) - my biggest calculated area had 256x256 T-cores ><

The problem was: ... my results had some unexpected superiority! - while my mentor awaited some tenfold optimisation, i received some thousands

We had some "numberchrunching benchmarks" ranging from a C-64 with 1800 minutes calculating time until an Cray XMP with 2 seconds - my nominal array of 64x64 tranputers (embedded in an Atari ATW (= Atari Transputer Workstation)) would need 2.2 seconds and should only cost 250 thousand Deutschmarks compared to more than a Million USD for the Cray

This was to much for him and he couldn't (or won't?) argue it, so he in fact dropped the results

Viktor

I am still sore that the transputer was discontinued by SGS after they acquired INMOS.

As a kid of the Lego generation the transputer was to computation and (something that I was really interested in) neural networks what Lego was to constructional toys.

I closely followed the newer DS link technology (it later influenced firewire) and eagerly awaited the T9 with its six links, it was all set for hyper cube territory.

Then died.

Must admit that I understood where folk were going with the link engines and routing, but for neural networks, if you exploited locality then they were'nt necessary. For other applications , yes of course they were important.

The T9's would have been great for modeling biological neural networks.

There is still the case for resuscitating a transputer alike on FPGA, particularly for this type of "Silicone Brain" application. There is little or no enthusiasm for it though out there.

A stack based processor with hardware scheduling.

It makes you cry to think about it.

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

... maybe it's all based on the same cause?

My first thoughts when diving into Transputer-specs and Occam were about overrunning some serious performance-problems i had with other computers and OSes ... maybe it was simply so good, that the other vendors tooks all neccessary steps to stop them from penetrating the market

In the last 30 years there were some similar occasions, where superior soft- or hardware vanished 2 years prior to hitting the market - e.g. holostore for data-storage or memevolution for 'human-based selfdeveloping/selfoptimizing software' ...

Viktor

My wife and I are Leonardo Da Vinci fans.

When you consider how far before his time he was and how much of his work was lost,then rediscovered, after a large part of it had been independently rediscovered by others, it makes you think.

IS it conspiracy or just plain stupidity and short sightedness.

Conspiracy theory is more fun but takes great organisation. There are very few organisations that are able to carry an agenda successfully forward across successive generations. Those that can are arguably not interested in the minutae of technologies that they don't understand. Much less fear.

Personaly I subscribe to the idea that a lot that is of great importance to humanitys forward progress, is lost to an unconspired conspiracy of independent greed, stupidity and short sightedness.

People appear to be blind to anything that is sufficiently ahead of their perception of it's time.

When you consider that businesses particularly those that have the clout to make things happen on a global scale think in quarter years. That profits (all that they are really interested in) are made and lost within that period. It si amazing that any progress is made at all.

More so when considering how the IP protection structures of Copyright and Patent are used to control markets and impose stasis.

Publishing to a wider audience though, through Open Sourcing and the internet should mitigate this to some degree.

How much remains to be seen.

aka47

PS sorry, I'll get of my soap box now.

---

Necessity hopefully becomes the absentee parent of successfully invented children.

aka47 Wrote:
-------------------------------------------------------
> Given that a divide operation takes the most
> processor cycles (how many depends on what your
> are dividing by what) the round robin shared
> memory access would make more sense with the
> divide, than without it.
>
> If a core is occupied dividing it would not be
> contending for round robin access to shared
> memory. Under a software divide the operating core
> is contending with the others for shared memory.
>
> Cheers
>
> aka47

Depends how its implemented on the core there are methods to do 1 clock divide

Jumping back to the inital posting here...

Interesting microcontroller indeed, but i would say it is not really what the reprap project currently needs to perform any of the tasks it has on his hands.
I am still very new to the project, so correct me if i'm wrong somewhere. As i can see it now, the microcontroller is only running down a "simple" task of moving the toolhead along a given path plus some kind of control loops to avoid colisions and keep the temperature at a given level. Everything that does involve some kind of more intelligent data processing is already precalculated on a very powerful (compared to every microcontroller) personal computer. So, i don't really see where you would need a big numbercrunching, expensive microcontroller on the machine itself.
As it currently is, you need the computer anyway to create the 3d drawings and shapes you want to print out, so it's from my point of view only logical to profit from the massive computing power at your hand and use that to directly generate all the data the reprap needs for printing.

The only place where this improved microcontroller could make sense is if you wanted to have a machine that can copy an existing object by scanning it and then automatically reproduce the scanned shape. But imho this is so far away from what the current project can handle that it is way to early to call for some processor decisions. By the time the technology around the scanning machine would be ready, this parallax propeller would probably be the limiting, outdated part of the machine...

If i would take a step away from the current microcontroller i'd probably go to a Cortex M3 or something in that range that is effective and easy to work with even if you don't have a bachelor degree in embedded software design. Haven't checked the pricetag on this propeller controller yet, but i would expect it to be quite high or at least high enough to hold off the current reprap community.

I don't think viktor is suggesting a processor change right now.

My understanding is that viktor is seeking to discuss and reviewing new technology as it appears.

With any newer technology I guess there is always the question of making the most of existing skills and skills sets, versus the steep learning curve of the new. ie Pain versus Gain.

For a new technology to be a good choice for a sudden change it has to have properties that far out weigh the disadvantages of a shift.

As I understand it viktor is posing the question/s would the rep rap project benefit from a parallel model of processing.

If so would something like the Propeller be worth a looking at ??

I guess there is also the quaduino to consider too....

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

... it's a bit from all

My basic idea was a sort of an optional 'video card' for the arduino capable of generating video-content for an embedded display and/or some other benefits too when needing much more processing power, than the Atmega core has.

The Arduino is much easier to handle, so for the everyday-user it should be the basic Arduino with some aditional libs and options, advanced programmers could hack some other potentialities though

Viktor

We appear to have lost a degree of parallel processing in moving from networked multi processors to a single motherboard.

I do fully understand the arguments re cost, ease and board count. Together with the argument that the inter-axis timing is more constrained and theoretically should produce more accurate results.

On the down side there is less processing power than in a multi processor machine and redevelopment of a sub part entails redevelopment of the whole motherboard. As opposed to plug and play subsystems.

Parallelism on a single processor clearly is not fully concurrent.

In considering a multi core processor to regain a measure of true concurrency we are still (Perhaps more so) heading down the road of the unitary motherboard.

Hmmm adding video capability is interesting. There is certainly an argument for adding a better controler with processing and video (I think Vik is using one of those funny laptops to do this) perhaps as well as breaking down the sub systems into more stand alone parts each with it's own processor.

Best true fine grained concurrency is achieved with asynchronous logic and FPGA's.... but again is single board oriented.

thoughts for what they are worth.

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.