memory DMA

[arnoldemu]

I was also wondering, is it possible for CPC to support DMA. I don't mean ASIC like "DMA" but DMA via some external DMA controller (like intels one).
I'm expecting the CPC timings would fall apart and the display would break???

It appears the Aleste can do it and it's CPC like...

[gerald]

I was also wondering, is it possible for CPC to support DMA. I don't mean ASIC like "DMA" but DMA via some external DMA controller (like intels one).
I'm expecting the CPC timings would fall apart and the display would break???

It appears the Aleste can do it and it's CPC like...

The CPC could support DMA, using the BUSREQ/BUSACK signal on the extension port. But there are some drawback :
- You steal the bus to the Z80, so your code will be stopped during the DMA
- The DMA controller must be GA friendly and watch for the WAITn signal and do not attempt more than one access between to WAITn

I've got some plan playing with this, but only plan yet 

[rpalmer]

You can connect a DMA to the CPC, but it best to use the Zilog DMA (if you can get them). If you any other make is used then special circuitry may be required, since the Zilog version was designed to work in tandem with the Z80.

The DMA itself MUST use the WAIT line to pause the DMA chip when it is asserted so as to not cause issue with memory access.
The other issue with DMA usage is the use of interrupts, since these will most likely interfere with the GA which make its more difficult to implement a DMA.

rpalmer

[rpalmer]

Also,  I have the Z80 DMA PDF and Applications Manuals (Volumes 1 and 2) for Zilog Devices which includes example logic diagrams for connecting a DMA to the Z80.

rpalmer

[TFM]

I was also wondering, is it possible for CPC to support DMA. I don't mean ASIC like "DMA" but DMA via some external DMA controller (like intels one).
I'm expecting the CPC timings would fall apart and the display would break???

It appears the Aleste can do it and it's CPC like...

The CPC can not use the Z80 DMA, because the expansion port does not provide the needed signals. You can put a DMA into the CPC with the Z80 moved to a daughter board though.

[TotO]

I think the question is more: Why using a DMA on CPC?
If an external device have to access the memory without using the Z80, it can use a dedicated RAM.

[Bryce]

There's lots of uses for DMA on the CPC, but it's rather complicated to implement because it needs access to many points within the CPC.

Bryce.

[arnoldemu]

Thanks all. Interesting.

I was asking why... I wasn't expecting anything to be implemented to use it.

[gerald]

The CPC can not use the Z80 DMA, because the expansion port does not provide the needed signals. You can put a DMA into the CPC with the Z80 moved to a daughter board though.

Can you elaborate ?

[gerald]

Also,  I have the Z80 DMA PDF and Applications Manuals (Volumes 1 and 2) for Zilog Devices which includes example logic diagrams for connecting a DMA to the Z80.

Are those Applications Manuals available in pdf form ?

[rpalmer]

sadly the manuals are in paper form.

[Velktron]

There's also the issue of the CRTC already "stealing" memory cycles from the Z80 and blowing up memory accesses to multiples of 4 cycles -any DMA controller would have to deal/cooperate with the CRTC and of course have direct access to the RAM, perhaps through a priority mechanism that would avoid collisions, and also bypass the Z80 for DMA access through the expansion ports, floppy controller, etc.

In other words, a CPC thus redesigned would be more like an 8-bit Amiga or a C64

Interestingly, the CRTC is the only hardware on the CPC motherboard (on the classic models, at least) which has "DMA" and can do its own memory accesses (too bad that it's not dual-channel with the Z80, heh).

Not all 8-bit computers drove their display "automagically" in that way: e.g. the Z80-based Galaksija had the Z80 drive the video output directly, and it would disable video output completely for intensive processing.

[gerald]

There's also the issue of the CRTC already "stealing" memory cycles from the Z80 and blowing up memory accesses to multiples of 4 cycles -any DMA controller would have to deal/cooperate with the CRTC and of course have direct access to the RAM, perhaps through a priority mechanism that would avoid collisions, and also bypass the Z80 for DMA access through the expansion ports, floppy controller, etc.

In other words, a CPC thus redesigned would be more like an 8-bit Amiga or a C64

Interestingly, the CRTC is the only hardware on the CPC motherboard (on the classic models, at least) which has "DMA" and can do its own memory accesses (too bad that it's not dual-channel with the Z80, heh).

Not all 8-bit computers drove their display "automagically" in that way: e.g. the Z80-based Galaksija had the Z80 drive the video output directly, and it would disable video output completely for intensive processing.

The CRTC does not have DMA access. The CRTC is only a programmable address generator tailored for screen display.
The only "DMA" on the CPC is the Gate array which statically shares the access to the base memory. One time slice for the screen, using CRTC generated address, one time slice for the Z80.
The Gate array is also managing the DRAM row and column addressing.
The WAITn signal on the Z80, driven by the GA is clearly indicating when a memory cycle is available for the Z80 bus.

[Velktron]

I stand corrected. Still, besides the Z80, the only device able to access memory more or less directly is the GA, at least on the classic CPC models. It's like a sort of "forced DMA", so any additional memory access mechanism would have to be able to "coexist" with it as well.

[gerald]

I stand corrected. Still, besides the Z80, the only device able to access memory more or less directly is the GA, at least on the classic CPC models. It's like a sort of "forced DMA", so any additional memory access mechanism would have to be able to "coexist" with it as well.

In fact, you can see the DRAM/CRTC/GA as a frame buffer, and the Z80 as a guest 
And since the Z80 can be inhibited by setting the BUSREQ signal, nothing prevent an other guest to access the DRAM as long as it follow the GA rules.

[Velktron]

Well, on the CPC motherboard, there's the problem of physically giving another "guest" access to the memory lines -I think the mobo is designed only for PIO, so everything has to go through the Z80. It would require a lot of soldering and of course, as you said, playing by the GA's rules.  Every new DMA-capable device would have its own address generator, iterator etc. but it would have to stall if the GA said so -plus there would need to be some sort of controller which would resolve conflicts between the GA and multiple DMA devices, allowing/disallowing specific ones, and of course letting the GA trump them all, as the big bad booty daddy that it is. 


In addition, the Z80 would now be disabled not only by the GA, but by every other DMA device present (by ORING the GA and the controller's "busy" signal). Talk about establishing a pecking order 


GA >> DMA devices >> Z80

[TotO]

Should be better to use the EXP RAM instead of the VGA RAM for DMA usages.

[MaV]

This is becoming a very theoretical and academic conversation.

The whole thing falls apart as soon as you can't guarantee that the Z80 is handed over control at exactly the right moment for the interrupt routines. Otherwise for example there's no reading the keyboard or handling sound, at least not in a predictable way (and wouldn't you hate typing something and not knowing if the system recognises your input?) The system would be pretty useless then.
And every piece of Z80 code that needs exact timing (i.e. FDC controller- or tape-routines) means the DMA devices need to hand over control to the Z80 beforehand (!) and for the duration of the routine as well.

It's either that or replace the above problems by daisy-chaining the keyboard, sound, disc, tape, etc. peripherals into the same DMA system.

Good bye compatibility and good bye CPC (because you're talking about a completely new Z80 system by then).

[Velktron]

Yup, the presence of such a disruptive DMA controlling scheme would throw software compatibility out of the window. It might work for code specifically written to account for its presence, possibly only by running under a controlled OS/monitor environment with a well-defined API and driver architecture, but as you said, then that wouldn't be a CPC anymore.

E.g. you couldn't just replace the basic PIO-based disk I/O access routines with a ROM containing  a set of "smarter" DMA-based ones, and soldering a new DMA-expansion board somewhere between the Z80, the FDC controller and main RAM, and expect it to have no side-effects whatsoever to timing-sensitive software (bye-bye, games) or even quite "vanilla" software if it decides to hog the memory bus at the wrong time.


It'd be quite risky to run software NOT designed to account for its presence, so such a peripheral (if you could still call it that) would need to be disabled easily at will so the user can switch between "normal CPC" and "super-duper enhanced CPC with DMA"

[TFM]

Can you elaborate ?

In brief, a DMA at the expansion port can't access DRAM on the PCB properly (ignoring timing of GA access here).

A solution could be a DMA for an external memory expansion, ok that's easy. But it wont help to access V-RAM quick.

Would love to see that you prove me wrong! 

[Velktron]

A solution could be a DMA for an external memory expansion, ok that's easy. But it wont help to access V-RAM quick.

Easy...and pointless. An external device can be as complex and free of the CPC's vagaries as we please, so for all we know it could be running a modified ZX Spectrum with an Amiga DMA controller, 5 SCSI hard disks, 10 external DD-I drives and 20 SID chips strapped on it, and of course it could move data from and to its own RAM as fast and with any access method it pleased, as long as it had a CPC expansion interface somewhere, of course. 


But the PC would still have to interface using good old PIO with it, so what's the point? 

[TFM]

Easy...and pointless. An external device can be as complex and free of the CPC's vagaries as we please, so for all we know it could be running a modified ZX Spectrum with an Amiga DMA controller, 5 SCSI hard disks, 10 external DD-I drives and 20 SID chips strapped on it, and of course it could move data from and to its own RAM as fast and with any access method it pleased, as long as it had a CPC expansion interface somewhere, of course. 


But the PC would still have to interface using good old PIO with it, so what's the point? 

That's exactly what I pointed out in my posts. Just less rude, but more efficient! 

[andycadley]

Should be better to use the EXP RAM instead of the VGA RAM for DMA usages.

Yes, except most of the obviously good uses of DMA, like a blitter, aren't much use without access to video memory.

[TFM]

Well, one idea could be to fill the E-RAM quick from mass media... ok, ok, I know what are you going to say now... it won't be a CPC any longer.... (as if we would built something like this anyway).

[Bruce Abbott]

it won't be a CPC any longer....

Yes. That's why it's pointless worrying about compatibility with games etc.

However we can imagine an expansion device for the CPC that uses DMA for some specialized task such as playing video, enhanced graphics or a second processor. It would be designed to maintain compatibility with existing CPC software and hardware when not active, and only use 'incompatible' DMA for its own purposes. Such a device was never produced during the CPC's  commercial lifetime, but that doesn't mean it couldn't have been.

Rather than trying to do it through the expansion port I would look at removing the Z80 and putting it on a daughter board. Then the DMA device (which might even be a different CPU) can get complete control of the bus. Even with the limitations imposed by the gate array, the DMA device could access memory much faster than the Z80. For example LDIR takes 6 cycles to copy one byte, when it could be done in 2 cycles by a device with no instruction overhead.

A daughter board could also have peripherals attached such as an SD card, which can be accessed without using any CPC RAM cycles. Data could be written from it to CPC RAM at a blistering 1 Megabyte per second. Imagine movies or 3D games running at 50fps with time to spare!