Sonnet 7200 PCI programming

[Hedeon]

Originally Posted by Hedeon

Crosspost. The Bigfoot PCIe is still not behaving as it should.]

I am now able to address the 128MB on the Bigfoot PCIe card.

Only one hurdle left to make it work. Assuming it will not behave like the K1/M1 PCI which does not work with Warp3D without a bridge in between.

[jbenam]

Originally Posted by Hedeon

I am now able to address the 128MB on the Bigfoot PCIe card.

Only one hurdle left to make it work. Assuming it will not behave like the K1/M1 PCI which does not work with Warp3D without a bridge in between.

Great progress. How does the Bigfoot PCIe compare to the other cards in terms of performance?

[Hedeon]

I would guess same as M1 with double the memory.

400MHz CPU, 266 MHz 128 MB DDR.

But it isn't running yet so i can't test.

[grelbfarlk]

But the good news is if it does work then we can look at other PCIe cards with ludicrous performance.

[cloverskull]

Hey friends, as I’ve recently gotten a PiStorm32-lite for my towerized A1200, I again find myself wondering if there will ever be an A1200 compatibility/implementation here Hope it’s cool to bring up some old stuff.

First off - happy to contribute to a bounty.

According to this issue: https://github.com/Sakura-IT/SonnetAmiga/issues/21 DvdBoon says

The assembly version of the library will not support the 1200TX mediator beyond what is implemented now. Refer to https://github.com/Sakura-IT/PowerPCAmiga for possible support in the new C version in the future.

I looked at the linked PowerPCAmiga project and it hasn’t been updated in three years. Additionally there’s an open issue for A1200/ZII compatibility https://github.com/Sakura-IT/PowerPCAmiga/issues/7 and DvdBoon says here

This feature will likely not be implemented due to technical difficulties.

So…is it possible? A1200s seem to be rather plentiful, and tower cases from Elbox are still available new. Is there a chance or is it not possible?

[trixster]

Suggest you search this thread, there’s lots of information as to why it didn’t work with the 1200TX

[cloverskull]

Yeah, I was reading through much of that. The gh issue I linked for the SonnetAmiga project also goes pretty deep on the technicals. However I was asking if it’s at all possible One can hope…

[Hedeon]

Only if there comes an A1200 accelerator with 32bit addressing to Zorro III (or better PCI(e)). I know Elbox made some busboards that do this (Z3i?), but I think it only gets activated if you populate the CPU slot (an A3000/A4000 one, not A1200) so still expensive for now.

It could also theoretically work with a G-Rex 1200 with 2 DMA slots and connected to a Blizzard mini-PCI slot, as the Blizzard can access the full PCI memory space. But these are like unobtanium.

[Hedeon]

Originally Posted by cloverskull

I looked at the linked PowerPCAmiga project and it hasn’t been updated in three years.

17.15 is in the works to support Bigfoot Killer 2100 for Prometheus. In combination with an update to the prometheus.library.

And if pci.library 14.0 ever comes out with support for config1 cycles, I'll implement it there too.

[jbenam]

Originally Posted by Hedeon

I would guess same as M1 with double the memory.

400MHz CPU, 266 MHz 128 MB DDR.

But it isn't running yet so i can't test.

Oh, so it has the same exact CPU as the M1 onboard? I hoped that it had a slightly newer architecture since it came out a bit later. Oh well, can’t complain about having double the RAM

[cloverskull]

Originally Posted by Hedeon

Only if there comes an A1200 accelerator with 32bit addressing to Zorro III (or better PCI(e)). I know Elbox made some busboards that do this (Z3i?), but I think it only gets activated if you populate the CPU slot (an A3000/A4000 one, not A1200) so still expensive for now.

It could also theoretically work with a G-Rex 1200 with 2 DMA slots and connected to a Blizzard mini-PCI slot, as the Blizzard can access the full PCI memory space. But these are like unobtanium.

Thanks for the explanation, and please excuse my ignorance, but is this something that can be accomplished with pistorm and emu68? To clarify, if emu68 can't do what's needed /now/ but can be added?

[Hedeon]

Originally Posted by cloverskull

Thanks for the explanation, and please excuse my ignorance, but is this something that can be accomplished with pistorm and emu68? To clarify, if emu68 can't do what's needed /now/ but can be added?

Adding PCI slots to the PiStorm

[Hedeon]

Originally Posted by jbenam

Oh, so it has the same exact CPU as the M1 onboard? I hoped that it had a slightly newer architecture since it came out a bit later. Oh well, can’t complain about having double the RAM

MPC8314E for the PCIe card vs MPC8343EA for the PCI card.

So not the exact same.

[cloverskull]

Ah gotcha. Hmm. Sounds tough.

Then again maybe not outside the realm of possibility. My pistorm uses a pi 4b. This is interesting: https://hackaday.com/2020/07/01/addi...he-easier-way/

Would /all/ pci devices need to live on this or would the mediator be fine for all other cards, and the ppc card directly connected to the pci bus on the pistorm?

[Hedeon]

Originally Posted by cloverskull

Ah gotcha. Hmm. Sounds tough.

Then again maybe not outside the realm of possibility. My pistorm uses a pi 4b. This is interesting: https://hackaday.com/2020/07/01/addi...he-easier-way/

Would /all/ pci devices need to live on this or would the mediator be fine for all other cards, and the ppc card directly connected to the pci bus on the pistorm?

All busses need to be connected otherwise it won't work.

[trixster]

Originally Posted by jbenam

Oh, so it has the same exact CPU as the M1 onboard? I hoped that it had a slightly newer architecture since it came out a bit later. Oh well, can’t complain about having double the RAM

Be aware that the 2100 when it’s connected to the PCIe adapter won’t fit in a 3000D or 4000D case.

[grelbfarlk]

I'm going to blow your mind.
If you are unsure if you can deal with seeing something that might question your beliefs or if you have heart problems, you might want to get your phone ready and dial 1-1 and if you feel faint press 1 again.


























Last chance...

[trixster]

You’ve seen my case, good luck trying to shoehorn a riser like that into it, with all the other shite that’s in there I’m actually using one of these at the moment to test 2100 whilst the aa3000+ is in pieces, it might explain why the connection is very temperamental..

[jbenam]

Originally Posted by trixster

Be aware that the 2100 when it’s connected to the PCIe adapter won’t fit in a 3000D or 4000D case.

I have got an open-ended PCIe x1 extender just for this reason - that way I can just shove it out of the way. I just have doubts on the reliability of the extender, but that will be a problem for another day

[Hedeon]

Don't go buying stuff just yet. There seems to be something not right. Transfers from the PCIe bridge are dropped by the MPC8314E CSB bridge, but only certain ones and it is 100% reproducable.

If I write a byte to Amiga $48000000 it get's translated by the Prometheus to $08000000 PCI which is transferred by the PCIe bridge to the bigfoot card and there it should be transferred to local memory $0 (Amiga address may vary on how the Prometheus auto-configured plus cards in the Prometheus).

Now if I write anything but $31, $32, $34, $37, $38, $3B, $3D, $3E (I tested from $00-$C0 before I stopped), the byte shows up in the memory of the bigfoot card. Any one the ones above the CSB gives an inbound bus error (on the MPC8314E) and the original data in that memory location persists.

With word writes and longword writes it becomes less clear to predict which data gives an error, but at least $52535445 does (RSTE) which is why I cannot get the device into a reset (this sequence written to a register is needed to prepare for a reset).

I also wrote a small PPC program to the memory and of the 4 opcodes, one of them gets dropped ($38630001).

There are probably more combinations. But it is 100% reproducable with the same data. Does not matter to which address is written, only the data written triggers it.


I got it confirmed by someone else on a different machine with different Prometheus and BigFoot.