Sonnet 7200 PCI programming

[Hedeon]

I am not seeing an increased amount of context switching, which leaves me to think that the 68K part (i.e. rtg.library) has become more CPU (68k) intensive which causes more time to elapse before returning to the PPC.

In other news, how much were you involved in OG W3D (also for OS4)? :-)

[Karlos]

Originally Posted by Hedeon

I am not seeing an increased amount of context switching, which leaves me to think that the 68K part (i.e. rtg.library) has become more CPU (68k) intensive which causes more time to elapse before returning to the PPC.

In other news, how much were you involved in OG W3D (also for OS4)? :-)

I fixed a bunch of bugs in the OG Permedia2 driver and improved performance but it never saw the light of day. However, some design choices (especially with respect to vertex arrays) did find their way into the general approach used for other drivers since (Voodoo, Radeon R100/200).

I got involved again with OS4 and got the Permedia2 driver for that (based on the unreleased OG improvements) working before spending time on R100/R200 drivers (as well as some tweaks in the corresponding chip drivers for P96). I also added some new features like having per-application settings due to the dependency on older software needing certain bugs and stuff.

All the RTG card swapping was the end of my A1XE

[BULI]

Originally Posted by trixster

These were my findings

https://eab.abime.net/showpost.php?p...&postcount=156

That why I still use old, tested software (Workbench 3.9 cool)

[Karlos]

If we entertain the idea, for the moment, that W3D pre-OS4 is essentially abandoned, to my mind it makes sense to provide an API compatible single library replacement.

What I mean by that is, for each Chip/RTG pairing there's a full Warp3D.library (and PPC counterpart). You simply install the pair that's relevant to your machine configuration.

The benefit of this is the removal of layers of indirection and better tailoring for the specific chip. For example, on the Permedia2, the allocation strategy described above is a bit of a bad fit (ISTR fixing that for the OS4 driver) because bitmaps can't be any arbitrary multiple of some common value wide. It has a more non-linear behaviour than that because widths are described by thee partial product values.

Managing all the potential duplication of effort in such a build strategy comes down to the source code organisation. You can use conditionally defined functions for the RTG aspect, to allocate memory and lock hardware, rather than a whole separate library invocation.

Just a thought.

[Hedeon]

Well, I don't know the legal status of W3D or any of the old gfx card drivers, specifically for R200, Voodoo and P2 which can do 3D.

What I do know is that they don't care about bug fixing anymore as I have seen that the 2D Radeon driver (OS4) has a wrong start-up leading to sometimes giving a scrambled output (on classic OS4 and.....classic). I put the linux source next to it and the error is obvious.

And wrong color setting in the voodoo W3D driver (OS4). And many more I am now forgetting as it has been years, but it was send to the betatest list. Or dev...I forgot that too.

Maybe they fixed the Radeon driver since it was transfered to AEon. I dunno.

[Karlos]

Well there's nothing illegal about making a new implementation that just supports the API. You can avoid a lot of pitfalls that way, too.

[Hedeon]

I have had insight in the source, it will always be a discussion.

[Hedeon]

I see that the Israeli seller caught up that his PrPMC cards are used for Amigas. Still 180 left ( take note, Lukzer)

[trixster]

Indeed. I wonder if this is the same guy we got those $50 ramix and apocs from a few years back.

[Hedeon]

I got them also from Israel including desert sand it seemed. Would be strange if these are different people both coming from Israel. The Apocs are now doubled in price at 120. He had over 200 and only a few sold. Now 50 sold i see.

[grelbfarlk]

Just to refresh my memory or whether there was any new information since then. The issue with using a PCIe card like the Killer 2100 on the Prometheus was that it didn't present its RAM to the system? Were all PCIe cards the same?

[Hedeon]

I am not sure I have a correctly working Killer 2100. I can see in the configuration block that things seem off. Linked registers don't change together when I change one of them. They do with the PCI M1.

You have one laying about that we could use for some testing? You also need a Firestorn and a PCI-PCIe bridge.

[Hedeon]

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

https://eab.abime.net/showpost.php?p...1&postcount=28

[Hedeon]

Originally Posted by grelbfarlk

There's some possibility* that the Prometheus could support PCIe cards by way of a PCI adapter.

*This possibility is likely to never come to fruition.

Another step closer

[grelbfarlk]

Interesting, I was actually looking at different PCIe cards altogether, but not those expensive QorIQ T-series, like much cheaper ones.

[cloverskull]

Hey friends, question regarding A1200 support. I know it is specifically unsupported but with the advent of pistorm, is there some way the linux side of things could help with compatibility?

I'm showing my ignorance here as I have no idea how the Sonnet Amiga project works :P

[jbenam]

Originally Posted by Hedeon

You have one laying about that we could use for some testing? You also need a Firestorn and a PCI-PCIe bridge.

I think I have everything needed - I have a PCIe Killer NIC, a StarTech PCIe-to-PCI and a Firestorm (that still needs assembling… this might give me motivation enough).

Which kind of PCIe-to-PCI bridges have been tested until now?

[Hedeon]

Only the one you see in the screenshots. But bridges should be (mostly) generic.

[Hedeon]

Originally Posted by cloverskull

Hey friends, question regarding A1200 support. I know it is specifically unsupported but with the advent of pistorm, is there some way the linux side of things could help with compatibility?

Only if the PiStorm has access to the full PCI space without having to bank through the mediator 8MB window.

[Hedeon]

Originally Posted by grelbfarlk

Interesting, I was actually looking at different PCIe cards altogether, but not those expensive QorIQ T-series, like much cheaper ones.

Got an AMC card laying around. Came with a carrier card but it is massive with its own 4 pin power connector. Which i cannot connect. It gives me power failure errors on the board, sadly. It was also a 2x PCIe, so i had to connect a 1x to 16x and connect the 2x to that.

I tested it with 1x->16x->1x (the audio card,) and that worked, but not with the 2x carrier card. But could still be not enough power

So still looking for a carrier card.