Sonnet 7200 PCI programming

[matthey]

Originally Posted by Hedeon

*) Maybe MPC8379E is the better choice. E600 core, 800 MHz. Not sure about the pci bridge however.

I believe this is still a 603 (G2) based CPU but with upgraded 32kB I/D caches.

603 = 8kB I/D
603e = 16kB I/D

Low end PPC like the 603(e) for the masses may be what killed the PPC. The 68060 performed nicely with 8kB I/D caches but a PPC?

The MPC85xx would get you an e500 core which is more like a G3. It is an efficient shallow pipeline PPC design from the '90s with large enough caches to let PPC breath (32kB I/D). The A-EON A1222 Tabor is based on the e500.

Edit: Ok, I see you figured out your mistake already.

[Hedeon]

The e500 core lacks a full FPU (the e500v1 and the e500v2). I wonder if the e500mc was so much more expensive that they didn't use that on the Tabor.
I rather like the e600 core (also has SIMD) but all the embedded ones have a PCI-e bridge instead of PCI as far as I can see (MPC8641 for example). I'm not sure if that is backwards compatible.

[strim]

As far as I have researched the currently available options... none of the integrated PCI bridges in newer CPUs feature slave mode, where you can control the CPU from PCI side of the bridge (like we currently do via MPC107). It is possible to build such logic with a custom FPGA-based solution. But that's not as easy as just dropping existing bridge onto PCB and programming it...

Also, no PCI Express is in no way compatible with PCI on a hardware level.

If we wanted to use MPC8641, then probably the most reasonable solution would be building custom local bus<->PCI bridge, plus some logic to control the CPU itself. If we wanted to do something this incredibly complicated, then we could as well design a whole new A3000/A4000 CPU card instead.

[grelbfarlk]

The TSI-107 and various versions are EOL, it is true but there are stocks out there. And consider how many people out there have a Mediator or Grex that want to drop $400 on an accelerator for them, a few hundred at best?
I think there should be enough "old" PPCs out there like the G4 series to satisfy what demand there is, assuming one could design a board that could take a fair range of different CPUs. If you want to go with the G3-750 compatible you have the PPC745 and PPC755 through the G4 7410, plenty of available stock there. But if you want to go for something higher like the a 7445, then you are locked into a different set of CPUs to be compatible with, above changing a couple resistors (or jumpers). Also that may mean designing for TSI-108 or another bridge.

[matthey]

Originally Posted by Hedeon

The e500 core lacks a full FPU (the e500v1 and the e500v2). I wonder if the e500mc was so much more expensive that they didn't use that on the Tabor.

I have heard that A-EON started the motherboard design before the e500mc was announced.

Originally Posted by Hedeon

I rather like the e600 core (also has SIMD) but all the embedded ones have a PCI-e bridge instead of PCI as far as I can see (MPC8641 for example). I'm not sure if that is backwards compatible.

The '90s PPC SIMD (and FPU) is an expendable feature while it seems like those fancy ARM processors get upgrades to the SIMD every 6 months. Maybe the ARMv8 loving Qualcomm will know what to do with the similar architecture PPC .

IDT still manufactures other PCI bridges like TSI108. Not studied it yet. So far I saw all PCI bridges adapt the PCI Agent mode + Config through PCI + Core controll by PCI. This TSI108 + 1GHz MPC7447.... Minimum 8 layer PCB, rather more. I think that then the PCI bus will be the bottleneck. And the performance would be scary high for Amiga environment.

[Hedeon]

I actually have had a 667 MHz 440GX card in my A3000. Had to modify the pci.library (again) but it is now recognized. The documentation is not very clear how to bootstrap the PPC or how to setup the memory on the card (SODIMM). So I gave up seeing it lacks an FPU. Can spend my time better.

[kev]

Originally Posted by zomos

IDT still manufactures other PCI bridges like TSI108. Not studied it yet. So far I saw all PCI bridges adapt the PCI Agent mode + Config through PCI + Core controll by PCI. This TSI108 + 1GHz MPC7447.... Minimum 8 layer PCB, rather more. I think that then the PCI bus will be the bottleneck. And the performance would be scary high for Amiga environment.

http://www.e2v-us.com is the latest as moto/tundra/atmel/e2v made the 106-110 pci bridge chips

http://www.e2v-us.com/products/semic...ors/interface/

Yep found that they still got the MPC107 and MPC755.

[kev]

if u want to make a 1 chip 603e pci card this be best bet as it has the bridge built in http://www.e2v-us.com/resources/acco...datasheet/2737

I got another question. How stable is Crescendo at A1200 Mediator ? What accelerator card is best ? Have anyone tested it in such configuration ?

[grelbfarlk]

The Crescendo doesn't really work on the A1200 mediator. Hedeon worked on this a while and didn't get it stable enough to say "it works". Currently only A3000, A4000, A4000T work and only with a Mediator, so no Prometheus or Grex, right now.

[Hedeon]

Just a heads up that I finally implemented some code which will patch the executables on the fly and transiently (meaning it does not alter the executables). It will mean that all the current patched stuff will be useless in the future. Also the name will change to powerpc.library instead of sonnet.library.

This will lead to WarpOS files working with the library without any alterations (renaming, patching etc).

It is currently in testing and needs some bug fixing.

Anyone knows if NewLoadSeg() is used anywhere btw (I mean in programs or the OS)?

[starship]

Originally Posted by Hedeon

Just a heads up that I finally implemented some code which will patch the executables on the fly and transiently (meaning it does not alter the executables). It will mean that all the current patched stuff will be useless in the future. Also the name will change to powerpc.library instead of sonnet.library.

This will lead to WarpOS files working with the library without any alterations (renaming, patching etc).

It is currently in testing and needs some bug fixing.

Anyone knows if NewLoadSeg() is used anywhere btw (I mean in programs or the OS)?

I found this;
NAME
NewLoadSeg -- Improved version of LoadSeg for stacksizes (V36)


SYNOPSIS
seglist = NewLoadSeg(file, tags)
D0 D1 D2


BPTR NewLoadSeg(STRPTR, struct TagItem *)


seglist = NewLoadSegTagList(file, tags)
D0 D1 D2


BPTR NewLoadSegTagList(STRPTR, struct TagItem *)


seglist = NewLoadSegTags(file, ...)


BPTR NewLoadSegTags(STRPTR, ...)


FUNCTION
Does a LoadSeg on a file, and takes additional actions based on the
tags supplied.


Clears unused portions of Code and Data hunks (as well as BSS hunks).
(This also applies to InternalLoadSeg() and LoadSeg()).


NOTE to overlay users: NewLoadSeg() does NOT return seglist in
both D0 and D1, as LoadSeg does. The current ovs.asm uses LoadSeg(),
and assumes returns are in D1. We will support this for LoadSeg()
ONLY.


INPUTS
file - Filename of file to load
tags - pointer to tagitem array


RESULT
seglist - Seglist loaded, or NULL

BPTR NewLoadSeg(CONST_STRPTR file, const struct TagItem *tags) (D1,D2)

http://www.alb42.de/fpc-docu/amigados/newloadseg.html
http://amigadev.elowar.com/read/ADCD.../node0194.html
https://github.com/robn/cake/blob/ma...wloadsegtags.c
https://trac.aros.org/trac/export/28...s/newloadseg.c
https://www.codewiz.org/projects/ami...eSrc/PlayWin.c

[grelbfarlk]

The powerpc.library replacement is getting very far along, most stuff working now, stay tuned.

[jack-3d]

Great news! Cant wait to test it ;o) Good job guys!

[S0ulA55a551n]

Careful you about to release the shark that elbox never could


Sent from my iPhone using Tapatalk

[jack-3d]

Why? This increases Mediator 3V sales a bit.

[grelbfarlk]

Nearly everything on the list that was working on the Sonnet.library is working on the Sonnet powered powerpc.library, but it's a long list to work through. A couple bugs remain which seem to be more limited to my system. I haven't tested everything yet.

New unimportant stuff that wasn't working before work now, like Sid4Amiga, some Hollywood apps (which are buggy and slow), Vmac Mac Plus Emulator (slow, old and only supports monochrome).

[Hedeon]

Pushed the changes to Git. You can still choose between sonnet.library or powerpc.library if you want. Also uploaded to Aminet.