Sonnet 7200 PCI programming

[spudje]

Thanks again for new pointers I can look into.

My CSPPC is a ClassicWB setup. My Sonnet isn't, that's "vanilla 3.9 BB1&2". (BTW all on same machine, simply different CF card )

I do have IDEfix97 installed on both setups, would that be interfering? I need it to get my CDROM working on 2nd IDE behind IDE splitter, it's a slimline IDE device that refuses to work on the primary IDE even when jumpering it to slave.

Versions, identical on both setups:
Voodoo.card file is v4.35
Warp3D files are v4.2
Picasso96API.library is v2.310

I don't remember exactly, but at least on the Sonnet setup I think I selected 060 when installing AHI.
BTW with memory fault I meant memory on the Voodoo not Sonnet, hence also my question to lower it's memory.

Added mediator reports from both setups.

[trixster]

I've got a spare voodoo 3 I can send you for testing if it helps?

Pm me your address if you want it.

[spudje]

Oh if you can miss it and you're sure it runs fine, that would be awesome. Borrowing or buying either way would be ok. Let's discuss tomorrow, off to bed now.

BTW added medication screen about Voodoo in CSPPC setup. Why it reports a Voodoo 3 2000 I don't understand, because I'm pretty sure it's a Voodoo 3 3000.

[grelbfarlk]

Originally Posted by spudje

Oh if you can miss it and you're sure it runs fine, that would be awesome. Borrowing or buying either way would be ok. Let's discuss tomorrow, off to bed now.

BTW added medication screen about Voodoo in CSPPC setup. Why it reports a Voodoo 3 2000 I don't understand, because I'm pretty sure it's a Voodoo 3 3000.

Don't know about medication, I don't use it, but if you see it has your memory clock at 166MHz, that means it's a V3-3000, the V3-2000 it would be at 143MHz. Vblank I think is VoodooInt which should be on. GrantDirectAccess should be on.

[Wrangler]

Spudje: I've compared your MedReport (for Sonnet) against mine and nothing obvious leaps out to me but here are a few differences:

• You're running Picasso96API.library ver 2.310 whereas I have ver 2.300 - I can't imagine that's a problem
• You're on ahi.device 4.180 whereas I have 6.7 - could be an issue?
• You're on 68060.library ver 46.15 whereas I have ver 46.3 - probably OK as yours is later?
• You have warp.library size -1 whereas I have 21236. Not sure whether I should though since WarpOS is bypassed!
• Your startup-sequence doesn't include InitPPC so I guess you don't have powerpc.library in libs: but start the Sonnet manually?
• You also have something that adds 256 meg on the Zorro bus eg BigRAMPlus or something? Have you tried without that? I can't imagine you need it on top of the Sonnet + 68060 mem!

Final thought: have you tried the 68060 version of Cowcat's Quake? That would test Warp3D and the Voodoo (inc its mem) but not require any of the PPC stuff. Might help to diagnose which of those things is giving you bother.

[Hedeon]

• You're running Picasso96API.library ver 2.310 whereas I have ver 2.300 - I can't imagine that's a problem. I have 2.300.
• You're on ahi.device 4.180 whereas I have 6.7 - could be an issue? Shouldn't be. 4.18 should give more performance over 6.7, however.
• You're on 68060.library ver 46.15 whereas I have ver 46.3 - probably OK as yours is later? Funny, I checked and currently I am using 46.2.
• You have warp.library size -1 whereas I have 21236. Not sure whether I should though since WarpOS is bypassed! -1 is correct. There should be no warp.library in libs:.
• Your startup-sequence doesn't include InitPPC so I guess you don't have powerpc.library in libs: but start the Sonnet manually?
• You also have something that adds 256 meg on the Zorro bus eg BigRAMPlus or something? Have you tried without that? I can't imagine you need it on top of the Sonnet + 68060 mem! Looking at the pciinfo output all memory fits correctly (600MB in total...nice). But for debugging reasons you could remove the BigRAMPlus to rule it out giving problems (or maybe you did already).

Edit:

I forgot: 46.15 is a phase5 lib i think. I used to develop using 46.7.

[spudje]

Thanks guys for the analysis!

No time for testing today unfortunately.

Where can I find AHI 6.7. I only found 6.0 on Amnet, and different onse for 68k and PPC which was a bit confusing:
http://aminet.net/package/driver/aud...amigaos-ahiusr
http://aminet.net/package/driver/aud...amigaos-ahiusr

060: Yes I go the p5 from here http://phase5.a1k.org/index.html
There's also a 46.3 and 46.7 there. Where can I find 46.15 and 46.20?

Yes I start the Sonnet manually. I also tried without BigRAMplus, same behaviour
I also have a Highway USB, no Poseidon installed in Sonnet setup, Poseidon is installed in CSPPC setup.

I'm probably gonna build a new setup again anyway.

[Wrangler]

I'd still suggest running Quake on '060 before changing drivers and especially a rebuild as it'll be quicker to see if it works or not.

[grelbfarlk]

I had crashes with ahi 6.7. 5.16 and 4.18 seem to work fine though.

[Hedeon]

This weekend I'm planning to transfer the Sonnet and PPC stuff to my A4000T with CSPPC. If you want me to test something with that config, let me know.

[spudje]

OK, I installed Clickboom's quake and added Blitzquake http://aminet.net/package/game/shoot/BlitzQuake68k_150, all in my Sonnet setup. I did not run initppc since the games uses the 060 correct?

This runs fine, apart from some stuttering. It does however look more like software rendering than Warp3D. Pretty fast though. So maybe I need to crank up some 3D setting arguments in the blitzquake launch script? Any help very much appreciated.

I also have no clue whether it uses default audio or AHI. SO any help enable AHI would also be cool.

Removing the -noaudiocd from the install script does not result in the cd audio being played. Do I need to start that in the game somehow? The cd audio works fine when I play it in Workbench.

[trixster]

Ok, I'm ready to admit defeat. After a week or more of testing loads of different versions of the library i cannot get the sonnet to remain completely stable without using setcmemdiv 4 or 5.

I thought i'd had a breakthrough with 284 when i tested removing setcmemdiv from ENV: entirely. This resulted in the sonnet working fine for quite a few long sessions and with longer periods of stability than with setcmemdiv at 0 or 3. However inevitably a crash would happen. Crashes in games manifest as sometimes a hard lock, sometimes a crash to workbench with either a Program exception or an Instruction Storage execption.

So this week i've spent houres trying loads of different library versions including the aminet one. Sometimes the machine will appear fine for many hours, sometimes a crash will occur after a few minutes, sometimes after a slightly longer period. Frustratingly, I found that eventually even the older builds like the aminet one would occasionally crash in the same way so I believe the issues im seeing arent necessarily introduced by the more recent builds.

What I have discovered is that the only way to get a totally stable set up is to run #284 with either 5 or 4. So this means my sonnet is not stable with the cache at the default 200mhz. How annoying! Surely not a second sonnet with duff hardware?!?!

[trixster]

Originally Posted by spudje

Removing the -noaudiocd from the install script does not result in the cd audio being played. Do I need to start that in the game somehow? The cd audio works fine when I play it in Workbench.

I don't know about the clickboom version, but i found that cowcat's blitzquake version needed some env: variables set. I had to make ENVARC:quake1/cd_device and ENVARC:quake1/cd_unit with atapi.device and 3 respectively (your variables may differ). Perhaps Clickboom quake is similar? Perhaps run snoopdos when starting the game to see if it's searching for a setting.

[Wrangler]

Originally Posted by trixster

... However inevitably a crash would happen... Sometimes the machine will appear fine for many hours, sometimes a crash will occur after a few minutes, sometimes after a slightly longer period...

How well cooled is your system trixster? When the system crashes, how hot is the Sonnet heatsink and the Voodoo heatsink (earth yourself first)? Can you touch them or are they too hot?

Another possibility is the power supply misbehaving. Have you got an alternate you could try?

It may not be the Sonnet that is the root cause here...

[Wrangler]

@spudje: checkout Cowcat's Quake as trixster says. The thread is on this forum: http://eab.abime.net/showthread.php?t=72310

IIRC there are typical framerates that others have got in the thread and if you're getting anywhere near those, Warp3D will be working.

[trixster]

It could be heat. However the lid of the A4000 is off and there's a fan blowing cool air over the voodoo and the sonnet.

The voodoo is not the issue I think as it's completely stable running 060 Quake, amiquake or amiquake2. The voodoo gets hot but I believe they all do! The sonnet and the voodoo have holes in the pci back plane to allow hot air to pass out the back of the case but as I said the lid is permanently off so I don't see airflow being an issue.

The other thing is the length of time the machine has been on does not appear to affect the time before a crash happens. So I might boot from cold and see a crash within 5 minutes. I'll then turn off, wait 10seconds, reboot and the machine will play quake2 for a 2 hour session. Then a crash.

Broadly speaking, using #284, if I set setcmemdiv with 0 or 3 I'll see a relatively quick crash. If I remove setcmemdiv from ENV: then I'll maybe get an hour or two. If I try the aminet library I can get several hours with no issues before a lock up. If I set #284 with 4 or 5 then it's totally stable.

So the time taken to see a crash is pretty variable. Making a new test is frustrating because sometimes the machine won't crash for ages and I'll think I've cracked it only for me to try again the next day and I'll see a lockup randomly within a few minutes.

So if it's heat then it's bloody weird! 1mb sonnets are designed to run the cache at 200mhz but it seems mine can't quite sustain it.

[Hedeon]

No setcmendiv and setting it to 0 is the same.

Older libraries had the divider standard on 3, so the 400/1mb versions ran at 133 MHz, but the Aminet release already has 200MHz.

Also instructions were not allowed to be cached in L2 in older versions. Also not applicable here.

But like I said before, don't test the options, test the versions. Soon, the version on Aminet will disappear.

Q2 will always crash due to a memory leak while switching levels. The Cowcat version at least. You'd better test with GlQuake from Cowcat.

[trixster]

Good info, thanks Hedeon!

[grelbfarlk]

Originally Posted by trixster

It could be heat. However the lid of the A4000 is off and there's a fan blowing cool air over the voodoo and the sonnet.

The voodoo is not the issue I think as it's completely stable running 060 Quake, amiquake or amiquake2. The voodoo gets hot but I believe they all do! The sonnet and the voodoo have holes in the pci back plane to allow hot air to pass out the back of the case but as I said the lid is permanently off so I don't see airflow being an issue.

The other thing is the length of time the machine has been on does not appear to affect the time before a crash happens. So I might boot from cold and see a crash within 5 minutes. I'll then turn off, wait 10seconds, reboot and the machine will play quake2 for a 2 hour session. Then a crash.

Broadly speaking, using #284, if I set setcmemdiv with 0 or 3 I'll see a relatively quick crash. If I remove setcmemdiv from ENV: then I'll maybe get an hour or two. If I try the aminet library I can get several hours with no issues before a lock up. If I set #284 with 4 or 5 then it's totally stable.

So the time taken to see a crash is pretty variable. Making a new test is frustrating because sometimes the machine won't crash for ages and I'll think I've cracked it only for me to try again the next day and I'll see a lockup randomly within a few minutes.

So if it's heat then it's bloody weird! 1mb sonnets are designed to run the cache at 200mhz but it seems mine can't quite sustain it.

Well maybe those chips have been replaced at some point with lower speed ones, or the factory screwed up. Take a high-res picture of the cache chip or get the part number off of it to confirm.

[trixster]

Cool, I'll have a look. Regardless, I think I'll just stick to setcmemdiv 4 as that works ok for me