English Amiga Board

English Amiga Board (http://eab.abime.net/index.php)
-   support.WinUAE (http://eab.abime.net/forumdisplay.php?f=5)
-   -   AOS 4.1 + WinUAE 3.3.0 b6 : Blizzard PPC + ppc_model=604e causes guru meditation (http://eab.abime.net/showthread.php?t=81566)

ccapublic 14 February 2016 22:31

AOS 4.1 + WinUAE 3.3.0 b6 : Blizzard PPC + ppc_model=604e causes guru meditation
 
1 Attachment(s)
Hi All,

I have a working AOS4.1U6 WinUAE configuration (3.3.0 beta 6). It emulates A4000 + Cyberstorm PPC.

To overcome the 128MB memory limit, I slighly modified the config to emulate a Blizzard PPC and boot hardfile from uaehf. As expected it runs fine but much slower. I did test the ppc_model=604e trick -> auto reboots once then guru meditation.

Did anyone make this trick work in order to have both the speed of cyberstorm emulation and the ram of Blizzard PPC ?

Thanks a lot for your help !

PS : Attached config & logs.

Locutus 14 February 2016 22:34

Seeing thats not a possible configuration with real hardware its not suprising that it crashes.

Keep in mind that a large part of why CSPPC emulation is faster is due to much lower CPU usage of SCSI I/O versus the IDE which has to be used with BlizzPPC.

I doubt that different CPU settings matter much with CPU speed for the actual QEMU JIT itself (i have never tried performance critical stuff with QEMU PPC softmmu but for other platforms ive never noticed a difference between cpu sub model).

ccapublic 14 February 2016 22:44

Quote:

Originally Posted by Locutus (Post 1070460)
Seeing thats not a possible configuration with real hardware its not suprising that it crashes.

I agree it's not a real config. That being said, it does not necessarily means crash + this is a trick I heard about here and there many times. So it must have run under "some" circumstances (even if not faster).

Quote:

Originally Posted by Locutus (Post 1070460)
Keep in mind that a large part of why CSPPC emulation is faster is due to much lower CPU usage of SCSI I/O versus the IDE which has to be used with BlizzPPC.

Since the introduction of OS4 extensions, uaehf can be used. It may still be slower than cyberstorm scsi and thus be the actual root cause for Blizzard config running slower, but I wanted to give a try to this "ppc_model" parameter trick just to see if its impact on speed is urban legend or true.

Thanks for your feedback !

Dunny 15 February 2016 00:53

Quote:

Originally Posted by ccapublic (Post 1070464)
I agree it's not a real config. That being said, it does not necessarily means crash + this is a trick I heard about here and there many times. So it must have run under "some" circumstances (even if not faster).

That doesn't matter. So long as the emulator itself doesn't crash, then it's emulating properly. If you try to emulate hardware that cannot possibly have worked in real life, then you're asking the emulator to do something that cannot be done.

The fact that it may or may not work for others is immaterial.

D.

ccapublic 15 February 2016 13:54

Quote:

Originally Posted by Dunny (Post 1070488)
If you try to emulate hardware that cannot possibly have worked in real life.

I do agree this hardware did not exist, but how do you know it would not have worked ? Thus how can you conclude the emulation "acts as the hardware would have acted" ?

That being said, I'm not saying there is an issue with emulation, nor asking the emulator to do this or this : I just asked if someone had it working as it is (was).

And I get answers from people who did not try but explain me it's normal... odd isn't it ? ;)

Thanks for your feedback though !

PS: I don't remember mediator having real full PCI dma, still AFAIK WinUAE emulates it, enabling sb128 support under os41 classic. So real hardware "exists or not" is not necessarily directly related to "works or not" under emulation.

Dunny 15 February 2016 14:56

I think that the answer here is that if the original hardware doesn't support it, you shouldn't be surprised when WinUAE doesn't - if Toni feels that it's worth doing, he'll do it - but it's probably going to be very low priority.

D.

Toni Wilen 15 February 2016 15:48

"ppc_model=604e_v2.4" works for me with BPPC config (with uaehf.device and uaegfx) but it was already confirmed that it won't fix BPPC speed issue. Reason is still unknown.

I think you should first use some more basic config with ppc_model and then add extra expansions.

ccapublic 15 February 2016 18:20

Quote:

Originally Posted by Dunny (Post 1070584)
I think that the answer here is that if the original hardware doesn't support it, you shouldn't be surprised when WinUAE doesn't

Please read me again :) I'm not surprised or anything like that. I was looking for feedback from people who tried, not hypothesis about whether or not it will/should/may work or not.

Quote:

Originally Posted by Dunny (Post 1070584)
if Toni feels that it's worth doing, he'll do it - but it's probably going to be very low priority.

Do what ? I did not ask for any modification.

Quote:

Originally Posted by Toni Wilen (Post 1070603)
"ppc_model=604e_v2.4" works for me with BPPC config (with uaehf.device and uaegfx) but it was already confirmed that it won't fix BPPC speed issue. Reason is still unknown.

I think you should first use some more basic config with ppc_model and then add extra expansions.

Thanks for your feedback. I will proceed this way. Though if it's proven to not speed things up the interest of doing it is pretty low.

I suppose there is nothing "a simple user" can do to help narrowing the reason for speed issue down ? (like trying different expansions one by one, combinations of emulated HW, parameters...)

Thanks for having taken time to read and advise !

ccapublic 15 February 2016 19:24

Just a quick report about my "testing"

- With the config I posted + ppc_model=604e_v2.4 instead of just ppc_model=604e => Same behaviour (reboot - so double pink screen - then guru)

- Started from a pretty empty config (A1200 Blizzard PPC / no PCI bridge / no emulated soundcard / no ethernet), just uaehf to boot my hardfile + Z3 UAE RTG : Same thing

- Tested various amounts of memory (RTG, Z3, Blizzard), enabling / disabling 68k JIT,... : Same

- From the same "pretty empty" config, Trying to boot from IDE => no crash but intuition message box claiming disk format is unknown

- From the same "pretty empty" config, Trying to boot from SCSI (auto) => no crash but boot screen (*AOS 4* floppy animation)

When it crashes, I don't know if it crashes before actually booting the hardfile or during boot process (Did not read winuae logs, but I'm pretty sure they would be pretty difficult to understand for me)

Just in case : not expecting any analyzis Toni, just sharing my (pretty useless) tests.

amyren 18 February 2016 08:07

Quote:

Originally Posted by ccapublic (Post 1070459)
Hi All,

I have a working AOS4.1U6 WinUAE configuration (3.3.0 beta 6). It emulates A4000 + Cyberstorm PPC.

To overcome the 128MB memory limit, I slighly modified the config to emulate a Blizzard PPC and boot hardfile from uaehf. As expected it runs fine but much slower. I did test the ppc_model=604e trick -> auto reboots once then guru meditation.

Did anyone make this trick work in order to have both the speed of cyberstorm emulation and the ram of Blizzard PPC ?

Thanks a lot for your help !

PS : Attached config & logs.

Your config works here, it boots up and load workbench and displays "AmigaOS 4.1 - Final Edition, 604e, 254MB FAST RAM, ???"
But the UAE zorro III rtg card doesnt seem to work, I can only use native resolutions. I tried selecting PicassoIV in the RTG settings, but then it fails to boot, getting HALT7.
In my cyberstormppc setup PicassoIV mode works fine, but any blizzardppc setup I have tried fail to work with rtg.

ccapublic 18 February 2016 08:54

Thanks for your feedback. Did you save the configuration before running it. Because I think I remember that saving it resets ppc_model to automatic and then it works.

Thanks for your testing ! (Weird because here it keeps crashing whatever I try).

amyren 18 February 2016 13:44

1 Attachment(s)
Quote:

Originally Posted by ccapublic (Post 1071404)
Thanks for your feedback. Did you save the configuration before running it. Because I think I remember that saving it resets ppc_model to automatic and then it works.

Thanks for your testing ! (Weird because here it keeps crashing whatever I try).

Got it working now, with proper screenmodes:)

I modified my own previous setup, adapted the mediator settings to match your settings. And edited the configfile to "ppc_model=604e"

Now it is much faster. Not as fast as cyberstormppc setup, but significantly faster than without the 604e setting.
I havent done any speed tests, but when booting up with 603e, it is so slow that the startup sound stops playing while the dock is loaded, and continues to play the tune afterwards. And it takes a while before the background image loads and the dock visualize itself at the bottom of the screen.
Then I go to workbench:prefs and open screenmode, it takes many seconds before the screenmode settings opens, and I can actually se parts of the windows getting drawn.
With 604e the startupsound plays without problems and the background image is loaded about the time the sound fades, shortly after this the dock apears. Opening the screenmode prefs takes only a few seconds now.

BTW, the 604e setting does not go back to automatic if I do changes and save the winue config.

I will attach my config file here with working A1200 604e

EDIT: btw. I'm using this config with the winuae 64 bit version

amyren 19 February 2016 14:53

Perhaps there are better test tools available for OS4, but here is what Ranger lists:

BlizzardPPC, ppc set to automatic: Ranger reports a 603e running at 95Mhz
BlizzardPPC with ppc set to 604e: Ranger reports a 604e running at 101Mhz
CyberstormPPC: Ranger reports a 604e running at 265Mhz

I dont know how much difference there is between 603e and 604e, but it certainly feels like the speed enchancement is more that just 95 to 101Mhz.

Edit: found some CPU data for these two CPU's:
Name Frequency Instructions L1 cache L2 cache
603e 100-300 MHz 2 per cycle 2x16 KB
604e 166-233 MHz 6 per cycle 2x32 KB external to 1 MB

Toni Wilen 19 February 2016 15:30

Always ignore all guessimated frequency values under emulation! (emulation that runs as fast as possible). They are total bogus.

Stop watch is the only reliable way to measure anything.

amyren 21 February 2016 12:58

Quote:

Originally Posted by Toni Wilen (Post 1071811)
Always ignore all guessimated frequency values under emulation! (emulation that runs as fast as possible). They are total bogus.

Stop watch is the only reliable way to measure anything.

I know, these frequency values cant be compared to real hardware. But compared to each other they did not seemed so far away from my user experience.

Did some tests though, clocking a few operations:

From WinUAE GUI to fully loaded workbench:
A1200, 603e: 3 minutes
A1200, 604e: 1 minute 10 sec.
A4000 CybPPC: 47 sec.

IBrowse startup:
A1200, 603e: 1 minute
A1200, 604e: 20 sec..
A4000 CybPPC: 8 sec.

OWB startup:
A1200, 603e: 2 minutes 10 sec.
A1200, 604e: 35 sec.
A4000 CybPPC: 8 sec.

I did only use a wall watch, so give or take a few seconds for accuracy.
Also note that the A4000 cyberppc setup does have an advantage that might affect speed, because the hardfile is SFS\02. The hardfile used for the A1200 blizzardppc tests is DOS\07.

I have tried to use SFS for the A1200 setups as well, but for some reason I have trouble getting it to work. When trying to use SFS\00 it fail to install the harddrive properly, because after rebooting back to the OS4.1 installation the harddrive still is listed as "Not installed". Setting it up as SFS 2 did work a bit better, and I was able to complete the installation, but after rebooting it returns guru meditation.
@ccapublic: Speaking of guru meditation, did you try using fastfilesystem for your setup?

If someone have a guide for setting up SFS harddrives that works on blizzardppc it would be nice.


All times are GMT +2. The time now is 19:49.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2019, vBulletin Solutions Inc.

Page generated in 0.04907 seconds with 11 queries