Quake 2 ported to AGA

[NovaCoder]

Originally Posted by JaYgRp

@novacoder: When you say in the doc: "You should always patch exec.library's copymem() copymemquick() ...", do you think to a particular hack to do this ?

edit: i actually use Oxyron patcher. Is this ok with it ?

Just do a search copymem on AmiNet and grab the fastest 060 version.

Are you sure it's not Oxy patched that's boosting your fps? Can you try disabling it and see what happens for me.

I've already searched on aminet, find and tryed some copymem 060 patch but the result is not better. Maybe worst i think. So i was wondering if oxy was not already patching the "copymem".

I'll try without oxy and tell you more about the fps.

[NovaCoder]

Originally Posted by JaYgRp

I've already searched on aminet, find and tryed some copymem 060 patch but the result is not better. Maybe worst i think. So i was wondering if oxy was not already patching the "copymem".

I'll try without oxy and tell you more about the fps.

Ok, thanks I'd be very interested to see if Oxy Patcher is helping your FPS

The two CopyMem() patches to try are:

CMQ060
CopyMem060

I think I'm using CMQ060 at the moment.

Yes it's very possible that Oxy Patcher is already patching CopyMem() for you.

[matthey]

Originally Posted by JaYgRp

I've already searched on aminet, find and tryed some copymem 060 patch but the result is not better. Maybe worst i think. So i was wondering if oxy was not already patching the "copymem".

Oxy should not be patching exec/CopyMem() but some other system patches will like MCP which gives less than 1/2 the speedup of the CopyMem patch. The speedup is about 35% over the AmigaOS 3.9 CopyMem() which is significant but not always noticeable. One place you can notice a difference is in AWeb when scrolling around. Try running with and without the patch and compare. Speed test results (as well as how to test yourself) are listed in the CopyMem patch readme on Aminet:

http://de3.aminet.net/util/boot/CopyMem.readme

Originally Posted by NovaCoder

Ok, thanks I'd be very interested to see if Oxy Patcher is helping your FPS

I didn't see a single trapped instruction in the AmiQuake 2 executable. It's possible that a library would have some but it's highly unlikely that Oxy would give a noticeable improvement.

Originally Posted by NovaCoder

The two CopyMem() patches to try are:

CMQ060
CopyMem060

I think I'm using CMQ060 at the moment.

Those are the 2 good ones .

Oxy did the trick ! Tested several times this morning:

With oxy:

Blizzard 1260: 060@85MHz
AmiQuake2 1.06
default 80% full screen
timedemo demomap q2demo1.dm2
fps : 13.3

Blizzard 1260: 060@85MHz
AmiQuake2 1.06
default 100% full screen
timedemo demomap q2demo1.dm2
fps : 12.6


Without oxy:

Blizzard 1260: 060@85MHz
AmiQuake2 1.06
default 80% full screen
timedemo demomap q2demo1.dm2
fps : 12.3

Blizzard 1260: 060@85MHz
AmiQuake2 1.06
default 100% full screen
timedemo demomap q2demo1.dm2
fps : 11.7

[hooverphonique]

that's strange.. is this port not compiled specifically for 060 and thus should not cause any traps?

Originally Posted by hooverphonique

that's strange.. is this port not compiled specifically for 060 and thus should not cause any traps?

Don't ask me why i can just report what i see. I've edited my previous post cause with and without oxy was inverted.

[NovaCoder]

Originally Posted by hooverphonique

that's strange.. is this port not compiled specifically for 060 and thus should not cause any traps?

Yes it is compiled for 060 + FPU but the Amiga compilers are not very good and sometimes insert 020 FPU instructions.

I will test myself with CyberPatcher.

Thanks



Update:

Tried both CyberPatcher and OxyPatcher 3.14, neither one had any effect on the FPS? Not sure what's going on, maybe I screwed up the install or maybe it conflicts with BlizKick?

Can anyone else confirm these results?

HSMATHLIBS certainly seemed to make it much more responsive for me, even though the didn't have much of an effect on FPS.

[matthey]

Originally Posted by JaYgRp

Don't ask me why i can just report what i see. I've edited my previous post cause with and without oxy was inverted.

Perhaps the Apollo 68060.library does not patch the utility.library. Try using this patch without oxy:

http://aminet.net/util/boot/UtilPatch.lha

Let us know if UtilPatch provides a similar (or better) speed than oxy.

Originally Posted by NovaCoder

Yes it is compiled for 060 + FPU but the Amiga compilers are not very good and sometimes insert 020 FPU instructions.

That is true but I didn't see any. Not even a FSIN or FCOS. It would take heavy use of trapped instructions to move the fps needle as we have observed.

Originally Posted by NovaCoder

I will test myself with CyberPatcher.

You can use CyberSnooper to detect trapped instructions. You will need Sushi or similar to see the debugging output.

Originally Posted by NovaCoder

Tried both CyberPatcher and OxyPatcher 3.14, neither one had any effect on the FPS? Not sure what's going on, maybe I screwed up the install or maybe it conflicts with BlizKick?

Or maybe his utility.library functions don't get patched like your 68060.library does. Do you use ThoR's mu 68060.library? It patches the utility.library with similar code to UtilPatch (UtilPatch becomes unnecessary).

Originally Posted by NovaCoder

HSMATHLIBS certainly seemed to make it much more responsive for me, even though the didn't have much of an effect on FPS.

That is possible. This version of GCC calls the IEEE fp libraries instead of using a few trapped FPU instructions like earlier versions.

[NovaCoder]

Originally Posted by matthey

Perhaps the Apollo 68060.library does not patch the utility.library. Try using this patch without oxy:

http://aminet.net/util/boot/UtilPatch.lha

Let us know if UtilPatch provides a similar (or better) speed than oxy.

Hiya Matt,

I'm currently using the last official Blizzard 68060 library, BlizKick, HSMATHLIBS (060) and your CopyMemPatch.

Do you think I would benefit from this Util Patch? Are there any other 060 patches are should be running?

This is what my 3.9 + BB2 Startup looks like.

; $VER: Startup-Sequence_HardDrive 45.2 (19.1.2001)
; Startup-Sequence for AmigaOS 3.9

C:BlizKick DEVS:rom_v45.57 EXTRESBUF=150000 MODULE exec.library bootmenu FileSystem.resource ram-handler scsi.device console.device shell mathffp.library mathieeesingbas.library BBlank NoClick LocalFast MoveVBR SpeedyIDE QUIET

;C:SetPatch QUIET
C:SetPatch NOROMUPDATE QUIET



C:Version >NIL:
C:AddBuffers >NIL: DF0: 15
FailAt 21

C:MakeDir RAM:T RAM:Clipboards RAM:ENV RAM:ENV/Sys
C:Copy >NIL: ENVARC: RAM:ENV ALL NOREQ

Resident >NIL: C:Assign PURE
Resident >NIL: C:Execute PURE

Assign >NIL: ENV: RAM:ENV
Assign >NIL: T: RAM:T
Assign >NIL: CLIPS: RAM:Clipboards
Assign >NIL: REXX: S:
Assign >NIL: PRINTERS: DEVS:Printers
Assign >NIL: KEYMAPS: DEVS:Keymaps
Assign >NIL: LOCALE: SYS:Locale
Assign >NIL: LIBS: SYS:Classes ADD
Assign >NIL: HELP: LOCALE:Help DEFER



;C:CMQ060
Run >NIL: CopyMem060




C:FBlit

BindDrivers
C:Mount >NIL: DEVS:DOSDrivers/~(#?.info)

C:LoadMonDrvs

SetEnv Language "english"
SetEnv Workbench $Workbench
SetEnv Kickstart $Kickstart
UnSet Workbench
UnSet Kickstart

C:AddDataTypes REFRESH QUIET
C:IPrefs

Run > NIL: "System:C/VisualPrefs"
System:C/FPPrefs


C:ConClip

Path >NIL: RAM: C: SYS:Utilities SYS:Rexxc SYS:System S: SYS:Prefs SYS:WBStartup SYS:Tools SYS:Tools/Commodities

SYS:System/REXXMast >NIL:

IF EXISTS S:User-Startup
  Execute S:User-Startup
EndIF

Resident Execute REMOVE
Resident Assign REMOVE


C:LoadWB
EndCLI >NIL:

[matthey]

Originally Posted by NovaCoder

I'm currently using the last official Blizzard 68060 library, BlizKick, HSMATHLIBS (060) and your CopyMemPatch.

The UtilPatch is very easy to try without installing. Just double click the icon to run and it will only last until you reboot. You can also drop in the WBStartup although you won't gain the benefits during boot. The UtilPatch results can vary significantly depending on the numbers being multiplied. Some versions of GCC do call the utility.library often. AmiQuake 2 in no exception but it looked like SMult32() is being called several times (should be an inlined 32x32=32 MULS.L instruction on 68060) for a 64x64=64 (much slower) instead of SMult64() meaning that the UtilPatch should offer no speedup in this case. The AmigaOS libraries do rarely call this function but I did a test using Snoopy (like SnoopDOS) and didn't see a single call during a Q2 run. It should give no speedup then. Running UtilPatch was an easy test to remove one possible reason why oxy would give a speedup on some systems. UtilPatch is not as beneficial as CopyMem because it's used much less. Every once in a while a program will use it repetitively. I recommend using ThoR's Mu 68060.library generally as it makes the patch obsolete. The Mu 68060.library is a little faster than the P5 version if "MuFastZero MOVESSP ON" is in the S:Startup-Sequence. It is more difficult to install though and the directions should be read carefully. The P5 CPU libraries are also good and what I used last.

Originally Posted by NovaCoder

Do you think I would benefit from this Util Patch? Are there any other 060 patches are should be running?

BlizKick and CopyMem are the only patches I use.

[vitux]

Quake 2 AGA 1.06 on my 1200 060 50 mhz
[ Show youtube player ]

[DDNI]

@vitux. Hmmm not great is it?

[vitux]

DDNI, NO :-)
just to show how runs

[NovaCoder]

V1.07 in action

[ Show youtube player ]

Minor speed increase and code tidy up.

I'll upload next week.

[AmigaFriend]

Originally Posted by vitux

Quake 2 AGA 1.06 on my 1200 060 50 mhz
[ Show youtube player ]

Slideshow. It seems that anything below a 80Mhz 060 is out of order here.

Originally Posted by matthey

Perhaps the Apollo 68060.library does not patch the utility.library. Try using this patch without oxy:

http://aminet.net/util/boot/UtilPatch.lha

Let us know if UtilPatch provides a similar (or better) speed than oxy.

I've just tryed with Utilpatch:

Blizzard 1260: 060@85MHz
AmiQuake2 1.06
default 80% full screen
timedemo demomap q2demo1.dm2
fps : 12.3

Blizzard 1260: 060@85MHz
AmiQuake2 1.06
default 100% full screen
timedemo demomap q2demo1.dm2
fps : 11.7

This don't help to gain extra frames.

[matthey]

Originally Posted by JaYgRp

I've just tryed with Utilpatch:

...

This don't help to gain extra frames.

Ok. I don't think the speed increase comes from avoiding trapped instructions. Oxy may turn on some caches or change some CPU settings. I think your system should be running at oxy speed most of the time without oxy. Next test is to run Scout before and after oxy for comparison.

http://aminet.net/util/moni/Scout_os3.lha

Click on "System" (or French equivalent) from the main menu. It will bring up a window with tabs at the top for "Hardware", "Registers", "Uptime", "Versions", "Memory" and "Last Alert". Click on "Registers". If VBR: $00000000, interrupts will likely be slow although this can be virtually remapped to a location in fast ram and still read $00000000 (Mu 68060.library does this and it's much better for compatibility with old games). I suspect that oxy moves the VBR to fast ram and that's why it gives you a speed increase. Also, my 68060 reads:

InstCache: enabled, burst
DataCache: enabled, burst
PCR: $04300601
BranchCache: enabled
68060: Superscalar, WriteBuffer, Load/Store Bypass enabled

Your settings should be the same with a rev 6 68060 or the Load/Store Bypass should be disabled and the PCR will be off by one bit. Everything else should be the same before and after running oxy.

[NovaCoder]

Ok, I've release v1.07 to AmiNet (should show up soon).

As I mentioned above, this build is not much quicker than v1.06 because I've run out of ideas

One thing I have done is to change the screen resize function, you now get three choices (status icons overlaid, icons not overlaid and letter boxed).

Here's a quick video of the release version in action on my A1200 -> [ Show youtube player ]

[Bamiga2002]

Ran timedemos on my setup:
demo1 - 7.5 FPS
demo2 - 6.7 FPS

Not bad for AGA