Performance patches for Amiga OS 3.2?

[shelter]

I have an Amiga with tf1230.

I've been looking for "safe" performance patches that works with Amiga OS 3.2, without making the system too unstable.

Currently I use this patches:
TRSaferPtch
CopyMemAIO
StackAttack2
FBlit (or P96 Native driver)
PeterK's icon.library - be sure to use his fblit.prefs too.
TLSFMem (Removed, see below in thread)
And PFS as filesystem

Added since original post:
MuMove4k
MuFastZero
MuFastROM
BlazeWCP
Executive (If you run OS 3.2 or OS 3.2.1 you need to set the ramlib task to NoSchedule in ExecutivePrefs, otherwise the system will hang)
SoftIEEE - Not really a performance patch, it adds a software FPU, it's very slow but might help in some cases.

I haven't patched scsi.device yet as I mostly use the ehide.device anyway.

The above patches seem to work OK and should give you a speed boost. While CopyMemAIO works, I'm not sure if it actually speed things up. I haven't ran a real benchmark test with it.

[alexh]

MMULib to use MUFastROM put ROM into FastRAM (tf1230 may have it's own feature for this) and MUFastZero to relocate address zero (the interrupt autovectors) from ChipRAM to FastRAM

[shelter]

I don't think MUFastROM is needed when you use the LoadModule ROMUPDATE feature.
I haven't touched the original 3.1 roms as I like to keep them, so it just does a ROMUPDATE during boot.

I am not sure about MUFastZero tho', if it's needed or not.

[musashi5150]

Originally Posted by shelter

I am not sure about MUFastZero tho', if it's needed or not.

Run SysInfo and look to see if the VBR is $0 - if it is your interrupt vectors are in ChipRAM.

As alexh says, there is something to be gained if you relocate them into FastRAM instead

[alexh]

Originally Posted by shelter

I don't think MUFastROM is needed when you use the LoadModule ROMUPDATE feature.

It depends how many modules the ROMUPDATE includes? Any that are not being updated will run from ROM and be slow?

That was my thought. Probably completely wrong. The documentation (for someone not familiar) is ambiguous.

I'm not an expert in this area, Thomas Richter will know for sure.

[patrik]

Originally Posted by alexh

It depends how many modules the ROMUPDATE includes. Any that are not being updated will run from ROM and be slow? That was my thought.

That is correct.

If you want, you can give the following work in progress project a try. It will conveniently list all system components not in highest prio memory:
http://megaburken.net/~patrik/pt/ResidentSpeed

Some docs with examples:
https://github.com/patrikaxelsson/Re...main/README.md

[shelter]

I've checked (with scout tho') and everything is in Fast Mem, except for expansion.library but that's apparently never in Fast mem.

HOWEVER, VBR was at $0 and running MuMove4k and MuFastZero actually moves VBR to within the Fast Mem range, so thanks for the hint.

I'll have to try patrik's program too, tack så mycket.

[shelter]

Well, that's interesting, I ran Patrik's program without MuFastROM, it showed some "issues", tho' almost everything was in either kick or Fast mem

Then I thought, what the heck, let's try MuFastROM, all issues disappeared and it only showed expansion.library being in chip mem at 20% speed.

So I guess MuFastROM actually does something even tho' the OS 3.2 roms are "softkicked" with LoadModule.


EDIT:
I also think TLSFMemPool crashes AIBB, so maybe it's too much of a hack to be functional? Or AIBB is doing illegal things, I don't know.

[alexh]

Originally Posted by shelter

EDIT:
I also think TLSFMemPool crashes AIBB, so maybe it's too much of a hack to be functional? Or AIBB is doing illegal things, I don't know.

It only works with exec 40.10 (e.g. KS3.1) and doesn't work with newer exec. Cosmos made a patch so it didn't crash if newer exec but I think all it does is turn it off.

I discussed with Thomas Richter and he says the need for TLSFMem is greatly reduced with AmigaOS 3.1.4+

http://eab.abime.net/showthread.php?...27#post1559427

Originally Posted by Thomas Richter

TLSFMem is probably no longer as much needed as it was in 3.1. In 3.1.4, the prime sources for memory segmentation were already removed, and those are ClipRects (layers V45 fixed that) and Regions and RegionRectangles (graphics V45 fixed that). Other than that, I would suggest PoolMem, which also keeps a scratch list for small memory allocations. That is probably already sufficient to speed up the rest as it buffers small blocks in a scratch list.

I think this is PoolMem that Thomas is talking about?

https://aminet.net/package/util/sys/PoolMem

[Estrayk]

http://aminet.net/package/util/boot/Blazewcp

is needed in 3.2.1?

[itsmedoofer]

Originally Posted by Estrayk

http://aminet.net/package/util/boot/Blazewcp

is needed in 3.2.1?

I'm using it with 3.2.1 still seems to work ok and still gives iBrowse a noticeable kick in the pants...

[patrik]

Originally Posted by shelter

Well, that's interesting, I ran Patrik's program without MuFastROM, it showed some "issues", tho' almost everything was in either kick or Fast mem

Then I thought, what the heck, let's try MuFastROM, all issues disappeared and it only showed expansion.library being in chip mem at 20% speed.

So I guess MuFastROM actually does something even tho' the OS 3.2 roms are "softkicked" with LoadModule.

Yes, that is correct. 3.2 does not have new versions of every component of the 3.1 ROM. Those with same version will not be replaced by LoadModule ROMUPDATE and if I recall correctly, they are not even provided as separate files on the modules disk.

Out of curiosity, could you paste the output ResidentSpeed produces without MuFastROM on your system?

[shelter]

Originally Posted by patrik

Out of curiosity, could you paste the output ResidentSpeed produces without MuFastROM on your system?

Type     Name                    Version Size    Loc  Speed   Speed %
Region   interrupt vector table            1.00k Chip   6.25M    23.8
Resident expansion.library        40.2       212 Kick  11.80M    54.5
LibBase  expansion.library        40.2       554 Chip   6.37M    24.9
Resident diag init                40.0     2.68k Kick  12.56M    47.5
Resident potgo.resource           37.4       310 Kick  12.17M    50.0
Resident cia.resource             39.1      1006 Kick  12.48M    48.0
Resident misc.resource            37.1       176 Kick  12.17M    50.0
Resident battmem.resource         39.2       484 Kick  12.48M    49.9
Resident keymap.library           40.4     3.12k Kick  12.56M    47.4
Resident romboot                  40.0       100 Kick  10.82M    50.0
Resident bootmenu                 47.11   15.27k Fast  25.87M    96.9
Resident alert.hook               40.0       327 Kick  12.17M    50.0
Resident strap                    40.1     3.74k Kick  12.56M    47.5
Resident ramlib                   40.2     1.03k Kick  12.53M    49.1
Resident workbench.task           39.1       194 Kick  11.80M    54.5

Originally Posted by alexh

It only works with exec 40.10 (e.g. KS3.1) and doesn't work with newer exec.

Yeah, I removed TLSFMemPool.

Originally Posted by Estrayk

http://aminet.net/package/util/boot/Blazewcp

is needed in 3.2.1?

I don't see it being mentioned in the OS 3.2 FAQ.
I have to try it out myself too.

[patrik]

Originally Posted by shelter

Type     Name                    Version Size    Loc  Speed   Speed %
Region   interrupt vector table            1.00k Chip   6.25M    23.8
Resident expansion.library        40.2       212 Kick  11.80M    54.5
LibBase  expansion.library        40.2       554 Chip   6.37M    24.9
Resident diag init                40.0     2.68k Kick  12.56M    47.5
Resident potgo.resource           37.4       310 Kick  12.17M    50.0
Resident cia.resource             39.1      1006 Kick  12.48M    48.0
Resident misc.resource            37.1       176 Kick  12.17M    50.0
Resident battmem.resource         39.2       484 Kick  12.48M    49.9
Resident keymap.library           40.4     3.12k Kick  12.56M    47.4
Resident romboot                  40.0       100 Kick  10.82M    50.0
Resident bootmenu                 47.11   15.27k Fast  25.87M    96.9
Resident alert.hook               40.0       327 Kick  12.17M    50.0
Resident strap                    40.1     3.74k Kick  12.56M    47.5
Resident ramlib                   40.2     1.03k Kick  12.53M    49.1
Resident workbench.task           39.1       194 Kick  11.80M    54.5

I would agree that it is indeed “issues”, none of the above components are used in a time critical manner as far as I know (please correct me if I am wrong).

Also given that the A1200 kickstart ROM is quite fast compared to your accelerator RAM at ~50% makes any impact small - it is not like a similar accelerator on an A500/A2000 where the ROM is 1/4 of the speed in the A1200.

Anyway, I guess you have plenty of memory, so no reason not to remap it.

Btw, bootmenu should not be there as it is in fast, but for some reason its result deviates too much .

[shelter]

Originally Posted by patrik

I would agree that it is indeed “issues”, none of the above components are used in a time critical manner as far as I know (please correct me if I am wrong).

The thing is, if I run MuFastROM the "issues" disappear. And if I run "ResidentSpeed ShowAll" it says workbench.task (for example) is in Kick but it's speed is 100% instead of 54%. (in fact almost everything is at 100%),

So MuFastROM seems to do something as I previously said, I'm not sure exactly what.

[shelter]

I noticed when running FBlit/FText and Copper-Demon, the" workbench refresh" gets worse, it's noticable when closing windows etc. It's caused by FBlit when there's a pattern on the Workbench. And from what I saw, running P96 with the Native driver did the same thing.

So I don't think it's related to Copper-Demon at all really, it's just that Copper-Demon happens to use a pattern for Workbench.

Anyway, I found a workaround and that is to increase the size of the WBPattern that Copper-Demon uses.

Included is a 64x64 pattern instead of the 16x16 pattern that comes with Best Classic WB. I think it takes slightly more chip ram but it's not much.

[ma693541]

@shelter If you use P96 with the Native driver from Aminet on AmigaOS3.2.X, there's a little incompability to it under AmigaOS3.2.X as you self found out and stated in #16.

[huggy70]

Although it gives slightly less free chip ram and ist considered a hack, I found fblit to be more than 2x as fast as p96 native.

[mfilos]

I use the following patches on my OS3.2.1 installation:

• StackAttack2 (sets minimum stack size for new processes)
• SaferPatches (records patches using "ShowPatch")
• PoolMem (memory defragmentizer)
• CopyMemAIO (fast replacement for CopyMem())
• NoMoreDiv0 (stops "Division by Zero" app crashes)
• BeepPatch (patches DisplayBeep() to flash the titlebar)

For visual stylesm eyecandy and ease of use:

• AfAOS (only afa_system, diskfont_lib, graphics_lib libraries though) for Antialiased TTF
• MCP for BBlank, AlertHistory, NoGuru, ReplaceTopaz and hotkey support
• TitleShadow for shadow in windows titles
• VisualPrefs for GUI styles
• FullPalette for locking the first 16 WB colors
• Birdie for window themes

I also use the following applications in WBStartup:

• ToolsDaemon/ToolsPrefs for custom menus. I know we have MenuTools, but until we get an UI, it's not so easy to use.
• Ranchero for using notification balloon for some events (like RoadShow start or stop etc).
• MagicMenu for menu enchancing
• ReqAttack for better requesters
• ScreenTime for having date/time in titlebar and click-on calendar
• WBDock as a replacement for Amidock but with a lot more options

Most of times I have PoolMem and CopyMemAIO commented (until I get a decent green core on my Vampire as I get lot of Deadends when using them).
Other than that, system runs extremely well with no issues using all the above so far. I have troubleshoot a lot of applications using SnoopDos and so far I didn't have any major problem.

It would be nice to know if BlazeWCP provides any benefit in RTG system as I used it always in non RTG systems without a single problem.

[shelter]

Originally Posted by ma693541

@shelter If you use P96 with the Native driver from Aminet on AmigaOS3.2.X, there's a little incompability to it under AmigaOS3.2.X as you self found out and stated in #16.

I have the most recent P96 version from iComp.

Originally Posted by huggy70

Although it gives slightly less free chip ram and ist considered a hack, I found fblit to be more than 2x as fast as p96 native.

Yes, FBlit/FText is way faster than P96/Native.
I haven't really seen any issues with it other than slow updates I mentioned 2 posts up.