JIT still crash ! Prob found ?!?

[Foul]

Just to remember :

http://eab.abime.net/showthread.php?...09#post1124009
http://eab.abime.net/showthread.php?t=84876
http://eab.abime.net/showthread.php?t=86521
http://eab.abime.net/showthread.php?t=80079

Activation of JIT causes me problems for a long time with my configuration. On my A1200/060 Setup... RAM is filled when I browse the directories and never free it ! from where at a given moment a crash due to lack of memory!
With latest dev fs-uae.. fs-uae boot and stop at mount command failled.



This does not happen with JIT disabled.

I decided today to make a minimal system config to test this prob.

I provide a small .hdf with a fresh install of WB3.1 .. nothing added !!!
+ fs-uae logs + default.fs-uae cfg...

run FS-UAE with JIT = Crash of the emulator
delete JIT lines =FS-UAE boot normaly

With some discutions with Frode, we found the prob is gone when i modify this line in the sources : src/jit/gencomp.cpp

*/
#define DISABLE_I_MVEL
/*

With this small change.. compiled myself, everything is working perfect !!! but iI do not know if it affects anything else.

Maybe can be corrected on next FS-UAE (WinUAE ?) release !!!

Thx for listening !

Amigatest.hdf : https://goo.gl/48FEA8 5.3 Mo

[jbl007]

Hi,

you should clean your configuration first. Use "official" options wherever possible. Remove

uae_cachesize

. Replace

uae_cpuboard_type = Blizzard1260

with

accelerator = blizzard-1260

and

uae_cpuboardmem1_size = 128

with

accelerator_memory = 131072

With these changes FS-UAE does NOT crash. Your Amigatest.hdf boots straight into a guru. This also happens with JIT disabled and in WinUAE. Probably due to CPU instructions/68060.library stuff - I don't know... [Edit: removed LIBS:68040.library, provided LIBS:68060.library, now it works fine]
It appears to work with your config with jit disabled. But if you look closely you'll see, the Blizzard (and the 060) isn't emulated at all.

Next: I don't think the Test Drive II issue is JIT related. It also happens without jit on my system. Looks like the typical cpu-is-to-fast-problem.

[Foul]

replaced uae_cpuboard_type = Blizzard1260 with accelerator = blizzard-1260 and
uae_cpuboardmem1_size = 128 with accelerator_memory = 131072

Still crash (Mount error) on my System here...
I'm sharing my CF card on FS-UAE and my A1200/060 with same config...

[jbl007]

So real CF/physical device like /dev/sdX? Is this even supported?
What happens if you slow down the cpu? Try

uae_cpu_throttle = -900.0

If this works we have our cpu-is-to-fast-problem again....

[Foul]

hard_drive_0 = /dev/sdx is working very well on an CF/usb reader

uae_cpu_throttle = -900.0 ... much slower to boot but same :



it would have been too good

ps : and working as usual with the modified compiled fs-uae

[Hedeon]

Failed returncode 20 is NOT a crash.

[Foul]

The Crash is with the "test" hdf.. here i'm using my real CF card. but it's not normal anyway...

Just for info : all starting from here : http://eab.abime.net/showthread.php?...09#post1124009

[Foul]

I'm not a coder but i think the question is : why with the little modification in the sourcecode, everything is working well ! and if it affect something else...

[Hedeon]

I'm not into the source, but it looks like the define does something with disabling illegal movs or something and fails if you try one. With commenting the define it's executed anyway, illegal or not.

[FrodeSolheim]

Originally Posted by Foul

I'm not a coder but i think the question is : why with the little modification in the sourcecode, everything is working well ! and if it affect something else...

What happens when you add

#define DISABLE_I_MVEL

is that you force UAE to run this instruction using the normal interpreter. So, if there is a bug with the JIT implementation of this instruction, then this will do as a workaround. However, there is some overhead in dropping out of JIT mode into interpreter mode and back, so obviously not a real solution.

Let's test this again later with merged code from WinUAE 4.2.0. There's been some JIT merges from Aranym, so maybe the problem is already fixed.

[Foul]

will test it for sure

thx !

[Foul]

Probs gone with latest winuae4210 commit !

[FrodeSolheim]

Yay

[Foul]

I think I spoke a little fast ... problem always present on the last dev ..

JIT OFF = No Prob at all



JIT ON = Memory Probs





when i open a folder, the ram goes down, and when i close the folder.. the ram is not released ...

edit: tested same thing with or without using maprom/blizkick

[Foul]

Sorry to unearth the topic, but the problem persists. Always crashes even with the latest version of fs-uae.

I put my CF card in my real A1200 / 060 .. it starts .. everything works perfectly ... I take this card, I put it in fs-uae without jit ... everything starts and works as it should ... I activate the JIT ... and I have exactly the same problem cited above

[PeterK]

That looks like random disk read errors for me. It seems that my icon.library gets damaged image data when it reads the icons, can't decode them successfully and then displays the internal default dot image instead.

Does the same happen when you display icons from the Ram Disk, a floppy or another disk? Check the icons that you copy to the Ram Disk for possible corruption with a compare before you open the Ram Disk drawer for displaying them.

[Foul]

i launched my system with icon.library via blizkick... without startup sequence, copied some icons to ram :



if i remove icon.library from blizkick.. same results...

edit : uhmm i replaced icon.library 46.4.527 in Libs by 44.543 and it boot normaly !!!

I had other problems in particular with Dopus, which I talked about on another forum .. and everything is back to normal once the icon.library is replaced ... : https://www.amigafrance.com/forums/t...lp/#post-83270 (in french)

[PeterK]

The icon.libraries v40, v44 and v45 from OS 3.1/3.5/3.9 are using buffered reads, I'm using unbuffered reads, which means reading directly from the filesystem.

Are that always the same icons which are displayed as dot images or other icons after each start? If these dots are always the same upload some of the "bad" icons, please.

Are all icons appearing correctly again with the v44 library?
And are they back again on the Ram Disk with my library when the JIT is off?

[Foul]

icon.library 46.4.527 :

Without JIT : icons normals but Dopus probs
With JIT : don't start, crash, icons with dots, dopus probs

icon.library 44.543 :

Without JIT : icons normals, Dopus OK
With JIT : icons normals, Dopus OK

[PeterK]

Are that always the same icons which are displayed as dot images or other icons after each start? If these dots are always the same upload some of the "bad" icons, please.

Try an older version <= 524 of my library. I've changed the decompression code recently.

And I still don't understand what's going wrong with my library in DOpus 4.