![]() |
2010-05-15
, 15:44
|
|
Posts: 903 |
Thanked: 632 times |
Joined on Apr 2008
|
#2
|
![]() |
2010-05-15
, 15:56
|
|
Posts: 2,355 |
Thanked: 5,249 times |
Joined on Jan 2009
@ Barcelona
|
#3
|
Doesn't matter how much memory I give it, it still crashes. I've noticed this with many of the more advanced Palm programs (for OS 5+), while OS 3.5 programs seem to work fine.
![]() |
2010-05-15
, 16:12
|
Posts: 87 |
Thanked: 13 times |
Joined on Dec 2009
|
#4
|
![]() |
2010-05-15
, 16:30
|
|
Posts: 2,355 |
Thanked: 5,249 times |
Joined on Jan 2009
@ Barcelona
|
#5
|
![]() |
2010-05-15
, 18:35
|
Posts: 87 |
Thanked: 13 times |
Joined on Dec 2009
|
#6
|
![]() |
2010-05-15
, 18:51
|
|
Posts: 903 |
Thanked: 632 times |
Joined on Apr 2008
|
#7
|
Actually this is completely on topic, as this is exactly the issue I'm experiencing; most "OS5+ only" programs contain ARMlets/PNOlets, and as I explain on my post, they fail to work on the N900 because of the OABI cacheflush call.
Solutions:
A ) ACCESS should be able to easily fix/workaround this, so poke them. I already sent them the above using their feedback form.
B) Use a kernel with CONFIG_OABI_COMPAT=y. I can build it (zImage only), but won't provide support for it, so better ask titan or somebody who will.
C) Binary patch GVM. Ugly, and hard.
And yes, performance of armlets seems improved on N900, as I get nearly 30% more frames per second on my PalmGL port.
![]() |
2010-05-16
, 17:33
|
|
Posts: 2,355 |
Thanked: 5,249 times |
Joined on Jan 2009
@ Barcelona
|
#8
|
Enabling CONFIG_OABI_COMPAT in kernel build config causes GVM to work again like on the N8x0.
Sorry for not providing a fix (but should be easy enough to fix if you're using a patched/"community" kernel). I think I'll email GVM about this to see if they can fix it in the next version.