View Single Post
Old 21 November 2015, 16:20   #29
meynaf
68k wisdom
meynaf's Avatar
 
Join Date: Nov 2007
Location: Lyon (France)
Age: 44
Posts: 2,362
Quote:
Originally Posted by Toni Wilen View Post
Ok, log shows some more clues, sprite corruption appears to be side-effect of something else..
Indeed it smells like a side-effect, like overwriting some other's memory or doing UMRs.

I tried playing with pointer prefs. Setting up an empty pointer (all area cleared) leaves me with vertical lines. Switching from lowres to hires gives half width and twice height.


Quote:
Originally Posted by Toni Wilen View Post
Your log has multiple "DMAL error!? xxxx" lines. I assume you tried to duplicate the sprite problem and it finally happened after 7 tries and each test logged that DMAL error message. (It is "this should never happen" message)
I didn't try to duplicate the sprite problem 7 times, only once. Perhaps it occurs after a set amount of time...
If it is "should never happen", then why does it happen nevertheless ?


Quote:
Originally Posted by Toni Wilen View Post
Do you get that log message more easily and faster than sprite corruption? (Run in windowed mode with -log command line to see the log in real time).
Doesn't appear so. Simply sending the app to the task bar isn't enough to see the message. Apparently they just come up together after a while (possibly several times so we see 7 messages but only one corruption of course).


Quote:
Originally Posted by Toni Wilen View Post
Does it only happen if JIT is enabled? JIT changes behavior of "DMAL" Paula to Agnus/Alice DMA request reporting.
Actually it's quite the opposite. It seems to happen faster with jit off (which looks odd to me because the cpu is actually idle in both cases).


Quote:
Originally Posted by Toni Wilen View Post
Unfortunately I haven't found any simple way (by force enabling/disabling AGA sprite feature bits) to duplicate the corrupted cursor without also breaking sprite positioning which probably means "DMAL" confusion may corrupt some variables. But on the other hand Amiga-side reset should re-initialize most of them.
Amiga-side reset doesn't remove the problem, whether done by software or ctrl-A-A.
However, clicking "restart" from the ui does remove the problem (why didn't i think of doing that earlier, silly me ).


Quote:
Originally Posted by Toni Wilen View Post
One really annoying method to debug this would be to create full dump of the process (dozes or even hundreds of megabytes in size..) and load it in debugger and check all variables... But that can wait..
You could do some kind of version with lots of debug outputs and send it to me. If it just outputs a selected set of variables and i do two versions, one before the problem, the other after, then the size can remain acceptable.
Just my suggestion of course, as you said you debug like you want


Quote:
Originally Posted by Toni Wilen View Post
Quickstart mode not active (no checkbox) and you didn't click "Set configuration" button?
Errmh... this is exact


Quote:
Originally Posted by Toni Wilen View Post
Hmm.. Even that DMAL message seems to be a side-effect of something else because DMA request variable is non-zero only when audio or disk DMA is active.

It probably is DMAL state variable corruption caused by some "external" reason because in normal operation high bits of variable are never set but they are set in your log.
Can "external reason" be "loss of directx surface", or "windows aggressively trimming the process" or something like that ?


Quote:
Originally Posted by Toni Wilen View Post
Try also without filter enabled just to be sure.
Current test running with filter set to none.
Note that changing the filter setting after getting the bug, doesn't restore normal operation.


Quote:
Originally Posted by esc View Post
This was an awfully polite way to respond, Toni. If I were you'd I'd have just said "I told you so!"
But he's had info from the log, not from the saved state
(And the bug is still far from being fixed, btw).
meynaf is offline  
 
Page generated in 0.05918 seconds with 9 queries