There's a fix for that being pushed through testing,

                                                                  Joe

Test results for these instances added to the test matrix.

I have two AP tasks that have been looping with repeated "exited with zero status but no 'finished' file."

Looking inside the result reports, they both say this:

ERROR: some exception inside long FFA, probably video-driver restart, restarting app...

The command line options are as follows:

-use_sleep -hp -unroll 16 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 8 1 -tune 2 64 8 1 

Since the GTX680 card that these tasks have run on behaves quite well with all other tasks, and still repeatedly denies to finish just these two, I think I may have stumbled upon a reproduceable error.

Anyone want to investigate? Any additional info needed?

Edit: Lunatics 0.43, running three tasks on the card.

That "ERROR: some exception inside long FFA..." is what shows when the app runs out of memory while trying to handle thousands of repetitive pulses above threshold. There's a fix for that being pushed through testing, meanwhile reducing how many signals have to be handled simultaneously by reducing the -ffa_block and -ffa_block_fetch is the way to go.

                                                                  Joe

Had a WU like that last night, don't know how many restarts it had, but I decided to abort it...

http://setiathome.berkeley.edu/workunit.php?wuid=1627971982

That's my point exactly, Richard.
I'm a software engineer myself, and when I see a reoccurring error like this, I know that a developer with a debugger can normally find out exactly what goes wrong.

I don't have a debugger, but I do have a GTX 670 - close match - and I'm starting to run tests under bench conditions, starting with default cmdline parameters. One thing to watch out for will be abnormally high memory usage on these tasks.

Edit - seemed to settle at 244 MB usage after a couple of minutes. High, but not as high as we've seen under bug conditions.

Edit2 - memory consumption over a gigabyte with the same command line, single instance. This does look like the bug which was already under investigation - will run with the bugfix version already under test next.

So, Mike's suggestion might be a temporary palliative while we wait for the bugfix to complete acceptance testing - or leave the commandline as it is if you're prepared to risk the same thing happening again, and supply additional test cases for the testing pool.

That's my point exactly, Richard.
I'm a software engineer myself, and when I see a reoccurring error like this, I know that a developer with a debugger can normally find out exactly what goes wrong.

So i dont have to worry any longer.

---

With each crime and every kindness we birth our future.

That's my point exactly, Richard.
I'm a software engineer myself, and when I see a reoccurring error like this, I know that a developer with a debugger can normally find out exactly what goes wrong.

---

Reduce unroll to 12 and ffa_block to 12288 6144.
See if this helps.

With 1061 valid AP v7 tasks so far, I think he knows how to drive the application. The question was why these two (and only these two, as I understand him) should have behaved differently.

Do you want to tell me how the app works ?

---

With each crime and every kindness we birth our future.

Reduce unroll to 12 and ffa_block to 12288 6144.
See if this helps.

With 1061 valid AP v7 tasks so far, I think he knows how to drive the application. The question was why these two (and only these two, as I understand him) should have behaved differently.

Reduce unroll to 12 and ffa_block to 12288 6144.
See if this helps.

---

With each crime and every kindness we birth our future.

I have two AP tasks ...

To get the hunt started:

http://boinc2.ssl.berkeley.edu/sah/download_fanout/1ba/ap_04jn14aa_B2_P0_00132_20141029_29201.wu
http://boinc2.ssl.berkeley.edu/sah/download_fanout/192/ap_04jn14aa_B0_P0_00236_20141029_29288.wu

I have two AP tasks that have been looping with repeated "exited with zero status but no 'finished' file."

Looking inside the result reports, they both say this:

ERROR: some exception inside long FFA, probably video-driver restart, restarting app...

The command line options are as follows:

-use_sleep -hp -unroll 16 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 8 1 -tune 2 64 8 1 

Since the GTX680 card that these tasks have run on behaves quite well with all other tasks, and still repeatedly denies to finish just these two, I think I may have stumbled upon a reproduceable error.

Anyone want to investigate? Any additional info needed?

Edit: Lunatics 0.43, running three tasks on the card.

---