Cannot initialize FFT code, errcode=1002

[kladner]

I am running P95 v 28.10 on an AMD FX-9590. Lately, I have had problems when changing number of workers and number of threads per worker. In the course of an experiment, I changed worker arrangements from my current standard of 2 workers, 4 threads each. There were several alternative arrangements which I tried, mostly to look at performance changes. When I tried to change back to the standard settings mentioned above, they would repeatedly revert to one of the alternates.

In the past, I found that I could get the desired settings back by making a drastic change, such as one thread each for 8 workers, saving, exiting, restarting, changing (while stopped) back to the "2 x 4" setup, saving, etc. etc. Eventually, that did not work, and I tried more direct means by messing with prime.txt and local.txt. I also copied a newly extracted P95 v 28.10 to my working folder. I checked if Read Only had sneaked in on the Prime95 folder: No. This folder is at the root of the drive it is on.

I now have things bollixed, in that P95 starts with the correct arrangement of workers and threads, but immediately gives the error message:

Cannot initialize FFT code, errcode=1002
Number sent to gwsetup is too large for the FFTs to handle

I am now at a loss as to how to recover. Suggestions, please.

EDIT: Also, though Options>CPU shows the correct information, when the program starts it says it is optimizing for Core i3/i5/i7.

[sdbardwick]

You aren't trying to run 64 bit P95 on a 32 bit OS, are you?

[kladner]

Originally Posted by sdbardwick

You aren't trying to run 64 bit P95 on a 32 bit OS, are you?

No. I should have mentioned Win 7 Pro 64 bit. This seems to have happened while I was struggling to get changes to the worker settings to remain as I set them.

[Prime95]

Make sure local.txt does not have any "CpuSupportsXXX" entries.

Make sure Options/CPU reports the correct CPU features for your CPU.

Check your worktodo.txt file. If "FFT2=xxxK" appears in any lines, remove it.

[kladner]

Originally Posted by Prime95

Make sure local.txt does not have any "CpuSupportsXXX" entries.

Make sure Options/CPU reports the correct CPU features for your CPU.

Check your worktodo.txt file. If "FFT2=xxxK" appears in any lines, remove it.

Unfortunately, No on local.txt
Yes on Options/CPU (I think. Screenshot attached.)
No on FFT2=

[Prime95]

Try adding "CpuArchitecture=102" in local.txt.
If no luck, try adding "CpuSupportsFMA3=0" to local.txt.

If I understand you, you're saying you used to run OK with 28.10 (as opposed to a previous version). What does your worktodo.txt file look like?

[kladner]

Originally Posted by Prime95

Try adding "CpuArchitecture=102" in local.txt.
If no luck, try adding "CpuSupportsFMA3=0" to local.txt.

If I understand you, you're saying you used to run OK with 28.10 (as opposed to a previous version). What does your worktodo.txt file look like?

On the first two, no change.
You are correct. I have had no problems with 28.10, nor with previous versions until I ran the test which resulted in 28.10
That was a special case, and did not affect my production work.

[0PolarBearsHere]

How big an exponent are you trying to test?

[kladner]

Originally Posted by 0PolarBearsHere

How big an exponent are you trying to test?

38.3M
EDIT: I have been doing 38-40M for months. This just came on last night, when I messed something up.
In the meantime, it is just as well that I run blend torture test to check up on stability in general.

[kladner]

I have it running again, at the cost of a few days' work. I built a new P95 folder and copied things over from the old. It seems that the problem involved a corrupted save file. At least, it recurred when I copied those over. I'm not sure how I messed this up, though it seems that I must have in some way.

Thanks for the responses and suggestions.