clean-up after failed execution

[xeiwoka]

when the volume sizes get too large, I sometimes get warnings like the following after execution of "tofu reco":
...
DEBUG: Assigning task 'filter' to node 0
DEBUG: Padded width: 4096
DEBUG: Padding mode: clamp_to_edge
DEBUG: Pass: 1, device: 1, region: [0.0, 900.0, 5.0]
Error querying for program binary sizes
Error querying for program binary sizes
Error querying for program binary sizes
Error querying for program binary sizes
ERROR: ufo-task-error-quark: Volume size doesn't fit to memory (1)
Warning: Program terminating, but clFFT resources not freed.
Please consider explicitly calling clfftTeardown( )

Could one not force that last command at program exit?

[tfarago]

@matze have you seen this before?

[matze]

Yes, I have and in the normal course of action everything should be cleaned up properly. The question is why the destructor is not called when an error happens elsewhere.

By the way, I find the AMD FFT to be questionable anyway and would prefer the bundled Apple FFT. For example, all these "Error querying for program binary sizes" are non-issues blurted out to stdout by the AMD lib for no good reason.

Could one not force that last command at program exit?

At program exit, everything is cleaned up anyway, so I would not worry about these things at that time.