Cause of error "Caught "std::exception" Exception message is: basic_string::substr: __pos (which is 18446744073709551615) > this->size() (which is 52)"

I get the following error message when using 'arrayfun()' to evaluate a function on the gpu. (I have often used this command without this error)

Caught "std::exception" Exception message is:
basic_string::substr: __pos (which is 18446744073709551615) > this->size() (which is 52)

I cannot understand the cause. Any ideas?

[Walter Roberson]

that number is negative one in 64 bits, and probably indicates that an allocation failed such as if you ran out of gpu memory

[Joss Knight]

This sounds like a bug, can you provide the code to reproduce this? Thanks.

[Robert Kirkby]

Code follows, but it is rather complex. I will try and create a stripped back version tomorrow.

[Is being run on Matlab 2020b in Kubuntu 20.04.2 LTS, gpu is an Nvidia Quadro P4000.]

The code is 'ConesaKitaoKrueger2009'. You will need both the attached and also 'Value Function Iteration (VFI) Toolkit for Matlab' (https://au.mathworks.com/matlabcentral/fileexchange/53478-value-function-iteration-vfi-toolkit-for-matlab).

The actual problem occurs when trying to evaluate 'ConesaKitaoKrueger2009_ReturnFn' on gpu grid using arrayfun() in line 113 of 'CreateReturnFnMatrix_Case1_Disc_Par2.m'.

[Line 188 of ConesaKitaoKrueger2009 calls ValueFnIter_Case1_FHorz_PType line 78 of which calls ValueFnIter_Case1_FHorz line 243 of which calls ValueFnIter_Case1_FHorz_raw line 47 of which calls CreateReturnFnMatrix_Case1_Disc_Par2 line 113 of which makes the aforementioned call to arrayfun() which results in the error.]

I will try and find time tomorrow to create a simpler version of this that results in the same error (without all the intermediate functions).

PS. I have used the same arrayfun() call many times will this VFI Toolkit without any problems. The novel parts versus previous calls are the ConesaKitaoKrueger2009 and ConesaKitaoKrueger2009_ReturnFn scripts, plus some minor updates to ValueFnIter_Case1_FHorz_PType.

[Robert Kirkby]

Attached is a shorter version of code that is enough to generate the error. I have attached the two scripts required. (No need for VFI Toolkit, I copy-pasted the minimum required lines.)

Not sure if it is relevant but when Matlab starts up it says "Warning: Name is nonexistent or not a directory: /tmp/Editor_xadmj "

[Joss Knight]

Thanks, I'll check it out.

[Joss Knight]

Unfortunately I cannot reproduce this. You might have to capture the value of the variables used in the arrayfun call in a mat-file so I can be sure of getting the exact same numbers.

This isn't the first problem we've seen on Ubuntu 20. It might be worth downgrading your GPU driver to a 455 to see if that's the problem. You can start here to search for archived drivers for your GPU.