I know this error has been discussed before, but my question concerns the actual causes of this error. I am currently porting a large chunk of Fortran code to CUDA. As soon as the last syntax error was removed I thought everything would be fine, but no! I got 5 new errors, all of them
fesft.cu(345): error: identifier “__eh_curr_region” is undefined
I ran a search on it in the forum (and on google) and it turned out that this error usually occurs when someone tries to allocate a chunk of shared memory that is not a constant size. I do not do that. Actually I do not fiddle with the memory hierarchies at all so far. This error occurs on the closing brackets of for-loops. I could not find any reference of this error or what “__eh_curr_region” is at all in the documentation. Can anybody help me on this?