I had a similar issue in the past. We discussed this a while ago. The problem may occur if your data show excessive collinearity, for instance. If this happens more than 10 times within the 999 bootstrap replications, gretl breaks the loop and gives this error messages.

First, I circumenvented this issue by increasing the number of possible unstable cases in the C-code, before I programmed it myself. Alternatively, one could increase the numbers of non-convergent cases to, let's say, 99 or so.