Fix: Segmentation fault whenever more than 2 CVs are used for METAD w… #399
Conversation
…ith REWEIGHTING_NGRID plumed#328 I tried to interpret and fix the code and for me it works nicely with 1 to 3 CVs. Here a brief description: the index i spans through the whole reweghting-grid (according to the mpi rank in case of mpi). rewf_grid_ contains the size on each dimension of the reweghting-grid. t_index is the array containing the coordinates of i in each dimension of the grid. The indexes had to be changed from i to j in line 1967, which indeed spans correctly the dimensions of the grid. In 1968 in order to get the correct index, we need to subtract the former index (ncv-2 -- instead of "ncv-1" which is yet not assigned -- from what is left of kk). I didn't attempt any further test. Please let me know if I had misinterpreted the code.
|
@valsson I think you first implemented these lines. Could you double check the change? If this is correct, I guess it should be backported to 2.4 at least (and perhaps 2.3, though we just declared it as non-supported). |
|
There is for sure some bug with the c(t) calculation for 3 or more CVs that has to be fixed. Now, it is actually @gtribello that implemented the integration where the bug/fix is. So it is better if he would take a look at the exact changes. Furthermore, I think it would be safer to change the unsigned int on lines 1964 and 1966 to Grid::index_t. Looking a travis, is seems that the change breaks the regtest (rt67) for 2 CVs, there are some small numerical difference. |
|
Hello all Sorry for taking a while to look at this. The code definitely looks like it is a proper fix for what is there. I am happy to merge this pull request. Thanks |
|
Since this is a bug fix, I'll merge it to v2.3 and fix the necessary regtests |
Description
Bugfix: Segmentation fault whenever more than 2 CVs are used for METAD with REWEIGHTING_NGRID #328
Target release
I would like my code to appear in release 2.5
Type of contribution
Copyright
COPYRIGHTfile with the correct license information. I also used the commandcd src && ./header.sh mymodulenamein order to make sure the headers of the module are correct.Tests