Override any OpenCL fast math JIT settings for born/coul/wolf{/cs}/gpu to resolve numerical deviations seen with some OpenCL implementations.
This commit is contained in:
@ -57,7 +57,7 @@ int BornCoulWolfT::init(const int ntypes, double **host_cutsq, double **host_rho
|
||||
const double alf, const double e_shift, const double f_shift) {
|
||||
int success;
|
||||
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split,
|
||||
_screen,born_coul_wolf,"k_born_coul_wolf");
|
||||
_screen,born_coul_wolf,"k_born_coul_wolf",1);
|
||||
if (success!=0)
|
||||
return success;
|
||||
|
||||
|
||||
Reference in New Issue
Block a user