ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@12588 f3b2605a-c512-4ea7-a41b-209d697bcdaa

Browse Source
This commit is contained in:
sjplimp
2014-10-06 22:59:05 +00:00
parent 4bb43ca885
commit 621fa7d600
122 changed files with 1934 additions and 1929 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
lib/cuda/pair_lj_sdk_cuda_kernel_nc.cu
View File

@ -21,28 +21,28 @@
This software is distributed under the GNU General Public License.
------------------------------------------------------------------------- */
__device__ inline F_FLOAT PairLJSDKCuda_Eval(const F_FLOAT &rsq, const int ij_type, F_FLOAT &factor_lj, int &eflag, ENERGY_FLOAT &evdwl) //0.11 of 0.4
__device__ inline F_CFLOAT PairLJSDKCuda_Eval(const F_CFLOAT &rsq, const int ij_type, F_CFLOAT &factor_lj, int &eflag, ENERGY_CFLOAT &evdwl) //0.11 of 0.4
{
const F_FLOAT r2inv = F_F(1.0) / rsq;
const F_CFLOAT r2inv = F_F(1.0) / rsq;
const int lj_type = _lj_type[ij_type];
const F_FLOAT r4inv = r2inv * r2inv;
const F_FLOAT rNinv_first = lj_type != CG_LJ9_6 ? r4inv : _RSQRT_(rsq);
const F_FLOAT rNinv_second = lj_type != CG_LJ12_4 ? -r2inv : -F_F(1.0);
const F_FLOAT forcelj = r4inv * (_lj1[ij_type] * r4inv * rNinv_first + _lj2[ij_type] * rNinv_second);
const F_CFLOAT r4inv = r2inv * r2inv;
const F_CFLOAT rNinv_first = lj_type != CG_LJ9_6 ? r4inv : _RSQRT_(rsq);
const F_CFLOAT rNinv_second = lj_type != CG_LJ12_4 ? -r2inv : -F_F(1.0);
const F_CFLOAT forcelj = r4inv * (_lj1[ij_type] * r4inv * rNinv_first + _lj2[ij_type] * rNinv_second);
if(eflag) evdwl += factor_lj * (r4inv * (_lj3[ij_type] * r4inv * rNinv_first + _lj4[ij_type] * rNinv_second) - _offset[ij_type]);
return factor_lj * forcelj * r2inv;
}
/*__device__ inline F_FLOAT PairLJSDKCuda_Eval(const F_FLOAT& rsq,const int ij_type,F_FLOAT& factor_lj,int& eflag, ENERGY_FLOAT& evdwl)
/*__device__ inline F_CFLOAT PairLJSDKCuda_Eval(const F_CFLOAT& rsq,const int ij_type,F_CFLOAT& factor_lj,int& eflag, ENERGY_CFLOAT& evdwl)
{
const int lj_type = tex1Dfetch(_coeff5_gm_tex,ij_type);
const F_FLOAT r2inv = F_F(1.0)/rsq;
const F_FLOAT r4inv = r2inv*r2inv;
const F_FLOAT rNinv_first = lj_type!=CG_LJ9_6?r4inv:_RSQRT_(rsq);
const F_FLOAT rNinv_second = lj_type!=CG_LJ12_4?r2inv:F_F(1.0);
const F_FLOAT forcelj = r4inv * (tex1Dfetch(_coeff1_gm_tex,ij_type)*r4inv*rNinv_first - tex1Dfetch(_coeff2_gm_tex,ij_type)*rNinv_second);
const F_CFLOAT r2inv = F_F(1.0)/rsq;
const F_CFLOAT r4inv = r2inv*r2inv;
const F_CFLOAT rNinv_first = lj_type!=CG_LJ9_6?r4inv:_RSQRT_(rsq);
const F_CFLOAT rNinv_second = lj_type!=CG_LJ12_4?r2inv:F_F(1.0);
const F_CFLOAT forcelj = r4inv * (tex1Dfetch(_coeff1_gm_tex,ij_type)*r4inv*rNinv_first - tex1Dfetch(_coeff2_gm_tex,ij_type)*rNinv_second);
if(eflag) evdwl += factor_lj*(r4inv*(tex1Dfetch(_coeff3_gm_tex,ij_type)*r4inv*rNinv_first-tex1Dfetch(_coeff4_gm_tex,ij_type)*rNinv_second));
return factor_lj*forcelj*r2inv;