moebiusband/MD-Bench
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Adjust NVCC flags to avoid issues with atomicAdd with doubles

Browse Source 
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
This commit is contained in:
Rafael Ravedutti
2022-11-14 18:01:46 +01:00
parent c70ebce4c1
commit 93188d1383
2 changed files with 10 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
gromacs/cuda/force_lj.cu
View File

@@ -196,10 +196,10 @@ __global__ void computeForceLJ_cuda_warp(MD_FLOAT *cuda_cl_x, MD_FLOAT *cuda_cl_
int cond;
#if CLUSTER_M == CLUSTER_N
cond = half_neigh ? (ci_cj0 != cj || cii_pos < cjj_pos) :
(ci_cj0 != cj || cii_pos != cjj_pos);
(ci_cj0 != cj || cii_pos != cjj_pos);
#elif CLUSTER_M < CLUSTER_N
cond = half_neigh ? (ci_cj0 != cj || cii_pos + CLUSTER_M * (ci_pos & 0x1) < cjj_pos) :
(ci_cj0 != cj || cii_pos + CLUSTER_M * (ci_pos & 0x1) != cjj_pos);
(ci_cj0 != cj || cii_pos + CLUSTER_M * (ci_pos & 0x1) != cjj_pos);
#endif
if(cond) {
MD_FLOAT delx = xtmp - cj_x[CL_X_OFFSET + cjj_pos];