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

Attempted to resolve the memory access runtime errors when acquiring single and mixed precision arrays from the GPU lib

Browse Source
This commit is contained in:
Trung Nguyen
2023-01-16 10:12:42 -06:00
parent 9dc0369cee
commit 973b46a907
3 changed files with 146 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
lib/gpu/lal_amoeba.cu
View File

@ -1631,9 +1631,9 @@ __kernel void k_amoeba_fphi_uind(const __global numtyp4 *restrict thetai1,
const __global numtyp4 *restrict thetai3,
const __global int *restrict igrid,
const __global numtyp2 *restrict grid,
__global numtyp *restrict fdip_phi1,
__global numtyp *restrict fdip_phi2,
__global numtyp *restrict fdip_sum_phi,
__global acctyp *restrict fdip_phi1,
__global acctyp *restrict fdip_phi2,
__global acctyp *restrict fdip_sum_phi,
const int bsorder, const int inum,
const int nzlo_out, const int nylo_out,
const int nxlo_out, const int ngridxy,
@ -1843,7 +1843,7 @@ __kernel void k_amoeba_fphi_uind(const __global numtyp4 *restrict thetai1,
}
int idx;
numtyp fdip_buf[20];
acctyp fdip_buf[20];
fdip_buf[0] = (numtyp)0.0;
fdip_buf[1] = tuv100_1;
@ -1917,7 +1917,7 @@ __kernel void k_amoeba_fphi_mpole(const __global numtyp4 *restrict thetai1,
const __global numtyp4 *restrict thetai3,
const __global int *restrict igrid,
const __global numtyp2 *restrict grid,
__global numtyp *restrict fphi,
__global acctyp *restrict fphi,
const int bsorder, const int inum, const numtyp felec,
const int nzlo_out, const int nylo_out,
const int nxlo_out, const int ngridxy,

2
lib/gpu/lal_base_amoeba.h
View File

@ -250,7 +250,7 @@ class BaseAmoeba {
UCL_Vector<numtyp4,numtyp4> _thetai1, _thetai2, _thetai3;
UCL_Vector<int,int> _igrid;
UCL_Vector<numtyp2,numtyp2> _cgrid_brick;
UCL_Vector<numtyp,numtyp> _fdip_phi1, _fdip_phi2, _fdip_sum_phi;
UCL_Vector<acctyp,acctyp> _fdip_phi1, _fdip_phi2, _fdip_sum_phi;
int _max_thetai_size;
int _nzlo_out, _nzhi_out, _nylo_out, _nyhi_out, _nxlo_out, _nxhi_out;
int _ngridx, _ngridy, _ngridz, _num_grid_points;