OilfieldToolsNet/LBPM
4
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

atomic add for GPU <6.0

Browse Source
This commit is contained in:
James E McClure 2018-08-28 20:45:12 -04:00
parent b06380b894
commit d17c59d772
2 changed files with 28 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
gpu/D3Q19.cu
View File

@ -30,6 +30,20 @@ __constant__ __device__ double mrt_V12=0.04166666666666666;
//__shared__ double Transform[722]=
// {};
#if !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 600
#else
__device__ double atomicAdd(double* address, double val) {
unsigned long long int* address_as_ull = (unsigned long long int*)address;
unsigned long long int old = *address_as_ull, assumed;
do {
assumed = old;
old = atomicCAS(address_as_ull, assumed, __double_as_longlong(val+__longlong_as_double(assumed)));
} while (assumed != old);
return __longlong_as_double(old);
}
#endif
using namespace cooperative_groups;
__device__ double reduce_sum(thread_group g, double *temp, double val)
{

14
gpu/dfh.cu
View File

@ -5,6 +5,20 @@
#define NBLOCKS 1024
#define NTHREADS 256
#if !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 600
#else
__device__ double atomicAdd(double* address, double val) {
unsigned long long int* address_as_ull = (unsigned long long int*)address;
unsigned long long int old = *address_as_ull, assumed;
do {
assumed = old;
old = atomicCAS(address_as_ull, assumed, __double_as_longlong(val+__longlong_as_double(assumed)));
} while (assumed != old);
return __longlong_as_double(old);
}
#endif
__global__ void dvc_ScaLBL_Gradient_Unpack(double weight, double Cqx, double Cqy, double Cqz,
int *list, int start, int count, double *recvbuf, double *phi, double *grad, int N){
//....................................................................................