bool setFillerCoords,
gmx_nbnxn_gpu_t *nb,
DeviceBuffer<float> d_x,
+ GpuEventSynchronizer *xReadyOnDevice,
const Nbnxm::AtomLocality locality,
int gridId,
int numColumnsMax)
// TODO: This will only work with CUDA
GMX_ASSERT(d_x, "Need a valid device pointer");
+ // ensure that coordinates are ready on the device before launching the kernel
+ GMX_ASSERT(xReadyOnDevice, "Need a valid GpuEventSynchronizer object");
+ xReadyOnDevice->enqueueWaitEvent(stream);
+
KernelLaunchConfig config;
config.blockSize[0] = c_bufOpsThreadsPerBlock;
config.blockSize[1] = 1;