Skip to main content

Browse the archive

https://github.com/N-BodyShop/changa
06 April 2024, 21:33:40 UTC
Revision 1a58a05e378186ae657a4ede9cfff00362485aef authored by Tom Quinn on 07 August 2018, 23:09:30 UTC, committed by Tim Haines on 02 October 2018, 06:15:11 UTC 
allocatePinnedHostMemory(): use size_t.
 
This allows larger than 2GB buffers to be allocated for the GPU.
Also use size_t in GPU Transfer functions.

Change-Id: I36bb3ec4156e4f7790ad24d0fc172b134a196c7e
1 parent b6ba8d6
Raw File
Tip revision: 1a58a05e378186ae657a4ede9cfff00362485aef authored by Tom Quinn on 07 August 2018, 23:09:30 UTC
allocatePinnedHostMemory(): use size_t.
Tip revision: 1a58a05
CudaFunctions.h 
#ifndef _CUDAFUNCTIONS_H_
#define _CUDAFUNCTIONS_H_

#ifdef CUDA
#include "hapi.h"
#include "HostCUDA.h"
void TreePieceCellListDataTransferBasic(CudaRequest *data, hapiWorkRequest *wr);
void TreePiecePartListDataTransferBasic(CudaRequest *data, hapiWorkRequest *wr);

#ifdef GPU_LOCAL_TREE_WALK
__global__ void gpuLocalTreeWalk(
    CompactPartData *particleCores,
    VariablePartData *particleVars,
    CudaMultipoleMoments* moments,
    int firstParticle,
    int lastParticle,
    int rootIdx,
    cudatype theta,
    cudatype thetaMono,
    int nReplicas,
    cudatype fperiod,
    cudatype fperiodY,
    cudatype fperiodZ);
#endif //GPU_LOCAL_TREE_WALK

__global__ void nodeGravityComputation(
		CompactPartData *particleCores,
		VariablePartData *particleVars,
		CudaMultipoleMoments *moments,
		ILCell *ils,
		int *ilmarks,
		int *bucketStarts,
		int *bucketSizes,
		cudatype fperiod
                );

#ifdef CUDA_2D_TB_KERNEL
__global__ void particleGravityComputation(
                CompactPartData *targetCores,
                VariablePartData *targetVars,
                CompactPartData *sourceCores,
                ILCell *ils,
                int *ilmarks,
		int *bucketStarts,
		int *bucketSizes,
		cudatype fperiod);
#else
__global__ void particleGravityComputation(
                CompactPartData *targetCores,
                VariablePartData *targetVars,
                CompactPartData *sourceCores,
                ILPart *ils,
                int *ilmarks,
		int *bucketStarts,
		int *bucketSizes,
		cudatype fperiod);
#endif

#endif

#endif
back to top