msmpot_cuda.h

00001 /***************************************************************************
00002 *cr
00003 *cr (C) Copyright 2008-2009 The Board of Trustees of the
00004 *cr University of Illinois
00005 *cr All Rights Reserved
00006 *cr
00007 ***************************************************************************/
00008 
00009 /***************************************************************************
00010 * RCS INFORMATION:
00011 *
00012 * $RCSfile: msmpot_cuda.h,v $
00013 * $Author: johns $ $Locker: $ $State: Exp $
00014 * $Revision: 1.3 $ $Date: 2010年06月03日 20:07:10 $
00015 *
00016 ***************************************************************************/
00017 
00018 #include "msmpot_internal.h"
00019 
00020 
00021 #ifndef MSMPOT_MSMCUDA_H
00022 #define MSMPOT_MSMCUDA_H
00023 
00024 /*
00025 * detect and report error from CUDA
00026 */
00027 #undef CUERR
00028 #define CUERR(errnum) \
00029 do { \
00030 cudaError_t cuerr = cudaGetLastError(); \
00031 if (cuerr != cudaSuccess) { \
00032 return ERROR(errnum); \
00033 } \
00034 } while (0)
00035 
00036 
00037 /*
00038 * Keep NBRLIST_MAXLEN of 3-tuples in GPU const cache memory:
00039 * (3 * 5333) ints + 1 int (giving use length) == 64000 bytes
00040 */
00041 #undef NBRLIST_MAXLEN
00042 #define NBRLIST_MAXLEN 5333
00043 
00044 
00045 #ifdef __cplusplus
00046 extern "C" {
00047 #endif
00048 
00049 struct MsmpotCuda_t {
00050 Msmpot *msmpot;
00051 
00052 /* get CUDA device info */
00053 #if 0
00054 struct cudaDeviceProp *dev;
00055 int ndevs;
00056 #endif
00057 int devnum; /* device number */
00058 
00059 /* CUDA short-range part ("binsmall") */
00060 int pmx, pmy, pmz; /* dimensions of padded epotmap */
00061 long maxpm; /* allocated points for padded map */ 
00062 float *padmap; /* padded epotmap for CUDA grid */
00063 
00064 float *dev_padmap; /* points to device memory */
00065 long dev_maxpm; /* allocated points on device */
00066 
00067 float4 *dev_bin; /* points to device memory */
00068 int dev_nbins; /* allocated bins on device */
00069 
00070 /* CUDA lattice cutoff */
00071 int lk_nlevels; /* number of levels for latcut kernel */
00072 int lk_srad; /* subcube radius for latcut kernel */
00073 int lk_padding; /* padding around internal array of subcubes */
00074 int subcube_total; /* total number of subcubes for compressed grids */
00075 int block_total; /* total number of thread blocks */
00076 /*
00077 * host_ --> memory allocated on host
00078 * device_ --> global memory allocated on device
00079 */
00080 int *host_sinfo; /* subcube info copy to device const mem */
00081 float *host_lfac; /* level factor copy to device const mem */
00082 int maxlevels;
00083 
00084 float *host_wt; /* weights copy to device const mem */
00085 int maxwts;
00086 
00087 float *host_qgrids; /* q-grid subcubes copy to device global mem */
00088 float *host_egrids; /* e-grid subcubes copy to device global mem */
00089 float *device_qgrids; /* q-grid subcubes allocate on device */
00090 float *device_egrids; /* e-grid subcubes allocate on device */
00091 long maxgridpts;
00092 
00093 };
00094 
00095 void Msmpot_cuda_cleanup(MsmpotCuda *);
00096 
00097 int Msmpot_cuda_setup_shortrng(MsmpotCuda *);
00098 void Msmpot_cuda_cleanup_shortrng(MsmpotCuda *);
00099 
00100 int Msmpot_cuda_setup_latcut(MsmpotCuda *);
00101 void Msmpot_cuda_cleanup_latcut(MsmpotCuda *);
00102 
00103 #ifdef __cplusplus
00104 }
00105 #endif
00106 
00107 
00108 #endif /* MSMPOT_MSMCUDA_H */