CudaUtils.C File Reference

#include <stdio.h>
#include "common.h"
#include "charm++.h"
#include "CudaUtils.h"

Go to the source code of this file.

Functions

void cudaDie (const char *msg, cudaError_t err)
void cudaNAMD_bug (const char *msg)
void clear_device_array_async_T (void *data, const int ndata, cudaStream_t stream, const size_t sizeofT)
void clear_device_array_T (void *data, const int ndata, const size_t sizeofT)
void allocate_host_T (void **pp, const int len, const size_t sizeofT)
void allocate_device_T (void **pp, const int len, const size_t sizeofT)
void deallocate_device_T (void **pp)
void deallocate_host_T (void **pp)
bool reallocate_device_T (void **pp, int *curlen, const int newlen, const float fac, const size_t sizeofT)
bool reallocate_host_T (void **pp, int *curlen, const int newlen, const float fac, const unsigned int flag, const size_t sizeofT)
void copy_HtoD_async_T (const void *h_array, void *d_array, int array_len, cudaStream_t stream, const size_t sizeofT)
void copy_HtoD_T (const void *h_array, void *d_array, int array_len, const size_t sizeofT)
void copy_DtoH_async_T (const void *d_array, void *h_array, const int array_len, cudaStream_t stream, const size_t sizeofT)
void copy_DtoH_T (const void *d_array, void *h_array, const int array_len, const size_t sizeofT)
void copy_DtoD_async_T (const void *d_src, void *d_dst, const int array_len, cudaStream_t stream, const size_t sizeofT)
void copy_DtoD_T (const void *d_src, void *d_dst, const int array_len, const size_t sizeofT)
void copy_PeerDtoD_async_T (const int src_dev, const int dst_dev, const void *d_src, void *d_dst, const int array_len, cudaStream_t stream, const size_t sizeofT)
void copy3D_HtoD_T (void *src_data, void *dst_data, int src_x0, int src_y0, int src_z0, size_t src_xsize, size_t src_ysize, int dst_x0, int dst_y0, int dst_z0, size_t dst_xsize, size_t dst_ysize, size_t width, size_t height, size_t depth, size_t sizeofT, cudaStream_t stream)
void copy3D_DtoH_T (void *src_data, void *dst_data, int src_x0, int src_y0, int src_z0, size_t src_xsize, size_t src_ysize, int dst_x0, int dst_y0, int dst_z0, size_t dst_xsize, size_t dst_ysize, size_t width, size_t height, size_t depth, size_t sizeofT, cudaStream_t stream)
void copy3D_DtoD_T (void *src_data, void *dst_data, int src_x0, int src_y0, int src_z0, size_t src_xsize, size_t src_ysize, int dst_x0, int dst_y0, int dst_z0, size_t dst_xsize, size_t dst_ysize, size_t width, size_t height, size_t depth, size_t sizeofT, cudaStream_t stream)
void copy3D_PeerDtoD_T (int src_dev, int dst_dev, void *src_data, void *dst_data, int src_x0, int src_y0, int src_z0, size_t src_xsize, size_t src_ysize, int dst_x0, int dst_y0, int dst_z0, size_t dst_xsize, size_t dst_ysize, size_t width, size_t height, size_t depth, size_t sizeofT, cudaStream_t stream)


Function Documentation

void allocate_device_T ( void **  pp,
const int  len,
const size_t  sizeofT 
)

Definition at line 58 of file CudaUtils.C.

References cudaCheck.

Referenced by allocate_device().

00058                                                                        {
00059   cudaCheck(cudaMalloc(pp, sizeofT*len));
00060 }

void allocate_host_T ( void **  pp,
const int  len,
const size_t  sizeofT 
)

Definition at line 48 of file CudaUtils.C.

References cudaCheck.

Referenced by allocate_host().

00048                                                                      {
00049   cudaCheck(cudaMallocHost(pp, sizeofT*len));
00050 }

void clear_device_array_async_T ( void *  data,
const int  ndata,
cudaStream_t  stream,
const size_t  sizeofT 
)

Definition at line 34 of file CudaUtils.C.

References cudaCheck.

Referenced by clear_device_array().

00034                                                                                                         {
00035   cudaCheck(cudaMemsetAsync(data, 0, sizeofT*ndata, stream));
00036 }

void clear_device_array_T ( void *  data,
const int  ndata,
const size_t  sizeofT 
)

Definition at line 38 of file CudaUtils.C.

References cudaCheck.

Referenced by clear_device_array_sync().

00038                                                                              {
00039   cudaCheck(cudaMemset(data, 0, sizeofT*ndata));
00040 }

void copy3D_DtoD_T ( void *  src_data,
void *  dst_data,
int  src_x0,
int  src_y0,
int  src_z0,
size_t  src_xsize,
size_t  src_ysize,
int  dst_x0,
int  dst_y0,
int  dst_z0,
size_t  dst_xsize,
size_t  dst_ysize,
size_t  width,
size_t  height,
size_t  depth,
size_t  sizeofT,
cudaStream_t  stream 
)

Definition at line 258 of file CudaUtils.C.

References cudaCheck.

Referenced by copy3D_DtoD().

00264                                        {
00265   cudaMemcpy3DParms parms = {0};
00266 
00267   parms.srcPos = make_cudaPos(sizeofT*src_x0, src_y0, src_z0);
00268   parms.srcPtr = make_cudaPitchedPtr(src_data, sizeofT*src_xsize, src_xsize, src_ysize);
00269 
00270   parms.dstPos = make_cudaPos(sizeofT*dst_x0, dst_y0, dst_z0);
00271   parms.dstPtr = make_cudaPitchedPtr(dst_data, sizeofT*dst_xsize, dst_xsize, dst_ysize);
00272 
00273   parms.extent = make_cudaExtent(sizeofT*width, height, depth);
00274   parms.kind = cudaMemcpyDeviceToDevice;
00275 
00276   cudaCheck(cudaMemcpy3DAsync(&parms, stream));
00277 }

void copy3D_DtoH_T ( void *  src_data,
void *  dst_data,
int  src_x0,
int  src_y0,
int  src_z0,
size_t  src_xsize,
size_t  src_ysize,
int  dst_x0,
int  dst_y0,
int  dst_z0,
size_t  dst_xsize,
size_t  dst_ysize,
size_t  width,
size_t  height,
size_t  depth,
size_t  sizeofT,
cudaStream_t  stream 
)

Definition at line 233 of file CudaUtils.C.

References cudaCheck.

Referenced by copy3D_DtoH().

00239                                        {
00240   cudaMemcpy3DParms parms = {0};
00241 
00242   parms.srcPos = make_cudaPos(sizeofT*src_x0, src_y0, src_z0);
00243   parms.srcPtr = make_cudaPitchedPtr(src_data, sizeofT*src_xsize, src_xsize, src_ysize);
00244 
00245   parms.dstPos = make_cudaPos(sizeofT*dst_x0, dst_y0, dst_z0);
00246   parms.dstPtr = make_cudaPitchedPtr(dst_data, sizeofT*dst_xsize, dst_xsize, dst_ysize);
00247 
00248   parms.extent = make_cudaExtent(sizeofT*width, height, depth);
00249   parms.kind = cudaMemcpyDeviceToHost;
00250 
00251   cudaCheck(cudaMemcpy3DAsync(&parms, stream));
00252 }

void copy3D_HtoD_T ( void *  src_data,
void *  dst_data,
int  src_x0,
int  src_y0,
int  src_z0,
size_t  src_xsize,
size_t  src_ysize,
int  dst_x0,
int  dst_y0,
int  dst_z0,
size_t  dst_xsize,
size_t  dst_ysize,
size_t  width,
size_t  height,
size_t  depth,
size_t  sizeofT,
cudaStream_t  stream 
)

Definition at line 208 of file CudaUtils.C.

References cudaCheck.

Referenced by copy3D_HtoD().

00214                                        {
00215   cudaMemcpy3DParms parms = {0};
00216 
00217   parms.srcPos = make_cudaPos(sizeofT*src_x0, src_y0, src_z0);
00218   parms.srcPtr = make_cudaPitchedPtr(src_data, sizeofT*src_xsize, src_xsize, src_ysize);
00219 
00220   parms.dstPos = make_cudaPos(sizeofT*dst_x0, dst_y0, dst_z0);
00221   parms.dstPtr = make_cudaPitchedPtr(dst_data, sizeofT*dst_xsize, dst_xsize, dst_ysize);
00222 
00223   parms.extent = make_cudaExtent(sizeofT*width, height, depth);
00224   parms.kind = cudaMemcpyHostToDevice;
00225 
00226   cudaCheck(cudaMemcpy3DAsync(&parms, stream));
00227 }

void copy3D_PeerDtoD_T ( int  src_dev,
int  dst_dev,
void *  src_data,
void *  dst_data,
int  src_x0,
int  src_y0,
int  src_z0,
size_t  src_xsize,
size_t  src_ysize,
int  dst_x0,
int  dst_y0,
int  dst_z0,
size_t  dst_xsize,
size_t  dst_ysize,
size_t  width,
size_t  height,
size_t  depth,
size_t  sizeofT,
cudaStream_t  stream 
)

Definition at line 283 of file CudaUtils.C.

References cudaCheck.

Referenced by copy3D_PeerDtoD().

00290                                        {
00291   cudaMemcpy3DPeerParms parms = {0};
00292 
00293   parms.srcDevice = src_dev;
00294   parms.dstDevice = dst_dev;
00295 
00296   parms.srcPos = make_cudaPos(sizeofT*src_x0, src_y0, src_z0);
00297   parms.srcPtr = make_cudaPitchedPtr(src_data, sizeofT*src_xsize, src_xsize, src_ysize);
00298 
00299   parms.dstPos = make_cudaPos(sizeofT*dst_x0, dst_y0, dst_z0);
00300   parms.dstPtr = make_cudaPitchedPtr(dst_data, sizeofT*dst_xsize, dst_xsize, dst_ysize);
00301 
00302   parms.extent = make_cudaExtent(sizeofT*width, height, depth);
00303 
00304   cudaCheck(cudaMemcpy3DPeerAsync(&parms, stream));
00305 }

void copy_DtoD_async_T ( const void *  d_src,
void *  d_dst,
const int  array_len,
cudaStream_t  stream,
const size_t  sizeofT 
)

Definition at line 185 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_DtoD().

00186                                  {
00187   cudaCheck(cudaMemcpyAsync(d_dst, d_src, sizeofT*array_len, cudaMemcpyDeviceToDevice, stream));
00188 }

void copy_DtoD_T ( const void *  d_src,
void *  d_dst,
const int  array_len,
const size_t  sizeofT 
)

Definition at line 190 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_DtoD_sync().

00190                                                                                             {
00191   cudaCheck(cudaMemcpy(d_dst, d_src, sizeofT*array_len, cudaMemcpyDeviceToDevice));
00192 }

void copy_DtoH_async_T ( const void *  d_array,
void *  h_array,
const int  array_len,
cudaStream_t  stream,
const size_t  sizeofT 
)

Definition at line 172 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_DtoH().

00173                                  {
00174   cudaCheck(cudaMemcpyAsync(h_array, d_array, sizeofT*array_len, cudaMemcpyDeviceToHost, stream));
00175 }

void copy_DtoH_T ( const void *  d_array,
void *  h_array,
const int  array_len,
const size_t  sizeofT 
)

Definition at line 177 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_DtoH_sync().

00177                                                                                                 {
00178   cudaCheck(cudaMemcpy(h_array, d_array, sizeofT*array_len, cudaMemcpyDeviceToHost));
00179 }

void copy_HtoD_async_T ( const void *  h_array,
void *  d_array,
int  array_len,
cudaStream_t  stream,
const size_t  sizeofT 
)

Definition at line 158 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_HtoD().

00159                                  {
00160   cudaCheck(cudaMemcpyAsync(d_array, h_array, sizeofT*array_len, cudaMemcpyHostToDevice, stream));
00161 }

void copy_HtoD_T ( const void *  h_array,
void *  d_array,
int  array_len,
const size_t  sizeofT 
)

Definition at line 163 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_HtoD_sync().

00164                            {
00165   cudaCheck(cudaMemcpy(d_array, h_array, sizeofT*array_len, cudaMemcpyHostToDevice));
00166 }

void copy_PeerDtoD_async_T ( const int  src_dev,
const int  dst_dev,
const void *  d_src,
void *  d_dst,
const int  array_len,
cudaStream_t  stream,
const size_t  sizeofT 
)

Definition at line 198 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_PeerDtoD().

00200                         {
00201   cudaCheck(cudaMemcpyPeerAsync(d_dst, dst_dev, d_src, src_dev, sizeofT*array_len, stream));
00202 }

void cudaDie ( const char *  msg,
cudaError_t  err 
)

Definition at line 9 of file CudaUtils.C.

References NAMD_die().

Referenced by DeviceCUDA::initialize(), and read_CUDA_ARCH().

00009                                                {
00010   char host[128];
00011 #ifdef NOHOSTNAME
00012   sprintf(host,"physical node %d", CmiPhysicalNodeID(CkMyPe()));
00013 #else
00014   gethostname(host, 128);  host[127] = 0;
00015 #endif
00016   char devstr[128] = "";
00017   int devnum;
00018   if ( cudaGetDevice(&devnum) == cudaSuccess ) {
00019     sprintf(devstr, " device %d", devnum);
00020   }
00021   char errmsg[1024];
00022   if (err == cudaSuccess) {
00023     sprintf(errmsg,"CUDA error %s on Pe %d (%s%s)", msg, CkMyPe(), host, devstr);
00024   } else {
00025     sprintf(errmsg,"CUDA error %s on Pe %d (%s%s): %s", msg, CkMyPe(), host, devstr, cudaGetErrorString(err));    
00026   }
00027   NAMD_die(errmsg);
00028 }

void cudaNAMD_bug ( const char *  msg  ) 

Definition at line 30 of file CudaUtils.C.

References NAMD_bug().

Referenced by CudaFFTCompute::backward(), CudaFFTCompute::forward(), gather_force(), and spread_charge().

00030 {NAMD_bug(msg);}

void deallocate_device_T ( void **  pp  ) 

Definition at line 67 of file CudaUtils.C.

References cudaCheck.

Referenced by deallocate_device().

00067                                     {
00068   
00069   if (*pp != NULL) {
00070     cudaCheck(cudaFree((void *)(*pp)));
00071     *pp = NULL;
00072   }
00073 
00074 }

void deallocate_host_T ( void **  pp  ) 

Definition at line 81 of file CudaUtils.C.

References cudaCheck.

Referenced by deallocate_host().

00081                                   {
00082   
00083   if (*pp != NULL) {
00084     cudaCheck(cudaFreeHost((void *)(*pp)));
00085     *pp = NULL;
00086   }
00087 
00088 }

bool reallocate_device_T ( void **  pp,
int *  curlen,
const int  newlen,
const float  fac,
const size_t  sizeofT 
)

Definition at line 100 of file CudaUtils.C.

References cudaCheck, and f.

Referenced by reallocate_device().

00100                                                                                                           {
00101 
00102   if (*pp != NULL && *curlen < newlen) {
00103     cudaCheck(cudaFree((void *)(*pp)));
00104     *pp = NULL;
00105   }
00106 
00107   if (*pp == NULL) {
00108     if (fac > 1.0f) {
00109       *curlen = (int)(((double)(newlen))*(double)fac);
00110     } else {
00111       *curlen = newlen;
00112     }
00113     cudaCheck(cudaMalloc(pp, sizeofT*(*curlen)));
00114     return true;
00115   }
00116 
00117   return false;
00118 }

bool reallocate_host_T ( void **  pp,
int *  curlen,
const int  newlen,
const float  fac,
const unsigned int  flag,
const size_t  sizeofT 
)

Definition at line 133 of file CudaUtils.C.

References cudaCheck, and f.

Referenced by reallocate_host().

00134                                                                                        {
00135 
00136   if (*pp != NULL && *curlen < newlen) {
00137     cudaCheck(cudaFreeHost((void *)(*pp)));
00138     *pp = NULL;
00139   }
00140 
00141   if (*pp == NULL) {
00142     if (fac > 1.0f) {
00143       *curlen = (int)(((double)(newlen))*(double)fac);
00144     } else {
00145       *curlen = newlen;
00146     }
00147     cudaCheck(cudaHostAlloc(pp, sizeofT*(*curlen), flag));
00148     return true;
00149   }
00150 
00151   return false;
00152 }


Generated on Mon Nov 20 01:17:15 2017 for NAMD by  doxygen 1.4.7