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 cuda_affinity_initialize ()
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 75 of file CudaUtils.C.

References cudaCheck.

Referenced by allocate_device().

00075                                                                        {
00076   cudaCheck(cudaMalloc(pp, sizeofT*len));
00077 }

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

Definition at line 65 of file CudaUtils.C.

References cudaCheck.

Referenced by allocate_host().

00065                                                                      {
00066   cudaCheck(cudaMallocHost(pp, sizeofT*len));
00067 }

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

Definition at line 51 of file CudaUtils.C.

References cudaCheck.

Referenced by clear_device_array().

00051                                                                                                         {
00052   cudaCheck(cudaMemsetAsync(data, 0, sizeofT*ndata, stream));
00053 }

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

Definition at line 55 of file CudaUtils.C.

References cudaCheck.

Referenced by clear_device_array_sync().

00055                                                                              {
00056   cudaCheck(cudaMemset(data, 0, sizeofT*ndata));
00057 }

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 275 of file CudaUtils.C.

References cudaCheck.

Referenced by copy3D_DtoD().

00281                                        {
00282   cudaMemcpy3DParms parms = {0};
00283 
00284   parms.srcPos = make_cudaPos(sizeofT*src_x0, src_y0, src_z0);
00285   parms.srcPtr = make_cudaPitchedPtr(src_data, sizeofT*src_xsize, src_xsize, src_ysize);
00286 
00287   parms.dstPos = make_cudaPos(sizeofT*dst_x0, dst_y0, dst_z0);
00288   parms.dstPtr = make_cudaPitchedPtr(dst_data, sizeofT*dst_xsize, dst_xsize, dst_ysize);
00289 
00290   parms.extent = make_cudaExtent(sizeofT*width, height, depth);
00291   parms.kind = cudaMemcpyDeviceToDevice;
00292 
00293   cudaCheck(cudaMemcpy3DAsync(&parms, stream));
00294 }

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 250 of file CudaUtils.C.

References cudaCheck.

Referenced by copy3D_DtoH().

00256                                        {
00257   cudaMemcpy3DParms parms = {0};
00258 
00259   parms.srcPos = make_cudaPos(sizeofT*src_x0, src_y0, src_z0);
00260   parms.srcPtr = make_cudaPitchedPtr(src_data, sizeofT*src_xsize, src_xsize, src_ysize);
00261 
00262   parms.dstPos = make_cudaPos(sizeofT*dst_x0, dst_y0, dst_z0);
00263   parms.dstPtr = make_cudaPitchedPtr(dst_data, sizeofT*dst_xsize, dst_xsize, dst_ysize);
00264 
00265   parms.extent = make_cudaExtent(sizeofT*width, height, depth);
00266   parms.kind = cudaMemcpyDeviceToHost;
00267 
00268   cudaCheck(cudaMemcpy3DAsync(&parms, stream));
00269 }

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 225 of file CudaUtils.C.

References cudaCheck.

Referenced by copy3D_HtoD().

00231                                        {
00232   cudaMemcpy3DParms parms = {0};
00233 
00234   parms.srcPos = make_cudaPos(sizeofT*src_x0, src_y0, src_z0);
00235   parms.srcPtr = make_cudaPitchedPtr(src_data, sizeofT*src_xsize, src_xsize, src_ysize);
00236 
00237   parms.dstPos = make_cudaPos(sizeofT*dst_x0, dst_y0, dst_z0);
00238   parms.dstPtr = make_cudaPitchedPtr(dst_data, sizeofT*dst_xsize, dst_xsize, dst_ysize);
00239 
00240   parms.extent = make_cudaExtent(sizeofT*width, height, depth);
00241   parms.kind = cudaMemcpyHostToDevice;
00242 
00243   cudaCheck(cudaMemcpy3DAsync(&parms, stream));
00244 }

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 300 of file CudaUtils.C.

References cudaCheck.

Referenced by copy3D_PeerDtoD().

00307                                        {
00308   cudaMemcpy3DPeerParms parms = {0};
00309 
00310   parms.srcDevice = src_dev;
00311   parms.dstDevice = dst_dev;
00312 
00313   parms.srcPos = make_cudaPos(sizeofT*src_x0, src_y0, src_z0);
00314   parms.srcPtr = make_cudaPitchedPtr(src_data, sizeofT*src_xsize, src_xsize, src_ysize);
00315 
00316   parms.dstPos = make_cudaPos(sizeofT*dst_x0, dst_y0, dst_z0);
00317   parms.dstPtr = make_cudaPitchedPtr(dst_data, sizeofT*dst_xsize, dst_xsize, dst_ysize);
00318 
00319   parms.extent = make_cudaExtent(sizeofT*width, height, depth);
00320 
00321   cudaCheck(cudaMemcpy3DPeerAsync(&parms, stream));
00322 }

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 202 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_DtoD().

00203                                  {
00204   cudaCheck(cudaMemcpyAsync(d_dst, d_src, sizeofT*array_len, cudaMemcpyDeviceToDevice, stream));
00205 }

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

Definition at line 207 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_DtoD_sync().

00207                                                                                             {
00208   cudaCheck(cudaMemcpy(d_dst, d_src, sizeofT*array_len, cudaMemcpyDeviceToDevice));
00209 }

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 189 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_DtoH().

00190                                  {
00191   cudaCheck(cudaMemcpyAsync(h_array, d_array, sizeofT*array_len, cudaMemcpyDeviceToHost, stream));
00192 }

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

Definition at line 194 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_DtoH_sync().

00194                                                                                                 {
00195   cudaCheck(cudaMemcpy(h_array, d_array, sizeofT*array_len, cudaMemcpyDeviceToHost));
00196 }

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 175 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_HtoD().

00176                                  {
00177   cudaCheck(cudaMemcpyAsync(d_array, h_array, sizeofT*array_len, cudaMemcpyHostToDevice, stream));
00178 }

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

Definition at line 180 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_HtoD_sync().

00181                            {
00182   cudaCheck(cudaMemcpy(d_array, h_array, sizeofT*array_len, cudaMemcpyHostToDevice));
00183 }

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 215 of file CudaUtils.C.

References cudaCheck.

Referenced by copy_PeerDtoD().

00217                         {
00218   cudaCheck(cudaMemcpyPeerAsync(d_dst, dst_dev, d_src, src_dev, sizeofT*array_len, stream));
00219 }

void cuda_affinity_initialize (  ) 

Definition at line 33 of file CudaUtils.C.

Referenced by all_init().

00033                                 {
00034   int devcnt = 0;
00035   cudaError_t err = cudaGetDeviceCount(&devcnt);
00036   if ( devcnt == 1 ) {  // only one device so it must be ours
00037     int *dummy;
00038     if ( err == cudaSuccess ) err = cudaSetDevice(0);
00039     if ( err == cudaSuccess ) err = cudaSetDeviceFlags(cudaDeviceMapHost);
00040     if ( err == cudaSuccess ) err = cudaMalloc(&dummy, 4);
00041   }
00042   if ( err != cudaSuccess ) {
00043     char host[128];
00044     gethostname(host, 128);  host[127] = 0;
00045     fprintf(stderr,"CUDA initialization error on %s: %s\n", host, cudaGetErrorString(err));
00046   }
00047 }

void cudaDie ( const char *  msg,
cudaError_t  err 
)

Definition at line 9 of file CudaUtils.C.

References NAMD_die().

Referenced by cuda_check_local_progress(), cuda_check_pme_charges(), cuda_check_pme_forces(), cuda_check_progress(), cuda_check_remote_progress(), DeviceCUDA::initialize(), and read_CUDA_ARCH().

00009                                                {
00010   char host[128];
00011   gethostname(host, 128);  host[127] = 0;
00012   char devstr[128] = "";
00013   int devnum;
00014   if ( cudaGetDevice(&devnum) == cudaSuccess ) {
00015     sprintf(devstr, " device %d", devnum);
00016   }
00017   cudaDeviceProp deviceProp;
00018   if ( cudaGetDeviceProperties(&deviceProp, devnum) == cudaSuccess ) {
00019     sprintf(devstr, " device %d pci %x:%x:%x", devnum,
00020       deviceProp.pciDomainID, deviceProp.pciBusID, deviceProp.pciDeviceID);
00021   }
00022   char errmsg[1024];
00023   if (err == cudaSuccess) {
00024     sprintf(errmsg,"CUDA error %s on Pe %d (%s%s)", msg, CkMyPe(), host, devstr);
00025   } else {
00026     sprintf(errmsg,"CUDA error %s on Pe %d (%s%s): %s", msg, CkMyPe(), host, devstr, cudaGetErrorString(err));    
00027   }
00028   NAMD_die(errmsg);
00029 }

void cudaNAMD_bug ( const char *  msg  ) 

Definition at line 31 of file CudaUtils.C.

References NAMD_bug().

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

00031 {NAMD_bug(msg);}

void deallocate_device_T ( void **  pp  ) 

Definition at line 84 of file CudaUtils.C.

References cudaCheck.

Referenced by deallocate_device().

00084                                     {
00085   
00086   if (*pp != NULL) {
00087     cudaCheck(cudaFree((void *)(*pp)));
00088     *pp = NULL;
00089   }
00090 
00091 }

void deallocate_host_T ( void **  pp  ) 

Definition at line 98 of file CudaUtils.C.

References cudaCheck.

Referenced by deallocate_host().

00098                                   {
00099   
00100   if (*pp != NULL) {
00101     cudaCheck(cudaFreeHost((void *)(*pp)));
00102     *pp = NULL;
00103   }
00104 
00105 }

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

Definition at line 117 of file CudaUtils.C.

References cudaCheck.

Referenced by reallocate_device().

00117                                                                                                           {
00118 
00119   if (*pp != NULL && *curlen < newlen) {
00120     cudaCheck(cudaFree((void *)(*pp)));
00121     *pp = NULL;
00122   }
00123 
00124   if (*pp == NULL) {
00125     if (fac > 1.0f) {
00126       *curlen = (int)(((double)(newlen))*(double)fac);
00127     } else {
00128       *curlen = newlen;
00129     }
00130     cudaCheck(cudaMalloc(pp, sizeofT*(*curlen)));
00131     return true;
00132   }
00133 
00134   return false;
00135 }

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 150 of file CudaUtils.C.

References cudaCheck.

Referenced by reallocate_host().

00151                                                                                        {
00152 
00153   if (*pp != NULL && *curlen < newlen) {
00154     cudaCheck(cudaFreeHost((void *)(*pp)));
00155     *pp = NULL;
00156   }
00157 
00158   if (*pp == NULL) {
00159     if (fac > 1.0f) {
00160       *curlen = (int)(((double)(newlen))*(double)fac);
00161     } else {
00162       *curlen = newlen;
00163     }
00164     cudaCheck(cudaHostAlloc(pp, sizeofT*(*curlen), flag));
00165     return true;
00166   }
00167 
00168   return false;
00169 }


Generated on 19 Sep 2020 for NAMD by  doxygen 1.6.1