jiri/LA_library
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Implemented deep copyonwrite() for nested types

Browse Source
This commit is contained in:
Jiri Pittner
2021-05-23 10:28:10 +02:00
parent 88628fb306
commit 80eb98411f
5 changed files with 55 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
mat.h
View File

@@ -1036,16 +1036,31 @@ void NRMat<T>::copyonwrite(bool detachonly) {
#ifdef MATPTR
T **newv = new T*[nn];
newv[0] = new T[(size_t)mm*nn];
if(!detachonly) memcpy(newv[0], v[0], (size_t)mm*nn*sizeof(T));
if(!detachonly)
{
if(LA_traits<T>::is_plaindata()) memcpy(newv[0], v[0], (size_t)mm*nn*sizeof(T));
else
{
for(int i=0; i<nn*mm; ++i) {newv[i]=v[i]; LA_traits<T>::copyonwrite(newv[i]);}
}
}
v = newv;
for(register int i=1; i<nn; i++) v[i] = v[i-1] + mm;
#else
T *newv = new T[(size_t)mm*nn];
if(!detachonly) memcpy(newv, v, (size_t)mm*nn*sizeof(T));
if(!detachonly)
{
if(LA_traits<T>::is_plaindata()) memcpy(newv, v, (size_t)mm*nn*sizeof(T));
else
{
for(int i=0; i<nn*mm; ++i) {newv[i]=v[i]; LA_traits<T>::copyonwrite(newv[i]);}
}
}
v = newv;
#endif
#ifdef CUDALA
}else{ //matrix is in GPU memory
if(!LA_traits<T>::is_plaindata()) laerror("nested types not supported on gpu memory");
T *newv = (T *) gpualloc((size_t)mm*nn*sizeof(T));
if(sizeof(T)%sizeof(float) != 0) laerror("cpu memcpy alignment problem");
if(!detachonly) cublasScopy(nn*mm*sizeof(T)/sizeof(float), (const float *) v, 1, (float *)newv, 1);