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

tensor: innercontraction implemented

Browse Source
This commit is contained in:
Jiri Pittner
2025-11-11 17:37:28 +01:00
parent 32242f18ed
commit 98ef46ad47
5 changed files with 322 additions and 165 deletions
Download Patch File Download Diff File Expand all files Collapse all files

97
vec.cc
View File

@@ -863,103 +863,6 @@ return -1;
/***************************************************************************//**
* extract block subvector
* @param[in] from starting position
* @param[in] to final position
* @return extracted block subvector
******************************************************************************/
template <typename T>
const NRVec<T> NRVec<T>::subvector(const int from, const int to) const
{
#ifdef DEBUG
if(from<0 || from>=nn|| to<0 || to>=nn || from>to){
laerror("invalid subvector specification");
}
#endif
//trivial case of whole vector for efficiency
if(from==0 && to == nn-1) return *this;
const int n = to - from + 1;
NRVec<T> r(n, getlocation());
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
#ifdef CUDALA
if(location == cpu){
#endif
memcpy(r.v, v+from, n*sizeof(T));
#ifdef CUDALA
}else{
if(sizeof(T)%sizeof(float) != 0) laerror("cpu memcpy alignment problem");
cublasScopy(n*sizeof(T)/sizeof(float), (const float *)(v+from), 1, (float*)r.v, 1);
TEST_CUBLAS("cublasScopy");
}
#endif
return r;
}
template <typename T>
const NRVec<T> NRVec<T>::subvector(const NRVec<int> &selection) const
{
NOT_GPU(*this);
const int n = selection.size();
NRVec<T> r(n);
for(int i=0; i<n; ++i)
{
int ii=selection[i];
if(ii<0||ii>=nn) laerror("bad row index in subvector");
r[i] = (*this)[ii];
}
return r;
}
/***************************************************************************//**
* places given vector as subvector at given position
* @param[in] from coordinate
* @param[in] rhs input vector
******************************************************************************/
template <typename T>
void NRVec<T>::storesubvector(const int from, const NRVec &rhs)
{
const int to = from + rhs.size() - 1;
#ifdef DEBUG
if(from<0 || from>=nn || to>=nn) laerror("bad indices in storesubvector");
#endif
SAME_LOC(*this, rhs);
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
#ifdef CUDALA
if(location == cpu){
#endif
memcpy(v+from, rhs.v, rhs.size()*sizeof(T));
#ifdef CUDALA
}else{
if(sizeof(T)%sizeof(float) != 0) laerror("cpu memcpy alignment problem");
cublasScopy(rhs.size()*sizeof(T)/sizeof(float), (const float *) (rhs.v), 1, (float *)(v + from), 1);
}
#endif
}
template <typename T>
void NRVec<T>::storesubvector(const NRVec<int> &selection, const NRVec &rhs)
{
NOT_GPU(*this);
const int n = selection.size();
if(n!=rhs.size()) laerror("size mismatch in storesubvector");
for(int i=0; i<n; ++i)
{
int ii=selection[i];
if(ii<0||ii>=nn) laerror("bad index in storesubvector");
(*this)[ii] = rhs[i];
}
}
/***************************************************************************//**
* conjugate this general vector
* @return reference to the (unmodified) matrix