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

added tests for plain data in constructors of vec mat smat

Browse Source
This commit is contained in:
Jiri Pittner 2021-10-28 20:17:32 +02:00
parent b50f9b36b1
commit d96531f340
5 changed files with 52 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
mat.cc
View File

@ -550,6 +550,7 @@ template <typename T>
const NRMat<T> NRMat<T>::operator&(const NRMat<T> &b) const { const NRMat<T> NRMat<T>::operator&(const NRMat<T> &b) const {
SAME_LOC(*this, b); SAME_LOC(*this, b);
NRMat<T> result((T)0, nn + b.nn, mm + b.mm, getlocation()); NRMat<T> result((T)0, nn + b.nn, mm + b.mm, getlocation());
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
#ifdef CUDALA #ifdef CUDALA
if(location == cpu){ if(location == cpu){
#endif #endif
@ -738,6 +739,7 @@ const NRMat<T> NRMat<T>::submatrix(const int fromrow, const int torow, const int
const int n = torow - fromrow + 1; const int n = torow - fromrow + 1;
const int m = tocol - fromcol + 1; const int m = tocol - fromcol + 1;
NRMat<T> r(n, m, getlocation()); NRMat<T> r(n, m, getlocation());
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
#ifdef CUDALA #ifdef CUDALA
if(location == cpu){ if(location == cpu){
@ -775,6 +777,7 @@ void NRMat<T>::storesubmatrix(const int fromrow, const int fromcol, const NRMat
if(fromrow<0 || fromrow>=nn || torow>=nn || fromcol<0 || fromcol>=mm || tocol>=mm) laerror("bad indices in storesubmatrix"); if(fromrow<0 || fromrow>=nn || torow>=nn || fromcol<0 || fromcol>=mm || tocol>=mm) laerror("bad indices in storesubmatrix");
#endif #endif
SAME_LOC(*this, rhs); SAME_LOC(*this, rhs);
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
const int m = tocol - fromcol + 1; const int m = tocol - fromcol + 1;
for(register int i = fromrow; i <= torow; ++i){ for(register int i = fromrow; i <= torow; ++i){

15
mat.h
View File

@ -478,7 +478,7 @@ NRMat<T>::NRMat(const T &a, const int n, const int m, const GPUID loc) : nn(n),
#else #else
p = v = new T[nm]; p = v = new T[nm];
#endif #endif
if (a != (T)0){ if (!LA_traits<T>::is_plaindata() || a != (T)0){
for (register int i=0; i<nm; i++) *p++ = a; for (register int i=0; i<nm; i++) *p++ = a;
}else{ }else{
memset(p, 0, nm*sizeof(T)); memset(p, 0, nm*sizeof(T));
@ -488,6 +488,7 @@ NRMat<T>::NRMat(const T &a, const int n, const int m, const GPUID loc) : nn(n),
if(sizeof(T)%sizeof(float) != 0)laerror("memory alignment error"); if(sizeof(T)%sizeof(float) != 0)laerror("memory alignment error");
v = (T*)gpualloc(nm*sizeof(T)); v = (T*)gpualloc(nm*sizeof(T));
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
smart_gpu_set(nm, a, v); smart_gpu_set(nm, a, v);
} }
#endif #endif
@ -522,7 +523,7 @@ inline NRMat<T>::NRMat(const T (&a)[R][C]) : count(new int) {
#endif #endif
#ifdef CUDALA #ifdef CUDALA
}else{ }else{
if!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data"); if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
v = (T*) gpualloc(nn*mm*sizeof(T)); v = (T*) gpualloc(nn*mm*sizeof(T));
cublasSetVector(nm, sizeof(T), a, 1, v, 1); cublasSetVector(nm, sizeof(T), a, 1, v, 1);
} }
@ -554,13 +555,14 @@ NRMat<T>::NRMat(const T &a, const int n, const int m) : nn(n), mm(m), count(new
#else #else
p = v = new T[nm]; p = v = new T[nm];
#endif #endif
if (a != (T)0) if (!LA_traits<T>::is_plaindata() || a != (T)0)
for (register int i=0; i<nm; i++) *p++ = a; for (register int i=0; i<nm; i++) *p++ = a;
else else
memset(p, 0, nm*sizeof(T)); memset(p, 0, nm*sizeof(T));
#ifdef CUDALA #ifdef CUDALA
}else{ }else{
v = (T*)gpualloc(nm*sizeof(T)); v = (T*)gpualloc(nm*sizeof(T));
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
smart_gpu_set(nm, a, v); smart_gpu_set(nm, a, v);
} }
#endif #endif
@ -588,14 +590,17 @@ NRMat<T>::NRMat(const T *a, const int n, const int m) : nn(n), mm(m), count(new
v = new T*[n]; v = new T*[n];
v[0] = new T[nm]; v[0] = new T[nm];
for (register int i=1; i<n; i++) v[i] = v[i-1] + m; for (register int i=1; i<n; i++) v[i] = v[i-1] + m;
memcpy(v[0], a, nm*sizeof(T)); if(LA_traits<T>::is_plaindata()) memcpy(v[0], a, nm*sizeof(T));
else for(int i=0; i<nm; ++i) v[0][i] = a[i];
#else #else
v = new T[nm]; v = new T[nm];
memcpy(v, a, nm*sizeof(T)); if(LA_traits<T>::is_plaindata()) memcpy(v, a, nm*sizeof(T));
else for(int i=0; i<nm; ++i) v[i] = a[i];
#endif #endif
#ifdef CUDALA #ifdef CUDALA
}else{ }else{
v = (T*) gpualloc(nm*sizeof(T)); v = (T*) gpualloc(nm*sizeof(T));
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
cublasSetVector(nm, sizeof(T), a, 1, v, 1); cublasSetVector(nm, sizeof(T), a, 1, v, 1);
} }
#endif #endif

7
smat.h
View File

@ -230,12 +230,13 @@ inline NRSMat<T>::NRSMat(const T& a, const int n) : nn(n), count(new int(1)) {
if(location == cpu){ if(location == cpu){
#endif #endif
v = new T[NN2]; v = new T[NN2];
if(a != (T)0) for(register size_t i = 0; i<NN2; i++) v[i] = a; if(!LA_traits<T>::is_plaindata() || a != (T)0) for(register size_t i = 0; i<NN2; i++) v[i] = a;
else memset(v, 0, NN2*sizeof(T)); else memset(v, 0, NN2*sizeof(T));
#ifdef CUDALA #ifdef CUDALA
}else{ }else{
v = (T*) gpualloc(NN2*sizeof(T)); v = (T*) gpualloc(NN2*sizeof(T));
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
cublasSetVector(NN2, sizeof(T), &a, 0, v, 1); cublasSetVector(NN2, sizeof(T), &a, 0, v, 1);
} }
#endif #endif
@ -253,10 +254,12 @@ inline NRSMat<T>::NRSMat(const T *a, const int n) : nn(n), count(new int(1)) {
location = DEFAULT_LOC; location = DEFAULT_LOC;
if(location == cpu){ if(location == cpu){
#endif #endif
memcpy(v, a, NN2*sizeof(T)); if(LA_traits<T>::is_plaindata()) memcpy(v, a, NN2*sizeof(T));
else for( int i=0; i<NN2; i++) v[i] = a[i];
#ifdef CUDALA #ifdef CUDALA
}else{ }else{
v = (T*) gpualloc(NN2*sizeof(T)); v = (T*) gpualloc(NN2*sizeof(T));
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
cublasSetVector(NN2, sizeof(T), a, 1, v, 1); cublasSetVector(NN2, sizeof(T), a, 1, v, 1);
} }
#endif #endif

18
vec.cc
View File

@ -443,7 +443,7 @@ NRVec<T>& NRVec<T>::operator=(const T &a){
NOT_GPU(*this); NOT_GPU(*this);
copyonwrite(); copyonwrite();
if(a != (T)0){ if(!LA_traits<T>::is_plaindata() || a != (T)0){
for(register int i=0; i<nn; i++) v[i] = a; for(register int i=0; i<nn; i++) v[i] = a;
}else{ }else{
memset(v, 0, nn*sizeof(T)); memset(v, 0, nn*sizeof(T));
@ -883,6 +883,22 @@ for(int cycle=1; cycle<=p.size(); ++cycle)
} }
} }
template<typename T>
const int NRVec<T>::find(const T &val) const
{
for(int i=0; i<nn; ++i) if(val==v[i]) return i;
return -1;
}
template<typename T>
const int NRVec<T>::findthr(const T &val, const typename LA_traits<T>::normtype &thr) const
{
for(int i=0; i<nn; ++i) if(MYABS(val-v[i])<thr) return i;
return -1;
}
/***************************************************************************//** /***************************************************************************//**
* forced instantization in the corespoding object file * forced instantization in the corespoding object file

25
vec.h
View File

@ -352,6 +352,10 @@ public:
//! get normalized copy of this vector //! get normalized copy of this vector
inline const NRVec unitvector() const; inline const NRVec unitvector() const;
//! find an element by value with threshold, first from left
const int find(const T &val) const;
const int findthr(const T &val, const typename LA_traits<T>::normtype &thr=0) const;
//! determine the maximal element (in the absolute value) of this vector //! determine the maximal element (in the absolute value) of this vector
inline const T amax() const; inline const T amax() const;
//! determine the minimal element (in the absolute value) of this vector //! determine the minimal element (in the absolute value) of this vector
@ -570,14 +574,15 @@ inline NRVec<T>::NRVec(const T& a, const int n): nn(n), count(new int) {
if(location == cpu){ if(location == cpu){
#endif #endif
v = new T[n]; v = new T[n];
if(a != (T)0){ if(!LA_traits<T>::is_plaindata() || a != (T)0){
for(register int i=0; i<n; i++) v[i] = a; for(register int i=0; i<n; i++) v[i] = a;
}else{ }else{
memset(v, 0, nn*sizeof(T)); memset(v, 0, nn*sizeof(T));
} }
#ifdef CUDALA #ifdef CUDALA
}else{ }else{
v = (T*) gpualloc(n*sizeof(T)); v = (T*) gpualloc(n*sizeof(T));
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
smart_gpu_set(n, a, v); smart_gpu_set(n, a, v);
} }
#endif #endif
@ -602,7 +607,7 @@ inline NRVec<T>::NRVec(const T (&a)[SIZE]) : count(new int) {
#ifdef CUDALA #ifdef CUDALA
}else{ }else{
v = (T*) gpualloc(nn*sizeof(T)); v = (T*) gpualloc(nn*sizeof(T));
if!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data"); if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
cublasSetVector(nn, sizeof(T), a, 1, v, 1); cublasSetVector(nn, sizeof(T), a, 1, v, 1);
TEST_CUBLAS("cublasSetVector"); TEST_CUBLAS("cublasSetVector");
} }
@ -626,10 +631,12 @@ inline NRVec<T>::NRVec(const T *a, const int n): nn(n), count(new int) {
#endif #endif
v = new T[n]; v = new T[n];
*count = 1; *count = 1;
memcpy(v, a, n*sizeof(T)); if(LA_traits<T>::is_plaindata()) memcpy(v, a, n*sizeof(T));
else for( int i=0; i<n; i++) v[i] = a[i];
#ifdef CUDALA #ifdef CUDALA
}else{ }else{
v = (T*) gpualloc(n*sizeof(T)); v = (T*) gpualloc(n*sizeof(T));
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
cublasSetVector(n, sizeof(T), a, 1, v, 1); cublasSetVector(n, sizeof(T), a, 1, v, 1);
TEST_CUBLAS("cublasSetVector"); TEST_CUBLAS("cublasSetVector");
} }
@ -653,10 +660,12 @@ inline NRVec<T>::NRVec(T *a, const int n, bool skeleton) : nn(n), count(new int)
#endif #endif
v = new T[n]; v = new T[n];
*count = 1; *count = 1;
memcpy(v, a, n*sizeof(T)); if(LA_traits<T>::is_plaindata()) memcpy(v, a, n*sizeof(T));
else for( int i=0; i<n; i++) v[i] = a[i];
#ifdef CUDALA #ifdef CUDALA
}else{ }else{
v= (T*) gpualloc(n*sizeof(T)); v= (T*) gpualloc(n*sizeof(T));
if(!LA_traits<T>::is_plaindata()) laerror("only implemented for plain data");
cublasSetVector(n, sizeof(T), a, 1, v, 1); cublasSetVector(n, sizeof(T), a, 1, v, 1);
TEST_CUBLAS("cublasSetVector"); TEST_CUBLAS("cublasSetVector");
} }