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

*** empty log message ***

Browse Source
This commit is contained in:
jiri
2013-11-04 14:56:39 +00:00
parent a9e30620f0
commit 80fe44fab2
18 changed files with 505 additions and 308 deletions
Download Patch File Download Diff File Expand all files Collapse all files

179
mat.cc
View File

@@ -26,11 +26,8 @@
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>
#include <unistd.h>
extern "C" {
extern ssize_t read(int, void *, size_t);
extern ssize_t write(int, const void *, size_t);
}
namespace LA {
@@ -77,14 +74,14 @@ const NRMat<T> NRMat<T>::otimes(const NRMat<T> &rhs, bool reversecolumns) const
{
T c = (*this)(i,j);
for(k=0;k<rhs.nn;k++) for(l=0;l<rhs.mm;l++)
r( i*rhs.nn + k, l*mm + j ) = c*rhs(k,l);
r( i*(size_t)rhs.nn + k, l*mm + j ) = c*rhs(k,l);
}
}else{
for(i=0;i<nn;i++) for(j=0;j<mm;j++)
{
T c=(*this)(i,j);
for(k=0;k<rhs.nn;k++) for(l=0;l<rhs.mm;l++)
r( i*rhs.nn+k, j*rhs.mm+l ) = c *rhs(k,l);
r( i*(size_t)rhs.nn+k, j*(size_t)rhs.mm+l ) = c *rhs(k,l);
}
}
@@ -108,7 +105,7 @@ const NRVec<T> NRMat<T>::row(const int i, int l) const {
#ifdef MATPTR
v[i]
#else
v + i*l
v + i*(size_t)l
#endif
, l);
return r;
@@ -144,7 +141,7 @@ void NRMat<T>::put(int fd, bool dim, bool transp) const {
#ifdef MATPTR
v[i][j]
#else
v[i*mm+j]
v[i*(size_t)mm+j]
#endif
,dim ,transp);
}
@@ -196,7 +193,7 @@ void NRMat<T>::get(int fd, bool dim, bool transp){
#ifdef MATPTR
v[i][j]
#else
v[i*mm+j]
v[i*(size_t)mm+j]
#endif
,dim,transp);
}
@@ -476,13 +473,13 @@ NRMat<T> & NRMat<T>::operator-=(const T &a) {
******************************************************************************/
template <>
const NRMat<double> NRMat<double>::operator-() const {
const int nm = nn*mm;
const size_t nm = (size_t)nn*mm;
NRMat<double> result(nn, mm, getlocation());
#ifdef CUDALA
if(location == cpu) {
#endif
#ifdef MATPTR
for(register int i=0; i<nm; i++) result.v[0][i] = -v[0][i];
for(register size_t i=0; i<nm; i++) result.v[0][i] = -v[0][i];
#else
memcpy(result.v, v, nm*sizeof(double));
cblas_dscal(nm, -1., result.v, 1);
@@ -506,13 +503,13 @@ const NRMat<double> NRMat<double>::operator-() const {
******************************************************************************/
template <>
const NRMat<complex<double> > NRMat<complex<double> >::operator-() const {
const int nm = nn*mm;
const size_t nm = (size_t)nn*mm;
NRMat<complex<double> > result(nn, mm, getlocation());
#ifdef CUDALA
if(location == cpu) {
#endif
#ifdef MATPTR
for(register int i=0; i<nm; i++) result.v[0][i]= -v[0][i];
for(register size_t i=0; i<nm; i++) result.v[0][i]= -v[0][i];
#else
memcpy(result.v, v, nm*sizeof(complex<double>));
cblas_zscal(nm, &CMONE, result.v, 1);
@@ -539,9 +536,9 @@ const NRMat<T> NRMat<T>::operator-() const {
NRMat<T> result(nn, mm, getlocation());
#ifdef MATPTR
for(register int i=0; i<nn*mm; i++) result.v[0][i] = -v[0][i];
for(register size_t i=0; i<(size_t)nn*mm; i++) result.v[0][i] = -v[0][i];
#else
for(register int i=0; i<nn*mm; i++) result.v[i] = -v[i];
for(register size_t i=0; i<(size_t)nn*mm; i++) result.v[i] = -v[i];
#endif
return result;
}
@@ -562,11 +559,11 @@ const NRMat<T> NRMat<T>::operator&(const NRMat<T> &b) const {
if(sizeof(T)%sizeof(float) != 0) laerror("memory alignment problem");
for(register int i=0; i<nn; i++){
cublasScopy(mm*sizeof(T)/sizeof(float), (float*)(v + i*mm), 1, (float*)(result.v + i*(mm + b.mm)), 1);
cublasScopy(mm*sizeof(T)/sizeof(float), (float*)(v + i*(size_t)mm), 1, (float*)(result.v + i*(size_t)(mm + b.mm)), 1);
TEST_CUBLAS("cublasScopy");
}
for(register int i=0; i<b.nn; i++){
cublasScopy(mm*sizeof(T)/sizeof(float), (float*)(b.v + i*b.mm), 1, (float*)(result.v + (nn + i)*(mm + b.mm)), 1);
cublasScopy(mm*sizeof(T)/sizeof(float), (float*)(b.v + i*(size_t)b.mm), 1, (float*)(result.v + (nn + i)*(mm + b.mm)), 1);
TEST_CUBLAS("cublasScopy");
}
}
@@ -582,7 +579,7 @@ const NRMat<T> NRMat<T>::operator|(const NRMat<T> &b) const {
for (int j=0; j<mm; j++)
for (int k=0; k<b.nn; k++)
for (int l=0; l<b.mm; l++)
result[i*b.nn+k][j*b.mm+l] = (*this)[i][j]*b[k][l];
result[i*(size_t)b.nn+k][j*(size_t)b.mm+l] = (*this)[i][j]*b[k][l];
return result;
}
@@ -689,7 +686,7 @@ const NRVec<double> NRMat<double>::rsum() const {
#ifdef CUDALA
}else{
for(register int i=0;i<nn;i++){
cublasDaxpy(mm, 1.0, v + i*mm, 1, result.v, 1);
cublasDaxpy(mm, 1.0, v + i*(size_t)mm, 1, result.v, 1);
TEST_CUBLAS("cublasDaxpy");
}
}
@@ -714,7 +711,7 @@ const NRVec<complex<double> > NRMat<complex<double> >::rsum() const {
#ifdef CUDALA
}else{
for(register int i=0;i<nn;i++){
cublasZaxpy(mm, CUONE, (cuDoubleComplex*)(v + i*mm), 1, (cuDoubleComplex*)(result.v), 1);
cublasZaxpy(mm, CUONE, (cuDoubleComplex*)(v + i*(size_t)mm), 1, (cuDoubleComplex*)(result.v), 1);
TEST_CUBLAS("cublasZaxpy");
}
}
@@ -748,14 +745,14 @@ const NRMat<T> NRMat<T>::submatrix(const int fromrow, const int torow, const int
#ifdef MATPTR
memcpy(r.v[i - fromrow], v[i] + fromcol, m*sizeof(T));
#else
memcpy(r.v+(i - fromrow)*m, v + i*mm + fromcol, m*sizeof(T));
memcpy(r.v+(i - fromrow)*m, v + i*(size_t)mm + fromcol, m*sizeof(T));
#endif
}
#ifdef CUDALA
}else{
if(sizeof(T)%sizeof(float) != 0) laerror("cpu memcpy alignment problem");
for(register int i=fromrow; i<=torow; ++i){
cublasScopy(m*sizeof(T)/sizeof(float), (const float *)(v + i*mm + fromcol), 1, (float*)(r.v + (i - fromrow)*m), 1);
cublasScopy(m*sizeof(T)/sizeof(float), (const float *)(v + i*(size_t)mm + fromcol), 1, (float*)(r.v + (i - fromrow)*m), 1);
TEST_CUBLAS("cublasScopy");
}
}
@@ -786,13 +783,13 @@ void NRMat<T>::storesubmatrix(const int fromrow, const int fromcol, const NRMat
#ifdef MATPTR
memcpy(v[i] + fromcol, rhs.v[i - fromrow], m*sizeof(T));
#else
memcpy(v + i*mm + fromcol, rhs.v + (i - fromrow)*m, m*sizeof(T));
memcpy(v + i*(size_t)mm + fromcol, rhs.v + (i - fromrow)*m, m*sizeof(T));
#endif
#ifdef CUDALA
}else{
if(sizeof(T)%sizeof(float) != 0) laerror("cpu memcpy alignment problem");
cublasScopy(m*sizeof(T)/sizeof(float), (const float *) (rhs.v + (i - fromrow)*m), 1, (float *)(v + i*mm + fromcol), 1);
cublasScopy(m*sizeof(T)/sizeof(float), (const float *) (rhs.v + (i - fromrow)*m), 1, (float *)(v + i*(size_t)mm + fromcol), 1);
}
#endif
}
@@ -821,8 +818,8 @@ NRMat<T>& NRMat<T>::transposeme(const int _n) {
v[j][i] = tmp;
#else
register int a, b;
a = i*mm + j;
b = j*mm + i;
a = i*(size_t)mm + j;
b = j*(size_t)mm + i;
T tmp = v[a];
v[a] = v[b];
v[b] = tmp;
@@ -847,7 +844,7 @@ NRMat<T>& NRMat<T>::transposeme(const int _n) {
******************************************************************************/
template<>
NRMat<complex<double> >::NRMat(const NRMat<double> &rhs, bool imagpart): nn(rhs.nrows()), mm(rhs.ncols()), count(new int(1)) {
const int nn_mm = nn*mm;
const size_t nn_mm = (size_t)nn*mm;
#ifdef CUDALA
if(location == cpu){
#endif
@@ -888,7 +885,7 @@ NRMat<complex<double> >::NRMat(const NRMat<double> &rhs, bool imagpart): nn(rhs.
******************************************************************************/
template<>
NRMat<double>::NRMat(const NRMat<complex<double> > &rhs, bool imagpart): nn(rhs.nrows()), mm(rhs.ncols()), count(new int(1)) {
const int nn_mm = nn*mm;
const size_t nn_mm = (size_t) nn*mm;
#ifdef CUDALA
if(location == cpu){
#endif
@@ -1079,7 +1076,7 @@ const NRSMat<double> NRMat<double>::timestransposed() const {
#ifdef MATPTR
r(i, j) = cblas_ddot(mm, v[i], 1, v[j], 1);
#else
r(i, j) = cblas_ddot(mm, v + i*mm, 1, v + j*mm, 1);
r(i, j) = cblas_ddot(mm, v + i*(size_t)mm, 1, v + j*(size_t)mm, 1);
#endif
}
}
@@ -1087,7 +1084,7 @@ const NRSMat<double> NRMat<double>::timestransposed() const {
}else{
for(i=0; i<nn; ++i){
for(j=0; j<=i; ++j){
r(i, j) = cublasDdot(nn, v + i*mm, 1, v + j*mm, 1);
r(i, j) = cublasDdot(nn, v + i*(size_t)mm, 1, v + j*(size_t)mm, 1);
TEST_CUBLAS("cublasDdot");
}
}
@@ -1113,7 +1110,7 @@ const NRSMat<complex<double> > NRMat<complex<double> >::timestransposed() const
#ifdef MATPTR
cblas_zdotc_sub(nn, v[i], 1, v[j], 1, &r(i,j));
#else
cblas_zdotc_sub(nn, v + i*mm, 1, v + j*mm, 1, &r(i,j));
cblas_zdotc_sub(nn, v + i*(size_t)mm, 1, v + j*(size_t)mm, 1, &r(i,j));
#endif
}
}
@@ -1121,7 +1118,7 @@ const NRSMat<complex<double> > NRMat<complex<double> >::timestransposed() const
}else{
for(i=0; i<mm; ++i){
for(j=0; j<=i; ++j){
cuDoubleComplex val = cublasZdotc(nn, (const cuDoubleComplex *)(v + i*mm), 1, (const cuDoubleComplex *)(v + j*mm), 1);
cuDoubleComplex val = cublasZdotc(nn, (const cuDoubleComplex *)(v + i*(size_t)mm), 1, (const cuDoubleComplex *)(v + j*(size_t)mm), 1);
TEST_CUBLAS("cublasZdotc");
r(i, j) = *(reinterpret_cast<complex<double>*> (&val));
}
@@ -1172,7 +1169,7 @@ void NRMat<double>::randomize(const double &x) {
}else{
NRMat<double> tmp(nn, mm, cpu);
double *tmp_data = tmp;
for(register int i=0; i<nn*mm; ++i){
for(register size_t i=0; i<(size_t)nn*mm; ++i){
tmp_data[i] = x*(2.*random()/(1. + RAND_MAX) - 1.);
}
tmp.moveto(this->location);
@@ -1203,7 +1200,7 @@ void NRMat<complex<double> >::randomize(const double &x) {
}else{
NRMat<complex<double> > tmp(nn, mm, cpu);
complex<double> *tmp_data = tmp;
for(register int i=0; i<nn*mm; ++i){
for(register size_t i=0; i<(size_t)nn*mm; ++i){
const double re = x*(2.*random()/(1. + RAND_MAX) - 1.);
const double im = x*(2.*random()/(1. + RAND_MAX) - 1.);
tmp_data[i] = complex<double>(re, im);
@@ -1226,10 +1223,10 @@ NRMat<double>& NRMat<double>::operator*=(const double &a) {
#ifdef CUDALA
if(location == cpu){
#endif
cblas_dscal(nn*mm, a, *this, 1);
cblas_dscal((size_t)nn*mm, a, *this, 1);
#ifdef CUDALA
}else{
cublasDscal(nn*mm, a, v, 1);
cublasDscal((size_t)nn*mm, a, v, 1);
TEST_CUBLAS("cublasDscal");
}
#endif
@@ -1249,11 +1246,11 @@ NRMat<complex<double> >::operator*=(const complex<double> &a) {
#ifdef CUDALA
if(location == cpu){
#endif
cblas_zscal(nn*mm, &a, (*this)[0], 1);
cblas_zscal((size_t)nn*mm, &a, (*this)[0], 1);
#ifdef CUDALA
}else{
const cuDoubleComplex fac = *(reinterpret_cast<const cuDoubleComplex*> (&a));
cublasZscal(nn*mm, fac, (cuDoubleComplex *)v, 1);
cublasZscal((size_t)nn*mm, fac, (cuDoubleComplex *)v, 1);
TEST_CUBLAS("cublasZscal");
}
#endif
@@ -1271,9 +1268,9 @@ NRMat<T> & NRMat<T>::operator*=(const T &a) {
NOT_GPU(*this);
copyonwrite();
#ifdef MATPTR
for(register int i=0; i< nn*mm; i++) v[0][i] *= a;
for(register size_t i=0; i< (size_t)nn*mm; i++) v[0][i] *= a;
#else
for(register int i=0; i< nn*mm; i++) v[i] *= a;
for(register size_t i=0; i< (size_t)nn*mm; i++) v[i] *= a;
#endif
return *this;
}
@@ -1294,10 +1291,10 @@ NRMat<double> & NRMat<double>::operator+=(const NRMat<double> &rhs) {
#ifdef CUDALA
if(location == cpu){
#endif
cblas_daxpy(nn*mm, 1.0, rhs, 1, *this, 1);
cblas_daxpy((size_t)nn*mm, 1.0, rhs, 1, *this, 1);
#ifdef CUDALA
}else{
cublasDaxpy(nn*mm, 1.0, rhs, 1, v, 1);
cublasDaxpy((size_t)nn*mm, 1.0, rhs, 1, v, 1);
TEST_CUBLAS("cublasDaxpy");
}
#endif
@@ -1320,10 +1317,10 @@ NRMat<complex<double> >::operator+=(const NRMat< complex<double> > &rhs) {
#ifdef CUDALA
if(location == cpu){
#endif
cblas_zaxpy(nn*mm, &CONE, rhs[0], 1, (*this)[0], 1);
cblas_zaxpy((size_t)nn*mm, &CONE, rhs[0], 1, (*this)[0], 1);
#ifdef CUDALA
}else{
cublasZaxpy(nn*mm, CUONE, (cuDoubleComplex*)(rhs[0]), 1, (cuDoubleComplex*)((*this)[0]), 1);
cublasZaxpy((size_t)nn*mm, CUONE, (cuDoubleComplex*)(rhs[0]), 1, (cuDoubleComplex*)((*this)[0]), 1);
}
#endif
return *this;
@@ -1345,9 +1342,9 @@ NRMat<T> & NRMat<T>::operator+=(const NRMat<T> &rhs) {
copyonwrite();
#ifdef MATPTR
for(int i=0; i< nn*mm; i++) v[0][i] += rhs.v[0][i];
for(size_t i=0; i< (size_t)nn*mm; i++) v[0][i] += rhs.v[0][i];
#else
for(int i=0; i< nn*mm; i++) v[i] += rhs.v[i];
for(size_t i=0; i< (size_t)nn*mm; i++) v[i] += rhs.v[i];
#endif
return *this;
}
@@ -1368,10 +1365,10 @@ NRMat<double> & NRMat<double>::operator-=(const NRMat<double> &rhs) {
#ifdef CUDALA
if(location == cpu){
#endif
cblas_daxpy(nn*mm, -1.0, rhs, 1, *this, 1);
cblas_daxpy((size_t)nn*mm, -1.0, rhs, 1, *this, 1);
#ifdef CUDALA
}else{
cublasDaxpy(nn*mm, -1.0, rhs, 1, v, 1);
cublasDaxpy((size_t)nn*mm, -1.0, rhs, 1, v, 1);
}
#endif
return *this;
@@ -1395,10 +1392,10 @@ NRMat< complex<double> >::operator-=(const NRMat< complex<double> > &rhs) {
#ifdef CUDALA
if(location == cpu){
#endif
cblas_zaxpy(nn*mm, &CMONE, rhs[0], 1, (*this)[0], 1);
cblas_zaxpy((size_t)nn*mm, &CMONE, rhs[0], 1, (*this)[0], 1);
#ifdef CUDALA
}else{
cublasZaxpy(nn*mm, CUMONE, (cuDoubleComplex*)(rhs[0]), 1, (cuDoubleComplex*)((*this)[0]), 1);
cublasZaxpy((size_t)nn*mm, CUMONE, (cuDoubleComplex*)(rhs[0]), 1, (cuDoubleComplex*)((*this)[0]), 1);
}
#endif
return *this;
@@ -1421,9 +1418,9 @@ NRMat<T> & NRMat<T>::operator-=(const NRMat<T> &rhs) {
copyonwrite();
#ifdef MATPTR
for(int i=0; i< nn*mm; i++) v[0][i] += rhs.v[0][i];
for(size_t i=0; i< (size_t)nn*mm; i++) v[0][i] += rhs.v[0][i];
#else
for(int i=0; i< nn*mm; i++) v[i] += rhs.v[i];
for(size_t i=0; i<(size_t) nn*mm; i++) v[i] += rhs.v[i];
#endif
return *this;
}
@@ -1693,10 +1690,10 @@ const double NRMat<double>::dot(const NRMat<double> &rhs) const {
#ifdef CUDALA
if(location == cpu){
#endif
ret = cblas_ddot(nn*mm, (*this)[0], 1, rhs[0], 1);
ret = cblas_ddot((size_t)nn*mm, (*this)[0], 1, rhs[0], 1);
#ifdef CUDALA
}else{
ret = cublasDdot(nn*mm, v, 1, rhs.v, 1);
ret = cublasDdot((size_t)nn*mm, v, 1, rhs.v, 1);
}
#endif
return ret;
@@ -1721,10 +1718,10 @@ NRMat<complex<double> >::dot(const NRMat<complex<double> > &rhs) const {
#ifdef CUDALA
if(location == cpu){
#endif
cblas_zdotc_sub(nn*mm, (*this)[0], 1, rhs[0], 1, &ret);
cblas_zdotc_sub((size_t)nn*mm, (*this)[0], 1, rhs[0], 1, &ret);
#ifdef CUDALA
}else{
cuDoubleComplex val = cublasZdotc(nn*mm, (cuDoubleComplex*)v, 1, (cuDoubleComplex*)(rhs.v), 1);
cuDoubleComplex val = cublasZdotc((size_t)nn*mm, (cuDoubleComplex*)v, 1, (cuDoubleComplex*)(rhs.v), 1);
ret = *(reinterpret_cast<complex<double>*> (&val));
}
#endif
@@ -1804,7 +1801,7 @@ void NRMat<double>::diagmultl(const NRVec<double> &rhs) {
for(register int i=0; i<nn; i++){ cblas_dscal(mm, rhs[i], (*this)[i], 1); }
#ifdef CUDALA
}else{
for(register int i=0; i<nn; i++){ cublasDscal(mm, rhs[i], v + i*mm, 1); }
for(register int i=0; i<nn; i++){ cublasDscal(mm, rhs[i], v + i*(size_t)mm, 1); }
}
#endif
}
@@ -1830,7 +1827,7 @@ void NRMat< complex<double> >::diagmultl(const NRVec< complex<double> > &rhs) {
}else{
for(register int i=0; i<nn; i++){
const cuDoubleComplex alpha = make_cuDoubleComplex(rhs[i].real(), rhs[i].imag());
cublasZscal(mm, alpha, (cuDoubleComplex*)(v + i*mm), 1);
cublasZscal(mm, alpha, (cuDoubleComplex*)(v + i*(size_t)mm), 1);
}
}
#endif
@@ -1913,7 +1910,7 @@ NRMat<double>::operator*(const NRSMat<double> &rhs) const {
#ifdef CUDALA
}else{
for(register int i=0; i<nn; i++){
cublasDspmv('U', mm, 1.0, rhs.v, v + i*mm, 1, 0.0, result.v + i*rhs_ncols, 1);
cublasDspmv('U', mm, 1.0, rhs.v, v + i*(size_t)mm, 1, 0.0, result.v + i*(size_t)rhs_ncols, 1);
}
}
#endif
@@ -1947,7 +1944,7 @@ NRMat< complex<double> >::operator*(const NRSMat< complex<double> > &rhs) const
#ifdef CUDALA
}else{
for(register int i=0; i<nn; i++){
cublasZhpmv('U', mm, CUONE, (cuDoubleComplex*)rhs.v, (cuDoubleComplex*)(v + i*mm), 1, CUZERO, (cuDoubleComplex*)(result.v + i*rhs_ncols), 1);
cublasZhpmv('U', mm, CUONE, (cuDoubleComplex*)rhs.v, (cuDoubleComplex*)(v + i*(size_t)mm), 1, CUZERO, (cuDoubleComplex*)(result.v + i*(size_t)rhs_ncols), 1);
}
}
#endif
@@ -1974,10 +1971,10 @@ NRMat<complex<double> >& NRMat<complex<double> >::conjugateme() {
#ifdef CUDALA
if(location == cpu){
#endif
cblas_dscal(mm*nn, -1.0, (double *)((*this)[0]) + 1, 2);
cblas_dscal((size_t)mm*nn, -1.0, (double *)((*this)[0]) + 1, 2);
#ifdef CUDALA
}else{
cublasDscal(mm*nn, -1.0, (double *)(this->v) + 1, 2);
cublasDscal((size_t)mm*nn, -1.0, (double *)(this->v) + 1, 2);
}
#endif
return *this;
@@ -2048,12 +2045,12 @@ void NRMat<double>::gemm(const double &beta, const NRMat<double> &a,
const char transa, const NRMat<double> &b, const char transb,
const double &alpha) {
int k(transa=='n'?a.mm:a.nn);
int k(tolower(transa)=='n'?a.mm:a.nn);
#ifdef DEBUG
int l(transa=='n'?a.nn:a.mm);
int kk(transb=='n'?b.nn:b.mm);
int ll(transb=='n'?b.mm:b.nn);
int l(tolower(transa)=='n'?a.nn:a.mm);
int kk(tolower(transb)=='n'?b.nn:b.mm);
int ll(tolower(transb)=='n'?b.mm:b.nn);
if (l!=nn || ll!=mm || k!=kk) laerror("incompatible matrices in NRMat<double>::gemm(...)");
if(b.mm <=0 || mm<=0) laerror("illegal matrix dimension in gemm");
#endif
@@ -2066,8 +2063,8 @@ void NRMat<double>::gemm(const double &beta, const NRMat<double> &a,
#ifdef CUDALA
if(location == cpu){
#endif
cblas_dgemm(CblasRowMajor, (transa=='n' ? CblasNoTrans : CblasTrans),
(transb=='n' ? CblasNoTrans : CblasTrans), nn, mm, k, alpha, a,
cblas_dgemm(CblasRowMajor, (tolower(transa)=='n' ? CblasNoTrans : CblasTrans),
(tolower(transb)=='n' ? CblasNoTrans : CblasTrans), nn, mm, k, alpha, a,
a.mm, b , b.mm, beta, *this , mm);
#ifdef CUDALA
}else{
@@ -2083,20 +2080,20 @@ void NRMat<complex<double> >::gemm(const complex<double> & beta,
const NRMat<complex<double> > & b, const char transb,
const complex<double> & alpha)
{
int k(transa=='n'?a.mm:a.nn);
int k(tolower(transa)=='n'?a.mm:a.nn);
#ifdef DEBUG
int l(transa=='n'?a.nn:a.mm);
int kk(transb=='n'?b.nn:b.mm);
int ll(transb=='n'?b.mm:b.nn);
int l(tolower(transa)=='n'?a.nn:a.mm);
int kk(tolower(transb)=='n'?b.nn:b.mm);
int ll(tolower(transb)=='n'?b.mm:b.nn);
if (l!=nn || ll!=mm || k!=kk) laerror("incompatible matrices in NRMat<complex<double> >::gemm(...)");
#endif
if (alpha==CZERO && beta==CONE) return;
copyonwrite();
cblas_zgemm(CblasRowMajor,
(transa=='n' ? CblasNoTrans : (transa=='c'?CblasConjTrans:CblasTrans)),
(transb=='n' ? CblasNoTrans : (transa=='c'?CblasConjTrans:CblasTrans)),
(tolower(transa)=='n' ? CblasNoTrans : (tolower(transa)=='c'?CblasConjTrans:CblasTrans)),
(tolower(transb)=='n' ? CblasNoTrans : (tolower(transb)=='c'?CblasConjTrans:CblasTrans)),
nn, mm, k, &alpha, a , a.mm, b , b.mm, &beta, *this , mm);
}
@@ -2113,10 +2110,10 @@ const double NRMat<double>::norm(const double scalar) const {
#ifdef CUDALA
if(location == cpu){
#endif
return cblas_dnrm2(nn*mm, (*this)[0], 1);
return cblas_dnrm2((size_t)nn*mm, (*this)[0], 1);
#ifdef CUDALA
}else{
return cublasDnrm2(nn*mm, v, 1);
return cublasDnrm2((size_t)nn*mm, v, 1);
}
#endif
}
@@ -2130,7 +2127,7 @@ const double NRMat<double>::norm(const double scalar) const {
#ifdef MATPTR
tmp = v[i][j];
#else
tmp = v[i*mm+j];
tmp = v[i*(size_t)mm+j];
#endif
if(i == j) tmp -= scalar;
sum += tmp*tmp;
@@ -2152,10 +2149,10 @@ const double NRMat<complex<double> >::norm(const complex<double> scalar) const {
#ifdef CUDALA
if(location == cpu){
#endif
return cblas_dznrm2(nn*mm, (*this)[0], 1);
return cblas_dznrm2((size_t)nn*mm, (*this)[0], 1);
#ifdef CUDALA
}else{
return cublasDznrm2(nn*mm, (cuDoubleComplex*)v, 1);
return cublasDznrm2((size_t)nn*mm, (cuDoubleComplex*)v, 1);
}
#endif
}
@@ -2168,7 +2165,7 @@ const double NRMat<complex<double> >::norm(const complex<double> scalar) const {
#ifdef MATPTR
tmp = v[i][j];
#else
tmp = v[i*mm+j];
tmp = v[i*(size_t)mm+j];
#endif
if(i == j) tmp -= scalar;
const double re = tmp.real();
@@ -2195,10 +2192,10 @@ void NRMat<double>::axpy(const double alpha, const NRMat<double> &mat) {
#ifdef CUDALA
if(location == cpu){
#endif
cblas_daxpy(nn*mm, alpha, mat, 1, *this, 1);
cblas_daxpy((size_t)nn*mm, alpha, mat, 1, *this, 1);
#ifdef CUDALA
}else{
cublasDaxpy(nn*mm, alpha, mat, 1, *this, 1);
cublasDaxpy((size_t)nn*mm, alpha, mat, 1, *this, 1);
}
#endif
}
@@ -2221,7 +2218,7 @@ void NRMat<complex<double> >::axpy(const complex<double> alpha,
#ifdef CUDALA
if(location == cpu){
#endif
cblas_zaxpy(nn*mm, &alpha, mat, 1, (*this)[0], 1);
cblas_zaxpy((size_t)nn*mm, &alpha, mat, 1, (*this)[0], 1);
#ifdef CUDALA
}else{
const cuDoubleComplex _alpha = make_cuDoubleComplex(alpha.real(), alpha.imag());
@@ -2245,7 +2242,7 @@ const T NRMat<T>::trace() const {
#ifdef MATPTR
for(register int i=0; i<nn; ++i) sum += v[i][i];
#else
for(register int i=0; i<nn*nn; i += (nn+1)) sum += v[i];
for(register size_t i=0; i<(size_t)nn*nn; i += (nn+1)) sum += v[i];
#endif
return sum;
}
@@ -2554,7 +2551,7 @@ NRMat<double>& NRMat<double>::swap_rows(){
#ifdef CUDALA
}else{
for(register int i=0; i<n_pul; i++){
cublasDswap(mm, v + i*mm, 1, v + (nn - i - 1)*mm, 1);
cublasDswap(mm, v + i*(size_t)mm, 1, v + (nn - i - 1)*mm, 1);
TEST_CUBLAS("cublasDswap");
}
}
@@ -2580,7 +2577,7 @@ NRMat<complex<double> >& NRMat<complex<double> >::swap_rows(){
#ifdef CUDALA
}else{
for(register int i=0; i<n_pul; i++){
cublasZswap(mm, (cuDoubleComplex*)(v + i*mm), 1, (cuDoubleComplex*)(v + (nn - i - 1)*mm), 1);
cublasZswap(mm, (cuDoubleComplex*)(v + i*(size_t)mm), 1, (cuDoubleComplex*)(v + (nn - i - 1)*mm), 1);
TEST_CUBLAS("cublasZswap");
}
}
@@ -2613,7 +2610,7 @@ NRMat<T>& NRMat<T>::swap_rows(){
}else{
if(sizeof(T)%sizeof(float) != 0) laerror("cpu memcpy alignment problem in NRMat<T>::swap_rows");
for(register int i=0; i<n_pul; i++){
cublasSswap(mm*sizeof(T)/sizeof(float), (float *)(v + i*mm), 1, (float *)(v + (nn - i - 1)*mm), 1);
cublasSswap(mm*sizeof(T)/sizeof(float), (float *)(v + i*(size_t)mm), 1, (float *)(v + (nn - i - 1)*mm), 1);
TEST_CUBLAS("cublasSswap");
}
}
@@ -2745,7 +2742,7 @@ NRMat<double>& NRMat<double>::swap_rows_cols(){
#ifdef CUDALA
}else{
for(register int i=0; i<n_pul; i++){
cublasDswap(mm, v + i*mm, 1, v + (nn - i - 1)*mm + mm - 1, -1);
cublasDswap(mm, v + i*(size_t)mm, 1, v + (nn - i - 1)*mm + mm - 1, -1);
TEST_CUBLAS("cublasDswap");
}
@@ -2792,7 +2789,7 @@ NRMat<complex<double> >& NRMat<complex<double> >::swap_rows_cols(){
#ifdef CUDALA
}else{
for(register int i=0;i<n_pul;i++){
cublasZswap(mm, (cuDoubleComplex*)(v + i*mm), 1, (cuDoubleComplex*)(v + (nn - i - 1)*mm + mm - 1), -1);
cublasZswap(mm, (cuDoubleComplex*)(v + i*(size_t)mm), 1, (cuDoubleComplex*)(v + (nn - i - 1)*mm + mm - 1), -1);
TEST_CUBLAS("cublasZswap");
}
if(nn & 1){
@@ -2817,7 +2814,7 @@ template<typename T>
NRMat<T>& NRMat<T>::swap_rows_cols(){
const int n_pul = nn >> 1;
const int m_pul = mm >> 1;
const int dim = nn*mm;
const size_t dim = (size_t)nn*mm;
T *data_ptr;
T tmp;
@@ -2837,7 +2834,7 @@ NRMat<T>& NRMat<T>::swap_rows_cols(){
}else{
if(sizeof(T)%sizeof(float) != 0) laerror("cpu memcpy alignment problem in NRMat<T>::swap_rows_cols");
for(register int i=0; i<n_pul; i++){
cublasSswap(mm*sizeof(T)/sizeof(float), (float *)(v + i*mm), 1, (float *)(v + (nn - i - 1)*mm) - 1, -1);
cublasSswap(mm*sizeof(T)/sizeof(float), (float *)(v + i*(size_t)mm), 1, (float *)(v + (nn - i - 1)*mm) - 1, -1);
TEST_CUBLAS("cublasSswap");
}