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

27
vec.cc
View File

@@ -27,11 +27,8 @@
#include <errno.h>
#include "vec.h"
#include "qsort.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 {
@@ -541,7 +538,7 @@ template<>
void NRVec<double>::gemv(const double beta, const NRMat<double> &A,
const char trans, const double alpha, const NRVec &x) {
#ifdef DEBUG
if((trans == 'n'?A.ncols():A.nrows()) != x.size()){ laerror("incompatible vectors"); }
if((tolower(trans) == 'n'?A.ncols():A.nrows()) != x.size()){ laerror("incompatible vectors"); }
#endif
SAME_LOC3(*this, x, A);
copyonwrite();
@@ -549,10 +546,10 @@ void NRVec<double>::gemv(const double beta, const NRMat<double> &A,
#ifdef CUDALA
if(location==cpu){
#endif
cblas_dgemv(CblasRowMajor, (trans=='n' ? CblasNoTrans:CblasTrans), A.nrows(), A.ncols(), alpha, A, A.ncols(), x.v, 1, beta, v, 1);
cblas_dgemv(CblasRowMajor, (tolower(trans)=='n' ? CblasNoTrans:CblasTrans), A.nrows(), A.ncols(), alpha, A, A.ncols(), x.v, 1, beta, v, 1);
#ifdef CUDALA
}else{
cublasDgemv((trans=='n'?'T':'N'), A.ncols(), A.nrows(), alpha, A, A.ncols(), x.v, 1, beta, v, 1);
cublasDgemv((tolower(trans)=='n'?'T':'N'), A.ncols(), A.nrows(), alpha, A, A.ncols(), x.v, 1, beta, v, 1);
TEST_CUBLAS("cublasDgemv");
}
#endif
@@ -572,7 +569,7 @@ template<>
void NRVec<complex<double> >::gemv(const double beta, const NRMat<double> &A,
const char trans, const double alpha, const NRVec<complex<double> > &x) {
#ifdef DEBUG
if ((trans == 'n'?A.ncols():A.nrows()) != x.size()){ laerror("incompatible vectors"); }
if ((tolower(trans) == 'n'?A.ncols():A.nrows()) != x.size()){ laerror("incompatible vectors"); }
#endif
SAME_LOC3(*this, x, A);
copyonwrite();
@@ -580,16 +577,16 @@ void NRVec<complex<double> >::gemv(const double beta, const NRMat<double> &A,
#ifdef CUDALA
if(location==cpu){
#endif
cblas_dgemv(CblasRowMajor, (trans=='n'?CblasNoTrans:CblasTrans),
cblas_dgemv(CblasRowMajor, (tolower(trans)=='n'?CblasNoTrans:CblasTrans),
A.nrows(), A.ncols(), alpha, A, A.ncols(), (double *)x.v, 2, beta, (double *)v, 2);
cblas_dgemv(CblasRowMajor, (trans=='n'?CblasNoTrans:CblasTrans),
cblas_dgemv(CblasRowMajor, (tolower(trans)=='n'?CblasNoTrans:CblasTrans),
A.nrows(), A.ncols(), alpha, A, A.ncols(), ((double *)x.v) + 1, 2, beta, ((double *)v)+1, 2);
#ifdef CUDALA
}else{
cublasDgemv((trans=='n'?'T':'N'), A.ncols(), A.nrows(), alpha, A, A.ncols(), (double*)(x.v), 2, beta, (double *)v, 2);
cublasDgemv((tolower(trans)=='n'?'T':'N'), A.ncols(), A.nrows(), alpha, A, A.ncols(), (double*)(x.v), 2, beta, (double *)v, 2);
TEST_CUBLAS("cublasDgemv");
cublasDgemv((trans=='n'?'T':'N'), A.ncols(), A.nrows(), alpha, A, A.ncols(), ((double *)x.v) + 1, 2, beta, ((double *)v)+1, 2);
cublasDgemv((tolower(trans)=='n'?'T':'N'), A.ncols(), A.nrows(), alpha, A, A.ncols(), ((double *)x.v) + 1, 2, beta, ((double *)v)+1, 2);
TEST_CUBLAS("cublasDgemv");
}
#endif
@@ -611,14 +608,14 @@ void NRVec<complex<double> >::gemv(const complex<double> beta,
const NRMat<complex<double> > &A, const char trans,
const complex<double> alpha, const NRVec<complex<double> > &x) {
#ifdef DEBUG
if ((trans == 'n'?A.ncols():A.nrows()) != x.size()){ laerror("incompatible vectors"); }
if ((tolower(trans) == 'n'?A.ncols():A.nrows()) != x.size()){ laerror("incompatible vectors"); }
#endif
SAME_LOC3(*this, x, A);
copyonwrite();
#ifdef CUDALA
if(location == cpu){
#endif
cblas_zgemv(CblasRowMajor, (trans=='n'?CblasNoTrans:CblasTrans),
cblas_zgemv(CblasRowMajor, (tolower(trans)=='n'?CblasNoTrans:CblasTrans),
A.nrows(), A.ncols(), &alpha, A, A.ncols(), x.v, 1, &beta, v, 1);
#ifdef CUDALA
}else{
@@ -626,7 +623,7 @@ void NRVec<complex<double> >::gemv(const complex<double> beta,
const cuDoubleComplex _alpha = make_cuDoubleComplex(alpha.real(), alpha.imag());
const cuDoubleComplex _beta = make_cuDoubleComplex(beta.real(), beta.imag());
cublasZgemv((trans=='n'?'T':'N'), A.ncols(), A.nrows(),
cublasZgemv((tolower(trans)=='n'?'T':'N'), A.ncols(), A.nrows(),
_alpha, (cuDoubleComplex*)(A[0]), A.ncols(), (cuDoubleComplex*)(x.v), 1, _beta, (cuDoubleComplex*)v, 1);
TEST_CUBLAS("cublasZgemv");
}