PARDISOSolver< valueType, zeroBased >::~PARDISOSolver()
{
int options = MKL_DSS_DEFAULTS;
dss_delete( m_handle, options );
}
void CPoissonExt::poissonExtend(cv::Mat &dst,int size)
{
cv::Mat gx=Mat::zeros(h+2*ex,w+2*ex,CV_32FC4);
cv::Mat gy=Mat::zeros(h+2*ex,w+2*ex,CV_32FC4);
#pragma omp parallel for
for(int y=0;y<h+ex*2;y++)
for (int x=0;x<w+ex*2;x++)
{
if(type[y*(w+2*ex)+x]>1)
{
Vec4f BGRA0;
Vec4f BGRA1;
BGRA1=dst.at<Vec4b>(y,x);
if(x>0)
{
if(type[y*(w+2*ex)+x-1]>1)
{
BGRA0=dst.at<Vec4b>(y,x-1);
if(BGRA0!=Vec4f(255,0,255,0)&&BGRA1!=Vec4f(255,0,255,0))
gx.at<Vec4f>(y,x)=BGRA1-BGRA0;
}
}
if(y>0)
{
if(type[(y-1)*(w+2*ex)+x]>1)
{
BGRA0=dst.at<Vec4b>(y-1,x);
if(BGRA0!=Vec4f(255,0,255,0)&&BGRA1!=Vec4f(255,0,255,0))
gy.at<Vec4f>(y,x)=BGRA1-BGRA0;
}
}
}
}
//matrix
float *A=new float[5*size];
_INTEGER_t *columns=new _INTEGER_t[5*size];
_INTEGER_t *rowindex=new _INTEGER_t[size+1];
float *B0=new float[size];
float *B1=new float[size];
float *B2=new float[size];
float *B3=new float[size];
float *X0=new float[size];
float *X1=new float[size];
float *X2=new float[size];
//float *X3=new float[size];
_INTEGER_t nNonZeros=0;
memset(A,0,5*size*sizeof(float));
memset(columns,0,5*size*sizeof(int));
memset(rowindex,0,(size+1)*sizeof(int));
memset(B0,0,size*sizeof(float));
memset(B1,0,size*sizeof(float));
memset(B2,0,size*sizeof(float));
memset(B3,0,size*sizeof(float));
memset(X0,0,size*sizeof(float));
memset(X1,0,size*sizeof(float));
memset(X2,0,size*sizeof(float));
//memset(X3,0,size*sizeof(float));
Vec4f BGRA;
for(int y=0;y<h+ex*2;y++)
for (int x=0;x<w+ex*2;x++)
{
float a[5];
a[0]=a[1]=a[2]=a[3]=a[4]=0.0f;
int ii=y*(w+2*ex)+x;
switch(type[ii])
{
case 0://inside
break;
case 1://boundary
a[2]+=1.0f;
BGRA=dst.at<Vec4b>(y,x);
B0[index[ii]]+=BGRA[0];
B1[index[ii]]+=BGRA[1];
B2[index[ii]]+=BGRA[2];
//B3[index[ii]]+=BGRA[3];
case 2://outside
if(y-1>=0&&type[ii-(w+2*ex)]>0)
{
a[2]+=1.0f;
a[0]-=1.0f;
BGRA=gy.at<Vec4f>(y,x);
B0[index[ii]]+=BGRA[0];
B1[index[ii]]+=BGRA[1];
B2[index[ii]]+=BGRA[2];
//B3[index[ii]]+=BGRA[3];
}
if(x-1>=0&&type[ii-1]>0)
{
a[2]+=1.0f;
a[1]-=1.0f;
BGRA=gx.at<Vec4f>(y,x);
B0[index[ii]]+=BGRA[0];
B1[index[ii]]+=BGRA[1];
B2[index[ii]]+=BGRA[2];
//B3[index[ii]]+=BGRA[3];
}
if(x+1<w+2*ex&&type[ii+1]>0)
{
a[2]+=1.0f;
a[3]-=1.0f;
BGRA=gx.at<Vec4f>(y,x+1);
B0[index[ii]]-=BGRA[0];
B1[index[ii]]-=BGRA[1];
B2[index[ii]]-=BGRA[2];
//B3[index[ii]]-=BGRA[3];
}
if(y+1<h+2*ex&&type[ii+(w+2*ex)]>0)
{
a[2]+=1.0f;
a[4]-=1.0f;
BGRA=gy.at<Vec4f>(y+1,x);
B0[index[ii]]-=BGRA[0];
B1[index[ii]]-=BGRA[1];
B2[index[ii]]-=BGRA[2];
//B3[index[ii]]-=BGRA[3];
}
//put into A
if(a[0]!=0)
{
A[nNonZeros]=a[0];
columns[nNonZeros]=index[ii-(w+2*ex)];
nNonZeros++;
}
if(a[1]!=0)
{
A[nNonZeros]=a[1];
columns[nNonZeros]=index[ii-1];
nNonZeros++;
}
if(a[2]!=0)
{
A[nNonZeros]=a[2];
columns[nNonZeros]=index[ii];
nNonZeros++;
}
if(a[3]!=0)
{
A[nNonZeros]=a[3];
columns[nNonZeros]=index[ii+1];
nNonZeros++;
}
if(a[4]!=0)
{
A[nNonZeros]=a[4];
columns[nNonZeros]=index[ii+(w+2*ex)];
nNonZeros++;
}
rowindex[index[ii]+1]=nNonZeros;
break;
}
}
//SOLVER
_INTEGER_t error;
_MKL_DSS_HANDLE_t solver;
_INTEGER_t opt=MKL_DSS_MSG_LVL_WARNING + MKL_DSS_TERM_LVL_ERROR + MKL_DSS_SINGLE_PRECISION + MKL_DSS_ZERO_BASED_INDEXING;
_INTEGER_t sym=MKL_DSS_NON_SYMMETRIC;
_INTEGER_t typ=MKL_DSS_POSITIVE_DEFINITE;
_INTEGER_t ord=MKL_DSS_AUTO_ORDER;
_INTEGER_t sov=MKL_DSS_DEFAULTS;
_INTEGER_t nRhs = 1;
_INTEGER_t size_l=size;
error= dss_create(solver, opt);
error = dss_define_structure(solver,sym, rowindex, size_l, size_l,columns, nNonZeros);
error = dss_reorder( solver,ord, 0);
error = dss_factor_real( solver, typ, A );
error = dss_solve_real( solver, sov, B0, nRhs, X0 );
error = dss_solve_real( solver, sov, B1, nRhs, X1 );
error = dss_solve_real( solver, sov, B2, nRhs, X2 );
//error = dss_solve_real( solver, sov, B3, nRhs, X3 );
error = dss_delete( solver, opt );
//paste
#pragma omp parallel for
for(int y=0;y<h+ex*2;y++)
for (int x=0;x<w+ex*2;x++)
{
int ii=y*(w+2*ex)+x;
if(type[ii]>0)
{
int B=MIN(MAX(X0[index[ii]],0),255);
int G=MIN(MAX(X1[index[ii]],0),255);
int R=MIN(MAX(X2[index[ii]],0),255);
int A=0;
dst.at<Vec4b>(y,x)=Vec4b(B,G,R,A);
}
}
delete[] A;
delete[] B0;
delete[] B1;
delete[] B2;
delete[] B3;
delete[] X0;
delete[] X1;
delete[] X2;
//delete[] X3;
delete[] columns;
delete[] rowindex;
}
