Moments ocl_moments(InputArray _contour) //for contour
{
CvMoments mom;
memset(&mom, 0, sizeof(mom));
Mat arr = _contour.getMat();
CvMat c_array = arr;
const void* array = &c_array;
CvSeq* contour = 0;
if( CV_IS_SEQ( array ))
{
contour = (CvSeq*)(array);
if( !CV_IS_SEQ_POINT_SET( contour ))
CV_Error( CV_StsBadArg, "The passed sequence is not a valid contour" );
}
int type, coi = 0;
CvMat stub, *mat = (CvMat*)(array);
CvContour contourHeader;
CvSeqBlock block;
if( !contour )
{
mat = cvGetMat( mat, &stub, &coi );
type = CV_MAT_TYPE( mat->type );
if( type == CV_32SC2 || type == CV_32FC2 )
{
contour = cvPointSeqFromMat(
CV_SEQ_KIND_CURVE | CV_SEQ_FLAG_CLOSED,
mat, &contourHeader, &block );
}
}
CV_Assert(contour);
icvContourMoments(contour, &mom);
return mom;
}
static void ocl_cvMoments( const void* array, CvMoments* mom, int binary )
{
const int TILE_SIZE = 256;
int type, depth, cn, coi = 0;
CvMat stub, *mat = (CvMat*)array;
CvContour contourHeader;
CvSeq* contour = 0;
CvSeqBlock block;
if( CV_IS_SEQ( array ))
{
contour = (CvSeq*)array;
if( !CV_IS_SEQ_POINT_SET( contour ))
CV_Error( CV_StsBadArg, "The passed sequence is not a valid contour" );
}
if( !moments )
CV_Error( CV_StsNullPtr, "" );
memset( mom, 0, sizeof(*mom));
if( !contour )
{
mat = cvGetMat( mat, &stub, &coi );
type = CV_MAT_TYPE( mat->type );
if( type == CV_32SC2 || type == CV_32FC2 )
{
contour = cvPointSeqFromMat(
CV_SEQ_KIND_CURVE | CV_SEQ_FLAG_CLOSED,
mat, &contourHeader, &block );
}
}
if( contour )
{
icvContourMoments( contour, mom );
return;
}
type = CV_MAT_TYPE( mat->type );
depth = CV_MAT_DEPTH( type );
cn = CV_MAT_CN( type );
cv::Size size = cvGetMatSize( mat );
if( cn > 1 && coi == 0 )
CV_Error( CV_StsBadArg, "Invalid image type" );
if( size.width <= 0 || size.height <= 0 )
return;
cv::Mat src0(mat);
cv::ocl::oclMat src(src0);
cv::Size tileSize;
int blockx,blocky;
if(size.width%TILE_SIZE == 0)
blockx = size.width/TILE_SIZE;
else
blockx = size.width/TILE_SIZE + 1;
if(size.height%TILE_SIZE == 0)
blocky = size.height/TILE_SIZE;
else
blocky = size.height/TILE_SIZE + 1;
cv::ocl::oclMat dst_m(blocky * 10, blockx, CV_64FC1);
cl_mem sum = openCLCreateBuffer(src.clCxt,CL_MEM_READ_WRITE,10*sizeof(double));
int tile_width = std::min(size.width,TILE_SIZE);
int tile_height = std::min(size.height,TILE_SIZE);
size_t localThreads[3] = { tile_height, 1, 1};
size_t globalThreads[3] = { size.height, blockx, 1};
std::vector<std::pair<size_t , const void *> > args,args_sum;
args.push_back( std::make_pair( sizeof(cl_mem) , (void *)&src.data ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&src.rows ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&src.cols ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&src.step ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&tileSize.width ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&tileSize.height ));
args.push_back( std::make_pair( sizeof(cl_mem) , (void *)&dst_m.data ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&dst_m.cols ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&dst_m.step ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&blocky ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&type ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&depth ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&cn ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&coi ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&binary ));
args.push_back( std::make_pair( sizeof(cl_int) , (void *)&TILE_SIZE ));
openCLExecuteKernel(dst_m.clCxt, &moments, "CvMoments", globalThreads, localThreads, args, -1, depth);
size_t localThreadss[3] = { 128, 1, 1};
size_t globalThreadss[3] = { 128, 1, 1};
args_sum.push_back( std::make_pair( sizeof(cl_int) , (void *)&src.rows ));
args_sum.push_back( std::make_pair( sizeof(cl_int) , (void *)&src.cols ));
args_sum.push_back( std::make_pair( sizeof(cl_int) , (void *)&tile_height ));
args_sum.push_back( std::make_pair( sizeof(cl_int) , (void *)&tile_width ));
args_sum.push_back( std::make_pair( sizeof(cl_int) , (void *)&TILE_SIZE ));
args_sum.push_back( std::make_pair( sizeof(cl_mem) , (void *)&sum ));
args_sum.push_back( std::make_pair( sizeof(cl_mem) , (void *)&dst_m.data ));
args_sum.push_back( std::make_pair( sizeof(cl_int) , (void *)&dst_m.step ));
openCLExecuteKernel(dst_m.clCxt, &moments, "dst_sum", globalThreadss, localThreadss, args_sum, -1, -1);
double* dstsum = new double[10];
memset(dstsum,0,10*sizeof(double));
openCLReadBuffer(dst_m.clCxt,sum,(void *)dstsum,10*sizeof(double));
mom->m00 = dstsum[0];
mom->m10 = dstsum[1];
mom->m01 = dstsum[2];
mom->m20 = dstsum[3];
mom->m11 = dstsum[4];
mom->m02 = dstsum[5];
mom->m30 = dstsum[6];
mom->m21 = dstsum[7];
mom->m12 = dstsum[8];
mom->m03 = dstsum[9];
delete [] dstsum;
icvCompleteMomentState( mom );
}
