generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

changes for GFTT

Browse Source
This commit is contained in:
Ilya Lavrenov 2014-03-13 14:57:13 +04:00
parent a5afcd9f11
commit 8e79de35a8
1 changed files with 14 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
modules/ocl/src/gftt.cpp
View File

@ -146,34 +146,33 @@ static void minMaxEig_caller(const oclMat &src, oclMat &dst, oclMat & tozero)
CV_Assert(groupnum != 0); CV_Assert(groupnum != 0);
int dbsize = groupnum * 2 * src.elemSize(); int dbsize = groupnum * 2 * src.elemSize();
ensureSizeIsEnough(1, dbsize, CV_8UC1, dst); ensureSizeIsEnough(1, dbsize, CV_8UC1, dst);
cl_mem dst_data = reinterpret_cast<cl_mem>(dst.data); cl_mem dst_data = reinterpret_cast<cl_mem>(dst.data);
int all_cols = src.step / src.elemSize(); int vElemSize = src.elemSize1();
int pre_cols = (src.offset % src.step) / src.elemSize(); int src_step = src.step / vElemSize, src_offset = src.offset / vElemSize;
int sec_cols = all_cols - (src.offset % src.step + src.cols * src.elemSize() - 1) / src.elemSize() - 1; int total = src.size().area();
int invalid_cols = pre_cols + sec_cols;
int cols = all_cols - invalid_cols , elemnum = cols * src.rows;
int offset = src.offset / src.elemSize();
{// first parallel pass {
// first parallel pass
vector<pair<size_t , const void *> > args; vector<pair<size_t , const void *> > args;
args.push_back( make_pair( sizeof(cl_mem) , (void *)&src.data)); args.push_back( make_pair( sizeof(cl_mem) , (void *)&src.data));
args.push_back( make_pair( sizeof(cl_mem) , (void *)&dst_data )); args.push_back( make_pair( sizeof(cl_int) , (void *)&src_offset));
args.push_back( make_pair( sizeof(cl_int) , (void *)&cols )); args.push_back( make_pair( sizeof(cl_int) , (void *)&src_step));
args.push_back( make_pair( sizeof(cl_int) , (void *)&invalid_cols )); args.push_back( make_pair( sizeof(cl_int) , (void *)&src.rows ));
args.push_back( make_pair( sizeof(cl_int) , (void *)&offset)); args.push_back( make_pair( sizeof(cl_int) , (void *)&src.cols ));
args.push_back( make_pair( sizeof(cl_int) , (void *)&elemnum)); args.push_back( make_pair( sizeof(cl_int) , (void *)&total));
args.push_back( make_pair( sizeof(cl_int) , (void *)&groupnum)); args.push_back( make_pair( sizeof(cl_int) , (void *)&groupnum));
args.push_back( make_pair( sizeof(cl_mem) , (void *)&dst_data ));
size_t globalThreads[3] = {groupnum * 256, 1, 1}; size_t globalThreads[3] = {groupnum * 256, 1, 1};
size_t localThreads[3] = {256, 1, 1}; size_t localThreads[3] = {256, 1, 1};
openCLExecuteKernel(src.clCxt, &arithm_minMax, "arithm_op_minMax", globalThreads, localThreads, openCLExecuteKernel(src.clCxt, &arithm_minMax, "arithm_op_minMax", globalThreads, localThreads,
args, -1, -1, "-D T=float -D DEPTH_5"); args, -1, -1, "-D T=float -D DEPTH_5 -D vlen=1");
} }
{// run final "serial" kernel to find accumulate results from threads and reset corner counter {
// run final "serial" kernel to find accumulate results from threads and reset corner counter
vector<pair<size_t , const void *> > args; vector<pair<size_t , const void *> > args;
args.push_back( make_pair( sizeof(cl_mem) , (void *)&dst_data )); args.push_back( make_pair( sizeof(cl_mem) , (void *)&dst_data ));
args.push_back( make_pair( sizeof(cl_int) , (void *)&groupnum )); args.push_back( make_pair( sizeof(cl_int) , (void *)&groupnum ));