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

fix the haar kernel problems on Nvidia and Intel OCL

Browse Source
This commit is contained in:
yao 2013-02-23 15:19:46 +08:00
parent e6dd4e840d
commit 720eaf1e1a
2 changed files with 220 additions and 215 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
modules/ocl/src/haar.cpp
View File

@ -926,7 +926,7 @@ CvSeq *cv::ocl::OclCascadeClassifier::oclHaarDetectObjects( oclMat &gimg, CvMemS
if( gimg.cols < minSize.width || gimg.rows < minSize.height )
CV_Error(CV_StsError, "Image too small");
if( (flags & CV_HAAR_SCALE_IMAGE) && gimg.clCxt->impl->devName.find("Intel(R) HD Graphics") == string::npos )
if( (flags & CV_HAAR_SCALE_IMAGE) )
{
CvSize winSize0 = cascade->orig_window_size;
//float scalefactor = 1.1f;

11
modules/ocl/src/kernels/haarobjectdetect.cl
View File

@ -9,6 +9,7 @@
// Niko Li, newlife20080214@gmail.com
// Wang Weiyan, wangweiyanster@gmail.com
// Jia Haipeng, jiahaipeng95@gmail.com
// Nathan, liujun@multicorewareinc.com
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
@ -299,6 +300,7 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
}
barrier(CLK_LOCAL_MEM_FENCE);
int queuecount = lclcount[0];
barrier(CLK_LOCAL_MEM_FENCE);
nodecounter = splitnode;
for(int stageloop = split_stage; stageloop< end_stage && queuecount>0; stageloop++)
{
@ -324,7 +326,8 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
int queue_pixel = mad24(((temp_coord & (int)0xffff0000)>>16),readwidth,temp_coord & 0xffff);
//barrier(CLK_LOCAL_MEM_FENCE);
if(lcl_compute_win_id < queuecount) {
if(lcl_compute_win_id < queuecount)
{
int tempnodecounter = lcl_compute_id;
float part_sum = 0.f;
@ -363,7 +366,8 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
partialsum[lcl_id]=part_sum;
}
barrier(CLK_LOCAL_MEM_FENCE);
if(lcl_compute_win_id < queuecount) {
if(lcl_compute_win_id < queuecount)
{
for(int i=0; i<lcl_compute_win && (lcl_compute_id==0); i++)
{
stage_sum += partialsum[lcl_id+i];
@ -378,8 +382,9 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
}
barrier(CLK_LOCAL_MEM_FENCE);
}//end for(int queueloop=0;queueloop<queuecount_loop;queueloop++)
barrier(CLK_LOCAL_MEM_FENCE);
//barrier(CLK_LOCAL_MEM_FENCE);
queuecount = lclcount[0];
barrier(CLK_LOCAL_MEM_FENCE);
nodecounter += stageinfo.x;
}//end for(int stageloop = splitstage; stageloop< endstage && queuecount>0;stageloop++)
//barrier(CLK_LOCAL_MEM_FENCE);