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

Optimize the performance of cascade OpenCL kernel.

Browse Source 
1. Use built-in mad() instead of += and *.
2. For stump stages, if weight.z == 0, avoid unnecessary calculation
because some features only have 2 rectangles..

It could improve
OCL_Cascade_Image_MinSize_CascadeClassifier.CascadeClassifier/*
about 10% mean values.

Signed-off-by: Yan Wang <yan.wang@linux.intel.com>
This commit is contained in:
Yan Wang 2015-03-23 17:47:40 +08:00
parent 5a0c7a1330
commit 433c3da7cf
1 changed files with 9 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
modules/objdetect/src/opencl/cascadedetect.cl
View File

@ -180,11 +180,11 @@ void runHaarClassifier(
int4 ofs = f->ofs[0];
sval = (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.x;
ofs = f->ofs[1];
sval += (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.y;
sval = mad((psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w]), weight.y, sval);
if( weight.z > 0 )
{
ofs = f->ofs[2];
sval += (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.z;
sval = mad((psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w]), weight.z, sval);
}
s += (sval < st.y*nf) ? st.z : st.w;
@ -204,11 +204,11 @@ void runHaarClassifier(
sval = (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.x;
ofs = f->ofs[1];
sval += (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.y;
sval = mad((psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w]), weight.y, sval);
if( weight.z > 0 )
{
ofs = f->ofs[2];
sval += (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.z;
sval = mad((psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w]), weight.z, sval);
}
idx = (sval < as_float(n.y)*nf) ? n.z : n.w;
@ -281,11 +281,12 @@ void runHaarClassifier(
int4 ofs = f->ofs[0];
float sval = (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.x;
ofs = f->ofs[1];
sval += (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.y;
sval = mad((psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w]), weight.y, sval);
//if( weight.z > 0 )
if( fabs(weight.z) > 0 )
{
ofs = f->ofs[2];
sval += (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.z;
sval = mad((psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w]), weight.z, sval);
}
partsum += (sval < st.y*nf) ? st.z : st.w;
@ -303,11 +304,11 @@ void runHaarClassifier(
float sval = (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.x;
ofs = f->ofs[1];
sval += (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.y;
sval = mad((psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w]), weight.y, sval);
if( weight.z > 0 )
{
ofs = f->ofs[2];
sval += (psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w])*weight.z;
sval = mad((psum[ofs.x] - psum[ofs.y] - psum[ofs.z] + psum[ofs.w]), weight.z, sval);
}
idx = (sval < as_float(n.y)*nf) ? n.z : n.w;