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

Avoid compile unnecessary OpenCL kernel.

Browse Source 
Signed-off-by: Yan Wang <yan.wang@linux.intel.com>
This commit is contained in:
Yan Wang 2015-01-14 19:33:12 +08:00
parent b2a5e66fc5
commit 07ab0c1c53
2 changed files with 12 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
modules/objdetect/src/cascadedetect.cpp
View File

@ -1072,10 +1072,10 @@ bool CascadeClassifierImpl::ocl_detectMultiScaleNoGrouping( const std::vector<fl
{ {
String opts; String opts;
if (lbufSize.area()) if (lbufSize.area())
opts = format("-D LOCAL_SIZE_X=%d -D LOCAL_SIZE_Y=%d -D SUM_BUF_SIZE=%d -D SUM_BUF_STEP=%d -D NODE_COUNT=%d -D SPLIT_STAGE=%d -D N_STAGES=%d -D MAX_FACES=%d", opts = format("-D LOCAL_SIZE_X=%d -D LOCAL_SIZE_Y=%d -D SUM_BUF_SIZE=%d -D SUM_BUF_STEP=%d -D NODE_COUNT=%d -D SPLIT_STAGE=%d -D N_STAGES=%d -D MAX_FACES=%d -D HAAR",
localsz.width, localsz.height, lbufSize.area(), lbufSize.width, data.maxNodesPerTree, splitstage_ocl, nstages, MAX_FACES); localsz.width, localsz.height, lbufSize.area(), lbufSize.width, data.maxNodesPerTree, splitstage_ocl, nstages, MAX_FACES);
else else
opts = format("-D LOCAL_SIZE_X=%d -D LOCAL_SIZE_Y=%d -D NODE_COUNT=%d -D SPLIT_STAGE=%d -D N_STAGES=%d -D MAX_FACES=%d", opts = format("-D LOCAL_SIZE_X=%d -D LOCAL_SIZE_Y=%d -D NODE_COUNT=%d -D SPLIT_STAGE=%d -D N_STAGES=%d -D MAX_FACES=%d -D HAAR",
localsz.width, localsz.height, data.maxNodesPerTree, splitstage_ocl, nstages, MAX_FACES); localsz.width, localsz.height, data.maxNodesPerTree, splitstage_ocl, nstages, MAX_FACES);
haarKernel.create("runHaarClassifier", ocl::objdetect::cascadedetect_oclsrc, opts); haarKernel.create("runHaarClassifier", ocl::objdetect::cascadedetect_oclsrc, opts);
if( haarKernel.empty() ) if( haarKernel.empty() )
@ -1112,10 +1112,10 @@ bool CascadeClassifierImpl::ocl_detectMultiScaleNoGrouping( const std::vector<fl
{ {
String opts; String opts;
if (lbufSize.area()) if (lbufSize.area())
opts = format("-D LOCAL_SIZE_X=%d -D LOCAL_SIZE_Y=%d -D SUM_BUF_SIZE=%d -D SUM_BUF_STEP=%d -D SPLIT_STAGE=%d -D N_STAGES=%d -D MAX_FACES=%d", opts = format("-D LOCAL_SIZE_X=%d -D LOCAL_SIZE_Y=%d -D SUM_BUF_SIZE=%d -D SUM_BUF_STEP=%d -D SPLIT_STAGE=%d -D N_STAGES=%d -D MAX_FACES=%d -D LBP",
localsz.width, localsz.height, lbufSize.area(), lbufSize.width, splitstage_ocl, nstages, MAX_FACES); localsz.width, localsz.height, lbufSize.area(), lbufSize.width, splitstage_ocl, nstages, MAX_FACES);
else else
opts = format("-D LOCAL_SIZE_X=%d -D LOCAL_SIZE_Y=%d -D SPLIT_STAGE=%d -D N_STAGES=%d -D MAX_FACES=%d", opts = format("-D LOCAL_SIZE_X=%d -D LOCAL_SIZE_Y=%d -D SPLIT_STAGE=%d -D N_STAGES=%d -D MAX_FACES=%d -D LBP",
localsz.width, localsz.height, splitstage_ocl, nstages, MAX_FACES); localsz.width, localsz.height, splitstage_ocl, nstages, MAX_FACES);
lbpKernel.create("runLBPClassifierStumpSimple", ocl::objdetect::cascadedetect_oclsrc, opts); lbpKernel.create("runLBPClassifierStumpSimple", ocl::objdetect::cascadedetect_oclsrc, opts);
if( lbpKernel.empty() ) if( lbpKernel.empty() )

9
modules/objdetect/src/opencl/cascadedetect.cl
View File

@ -12,19 +12,22 @@
// Erping Pang, erping@multicorewareinc.com // Erping Pang, erping@multicorewareinc.com
// //
#ifdef HAAR
typedef struct __attribute__((aligned(4))) OptHaarFeature typedef struct __attribute__((aligned(4))) OptHaarFeature
{ {
int4 ofs[3] __attribute__((aligned (4))); int4 ofs[3] __attribute__((aligned (4)));
float4 weight __attribute__((aligned (4))); float4 weight __attribute__((aligned (4)));
} }
OptHaarFeature; OptHaarFeature;
#endif
#ifdef LBP
typedef struct __attribute__((aligned(4))) OptLBPFeature typedef struct __attribute__((aligned(4))) OptLBPFeature
{ {
int16 ofs __attribute__((aligned (4))); int16 ofs __attribute__((aligned (4)));
} }
OptLBPFeature; OptLBPFeature;
#endif
typedef struct __attribute__((aligned(4))) Stump typedef struct __attribute__((aligned(4))) Stump
{ {
@ -64,6 +67,7 @@ ScaleData;
#define NODE_COUNT 1 #define NODE_COUNT 1
#endif #endif
#ifdef HAAR
__kernel __attribute__((reqd_work_group_size(LOCAL_SIZE_X,LOCAL_SIZE_Y,1))) __kernel __attribute__((reqd_work_group_size(LOCAL_SIZE_X,LOCAL_SIZE_Y,1)))
void runHaarClassifier( void runHaarClassifier(
int nscales, __global const ScaleData* scaleData, int nscales, __global const ScaleData* scaleData,
@ -352,7 +356,9 @@ void runHaarClassifier(
} }
} }
} }
#endif
#ifdef LBP
#undef CALC_SUM_OFS_ #undef CALC_SUM_OFS_
#define CALC_SUM_OFS_(p0, p1, p2, p3, ptr) \ #define CALC_SUM_OFS_(p0, p1, p2, p3, ptr) \
((ptr)[p0] - (ptr)[p1] - (ptr)[p2] + (ptr)[p3]) ((ptr)[p0] - (ptr)[p1] - (ptr)[p2] + (ptr)[p3])
@ -651,3 +657,4 @@ void runLBPClassifierStump(
} }
} }
} }
#endif