mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-06-10 19:24:07 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Fix white-spacing

Browse Source
This commit is contained in:
peng xiao 2013-06-27 09:57:42 +08:00
parent 9270205947
commit dcc4766129
3 changed files with 207 additions and 197 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
modules/ocl/perf/perf_opticalflow.cpp
View File

@ -136,11 +136,13 @@ PERFTEST(PyrLKOpticalFlow)
size_t mismatch = 0; size_t mismatch = 0;
for (int i = 0; i < (int)nextPts.size(); ++i) for (int i = 0; i < (int)nextPts.size(); ++i)
{ {
if(status[i] != ocl_status.at<unsigned char>(0, i)){ if(status[i] != ocl_status.at<unsigned char>(0, i))
{
mismatch++; mismatch++;
continue; continue;
} }
if(status[i]){ if(status[i])
{
Point2f gpu_rst = ocl_nextPts.at<Point2f>(0, i); Point2f gpu_rst = ocl_nextPts.at<Point2f>(0, i);
Point2f cpu_rst = nextPts[i]; Point2f cpu_rst = nextPts[i];
if(fabs(gpu_rst.x - cpu_rst.x) >= 1. || fabs(gpu_rst.y - cpu_rst.y) >= 1.) if(fabs(gpu_rst.x - cpu_rst.x) >= 1. || fabs(gpu_rst.y - cpu_rst.y) >= 1.)
@ -193,7 +195,7 @@ PERFTEST(tvl1flow)
WARMUP_ON; WARMUP_ON;
d_alg(d0, d1, d_flowx, d_flowy); d_alg(d0, d1, d_flowx, d_flowy);
WARMUP_OFF; WARMUP_OFF;
/* /*
double diff1 = 0.0, diff2 = 0.0; double diff1 = 0.0, diff2 = 0.0;
if(ExceptedMatSimilar(gold[0], cv::Mat(d_flowx), 3e-3, diff1) == 1 if(ExceptedMatSimilar(gold[0], cv::Mat(d_flowx), 3e-3, diff1) == 1
&&ExceptedMatSimilar(gold[1], cv::Mat(d_flowy), 3e-3, diff2) == 1) &&ExceptedMatSimilar(gold[1], cv::Mat(d_flowy), 3e-3, diff2) == 1)
@ -203,7 +205,7 @@ PERFTEST(tvl1flow)
TestSystem::instance().setDiff(diff1); TestSystem::instance().setDiff(diff1);
TestSystem::instance().setDiff(diff2); TestSystem::instance().setDiff(diff2);
*/ */
GPU_ON; GPU_ON;
@ -352,4 +354,3 @@ PERFTEST(FarnebackOpticalFlow)
} }
} }
} }

30
modules/ocl/src/opencl/optical_flow_farneback.cl
View File

@ -223,7 +223,8 @@ __kernel void updateMatrices(__global float * M,
int x1 = convert_int(floor(fx)); int x1 = convert_int(floor(fx));
int y1 = convert_int(floor(fy)); int y1 = convert_int(floor(fy));
fx -= x1; fy -= y1; fx -= x1;
fy -= y1;
float r2, r3, r4, r5, r6; float r2, r3, r4, r5, r6;
@ -283,8 +284,11 @@ __kernel void updateMatrices(__global float * M,
c_border[min(width - x - 1, BORDER_SIZE)] * c_border[min(width - x - 1, BORDER_SIZE)] *
c_border[min(height - y - 1, BORDER_SIZE)]; c_border[min(height - y - 1, BORDER_SIZE)];
r2 *= scale; r3 *= scale; r4 *= scale; r2 *= scale;
r5 *= scale; r6 *= scale; r3 *= scale;
r4 *= scale;
r5 *= scale;
r6 *= scale;
M[mad24(y, mStep, x)] = r4*r4 + r6*r6; M[mad24(y, mStep, x)] = r4*r4 + r6*r6;
M[mad24(height + y, mStep, x)] = (r4 + r5)*r6; M[mad24(height + y, mStep, x)] = (r4 + r5)*r6;
@ -319,12 +323,12 @@ __kernel void boxFilter5(__global float * dst,
int xExt = (int)(bx * bdx) + i - ksizeHalf; int xExt = (int)(bx * bdx) + i - ksizeHalf;
xExt = min(max(xExt, 0), width - 1); xExt = min(max(xExt, 0), width - 1);
#pragma unroll #pragma unroll
for (int k = 0; k < 5; ++k) for (int k = 0; k < 5; ++k)
row[k*smw + i] = src[mad24(k*height + y, srcStep, xExt)]; row[k*smw + i] = src[mad24(k*height + y, srcStep, xExt)];
for (int j = 1; j <= ksizeHalf; ++j) for (int j = 1; j <= ksizeHalf; ++j)
#pragma unroll #pragma unroll
for (int k = 0; k < 5; ++k) for (int k = 0; k < 5; ++k)
row[k*smw + i] += row[k*smw + i] +=
src[mad24(k*height + max(y - j, 0), srcStep, xExt)] + src[mad24(k*height + max(y - j, 0), srcStep, xExt)] +
@ -341,16 +345,16 @@ __kernel void boxFilter5(__global float * dst,
row += tx + ksizeHalf; row += tx + ksizeHalf;
float res[5]; float res[5];
#pragma unroll #pragma unroll
for (int k = 0; k < 5; ++k) for (int k = 0; k < 5; ++k)
res[k] = row[k*smw]; res[k] = row[k*smw];
for (int i = 1; i <= ksizeHalf; ++i) for (int i = 1; i <= ksizeHalf; ++i)
#pragma unroll #pragma unroll
for (int k = 0; k < 5; ++k) for (int k = 0; k < 5; ++k)
res[k] += row[k*smw - i] + row[k*smw + i]; res[k] += row[k*smw - i] + row[k*smw + i];
#pragma unroll #pragma unroll
for (int k = 0; k < 5; ++k) for (int k = 0; k < 5; ++k)
dst[mad24(k*height + y, dstStep, x)] = res[k] * boxAreaInv; dst[mad24(k*height + y, dstStep, x)] = res[k] * boxAreaInv;
} }
@ -408,12 +412,12 @@ __kernel void gaussianBlur5(__global float * dst,
int xExt = (int)(bx * bdx) + i - ksizeHalf; int xExt = (int)(bx * bdx) + i - ksizeHalf;
xExt = idx_col(xExt, width - 1); xExt = idx_col(xExt, width - 1);
#pragma unroll #pragma unroll
for (int k = 0; k < 5; ++k) for (int k = 0; k < 5; ++k)
row[k*smw + i] = src[mad24(k*height + y, srcStep, xExt)] * c_gKer[0]; row[k*smw + i] = src[mad24(k*height + y, srcStep, xExt)] * c_gKer[0];
for (int j = 1; j <= ksizeHalf; ++j) for (int j = 1; j <= ksizeHalf; ++j)
#pragma unroll #pragma unroll
for (int k = 0; k < 5; ++k) for (int k = 0; k < 5; ++k)
row[k*smw + i] += row[k*smw + i] +=
(src[mad24(k*height + idx_row_low(y - j, height - 1), srcStep, xExt)] + (src[mad24(k*height + idx_row_low(y - j, height - 1), srcStep, xExt)] +
@ -430,16 +434,16 @@ __kernel void gaussianBlur5(__global float * dst,
row += tx + ksizeHalf; row += tx + ksizeHalf;
float res[5]; float res[5];
#pragma unroll #pragma unroll
for (int k = 0; k < 5; ++k) for (int k = 0; k < 5; ++k)
res[k] = row[k*smw] * c_gKer[0]; res[k] = row[k*smw] * c_gKer[0];
for (int i = 1; i <= ksizeHalf; ++i) for (int i = 1; i <= ksizeHalf; ++i)
#pragma unroll #pragma unroll
for (int k = 0; k < 5; ++k) for (int k = 0; k < 5; ++k)
res[k] += (row[k*smw - i] + row[k*smw + i]) * c_gKer[i]; res[k] += (row[k*smw - i] + row[k*smw + i]) * c_gKer[i];
#pragma unroll #pragma unroll
for (int k = 0; k < 5; ++k) for (int k = 0; k < 5; ++k)
dst[mad24(k*height + y, dstStep, x)] = res[k]; dst[mad24(k*height + y, dstStep, x)] = res[k];
} }

83
modules/ocl/src/optical_flow_farneback.cpp
View File

@ -55,35 +55,37 @@ using namespace cv::ocl;
namespace cv namespace cv
{ {
namespace ocl namespace ocl
{ {
///////////////////////////OpenCL kernel strings/////////////////////////// ///////////////////////////OpenCL kernel strings///////////////////////////
extern const char *optical_flow_farneback; extern const char *optical_flow_farneback;
} }
} }
namespace cv { namespace ocl { namespace optflow_farneback namespace cv {
namespace ocl {
namespace optflow_farneback
{ {
oclMat g; oclMat g;
oclMat xg; oclMat xg;
oclMat xxg; oclMat xxg;
oclMat gKer; oclMat gKer;
float ig[4]; float ig[4];
inline int divUp(int total, int grain) inline int divUp(int total, int grain)
{ {
return (total + grain - 1) / grain; return (total + grain - 1) / grain;
} }
inline void setGaussianBlurKernel(const float *c_gKer, int ksizeHalf) inline void setGaussianBlurKernel(const float *c_gKer, int ksizeHalf)
{ {
cv::Mat t_gKer(1, ksizeHalf + 1, CV_32FC1, const_cast<float *>(c_gKer)); cv::Mat t_gKer(1, ksizeHalf + 1, CV_32FC1, const_cast<float *>(c_gKer));
gKer.upload(t_gKer); gKer.upload(t_gKer);
} }
static void gaussianBlurOcl(const oclMat &src, int ksizeHalf, oclMat &dst) static void gaussianBlurOcl(const oclMat &src, int ksizeHalf, oclMat &dst)
{ {
string kernelName("gaussianBlur"); string kernelName("gaussianBlur");
size_t localThreads[3] = { 256, 1, 1 }; size_t localThreads[3] = { 256, 1, 1 };
size_t globalThreads[3] = { divUp(src.cols, localThreads[0]) * localThreads[0], src.rows, 1 }; size_t globalThreads[3] = { divUp(src.cols, localThreads[0]) * localThreads[0], src.rows, 1 };
@ -103,10 +105,10 @@ namespace cv { namespace ocl { namespace optflow_farneback
openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName, openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName,
globalThreads, localThreads, args, -1, -1); globalThreads, localThreads, args, -1, -1);
} }
static void polynomialExpansionOcl(const oclMat &src, int polyN, oclMat &dst) static void polynomialExpansionOcl(const oclMat &src, int polyN, oclMat &dst)
{ {
string kernelName("polynomialExpansion"); string kernelName("polynomialExpansion");
size_t localThreads[3] = { 256, 1, 1 }; size_t localThreads[3] = { 256, 1, 1 };
size_t globalThreads[3] = { divUp(src.cols, localThreads[0] - 2*polyN) * localThreads[0], src.rows, 1 }; size_t globalThreads[3] = { divUp(src.cols, localThreads[0] - 2*polyN) * localThreads[0], src.rows, 1 };
@ -130,15 +132,16 @@ namespace cv { namespace ocl { namespace optflow_farneback
openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName, openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName,
globalThreads, localThreads, args, -1, -1, opt); globalThreads, localThreads, args, -1, -1, opt);
} }
static void updateMatricesOcl(const oclMat &flowx, const oclMat &flowy, const oclMat &R0, const oclMat &R1, oclMat &M) static void updateMatricesOcl(const oclMat &flowx, const oclMat &flowy, const oclMat &R0, const oclMat &R1, oclMat &M)
{ {
string kernelName("updateMatrices"); string kernelName("updateMatrices");
size_t localThreads[3] = { 32, 8, 1 }; size_t localThreads[3] = { 32, 8, 1 };
size_t globalThreads[3] = { divUp(flowx.cols, localThreads[0]) * localThreads[0], size_t globalThreads[3] = { divUp(flowx.cols, localThreads[0]) * localThreads[0],
divUp(flowx.rows, localThreads[1]) * localThreads[1], divUp(flowx.rows, localThreads[1]) * localThreads[1],
1 }; 1
};
std::vector< std::pair<size_t, const void *> > args; std::vector< std::pair<size_t, const void *> > args;
args.push_back(std::make_pair(sizeof(cl_mem), (void *)&M.data)); args.push_back(std::make_pair(sizeof(cl_mem), (void *)&M.data));
@ -156,10 +159,10 @@ namespace cv { namespace ocl { namespace optflow_farneback
openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName, openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName,
globalThreads, localThreads, args, -1, -1); globalThreads, localThreads, args, -1, -1);
} }
static void boxFilter5Ocl(const oclMat &src, int ksizeHalf, oclMat &dst) static void boxFilter5Ocl(const oclMat &src, int ksizeHalf, oclMat &dst)
{ {
string kernelName("boxFilter5"); string kernelName("boxFilter5");
int height = src.rows / 5; int height = src.rows / 5;
size_t localThreads[3] = { 256, 1, 1 }; size_t localThreads[3] = { 256, 1, 1 };
@ -178,16 +181,17 @@ namespace cv { namespace ocl { namespace optflow_farneback
openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName, openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName,
globalThreads, localThreads, args, -1, -1); globalThreads, localThreads, args, -1, -1);
} }
static void updateFlowOcl(const oclMat &M, oclMat &flowx, oclMat &flowy) static void updateFlowOcl(const oclMat &M, oclMat &flowx, oclMat &flowy)
{ {
string kernelName("updateFlow"); string kernelName("updateFlow");
int cols = divUp(flowx.cols, 4); int cols = divUp(flowx.cols, 4);
size_t localThreads[3] = { 32, 8, 1 }; size_t localThreads[3] = { 32, 8, 1 };
size_t globalThreads[3] = { divUp(cols, localThreads[0]) * localThreads[0], size_t globalThreads[3] = { divUp(cols, localThreads[0]) * localThreads[0],
divUp(flowx.rows, localThreads[1]) * localThreads[0], divUp(flowx.rows, localThreads[1]) * localThreads[0],
1 }; 1
};
std::vector< std::pair<size_t, const void *> > args; std::vector< std::pair<size_t, const void *> > args;
args.push_back(std::make_pair(sizeof(cl_mem), (void *)&flowx.data)); args.push_back(std::make_pair(sizeof(cl_mem), (void *)&flowx.data));
@ -201,10 +205,10 @@ namespace cv { namespace ocl { namespace optflow_farneback
openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName, openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName,
globalThreads, localThreads, args, -1, -1); globalThreads, localThreads, args, -1, -1);
} }
static void gaussianBlur5Ocl(const oclMat &src, int ksizeHalf, oclMat &dst) static void gaussianBlur5Ocl(const oclMat &src, int ksizeHalf, oclMat &dst)
{ {
string kernelName("gaussianBlur5"); string kernelName("gaussianBlur5");
int height = src.rows / 5; int height = src.rows / 5;
int width = src.cols; int width = src.cols;
@ -225,8 +229,10 @@ namespace cv { namespace ocl { namespace optflow_farneback
openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName, openCLExecuteKernel(Context::getContext(), &optical_flow_farneback, kernelName,
globalThreads, localThreads, args, -1, -1); globalThreads, localThreads, args, -1, -1);
} }
}}} // namespace cv { namespace ocl { namespace optflow_farneback }
}
} // namespace cv { namespace ocl { namespace optflow_farneback
static oclMat allocMatFromBuf(int rows, int cols, int type, oclMat &mat) static oclMat allocMatFromBuf(int rows, int cols, int type, oclMat &mat)
{ {
@ -504,4 +510,3 @@ void cv::ocl::FarnebackOpticalFlow::operator ()(
flowx = curFlowX; flowx = curFlowX;
flowy = curFlowY; flowy = curFlowY;
} }