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

1) more convenient naming for samples gpu

Browse Source 
2) added mask support to device 'transform' function 
3) sample hog gpu: waitKey(1) -> waitKey(3), in other case image is not displayed.
This commit is contained in:
Anatoly Baksheev 2010-11-24 09:43:17 +00:00
parent 790cd2ef28
commit 0e43976259
4 changed files with 33 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
modules/gpu/src/cuda/mathfunc.cu
View File

@ -217,7 +217,7 @@ namespace cv { namespace gpu { namespace mathfunc
template <typename T1, typename T2>
struct NotEqual
{
__device__ uchar operator()(const T1& src1, const T2& src2, int, int)
__device__ uchar operator()(const T1& src1, const T2& src2)
{
return static_cast<uchar>(static_cast<int>(src1 != src2) * 255);
}

38
modules/gpu/src/cuda/transform.hpp
View File

@ -47,29 +47,47 @@
namespace cv { namespace gpu { namespace device
{
template <typename T, typename D, typename UnOp>
static __global__ void transform(const DevMem2D_<T> src, PtrStep_<D> dst, UnOp op)
//! Mask accessor
template<class T> struct MaskReader_
{
PtrStep_<T> mask;
explicit MaskReader_(PtrStep_<T> mask): mask(mask) {}
__device__ bool operator()(int y, int x) const { return mask.ptr(y)[x]; }
};
//! Stub mask accessor
struct NoMask
{
__device__ bool operator()(int y, int x) const { return true; }
};
//! Transform kernels
template <typename T, typename D, typename Mask, typename UnOp>
static __global__ void transform(const DevMem2D_<T> src, PtrStep_<D> dst, const Mask mask, UnOp op)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < src.cols && y < src.rows)
if (x < src.cols && y < src.rows && mask(y, x))
{
T src_data = src.ptr(y)[x];
dst.ptr(y)[x] = op(src_data, x, y);
dst.ptr(y)[x] = op(src_data);
}
}
template <typename T1, typename T2, typename D, typename BinOp>
static __global__ void transform(const DevMem2D_<T1> src1, const PtrStep_<T2> src2, PtrStep_<D> dst, BinOp op)
template <typename T1, typename T2, typename D, typename Mask, typename BinOp>
static __global__ void transform(const DevMem2D_<T1> src1, const PtrStep_<T2> src2, PtrStep_<D> dst, const Mask mask, BinOp op)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < src1.cols && y < src1.rows)
if (x < src1.cols && y < src1.rows && mask(y, x))
{
T1 src1_data = src1.ptr(y)[x];
T2 src2_data = src2.ptr(y)[x];
dst.ptr(y)[x] = op(src1_data, src2_data, x, y);
dst.ptr(y)[x] = op(src1_data, src2_data);
}
}
}}}
@ -87,7 +105,7 @@ namespace cv
grid.x = divUp(src.cols, threads.x);
grid.y = divUp(src.rows, threads.y);
device::transform<T, D, UnOp><<<grid, threads, 0, stream>>>(src, dst, op);
device::transform<T, D, UnOp><<<grid, threads, 0, stream>>>(src, dst, device::NoMask(), op);
if (stream == 0)
cudaSafeCall( cudaThreadSynchronize() );
@ -101,7 +119,7 @@ namespace cv
grid.x = divUp(src1.cols, threads.x);
grid.y = divUp(src1.rows, threads.y);
device::transform<T1, T2, D, BinOp><<<grid, threads, 0, stream>>>(src1, src2, dst, op);
device::transform<T1, T2, D><<<grid, threads, 0, stream>>>(src1, src2, dst, device::NoMask(), op);
if (stream == 0)
cudaSafeCall( cudaThreadSynchronize() );

4
samples/gpu/CMakeLists.txt
View File

@ -24,11 +24,11 @@ if (BUILD_EXAMPLES)
# Define executable targets
# ---------------------------------------------
MACRO(MY_DEFINE_EXAMPLE name srcs)
set(the_target "example_${name}")
set(the_target "example_gpu_${name}")
add_executable(${the_target} ${srcs})
set_target_properties(${the_target} PROPERTIES
OUTPUT_NAME "${name}"
PROJECT_LABEL "(EXAMPLE) ${name}")
PROJECT_LABEL "(EXAMPLE_GPU) ${name}")
add_dependencies(${the_target} opencv_core opencv_flann opencv_imgproc opencv_highgui
opencv_ml opencv_video opencv_objdetect opencv_features2d
opencv_calib3d opencv_legacy opencv_contrib opencv_gpu)

2
samples/gpu/hog.cpp
View File

@ -283,7 +283,7 @@ void App::RunOpencvGui()
// Show results
putText(img_to_show, GetPerformanceSummary(), Point(5, 25), FONT_HERSHEY_SIMPLEX, 1.0, Scalar(0, 0, 255), 2);
imshow("opencv_gpu_hog", img_to_show);
HandleKey((char)waitKey(1));
HandleKey((char)waitKey(3));
if (settings.src_is_video)
{