Chiebot-Mirror
/
opencv
mirror of https://github.com/opencv/opencv.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

fixed several bugs in gpu arithm functions

Browse Source 
refactored tests for them
pull/13383/head
Vladislav Vinogradov 13 years ago
parent f58c40bfab
commit 844bdea5ac
5 changed files with 1601 additions and 780 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 4
      modules/gpu/include/opencv2/gpu/gpu.hpp
  2. 4
      modules/gpu/src/cuda/element_operations.cu
  3. 26
      modules/gpu/src/element_operations.cpp
  4. 1804
      modules/gpu/test/test_arithm.cpp
  5. 31
      modules/gpu/test/utility.hpp

4
modules/gpu/include/opencv2/gpu/gpu.hpp
Unescape View File

@ -638,11 +638,11 @@ CV_EXPORTS void bitwise_xor(const GpuMat& src1, const Scalar& sc, GpuMat& dst, S
//! pixel by pixel right shift of an image by a constant value //! pixel by pixel right shift of an image by a constant value
//! supports 1, 3 and 4 channels images with integers elements //! supports 1, 3 and 4 channels images with integers elements
CV_EXPORTS void rshift(const GpuMat& src, const Scalar& sc, GpuMat& dst, Stream& stream = Stream::Null()); CV_EXPORTS void rshift(const GpuMat& src, Scalar_<int> sc, GpuMat& dst, Stream& stream = Stream::Null());
//! pixel by pixel left shift of an image by a constant value //! pixel by pixel left shift of an image by a constant value
//! supports 1, 3 and 4 channels images with CV_8U, CV_16U or CV_32S depth //! supports 1, 3 and 4 channels images with CV_8U, CV_16U or CV_32S depth
CV_EXPORTS void lshift(const GpuMat& src, const Scalar& sc, GpuMat& dst, Stream& stream = Stream::Null()); CV_EXPORTS void lshift(const GpuMat& src, Scalar_<int> sc, GpuMat& dst, Stream& stream = Stream::Null());
//! computes per-element minimum of two arrays (dst = min(src1, src2)) //! computes per-element minimum of two arrays (dst = min(src1, src2))
CV_EXPORTS void min(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream = Stream::Null()); CV_EXPORTS void min(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream = Stream::Null());

4
modules/gpu/src/cuda/element_operations.cu
Unescape View File

@ -706,8 +706,8 @@ namespace cv { namespace gpu { namespace device
{ {
__device__ __forceinline__ short4 operator ()(short4 a, float b) const __device__ __forceinline__ short4 operator ()(short4 a, float b) const
{ {
return b != 0 ? make_short4(saturate_cast<short>(a.x / b), saturate_cast<uchar>(a.y / b), return b != 0 ? make_short4(saturate_cast<short>(a.x / b), saturate_cast<short>(a.y / b),
saturate_cast<short>(a.z / b), saturate_cast<uchar>(a.w / b)) saturate_cast<short>(a.z / b), saturate_cast<short>(a.w / b))
: make_short4(0,0,0,0); : make_short4(0,0,0,0);
} }
}; };

26
modules/gpu/src/element_operations.cpp
Unescape View File

@ -71,8 +71,8 @@ void cv::gpu::bitwise_and(const GpuMat&, const GpuMat&, GpuMat&, const GpuMat&,
void cv::gpu::bitwise_and(const GpuMat&, const Scalar&, GpuMat&, Stream&) { throw_nogpu(); } void cv::gpu::bitwise_and(const GpuMat&, const Scalar&, GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::bitwise_xor(const GpuMat&, const GpuMat&, GpuMat&, const GpuMat&, Stream&) { throw_nogpu(); } void cv::gpu::bitwise_xor(const GpuMat&, const GpuMat&, GpuMat&, const GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::bitwise_xor(const GpuMat&, const Scalar&, GpuMat&, Stream&) { throw_nogpu(); } void cv::gpu::bitwise_xor(const GpuMat&, const Scalar&, GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::rshift(const GpuMat&, const Scalar&, GpuMat&, Stream&) { throw_nogpu(); } void cv::gpu::rshift(const GpuMat&, Scalar_<int>, GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::lshift(const GpuMat&, const Scalar&, GpuMat&, Stream&) { throw_nogpu(); } void cv::gpu::lshift(const GpuMat&, Scalar_<int>, GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::min(const GpuMat&, const GpuMat&, GpuMat&, Stream&) { throw_nogpu(); } void cv::gpu::min(const GpuMat&, const GpuMat&, GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::min(const GpuMat&, double, GpuMat&, Stream&) { throw_nogpu(); } void cv::gpu::min(const GpuMat&, double, GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::max(const GpuMat&, const GpuMat&, GpuMat&, Stream&) { throw_nogpu(); } void cv::gpu::max(const GpuMat&, const GpuMat&, GpuMat&, Stream&) { throw_nogpu(); }
@ -462,15 +462,14 @@ void cv::gpu::subtract(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, cons
{0/*subtract_gpu<double, unsigned char>*/, 0/*subtract_gpu<double, signed char>*/, 0/*subtract_gpu<double, unsigned short>*/, 0/*subtract_gpu<double, short>*/, 0/*subtract_gpu<double, int>*/, 0/*subtract_gpu<double, float>*/, subtract_gpu<double, double>} {0/*subtract_gpu<double, unsigned char>*/, 0/*subtract_gpu<double, signed char>*/, 0/*subtract_gpu<double, unsigned short>*/, 0/*subtract_gpu<double, short>*/, 0/*subtract_gpu<double, int>*/, 0/*subtract_gpu<double, float>*/, subtract_gpu<double, double>}
}; };
static const func_t npp_funcs[7] = static const func_t npp_funcs[6] =
{ {
NppArithm<CV_8U, nppiSub_8u_C1RSfs>::call, NppArithm<CV_8U, nppiSub_8u_C1RSfs>::call,
0, 0,
NppArithm<CV_16U, nppiSub_16u_C1RSfs>::call, NppArithm<CV_16U, nppiSub_16u_C1RSfs>::call,
NppArithm<CV_16S, nppiSub_16s_C1RSfs>::call, NppArithm<CV_16S, nppiSub_16s_C1RSfs>::call,
NppArithm<CV_32S, nppiSub_32s_C1RSfs>::call, NppArithm<CV_32S, nppiSub_32s_C1RSfs>::call,
NppArithm<CV_32F, nppiSub_32f_C1R>::call, NppArithm<CV_32F, nppiSub_32f_C1R>::call
subtract_gpu<double, double>
}; };
CV_Assert(src1.type() != CV_8S); CV_Assert(src1.type() != CV_8S);
@ -484,7 +483,7 @@ void cv::gpu::subtract(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, cons
cudaStream_t stream = StreamAccessor::getStream(s); cudaStream_t stream = StreamAccessor::getStream(s);
if (mask.empty() && dst.type() == src1.type()) if (mask.empty() && dst.type() == src1.type() && src1.depth() <= CV_32F)
{ {
npp_funcs[src1.depth()](src2.reshape(1), src1.reshape(1), dst.reshape(1), PtrStepb(), stream); npp_funcs[src1.depth()](src2.reshape(1), src1.reshape(1), dst.reshape(1), PtrStepb(), stream);
return; return;
@ -734,15 +733,14 @@ void cv::gpu::divide(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, double
{0/*divide_gpu<double, unsigned char>*/, 0/*divide_gpu<double, signed char>*/, 0/*divide_gpu<double, unsigned short>*/, 0/*divide_gpu<double, short>*/, 0/*divide_gpu<double, int>*/, 0/*divide_gpu<double, float>*/, divide_gpu<double, double>} {0/*divide_gpu<double, unsigned char>*/, 0/*divide_gpu<double, signed char>*/, 0/*divide_gpu<double, unsigned short>*/, 0/*divide_gpu<double, short>*/, 0/*divide_gpu<double, int>*/, 0/*divide_gpu<double, float>*/, divide_gpu<double, double>}
}; };
static const func_t npp_funcs[7] = static const func_t npp_funcs[6] =
{ {
NppArithm<CV_8U, nppiDiv_8u_C1RSfs>::call, NppArithm<CV_8U, nppiDiv_8u_C1RSfs>::call,
0, 0,
NppArithm<CV_16U, nppiDiv_16u_C1RSfs>::call, NppArithm<CV_16U, nppiDiv_16u_C1RSfs>::call,
NppArithm<CV_16S, nppiDiv_16s_C1RSfs>::call, NppArithm<CV_16S, nppiDiv_16s_C1RSfs>::call,
NppArithm<CV_32S, nppiDiv_32s_C1RSfs>::call, NppArithm<CV_32S, nppiDiv_32s_C1RSfs>::call,
NppArithm<CV_32F, nppiDiv_32f_C1R>::call, NppArithm<CV_32F, nppiDiv_32f_C1R>::call
divide_gpu<double, double>
}; };
cudaStream_t stream = StreamAccessor::getStream(s); cudaStream_t stream = StreamAccessor::getStream(s);
@ -753,7 +751,7 @@ void cv::gpu::divide(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, double
dst.create(src1.size(), src1.type()); dst.create(src1.size(), src1.type());
multiply_gpu(static_cast<DevMem2D_<uchar4> >(src1), static_cast<DevMem2Df>(src2), static_cast<DevMem2D_<uchar4> >(dst), stream); divide_gpu(static_cast<DevMem2D_<uchar4> >(src1), static_cast<DevMem2Df>(src2), static_cast<DevMem2D_<uchar4> >(dst), stream);
} }
else if (src1.type() == CV_16SC4 && src2.type() == CV_32FC1) else if (src1.type() == CV_16SC4 && src2.type() == CV_32FC1)
{ {
@ -761,7 +759,7 @@ void cv::gpu::divide(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, double
dst.create(src1.size(), src1.type()); dst.create(src1.size(), src1.type());
multiply_gpu(static_cast<DevMem2D_<short4> >(src1), static_cast<DevMem2Df>(src2), static_cast<DevMem2D_<short4> >(dst), stream); divide_gpu(static_cast<DevMem2D_<short4> >(src1), static_cast<DevMem2Df>(src2), static_cast<DevMem2D_<short4> >(dst), stream);
} }
else else
{ {
@ -773,7 +771,7 @@ void cv::gpu::divide(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, double
dst.create(src1.size(), CV_MAKE_TYPE(CV_MAT_DEPTH(dtype), src1.channels())); dst.create(src1.size(), CV_MAKE_TYPE(CV_MAT_DEPTH(dtype), src1.channels()));
if (scale == 1 && dst.type() == src1.type()) if (scale == 1 && dst.type() == src1.type() && src1.depth() <= CV_32F)
{ {
npp_funcs[src1.depth()](src2.reshape(1), src1.reshape(1), dst.reshape(1), 1, stream); npp_funcs[src1.depth()](src2.reshape(1), src1.reshape(1), dst.reshape(1), 1, stream);
return; return;
@ -1729,7 +1727,7 @@ namespace
}; };
} }
void cv::gpu::rshift(const GpuMat& src, const Scalar& sc, GpuMat& dst, Stream& stream) void cv::gpu::rshift(const GpuMat& src, Scalar_<int> sc, GpuMat& dst, Stream& stream)
{ {
typedef void (*func_t)(const GpuMat& src, Scalar_<Npp32u> sc, GpuMat& dst, cudaStream_t stream); typedef void (*func_t)(const GpuMat& src, Scalar_<Npp32u> sc, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[5][4] = static const func_t funcs[5][4] =
@ -1749,7 +1747,7 @@ void cv::gpu::rshift(const GpuMat& src, const Scalar& sc, GpuMat& dst, Stream& s
funcs[src.depth()][src.channels() - 1](src, sc, dst, StreamAccessor::getStream(stream)); funcs[src.depth()][src.channels() - 1](src, sc, dst, StreamAccessor::getStream(stream));
} }
void cv::gpu::lshift(const GpuMat& src, const Scalar& sc, GpuMat& dst, Stream& stream) void cv::gpu::lshift(const GpuMat& src, Scalar_<int> sc, GpuMat& dst, Stream& stream)
{ {
typedef void (*func_t)(const GpuMat& src, Scalar_<Npp32u> sc, GpuMat& dst, cudaStream_t stream); typedef void (*func_t)(const GpuMat& src, Scalar_<Npp32u> sc, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[5][4] = static const func_t funcs[5][4] =

1804
modules/gpu/test/test_arithm.cpp
View File

File diff suppressed because it is too large Load Diff

31
modules/gpu/test/utility.hpp
Unescape View File

@ -162,10 +162,37 @@ CV_FLAGS(DftFlags, cv::DFT_INVERSE, cv::DFT_SCALE, cv::DFT_ROWS, cv::DFT_COMPLEX
#define ALL_DEVICES testing::ValuesIn(devices()) #define ALL_DEVICES testing::ValuesIn(devices())
#define DEVICES(feature) testing::ValuesIn(devices(feature)) #define DEVICES(feature) testing::ValuesIn(devices(feature))
#define DIFFERENT_SIZES testing::Values(cv::Size(128, 128), cv::Size(113, 113))
#define ALL_DEPTH testing::Values(MatDepth(CV_8U), MatDepth(CV_8S), MatDepth(CV_16U), MatDepth(CV_16S), MatDepth(CV_32S), MatDepth(CV_32F), MatDepth(CV_64F))
#define ALL_TYPES testing::ValuesIn(all_types()) #define ALL_TYPES testing::ValuesIn(all_types())
#define TYPES(depth_start, depth_end, cn_start, cn_end) testing::ValuesIn(types(depth_start, depth_end, cn_start, cn_end)) #define TYPES(depth_start, depth_end, cn_start, cn_end) testing::ValuesIn(types(depth_start, depth_end, cn_start, cn_end))
#define DIFFERENT_SIZES testing::Values(cv::Size(128, 128), cv::Size(113, 113)) #define DEPTH_PAIRS testing::Values(std::make_pair(MatDepth(CV_8U), MatDepth(CV_8U)), \
std::make_pair(MatDepth(CV_8U), MatDepth(CV_16U)), \
std::make_pair(MatDepth(CV_8U), MatDepth(CV_16S)), \
std::make_pair(MatDepth(CV_8U), MatDepth(CV_32S)), \
std::make_pair(MatDepth(CV_8U), MatDepth(CV_32F)), \
std::make_pair(MatDepth(CV_8U), MatDepth(CV_64F)), \
\
std::make_pair(MatDepth(CV_16U), MatDepth(CV_16U)), \
std::make_pair(MatDepth(CV_16U), MatDepth(CV_32S)), \
std::make_pair(MatDepth(CV_16U), MatDepth(CV_32F)), \
std::make_pair(MatDepth(CV_16U), MatDepth(CV_64F)), \
\
std::make_pair(MatDepth(CV_16S), MatDepth(CV_16S)), \
std::make_pair(MatDepth(CV_16S), MatDepth(CV_32S)), \
std::make_pair(MatDepth(CV_16S), MatDepth(CV_32F)), \
std::make_pair(MatDepth(CV_16S), MatDepth(CV_64F)), \
\
std::make_pair(MatDepth(CV_32S), MatDepth(CV_32S)), \
std::make_pair(MatDepth(CV_32S), MatDepth(CV_32F)), \
std::make_pair(MatDepth(CV_32S), MatDepth(CV_64F)), \
\
std::make_pair(MatDepth(CV_32F), MatDepth(CV_32F)), \
std::make_pair(MatDepth(CV_32F), MatDepth(CV_64F)), \
\
std::make_pair(MatDepth(CV_64F), MatDepth(CV_64F)))
#define WHOLE testing::Values(UseRoi(false)) #define WHOLE testing::Values(UseRoi(false))
#define SUBMAT testing::Values(UseRoi(true)) #define SUBMAT testing::Values(UseRoi(true))
@ -173,4 +200,6 @@ CV_FLAGS(DftFlags, cv::DFT_INVERSE, cv::DFT_SCALE, cv::DFT_ROWS, cv::DFT_COMPLEX
#define DIRECT_INVERSE testing::Values(Inverse(false), Inverse(true)) #define DIRECT_INVERSE testing::Values(Inverse(false), Inverse(true))
#define ALL_CMP_CODES testing::Values(CmpCode(cv::CMP_EQ), CmpCode(cv::CMP_NE), CmpCode(cv::CMP_GT), CmpCode(cv::CMP_GE), CmpCode(cv::CMP_LT), CmpCode(cv::CMP_LE))
#endif // __OPENCV_TEST_UTILITY_HPP__ #endif // __OPENCV_TEST_UTILITY_HPP__

Write Preview
Loading…
Cancel
Save