Skip to content

O
opencv
Commit 0c325cac authored by Anton Obukhov's avatar Anton Obukhov
Browse files
Options

[~] Minor refactoring, clean-up 

[+] Added 128-bit transpose
parent e2caf4a3
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 166 additions and 95 deletions
modules/gpu/src/nvidia/NPP_staging/NPP_staging.hpp
View file @ 0c325cac
......@@ -96,65 +96,65 @@ cudaStream_t nppStSetActiveCUDAstream(cudaStream_t cudaStream);
* \return NCV status code
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_32u_C1R(Ncv32u *d_src, Ncv32u srcStep,
Ncv32u *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
NCVStatus nppiStDecimate_32u_C1R(Ncv32u *d_src, Ncv32u srcStep,
Ncv32u *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 32-bit signed pixels, single channel.
* \see nppiStDownsampleNearest_32u_C1R
* \see nppiStDecimate_32u_C1R
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_32s_C1R(Ncv32s *d_src, Ncv32u srcStep,
Ncv32s *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
NCVStatus nppiStDecimate_32s_C1R(Ncv32s *d_src, Ncv32u srcStep,
Ncv32s *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 32-bit float pixels, single channel.
* \see nppiStDownsampleNearest_32u_C1R
* \see nppiStDecimate_32u_C1R
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_32f_C1R(Ncv32f *d_src, Ncv32u srcStep,
Ncv32f *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
NCVStatus nppiStDecimate_32f_C1R(Ncv32f *d_src, Ncv32u srcStep,
Ncv32f *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit unsigned pixels, single channel.
* \see nppiStDownsampleNearest_32u_C1R
* \see nppiStDecimate_32u_C1R
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_64u_C1R(Ncv64u *d_src, Ncv32u srcStep,
Ncv64u *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
NCVStatus nppiStDecimate_64u_C1R(Ncv64u *d_src, Ncv32u srcStep,
Ncv64u *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit signed pixels, single channel.
* \see nppiStDownsampleNearest_32u_C1R
* \see nppiStDecimate_32u_C1R
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_64s_C1R(Ncv64s *d_src, Ncv32u srcStep,
Ncv64s *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
NCVStatus nppiStDecimate_64s_C1R(Ncv64s *d_src, Ncv32u srcStep,
Ncv64s *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit float pixels, single channel.
* \see nppiStDownsampleNearest_32u_C1R
* \see nppiStDecimate_32u_C1R
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_64f_C1R(Ncv64f *d_src, Ncv32u srcStep,
Ncv64f *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
NCVStatus nppiStDecimate_64f_C1R(Ncv64f *d_src, Ncv32u srcStep,
Ncv64f *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
......@@ -170,59 +170,59 @@ NCVStatus nppiStDownsampleNearest_64f_C1R(Ncv64f *d_src, Ncv32u srcStep,
* \return NCV status code
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_32u_C1R_host(Ncv32u *h_src, Ncv32u srcStep,
Ncv32u *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
NCVStatus nppiStDecimate_32u_C1R_host(Ncv32u *h_src, Ncv32u srcStep,
Ncv32u *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 32-bit signed pixels, single channel. Host implementation.
* \see nppiStDownsampleNearest_32u_C1R_host
* \see nppiStDecimate_32u_C1R_host
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_32s_C1R_host(Ncv32s *h_src, Ncv32u srcStep,
Ncv32s *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
NCVStatus nppiStDecimate_32s_C1R_host(Ncv32s *h_src, Ncv32u srcStep,
Ncv32s *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 32-bit float pixels, single channel. Host implementation.
* \see nppiStDownsampleNearest_32u_C1R_host
* \see nppiStDecimate_32u_C1R_host
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_32f_C1R_host(Ncv32f *h_src, Ncv32u srcStep,
Ncv32f *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
NCVStatus nppiStDecimate_32f_C1R_host(Ncv32f *h_src, Ncv32u srcStep,
Ncv32f *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit unsigned pixels, single channel. Host implementation.
* \see nppiStDownsampleNearest_32u_C1R_host
* \see nppiStDecimate_32u_C1R_host
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_64u_C1R_host(Ncv64u *h_src, Ncv32u srcStep,
Ncv64u *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
NCVStatus nppiStDecimate_64u_C1R_host(Ncv64u *h_src, Ncv32u srcStep,
Ncv64u *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit signed pixels, single channel. Host implementation.
* \see nppiStDownsampleNearest_32u_C1R_host
* \see nppiStDecimate_32u_C1R_host
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_64s_C1R_host(Ncv64s *h_src, Ncv32u srcStep,
Ncv64s *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
NCVStatus nppiStDecimate_64s_C1R_host(Ncv64s *h_src, Ncv32u srcStep,
Ncv64s *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit float pixels, single channel. Host implementation.
* \see nppiStDownsampleNearest_32u_C1R_host
* \see nppiStDecimate_32u_C1R_host
*/
NCV_EXPORTS
NCVStatus nppiStDownsampleNearest_64f_C1R_host(Ncv64f *h_src, Ncv32u srcStep,
Ncv64f *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
NCVStatus nppiStDecimate_64f_C1R_host(Ncv64f *h_src, Ncv32u srcStep,
Ncv64f *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
......@@ -333,6 +333,15 @@ NCVStatus nppiStTranspose_64f_C1R(Ncv64f *d_src, Ncv32u srcStride,
Ncv64f *d_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 128-bit pixels of any type, single channel
* \see nppiStTranspose_32u_C1R
*/
NCV_EXPORTS
NCVStatus nppiStTranspose_128_C1R(void *d_src, Ncv32u srcStep,
void *d_dst, Ncv32u dstStep, NcvSize32u srcRoi);
/**
* Transposes an image. 32-bit unsigned pixels, single channel. Host implementation
*
......@@ -394,6 +403,15 @@ NCVStatus nppiStTranspose_64f_C1R_host(Ncv64f *h_src, Ncv32u srcStride,
Ncv64f *h_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 128-bit pixels of any type, single channel. Host implementation
* \see nppiStTranspose_32u_C1R_host
*/
NCV_EXPORTS
NCVStatus nppiStTranspose_128_C1R_host(void *d_src, Ncv32u srcStep,
void *d_dst, Ncv32u dstStep, NcvSize32u srcRoi);
/**
* Calculates the size of the temporary buffer for integral image creation
*
......
modules/gpu/src/nvidia/core/NCV.hpp
View file @ 0c325cac
......@@ -129,8 +129,8 @@ struct NcvRect8u
Ncv8u y;
Ncv8u width;
Ncv8u height;
NcvRect8u() : x(0), y(0), width(0), height(0) {};
NcvRect8u(Ncv8u x, Ncv8u y, Ncv8u width, Ncv8u height) : x(x), y(y), width(width), height(height) {}
__host__ __device__ NcvRect8u() : x(0), y(0), width(0), height(0) {};
__host__ __device__ NcvRect8u(Ncv8u x, Ncv8u y, Ncv8u width, Ncv8u height) : x(x), y(y), width(width), height(height) {}
};
......@@ -140,8 +140,8 @@ struct NcvRect32s
Ncv32s y; ///< y-coordinate of upper left corner.
Ncv32s width; ///< Rectangle width.
Ncv32s height; ///< Rectangle height.
NcvRect32s() : x(0), y(0), width(0), height(0) {};
NcvRect32s(Ncv32s x, Ncv32s y, Ncv32s width, Ncv32s height) : x(x), y(y), width(width), height(height) {}
__host__ __device__ NcvRect32s() : x(0), y(0), width(0), height(0) {};
__host__ __device__ NcvRect32s(Ncv32s x, Ncv32s y, Ncv32s width, Ncv32s height) : x(x), y(y), width(width), height(height) {}
};
......@@ -151,8 +151,8 @@ struct NcvRect32u
Ncv32u y; ///< y-coordinate of upper left corner.
Ncv32u width; ///< Rectangle width.
Ncv32u height; ///< Rectangle height.
NcvRect32u() : x(0), y(0), width(0), height(0) {};
NcvRect32u(Ncv32u x, Ncv32u y, Ncv32u width, Ncv32u height) : x(x), y(y), width(width), height(height) {}
__host__ __device__ NcvRect32u() : x(0), y(0), width(0), height(0) {};
__host__ __device__ NcvRect32u(Ncv32u x, Ncv32u y, Ncv32u width, Ncv32u height) : x(x), y(y), width(width), height(height) {}
};
......@@ -160,8 +160,8 @@ struct NcvSize32s
{
Ncv32s width; ///< Rectangle width.
Ncv32s height; ///< Rectangle height.
NcvSize32s() : width(0), height(0) {};
NcvSize32s(Ncv32s width, Ncv32s height) : width(width), height(height) {}
__host__ __device__ NcvSize32s() : width(0), height(0) {};
__host__ __device__ NcvSize32s(Ncv32s width, Ncv32s height) : width(width), height(height) {}
};
......@@ -169,8 +169,8 @@ struct NcvSize32u
{
Ncv32u width; ///< Rectangle width.
Ncv32u height; ///< Rectangle height.
NcvSize32u() : width(0), height(0) {};
NcvSize32u(Ncv32u width, Ncv32u height) : width(width), height(height) {}
__host__ __device__ NcvSize32u() : width(0), height(0) {};
__host__ __device__ NcvSize32u(Ncv32u width, Ncv32u height) : width(width), height(height) {}
};
......@@ -275,6 +275,7 @@ enum NCVStatus
{
//NCV statuses
NCV_SUCCESS,
NCV_UNKNOWN_ERROR,
NCV_CUDA_ERROR,
NCV_NPP_ERROR,
......@@ -501,13 +502,18 @@ private:
/**
* Copy dispatcher
* Copy dispatchers
*/
NCV_EXPORTS NCVStatus memSegCopyHelper(void *dst, NCVMemoryType dstType,
const void *src, NCVMemoryType srcType,
size_t sz, cudaStream_t cuStream);
NCV_EXPORTS NCVStatus memSegCopyHelper2D(void *dst, Ncv32u dstPitch, NCVMemoryType dstType,
const void *src, Ncv32u srcPitch, NCVMemoryType srcType,
Ncv32u widthbytes, Ncv32u height, cudaStream_t cuStream);
/**
* NCVVector (1D)
*/
......@@ -532,7 +538,7 @@ public:
_memtype = NCVMemoryTypeNone;
}
NCVStatus copySolid(NCVVector<T> &dst, cudaStream_t cuStream, size_t howMuch=0)
NCVStatus copySolid(NCVVector<T> &dst, cudaStream_t cuStream, size_t howMuch=0) const
{
if (howMuch == 0)
{
......@@ -600,7 +606,6 @@ public:
this->_memtype = this->allocatedMem.begin.memtype;
}
~NCVVectorAlloc()
{
NCVStatus ncvStat;
......@@ -611,25 +616,22 @@ public:
this->clear();
}
NcvBool isMemAllocated() const
{
return (this->allocatedMem.begin.ptr != NULL) || (this->allocator.isCounting());
}
Ncv32u getAllocatorsAlignment() const
{
return allocator.alignment();
}
NCVMemSegment getSegment() const
{
return allocatedMem;
}
private:
private:
INCVMemAllocator &allocator;
NCVMemSegment allocatedMem;
};
......@@ -658,7 +660,6 @@ public:
this->bReused = true;
}
NCVVectorReuse(const NCVMemSegment &memSegment, Ncv32u length)
{
this->bReused = false;
......@@ -674,7 +675,6 @@ public:
this->bReused = true;
}
NcvBool isMemReused() const
{
return this->bReused;
......@@ -703,7 +703,6 @@ public:
virtual ~NCVMatrix() {}
void clear()
{
_ptr = NULL;
......@@ -713,14 +712,13 @@ public:
_memtype = NCVMemoryTypeNone;
}
Ncv32u stride() const
{
return _pitch / sizeof(T);
}
NCVStatus copySolid(NCVMatrix<T> &dst, cudaStream_t cuStream, size_t howMuch=0)
//a side effect of this function is that it copies everything in a single chunk, so the "padding" will be overwritten
NCVStatus copySolid(NCVMatrix<T> &dst, cudaStream_t cuStream, size_t howMuch=0) const
{
if (howMuch == 0)
{
......@@ -748,6 +746,24 @@ public:
return ncvStat;
}
NCVStatus copy2D(NCVMatrix<T> &dst, NcvSize32u roi, cudaStream_t cuStream) const
{
ncvAssertReturn(this->width() >= roi.width && this->height() >= roi.height &&
dst.width() >= roi.width && dst.height() >= roi.height, NCV_MEM_COPY_ERROR);
ncvAssertReturn((this->_ptr != NULL || this->_memtype == NCVMemoryTypeNone) &&
(dst._ptr != NULL || dst._memtype == NCVMemoryTypeNone), NCV_NULL_PTR);
NCVStatus ncvStat = NCV_SUCCESS;
if (this->_memtype != NCVMemoryTypeNone)
{
ncvStat = memSegCopyHelper2D(dst._ptr, dst._pitch, dst._memtype,
this->_ptr, this->_pitch, this->_memtype,
roi.width * sizeof(T), roi.height, cuStream);
}
return ncvStat;
}
T *ptr() const {return this->_ptr;}
Ncv32u width() const {return this->_width;}
Ncv32u height() const {return this->_height;}
......@@ -817,19 +833,16 @@ public:
this->clear();
}
NcvBool isMemAllocated() const
{
return (this->allocatedMem.begin.ptr != NULL) || (this->allocator.isCounting());
}
Ncv32u getAllocatorsAlignment() const
{
return allocator.alignment();
}
NCVMemSegment getSegment() const
{
return allocatedMem;
......@@ -888,6 +901,23 @@ public:
this->bReused = true;
}
NCVMatrixReuse(const NCVMatrix<T> &mat, NcvRect32u roi)
{
this->bReused = false;
this->clear();
ncvAssertPrintReturn(roi.x < mat.width() && roi.y < mat.height() && \
roi.x + roi.width <= mat.width() && roi.y + roi.height <= mat.height(),
"NCVMatrixReuse ctor:: memory binding failed due to mismatching ROI and source matrix dims", );
this->_width = roi.width;
this->_height = roi.height;
this->_pitch = mat.pitch();
this->_ptr = mat.ptr() + roi.y * mat.stride() + roi.x;
this->_memtype = mat.memType();
this->bReused = true;
}
NcvBool isMemReused() const
{
......@@ -899,4 +929,27 @@ private:
NcvBool bReused;
};
/**
* Operations with rectangles
*/
NCV_EXPORTS NCVStatus ncvGroupRectangles_host(NCVVector<NcvRect32u> &hypotheses, Ncv32u &numHypotheses,
Ncv32u minNeighbors, Ncv32f intersectEps, NCVVector<Ncv32u> *hypothesesWeights);
NCV_EXPORTS NCVStatus ncvDrawRects_8u_host(Ncv8u *h_dst, Ncv32u dstStride, Ncv32u dstWidth, Ncv32u dstHeight,
NcvRect32u *h_rects, Ncv32u numRects, Ncv8u color);
NCV_EXPORTS NCVStatus ncvDrawRects_32u_host(Ncv32u *h_dst, Ncv32u dstStride, Ncv32u dstWidth, Ncv32u dstHeight,
NcvRect32u *h_rects, Ncv32u numRects, Ncv32u color);
NCV_EXPORTS NCVStatus ncvDrawRects_8u_device(Ncv8u *d_dst, Ncv32u dstStride, Ncv32u dstWidth, Ncv32u dstHeight,
NcvRect32u *d_rects, Ncv32u numRects, Ncv8u color, cudaStream_t cuStream);
NCV_EXPORTS NCVStatus ncvDrawRects_32u_device(Ncv32u *d_dst, Ncv32u dstStride, Ncv32u dstWidth, Ncv32u dstHeight,
NcvRect32u *d_rects, Ncv32u numRects, Ncv32u color, cudaStream_t cuStream);
#endif // _ncv_hpp_
modules/gpu/src/nvidia/core/NCVRuntimeTemplates.hpp
View file @ 0c325cac
......@@ -150,14 +150,14 @@ namespace NCVRuntimeTemplateBool
{
//Convenience function used by the user
//Takes a variable argument list, transforms it into a list
static void call(Func *functor, int dummy, ...)
static void call(Func *functor, ...)
{
//Vector used to collect arguments
std::vector<int> templateParamList;
//Variable argument list manipulation
va_list listPointer;
va_start(listPointer, dummy);
va_start(listPointer, functor);
//Collect parameters into the list
for(int i=0; i<NumArguments; i++)
{
......
modules/gpu/test/nvidia/TestHypothesesFilter.cpp
View file @ 0c325cac
......@@ -134,7 +134,7 @@ bool TestHypothesesFilter::process()
Ncv32u numHypothesesSrc = h_vecSrc.length();
NCV_SKIP_COND_BEGIN
ncvStat = ncvFilterHypotheses_host(h_vecSrc, numHypothesesSrc, this->minNeighbors, this->eps, NULL);
ncvStat = ncvGroupRectangles_host(h_vecSrc, numHypothesesSrc, this->minNeighbors, this->eps, NULL);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
NCV_SKIP_COND_END
......
modules/gpu/test/nvidia/TestResize.cpp
View file @ 0c325cac
......@@ -83,17 +83,17 @@ bool TestResize<T>::process()
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = nppiStDownsampleNearest_32u_C1R((Ncv32u *)d_img.ptr(), d_img.pitch(),
(Ncv32u *)d_small.ptr(), d_small.pitch(),
srcSize, this->scaleFactor,
this->bTextureCache);
ncvStat = nppiStDecimate_32u_C1R((Ncv32u *)d_img.ptr(), d_img.pitch(),
(Ncv32u *)d_small.ptr(), d_small.pitch(),
srcSize, this->scaleFactor,
this->bTextureCache);
}
else if (sizeof(T) == sizeof(Ncv64u))
{
ncvStat = nppiStDownsampleNearest_64u_C1R((Ncv64u *)d_img.ptr(), d_img.pitch(),
(Ncv64u *)d_small.ptr(), d_small.pitch(),
srcSize, this->scaleFactor,
this->bTextureCache);
ncvStat = nppiStDecimate_64u_C1R((Ncv64u *)d_img.ptr(), d_img.pitch(),
(Ncv64u *)d_small.ptr(), d_small.pitch(),
srcSize, this->scaleFactor,
this->bTextureCache);
}
else
{
......@@ -107,15 +107,15 @@ bool TestResize<T>::process()
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = nppiStDownsampleNearest_32u_C1R_host((Ncv32u *)h_img.ptr(), h_img.pitch(),
(Ncv32u *)h_small.ptr(), h_small.pitch(),
srcSize, this->scaleFactor);
ncvStat = nppiStDecimate_32u_C1R_host((Ncv32u *)h_img.ptr(), h_img.pitch(),
(Ncv32u *)h_small.ptr(), h_small.pitch(),
srcSize, this->scaleFactor);
}
else if (sizeof(T) == sizeof(Ncv64u))
{
ncvStat = nppiStDownsampleNearest_64u_C1R_host((Ncv64u *)h_img.ptr(), h_img.pitch(),
(Ncv64u *)h_small.ptr(), h_small.pitch(),
srcSize, this->scaleFactor);
ncvStat = nppiStDecimate_64u_C1R_host((Ncv64u *)h_img.ptr(), h_img.pitch(),
(Ncv64u *)h_small.ptr(), h_small.pitch(),
srcSize, this->scaleFactor);
}
else
{
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment