Skip to content

O
opencv
Commit 0747f2d8 authored by Anatoly Baksheev's avatar Anatoly Baksheev
Browse files
Options

1) NPP_staging as sources. Binaries removed. 

2) NVidia tests for GPU
3) FD sample that uses NVidia's interface.
parent 811f6fbe
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 713 additions and 141 deletions
modules/gpu/CMakeLists.txt
View file @ 0747f2d8
......@@ -36,10 +36,12 @@ file(GLOB lib_device_hdrs "src/opencv2/gpu/device/*.h*")
source_group("Device" FILES ${lib_device_hdrs})
if (HAVE_CUDA AND MSVC)
file(GLOB ncv_srcs "src/nvidia/*.cpp")
file(GLOB ncv_hdrs "src/nvidia/*.h*")
file(GLOB ncv_cuda "src/nvidia/*.cu")
source_group("Src\\NVidia" FILES ${ncv_srcs} ${ncv_hdrs} ${ncv_cuda})
file(GLOB_RECURSE ncv_srcs "src/nvidia/*.cpp")
file(GLOB_RECURSE ncv_cuda "src/nvidia/*.cu")
file(GLOB_RECURSE ncv_hdr1 "src/nvidia/*.hpp")
file(GLOB_RECURSE ncv_hdr2 "src/nvidia/*.h")
source_group("Src\\NVidia" FILES ${ncv_srcs} ${ncv_hdr1} ${ncv_hdr2} ${ncv_cuda})
include_directories("src/nvidia/core" "src/nvidia/NPP_staging")
endif()
if (HAVE_CUDA)
......@@ -74,17 +76,13 @@ if (HAVE_CUDA)
string(REPLACE "/EHsc-" "/EHs" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
string(REPLACE "/EHsc-" "/EHs" CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE}")
string(REPLACE "/EHsc-" "/EHs" CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG}")
endif()
include(FindNPP_staging.cmake)
include_directories(${NPPST_INC})
endif()
CUDA_COMPILE(cuda_objs ${lib_cuda} ${ncv_cuda})
#CUDA_BUILD_CLEAN_TARGET()
endif()
add_library(${the_target} ${lib_srcs} ${lib_hdrs} ${lib_int_hdrs} ${lib_cuda} ${lib_cuda_hdrs} ${lib_device_hdrs} ${ncv_srcs} ${ncv_hdrs} ${ncv_cuda} ${cuda_objs})
add_library(${the_target} ${lib_srcs} ${lib_hdrs} ${lib_int_hdrs} ${lib_cuda} ${lib_cuda_hdrs} ${lib_device_hdrs} ${ncv_srcs} ${ncv_hdr1} ${ncv_hdr2} ${ncv_cuda} ${cuda_objs})
if(PCHSupport_FOUND)
set(pch_header ${CMAKE_CURRENT_SOURCE_DIR}/src/precomp.hpp)
......@@ -117,8 +115,7 @@ set_target_properties(${the_target} PROPERTIES
target_link_libraries(${the_target} ${OPENCV_LINKER_LIBS} ${IPP_LIBS} ${DEPS} )
if (HAVE_CUDA)
target_link_libraries(${the_target} ${CUDA_LIBRARIES} ${CUDA_NPP_LIBRARIES})
target_link_libraries(${the_target} ${NPPST_LIB})
target_link_libraries(${the_target} ${CUDA_LIBRARIES} ${CUDA_NPP_LIBRARIES})
CUDA_ADD_CUFFT_TO_TARGET(${the_target})
endif()
......
modules/gpu/FindNPP_staging.cmake deleted 100644 → 0
View file @ 811f6fbe
if(CMAKE_SIZEOF_VOID_P EQUAL 4)
set(BIT_SUFF 32)
else()
set(BIT_SUFF 64)
endif()
if (APPLE)
set(PLATFORM_SUFF Darwin)
elseif (UNIX)
set(PLATFORM_SUFF Linux)
else()
set(PLATFORM_SUFF Windows)
endif()
set(LIB_FILE NPP_staging_static_${PLATFORM_SUFF}_${BIT_SUFF}_v1)
find_library(NPPST_LIB
NAMES "${LIB_FILE}" "lib${LIB_FILE}"
PATHS "${CMAKE_SOURCE_DIR}/3rdparty/NPP_staging"
DOC "NPP staging library"
)
SET(NPPST_INC "${CMAKE_SOURCE_DIR}//3rdparty/NPP_staging")
\ No newline at end of file
modules/gpu/src/arithm.cpp
View file @ 0747f2d8
......@@ -83,25 +83,25 @@ void cv::gpu::transpose(const GpuMat& src, GpuMat& dst)
sz.width = src.cols;
sz.height = src.rows;
nppSafeCall( nppiTranspose_8u_C1R(src.ptr<Npp8u>(), src.step, dst.ptr<Npp8u>(), dst.step, sz) );
nppSafeCall( nppiTranspose_8u_C1R(src.ptr<Npp8u>(), src.step, dst.ptr<Npp8u>(), dst.step, sz) );
}
else if (src.elemSize() == 4)
{
NppStSize32u sz;
NcvSize32u sz;
sz.width = src.cols;
sz.height = src.rows;
nppSafeCall( nppiStTranspose_32u_C1R(const_cast<NppSt32u*>(src.ptr<NppSt32u>()), src.step,
dst.ptr<NppSt32u>(), dst.step, sz) );
nppSafeCall( nppiStTranspose_32u_C1R(const_cast<Ncv32u*>(src.ptr<Ncv32u>()), src.step,
dst.ptr<Ncv32u>(), dst.step, sz) );
}
else // if (src.elemSize() == 8)
{
NppStSize32u sz;
NcvSize32u sz;
sz.width = src.cols;
sz.height = src.rows;
nppSafeCall( nppiStTranspose_64u_C1R(const_cast<NppSt64u*>(src.ptr<NppSt64u>()), src.step,
dst.ptr<NppSt64u>(), dst.step, sz) );
nppSafeCall( nppiStTranspose_64u_C1R(const_cast<Ncv64u*>(src.ptr<Ncv64u>()), src.step,
dst.ptr<Ncv64u>(), dst.step, sz) );
}
cudaSafeCall( cudaThreadSynchronize() );
......
modules/gpu/src/cascadeclassifier.cpp
View file @ 0747f2d8
......@@ -126,7 +126,7 @@ struct cv::gpu::CascadeClassifier_GPU::CascadeClassifierImpl
minNeighbors,
scaleStep, 1,
flags,
*gpuAllocator, *cpuAllocator, devProp.major, devProp.minor, 0);
*gpuAllocator, *cpuAllocator, devProp, 0);
ncvAssertReturnNcvStat(ncvStat);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), NCV_CUDA_ERROR);
......@@ -146,8 +146,8 @@ private:
ncvAssertCUDAReturn(cudaGetDeviceProperties(&devProp, devId), NCV_CUDA_ERROR);
// Load the classifier from file (assuming its size is about 1 mb) using a simple allocator
gpuCascadeAllocator = new NCVMemNativeAllocator(NCVMemoryTypeDevice);
cpuCascadeAllocator = new NCVMemNativeAllocator(NCVMemoryTypeHostPinned);
gpuCascadeAllocator = new NCVMemNativeAllocator(NCVMemoryTypeDevice, devProp.textureAlignment);
cpuCascadeAllocator = new NCVMemNativeAllocator(NCVMemoryTypeHostPinned, devProp.textureAlignment);
ncvAssertPrintReturn(gpuCascadeAllocator->isInitialized(), "Error creating cascade GPU allocator", NCV_CUDA_ERROR);
ncvAssertPrintReturn(cpuCascadeAllocator->isInitialized(), "Error creating cascade CPU allocator", NCV_CUDA_ERROR);
......@@ -212,7 +212,7 @@ private:
roi.height = d_src.height();
Ncv32u numDetections;
ncvStat = ncvDetectObjectsMultiScale_device(d_src, roi, d_rects, numDetections, haar, *h_haarStages,
*d_haarStages, *d_haarNodes, *d_haarFeatures, haar.ClassifierSize, 4, 1.2f, 1, 0, gpuCounter, cpuCounter, devProp.major, devProp.minor, 0);
*d_haarStages, *d_haarNodes, *d_haarFeatures, haar.ClassifierSize, 4, 1.2f, 1, 0, gpuCounter, cpuCounter, devProp, 0);
ncvAssertReturnNcvStat(ncvStat);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), NCV_CUDA_ERROR);
......
modules/gpu/src/imgproc_gpu.cpp
View file @ 0747f2d8
......@@ -560,16 +560,19 @@ void cv::gpu::integralBuffered(const GpuMat& src, GpuMat& sum, GpuMat& buffer)
sum.create(src.rows + 1, src.cols + 1, CV_32S);
NppStSize32u roiSize;
NcvSize32u roiSize;
roiSize.width = src.cols;
roiSize.height = src.rows;
NppSt32u bufSize;
nppSafeCall( nppiStIntegralGetSize_8u32u(roiSize, &bufSize) );
cudaDeviceProp prop;
cudaSafeCall( cudaGetDeviceProperties(&prop, cv::gpu::getDevice()) );
Ncv32u bufSize;
nppSafeCall( nppiStIntegralGetSize_8u32u(roiSize, &bufSize, prop) );
ensureSizeIsEnough(1, bufSize, CV_8UC1, buffer);
nppSafeCall( nppiStIntegral_8u32u_C1R(const_cast<NppSt8u*>(src.ptr<NppSt8u>()), src.step,
sum.ptr<NppSt32u>(), sum.step, roiSize, buffer.ptr<NppSt8u>(), bufSize) );
nppSafeCall( nppiStIntegral_8u32u_C1R(const_cast<Ncv8u*>(src.ptr<Ncv8u>()), src.step,
sum.ptr<Ncv32u>(), sum.step, roiSize, buffer.ptr<Ncv8u>(), bufSize, prop) );
cudaSafeCall( cudaThreadSynchronize() );
}
......@@ -600,19 +603,20 @@ void cv::gpu::sqrIntegral(const GpuMat& src, GpuMat& sqsum)
{
CV_Assert(src.type() == CV_8U);
NppStSize32u roiSize;
NcvSize32u roiSize;
roiSize.width = src.cols;
roiSize.height = src.rows;
NppSt32u bufSize;
nppSafeCall(nppiStSqrIntegralGetSize_8u64u(roiSize, &bufSize));
cudaDeviceProp prop;
cudaSafeCall( cudaGetDeviceProperties(&prop, cv::gpu::getDevice()) );
Ncv32u bufSize;
nppSafeCall(nppiStSqrIntegralGetSize_8u64u(roiSize, &bufSize, prop));
GpuMat buf(1, bufSize, CV_8U);
sqsum.create(src.rows + 1, src.cols + 1, CV_64F);
nppSafeCall(nppiStSqrIntegral_8u64u_C1R(
const_cast<NppSt8u*>(src.ptr<NppSt8u>(0)), src.step,
sqsum.ptr<NppSt64u>(0), sqsum.step, roiSize,
buf.ptr<NppSt8u>(0), bufSize));
nppSafeCall(nppiStSqrIntegral_8u64u_C1R(const_cast<Ncv8u*>(src.ptr<Ncv8u>(0)), src.step,
sqsum.ptr<Ncv64u>(0), sqsum.step, roiSize, buf.ptr<Ncv8u>(0), bufSize, prop));
cudaSafeCall( cudaThreadSynchronize() );
}
......
modules/gpu/src/nvidia/NCVHaarObjectDetection.hpp
View file @ 0747f2d8
......@@ -75,13 +75,13 @@ struct HaarFeature64
#define HaarFeature64_CreateCheck_MaxRectField 0xFF
__host__ NCVStatus setRect(Ncv32u rectX, Ncv32u rectY, Ncv32u rectWidth, Ncv32u rectHeight, Ncv32u clsWidth, Ncv32u clsHeight)
__host__ NCVStatus setRect(Ncv32u rectX, Ncv32u rectY, Ncv32u rectWidth, Ncv32u rectHeight, Ncv32u /*clsWidth*/, Ncv32u /*clsHeight*/)
{
ncvAssertReturn(rectWidth <= HaarFeature64_CreateCheck_MaxRectField && rectHeight <= HaarFeature64_CreateCheck_MaxRectField, NCV_HAAR_TOO_LARGE_FEATURES);
((NcvRect8u*)&(this->_ui2.x))->x = rectX;
((NcvRect8u*)&(this->_ui2.x))->y = rectY;
((NcvRect8u*)&(this->_ui2.x))->width = rectWidth;
((NcvRect8u*)&(this->_ui2.x))->height = rectHeight;
((NcvRect8u*)&(this->_ui2.x))->x = (Ncv8u)rectX;
((NcvRect8u*)&(this->_ui2.x))->y = (Ncv8u)rectY;
((NcvRect8u*)&(this->_ui2.x))->width = (Ncv8u)rectWidth;
((NcvRect8u*)&(this->_ui2.x))->height = (Ncv8u)rectHeight;
return NCV_SUCCESS;
}
......@@ -306,11 +306,11 @@ struct HaarStage64
};
NPPST_CT_ASSERT(sizeof(HaarFeature64) == 8);
NPPST_CT_ASSERT(sizeof(HaarFeatureDescriptor32) == 4);
NPPST_CT_ASSERT(sizeof(HaarClassifierNodeDescriptor32) == 4);
NPPST_CT_ASSERT(sizeof(HaarClassifierNode128) == 16);
NPPST_CT_ASSERT(sizeof(HaarStage64) == 8);
NCV_CT_ASSERT(sizeof(HaarFeature64) == 8);
NCV_CT_ASSERT(sizeof(HaarFeatureDescriptor32) == 4);
NCV_CT_ASSERT(sizeof(HaarClassifierNodeDescriptor32) == 4);
NCV_CT_ASSERT(sizeof(HaarClassifierNode128) == 16);
NCV_CT_ASSERT(sizeof(HaarStage64) == 8);
//==============================================================================
......@@ -347,7 +347,7 @@ enum
NCVPipeObjDet_VisualizeInPlace = 0x004,
};
NCV_EXPORTS
NCVStatus ncvDetectObjectsMultiScale_device(NCVMatrix<Ncv8u> &d_srcImg,
NcvSize32u srcRoi,
NCVVector<NcvRect32u> &d_dstRects,
......@@ -367,15 +367,14 @@ NCVStatus ncvDetectObjectsMultiScale_device(NCVMatrix<Ncv8u> &d_srcImg,
INCVMemAllocator &gpuAllocator,
INCVMemAllocator &cpuAllocator,
Ncv32u devPropMajor,
Ncv32u devPropMinor,
cudaDeviceProp &devProp,
cudaStream_t cuStream);
#define OBJDET_MASK_ELEMENT_INVALID_32U 0xFFFFFFFF
#define HAAR_STDDEV_BORDER 1
NCV_EXPORTS
NCVStatus ncvApplyHaarClassifierCascade_device(NCVMatrix<Ncv32u> &d_integralImage,
NCVMatrix<Ncv32f> &d_weights,
NCVMatrixAlloc<Ncv32u> &d_pixelMask,
......@@ -391,11 +390,10 @@ NCVStatus ncvApplyHaarClassifierCascade_device(NCVMatrix<Ncv32u> &d_integralImag
Ncv32f scaleArea,
INCVMemAllocator &gpuAllocator,
INCVMemAllocator &cpuAllocator,
Ncv32u devPropMajor,
Ncv32u devPropMinor,
cudaDeviceProp &devProp,
cudaStream_t cuStream);
NCV_EXPORTS
NCVStatus ncvApplyHaarClassifierCascade_host(NCVMatrix<Ncv32u> &h_integralImage,
NCVMatrix<Ncv32f> &h_weights,
NCVMatrixAlloc<Ncv32u> &h_pixelMask,
......@@ -409,7 +407,7 @@ NCVStatus ncvApplyHaarClassifierCascade_host(NCVMatrix<Ncv32u> &h_integralImage,
Ncv32u pixelStep,
Ncv32f scaleArea);
NCV_EXPORTS
NCVStatus ncvDrawRects_8u_device(Ncv8u *d_dst,
Ncv32u dstStride,
Ncv32u dstWidth,
......@@ -419,7 +417,7 @@ NCVStatus ncvDrawRects_8u_device(Ncv8u *d_dst,
Ncv8u color,
cudaStream_t cuStream);
NCV_EXPORTS
NCVStatus ncvDrawRects_32u_device(Ncv32u *d_dst,
Ncv32u dstStride,
Ncv32u dstWidth,
......@@ -429,7 +427,7 @@ NCVStatus ncvDrawRects_32u_device(Ncv32u *d_dst,
Ncv32u color,
cudaStream_t cuStream);
NCV_EXPORTS
NCVStatus ncvDrawRects_8u_host(Ncv8u *h_dst,
Ncv32u dstStride,
Ncv32u dstWidth,
......@@ -438,7 +436,7 @@ NCVStatus ncvDrawRects_8u_host(Ncv8u *h_dst,
Ncv32u numRects,
Ncv8u color);
NCV_EXPORTS
NCVStatus ncvDrawRects_32u_host(Ncv32u *h_dst,
Ncv32u dstStride,
Ncv32u dstWidth,
......@@ -450,7 +448,7 @@ NCVStatus ncvDrawRects_32u_host(Ncv32u *h_dst,
#define RECT_SIMILARITY_PROPORTION 0.2f
NCV_EXPORTS
NCVStatus ncvGrowDetectionsVector_device(NCVVector<Ncv32u> &pixelMask,
Ncv32u numPixelMaskDetections,
NCVVector<NcvRect32u> &hypotheses,
......@@ -461,7 +459,7 @@ NCVStatus ncvGrowDetectionsVector_device(NCVVector<Ncv32u> &pixelMask,
Ncv32f curScale,
cudaStream_t cuStream);
NCV_EXPORTS
NCVStatus ncvGrowDetectionsVector_host(NCVVector<Ncv32u> &pixelMask,
Ncv32u numPixelMaskDetections,
NCVVector<NcvRect32u> &hypotheses,
......@@ -471,18 +469,18 @@ NCVStatus ncvGrowDetectionsVector_host(NCVVector<Ncv32u> &pixelMask,
Ncv32u rectHeight,
Ncv32f curScale);
NCV_EXPORTS
NCVStatus ncvFilterHypotheses_host(NCVVector<NcvRect32u> &hypotheses,
Ncv32u &numHypotheses,
Ncv32u minNeighbors,
Ncv32f intersectEps,
NCVVector<Ncv32u> *hypothesesWeights);
NCV_EXPORTS
NCVStatus ncvHaarGetClassifierSize(const std::string &filename, Ncv32u &numStages,
Ncv32u &numNodes, Ncv32u &numFeatures);
NCV_EXPORTS
NCVStatus ncvHaarLoadFromFile_host(const std::string &filename,
HaarClassifierCascadeDescriptor &haar,
NCVVector<HaarStage64> &h_HaarStages,
......@@ -490,6 +488,7 @@ NCVStatus ncvHaarLoadFromFile_host(const std::string &filename,
NCVVector<HaarFeature64> &h_HaarFeatures);
NCV_EXPORTS
NCVStatus ncvHaarStoreNVBIN_host(const std::string &filename,
HaarClassifierCascadeDescriptor haar,
NCVVector<HaarStage64> &h_HaarStages,
......
modules/gpu/src/nvidia/NPP_staging/NPP_staging.cu 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
modules/gpu/src/nvidia/NCV.cpp → modules/gpu/src/nvidia/core/NCV.cpp
View file @ 0747f2d8
......@@ -40,15 +40,13 @@
//M*/
#include <precomp.hpp>
#if !defined (HAVE_CUDA)
#else /* !defined (HAVE_CUDA) */
#include <ios>
#include <stdarg.h>
#include "NCV.hpp"
......@@ -94,17 +92,6 @@ void ncvSetDebugOutputHandler(NCVDebugOutputHandler *func)
//==============================================================================
NCVStatus GPUAlignmentValue(Ncv32u &alignment)
{
int curDev;
cudaDeviceProp curProp;
ncvAssertCUDAReturn(cudaGetDevice(&curDev), NCV_CUDA_ERROR);
ncvAssertCUDAReturn(cudaGetDeviceProperties(&curProp, curDev), NCV_CUDA_ERROR);
alignment = curProp.textureAlignment; //GPUAlignmentValue(curProp.major);
return NCV_SUCCESS;
}
Ncv32u alignUp(Ncv32u what, Ncv32u alignment)
{
Ncv32u alignMask = alignment-1;
......@@ -216,7 +203,7 @@ NCVMemStackAllocator::NCVMemStackAllocator(Ncv32u alignment)
}
NCVMemStackAllocator::NCVMemStackAllocator(NCVMemoryType memT, size_t capacity, Ncv32u alignment)
NCVMemStackAllocator::NCVMemStackAllocator(NCVMemoryType memT, size_t capacity, Ncv32u alignment, void *reusePtr)
:
currentSize(0),
_maxSize(0),
......@@ -229,17 +216,26 @@ NCVMemStackAllocator::NCVMemStackAllocator(NCVMemoryType memT, size_t capacity,
allocBegin = NULL;
switch (memT)
if (reusePtr == NULL)
{
case NCVMemoryTypeDevice:
ncvAssertCUDAReturn(cudaMalloc(&allocBegin, capacity), );
break;
case NCVMemoryTypeHostPinned:
ncvAssertCUDAReturn(cudaMallocHost(&allocBegin, capacity), );
break;
case NCVMemoryTypeHostPageable:
allocBegin = (Ncv8u *)malloc(capacity);
break;
bReusesMemory = false;
switch (memT)
{
case NCVMemoryTypeDevice:
ncvAssertCUDAReturn(cudaMalloc(&allocBegin, capacity), );
break;
case NCVMemoryTypeHostPinned:
ncvAssertCUDAReturn(cudaMallocHost(&allocBegin, capacity), );
break;
case NCVMemoryTypeHostPageable:
allocBegin = (Ncv8u *)malloc(capacity);
break;
}
}
else
{
bReusesMemory = true;
allocBegin = (Ncv8u *)reusePtr;
}
if (capacity == 0)
......@@ -260,18 +256,23 @@ NCVMemStackAllocator::~NCVMemStackAllocator()
if (allocBegin != NULL)
{
ncvAssertPrintCheck(currentSize == 0, "NCVMemStackAllocator dtor:: not all objects were deallocated properly, forcing destruction");
switch (_memType)
if (!bReusesMemory)
{
case NCVMemoryTypeDevice:
ncvAssertCUDAReturn(cudaFree(allocBegin), );
break;
case NCVMemoryTypeHostPinned:
ncvAssertCUDAReturn(cudaFreeHost(allocBegin), );
break;
case NCVMemoryTypeHostPageable:
free(allocBegin);
break;
switch (_memType)
{
case NCVMemoryTypeDevice:
ncvAssertCUDAReturn(cudaFree(allocBegin), );
break;
case NCVMemoryTypeHostPinned:
ncvAssertCUDAReturn(cudaFreeHost(allocBegin), );
break;
case NCVMemoryTypeHostPageable:
free(allocBegin);
break;
}
}
allocBegin = NULL;
}
}
......@@ -356,14 +357,14 @@ size_t NCVMemStackAllocator::maxSize(void) const
//===================================================================
NCVMemNativeAllocator::NCVMemNativeAllocator(NCVMemoryType memT)
NCVMemNativeAllocator::NCVMemNativeAllocator(NCVMemoryType memT, Ncv32u alignment)
:
currentSize(0),
_maxSize(0),
_memType(memT)
_memType(memT),
_alignment(alignment)
{
ncvAssertPrintReturn(memT != NCVMemoryTypeNone, "NCVMemNativeAllocator ctor:: counting not permitted for this allocator type", );
ncvAssertPrintReturn(NCV_SUCCESS == GPUAlignmentValue(this->_alignment), "NCVMemNativeAllocator ctor:: couldn't get device _alignment", );
}
......
modules/gpu/src/nvidia/NCVRuntimeTemplates.hpp → modules/gpu/src/nvidia/core/NCVRuntimeTemplates.hpp
View file @ 0747f2d8
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2009-2010, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _ncvruntimetemplates_hpp_
#define _ncvruntimetemplates_hpp_
#include <stdarg.h>
#include <vector>
////////////////////////////////////////////////////////////////////////////////
// The Loki Library
// Copyright (c) 2001 by Andrei Alexandrescu
......@@ -14,13 +62,6 @@
// http://loki-lib.sourceforge.net/index.php?n=Main.License
////////////////////////////////////////////////////////////////////////////////
#ifndef _ncvruntimetemplates_hpp_
#define _ncvruntimetemplates_hpp_
#include <stdarg.h>
#include <vector>
namespace Loki
{
//==============================================================================
......
modules/gpu/src/precomp.hpp
View file @ 0747f2d8
......@@ -69,9 +69,9 @@
#include "cufft.h"
#include "opencv2/gpu/stream_accessor.hpp"
#include "npp.h"
#include "npp_staging.h"
#include "nvidia/NCV.hpp"
#include "nvidia/core/NCV.hpp"
#include "nvidia/NPP_staging/npp_staging.hpp"
#include "nvidia/NCVHaarObjectDetection.hpp"
#define CUDART_MINIMUM_REQUIRED_VERSION 3020
......
modules/gtest/src/gtestcv.cpp
View file @ 0747f2d8
......@@ -1378,7 +1378,7 @@ cmpEpsFlt_(const _Tp* src1, const _Tp* src2, size_t total, int imaxdiff, size_t
{
_Tp a = src1[i], b = src2[i];
if( a < 0 ) a ^= C; if( b < 0 ) b ^= C;
_Tp d = std::abs(a - b);
_Tp d = std::abs(double(a - b));
if( d > imaxdiff )
{
idx = i + startidx;
......
samples/CMakeLists.txt
View file @ 0747f2d8
......@@ -3,8 +3,8 @@
#
# ----------------------------------------------------------------------------
add_subdirectory(c)
add_subdirectory(cpp)
#add_subdirectory(c)
#add_subdirectory(cpp)
add_subdirectory(gpu)
if(0)
......
samples/gpu/CMakeLists.txt
View file @ 0747f2d8
......@@ -14,10 +14,12 @@ if (BUILD_EXAMPLES)
"${CMAKE_SOURCE_DIR}/modules/legacy/include"
"${CMAKE_SOURCE_DIR}/modules/contrib/include"
"${CMAKE_SOURCE_DIR}/modules/gpu/include"
"${CMAKE_SOURCE_DIR}/modules/gpu/src/nvidia"
"${CMAKE_SOURCE_DIR}/modules/gpu/src/nvidia/core"
)
if(HAVE_CUDA)
include_directories(${CUDA_INCLUDE_DIRS})
include_directories(${CUDA_INCLUDE_DIRS} ${CMAKE_SOURCE_DIR}/modules/gpu/src/nvidia ${CMAKE_SOURCE_DIR}/modules/gpu/src/nvidia/core)
endif()
if(CMAKE_COMPILER_IS_GNUCXX)
......
tests/gpu/CMakeLists.txt
View file @ 0747f2d8
......@@ -3,12 +3,15 @@
# ----------------------------------------------------------------------------
project(opencv_test_gpu)
set(the_target "opencv_test_gpu")
file(GLOB test_srcs "src/*.cpp")
source_group("Src" FILES ${test_srcs})
file(GLOB test_hdrs "src/*.h*")
source_group("Src" FILES ${test_srcs})
source_group("Include" FILES ${test_hdrs})
set(the_target "opencv_test_gpu")
include_directories (
"${CMAKE_SOURCE_DIR}/include/opencv"
......@@ -26,11 +29,21 @@ include_directories (
"${CMAKE_SOURCE_DIR}/modules/ml/include"
"${CMAKE_CURRENT_SOURCE_DIR}/src"
"${CMAKE_CURRENT_BINARY_DIR}"
"${CMAKE_SOURCE_DIR}/modules/gpu/src/nvidia"
"${CMAKE_SOURCE_DIR}/modules/gpu/src/nvidia/core"
)
include_directories(../cxts)
add_executable(${the_target} ${test_srcs} ${test_hdrs})
if(HAVE_CUDA)
include_directories(${CUDA_INCLUDE_DIRS} ${CMAKE_SOURCE_DIR}/modules/gpu/src/nvidia ${CMAKE_SOURCE_DIR}/modules/gpu/src/nvidia/core ${CMAKE_SOURCE_DIR}/modules/gpu/src/nvidia/NPP_staging)
file(GLOB nvidia "src/nvidia/*.*")
SET(ncv_cpp ../../modules/gpu/src/nvidia/core/NCV.cpp)
source_group("Src\\nvidia" FILES ${nvidia})
endif()
add_executable(${the_target} ${test_srcs} ${test_hdrs} ${nvidia} ${ncv_cpp})
# Additional target properties
set_target_properties(${the_target} PROPERTIES
......
tests/gpu/src/nvidia/NCVAutoTestLister.hpp 0 → 100644
View file @ 0747f2d8
/*
* Copyright 1993-2010 NVIDIA Corporation. All rights reserved.
*
* NVIDIA Corporation and its licensors retain all intellectual
* property and proprietary rights in and to this software and
* related documentation and any modifications thereto.
* Any use, reproduction, disclosure, or distribution of this
* software and related documentation without an express license
* agreement from NVIDIA Corporation is strictly prohibited.
*/
#ifndef _ncvautotestlister_hpp_
#define _ncvautotestlister_hpp_
#include <vector>
#include "NCVTest.hpp"
class NCVAutoTestLister
{
public:
NCVAutoTestLister(std::string testSuiteName, NcvBool bStopOnFirstFail=false, NcvBool bCompactOutput=true)
:
testSuiteName(testSuiteName),
bStopOnFirstFail(bStopOnFirstFail),
bCompactOutput(bCompactOutput)
{
}
void add(INCVTest *test)
{
this->tests.push_back(test);
}
void invoke()
{
Ncv32u nPassed = 0;
Ncv32u nFailed = 0;
Ncv32u nFailedMem = 0;
if (bCompactOutput)
{
printf("Test suite '%s' with %d tests\n",
testSuiteName.c_str(),
(int)(this->tests.size()));
}
for (Ncv32u i=0; i<this->tests.size(); i++)
{
INCVTest &curTest = *tests[i];
NCVTestReport curReport;
bool res = curTest.executeTest(curReport);
if (!bCompactOutput)
{
printf("Test %3i %16s; Consumed mem GPU = %8d, CPU = %8d; %s\n",
i,
curTest.getName().c_str(),
curReport.statsNums["MemGPU"],
curReport.statsNums["MemCPU"],
curReport.statsText["rcode"].c_str());
}
if (res)
{
nPassed++;
if (bCompactOutput)
{
printf(".");
}
}
else
{
if (!curReport.statsText["rcode"].compare("FAILED"))
{
nFailed++;
if (bCompactOutput)
{
printf("x");
}
if (bStopOnFirstFail)
{
break;
}
}
else
{
nFailedMem++;
if (bCompactOutput)
{
printf("m");
}
}
}
fflush(stdout);
}
if (bCompactOutput)
{
printf("\n");
}
printf("Test suite '%s' complete: %d total, %d passed, %d memory errors, %d failed\n\n",
testSuiteName.c_str(),
(int)(this->tests.size()),
nPassed,
nFailedMem,
nFailed);
}
~NCVAutoTestLister()
{
for (Ncv32u i=0; i<this->tests.size(); i++)
{
delete tests[i];
}
}
private:
NcvBool bStopOnFirstFail;
NcvBool bCompactOutput;
std::string testSuiteName;
std::vector<INCVTest *> tests;
};
#endif // _ncvautotestlister_hpp_
tests/gpu/src/nvidia/NCVTest.hpp 0 → 100644
View file @ 0747f2d8
/*
* Copyright 1993-2010 NVIDIA Corporation. All rights reserved.
*
* NVIDIA Corporation and its licensors retain all intellectual
* property and proprietary rights in and to this software and
* related documentation and any modifications thereto.
* Any use, reproduction, disclosure, or distribution of this
* software and related documentation without an express license
* agreement from NVIDIA Corporation is strictly prohibited.
*/
#ifndef _ncvtest_hpp_
#define _ncvtest_hpp_
#pragma warning( disable : 4201 4408 4127 4100)
#include <string>
#include <vector>
#include <map>
#include <memory>
#include <algorithm>
#include <fstream>
#include <cuda_runtime.h>
#include "NPP_staging.hpp"
struct NCVTestReport
{
std::map<std::string, Ncv32u> statsNums;
std::map<std::string, std::string> statsText;
};
class INCVTest
{
public:
virtual bool executeTest(NCVTestReport &report) = 0;
virtual std::string getName() const = 0;
};
class NCVTestProvider : public INCVTest
{
public:
NCVTestProvider(std::string testName)
:
testName(testName)
{
int devId;
ncvAssertPrintReturn(cudaSuccess == cudaGetDevice(&devId), "Error returned from cudaGetDevice", );
ncvAssertPrintReturn(cudaSuccess == cudaGetDeviceProperties(&this->devProp, devId), "Error returned from cudaGetDeviceProperties", );
}
virtual bool init() = 0;
virtual bool process() = 0;
virtual bool deinit() = 0;
virtual bool toString(std::ofstream &strOut) = 0;
virtual std::string getName() const
{
return this->testName;
}
virtual ~NCVTestProvider()
{
deinitMemory();
}
virtual bool executeTest(NCVTestReport &report)
{
bool res;
report.statsText["rcode"] = "FAILED";
res = initMemory(report);
if (!res)
{
dumpToFile(report);
deinitMemory();
return false;
}
res = init();
if (!res)
{
dumpToFile(report);
deinit();
deinitMemory();
return false;
}
res = process();
if (!res)
{
dumpToFile(report);
deinit();
deinitMemory();
return false;
}
res = deinit();
if (!res)
{
dumpToFile(report);
deinitMemory();
return false;
}
deinitMemory();
report.statsText["rcode"] = "Passed";
return true;
}
protected:
cudaDeviceProp devProp;
std::auto_ptr<INCVMemAllocator> allocatorGPU;
std::auto_ptr<INCVMemAllocator> allocatorCPU;
private:
std::string testName;
bool initMemory(NCVTestReport &report)
{
this->allocatorGPU.reset(new NCVMemStackAllocator(devProp.textureAlignment));
this->allocatorCPU.reset(new NCVMemStackAllocator(devProp.textureAlignment));
if (!this->allocatorGPU.get()->isInitialized() ||
!this->allocatorCPU.get()->isInitialized())
{
report.statsText["rcode"] = "Memory FAILED";
return false;
}
if (!this->process())
{
report.statsText["rcode"] = "Memory FAILED";
return false;
}
Ncv32u maxGPUsize = (Ncv32u)this->allocatorGPU.get()->maxSize();
Ncv32u maxCPUsize = (Ncv32u)this->allocatorCPU.get()->maxSize();
report.statsNums["MemGPU"] = maxGPUsize;
report.statsNums["MemCPU"] = maxCPUsize;
this->allocatorGPU.reset(new NCVMemStackAllocator(NCVMemoryTypeDevice, maxGPUsize, devProp.textureAlignment));
this->allocatorCPU.reset(new NCVMemStackAllocator(NCVMemoryTypeHostPinned, maxCPUsize, devProp.textureAlignment));
if (!this->allocatorGPU.get()->isInitialized() ||
!this->allocatorCPU.get()->isInitialized())
{
report.statsText["rcode"] = "Memory FAILED";
return false;
}
return true;
}
void deinitMemory()
{
this->allocatorGPU.reset();
this->allocatorCPU.reset();
}
void dumpToFile(NCVTestReport &report)
{
bool bReasonMem = (0 == report.statsText["rcode"].compare("Memory FAILED"));
std::string fname = "TestDump_";
fname += (bReasonMem ? "m_" : "") + this->testName + ".log";
std::ofstream stream(fname.c_str(), std::ios::trunc | std::ios::out);
if (!stream.is_open()) return;
stream << "NCV Test Failure Log: " << this->testName << std::endl;
stream << "====================================================" << std::endl << std::endl;
stream << "Test initialization report: " << std::endl;
for (std::map<std::string,std::string>::iterator it=report.statsText.begin();
it != report.statsText.end(); it++)
{
stream << it->first << "=" << it->second << std::endl;
}
for (std::map<std::string,Ncv32u>::iterator it=report.statsNums.begin();
it != report.statsNums.end(); it++)
{
stream << it->first << "=" << it->second << std::endl;
}
stream << std::endl;
stream << "Test initialization parameters: " << std::endl;
bool bSerializeRes = false;
try
{
bSerializeRes = this->toString(stream);
}
catch (...)
{
}
if (!bSerializeRes)
{
stream << "Couldn't retrieve object dump" << std::endl;
}
stream.flush();
}
};
#endif // _ncvtest_hpp_
tests/gpu/src/nvidia/NCVTestSourceProvider.hpp 0 → 100644
View file @ 0747f2d8
/*
* Copyright 1993-2010 NVIDIA Corporation. All rights reserved.
*
* NVIDIA Corporation and its licensors retain all intellectual
* property and proprietary rights in and to this software and
* related documentation and any modifications thereto.
* Any use, reproduction, disclosure, or distribution of this
* software and related documentation without an express license
* agreement from NVIDIA Corporation is strictly prohibited.
*/
#ifndef _ncvtestsourceprovider_hpp_
#define _ncvtestsourceprovider_hpp_
#include <memory>
#include "NCV.hpp"
#include <opencv2/highgui/highgui.hpp>
template <class T>
class NCVTestSourceProvider
{
public:
NCVTestSourceProvider(Ncv32u seed, T rangeLow, T rangeHigh, Ncv32u maxWidth, Ncv32u maxHeight)
:
bInit(false)
{
ncvAssertPrintReturn(rangeLow < rangeHigh, "NCVTestSourceProvider ctor:: Invalid range", );
int devId;
cudaDeviceProp devProp;
ncvAssertPrintReturn(cudaSuccess == cudaGetDevice(&devId), "Error returned from cudaGetDevice", );
ncvAssertPrintReturn(cudaSuccess == cudaGetDeviceProperties(&devProp, devId), "Error returned from cudaGetDeviceProperties", );
//Ncv32u maxWpitch = alignUp(maxWidth * sizeof(T), devProp.textureAlignment);
allocatorCPU.reset(new NCVMemNativeAllocator(NCVMemoryTypeHostPinned, devProp.textureAlignment));
data.reset(new NCVMatrixAlloc<T>(*this->allocatorCPU.get(), maxWidth, maxHeight));
ncvAssertPrintReturn(data.get()->isMemAllocated(), "NCVTestSourceProvider ctor:: Matrix not allocated", );
this->dataWidth = maxWidth;
this->dataHeight = maxHeight;
srand(seed);
for (Ncv32u i=0; i<maxHeight; i++)
{
for (Ncv32u j=0; j<data.get()->stride(); j++)
{
data.get()->ptr()[i * data.get()->stride() + j] =
(T)(((1.0 * rand()) / RAND_MAX) * (rangeHigh - rangeLow) + rangeLow);
}
}
this->bInit = true;
}
NCVTestSourceProvider(std::string pgmFilename)
:
bInit(false)
{
ncvAssertPrintReturn(sizeof(T) == 1, "NCVTestSourceProvider ctor:: PGM constructor complies only with 8bit types", );
cv::Mat image = cv::imread(pgmFilename);
ncvAssertPrintReturn(!image.empty(), "NCVTestSourceProvider ctor:: PGM file error", );
int devId;
cudaDeviceProp devProp;
ncvAssertPrintReturn(cudaSuccess == cudaGetDevice(&devId), "Error returned from cudaGetDevice", );
ncvAssertPrintReturn(cudaSuccess == cudaGetDeviceProperties(&devProp, devId), "Error returned from cudaGetDeviceProperties", );
allocatorCPU.reset(new NCVMemNativeAllocator(NCVMemoryTypeHostPinned, devProp.textureAlignment));
data.reset(new NCVMatrixAlloc<T>(*this->allocatorCPU.get(), image.cols, image.rows));
ncvAssertPrintReturn(data.get()->isMemAllocated(), "NCVTestSourceProvider ctor:: Matrix not allocated", );
this->dataWidth = image.cols;
this->dataHeight = image.rows;
cv::Mat hdr(image.size(), CV_8UC1, data.get()->ptr(), data.get()->pitch());
image.copyTo(hdr);
this->bInit = true;
}
NcvBool fill(NCVMatrix<T> &dst)
{
ncvAssertReturn(this->isInit() &&
dst.memType() == allocatorCPU.get()->memType(), false);
if (dst.width() == 0 || dst.height() == 0)
{
return true;
}
for (Ncv32u i=0; i<dst.height(); i++)
{
Ncv32u srcLine = i % this->dataHeight;
Ncv32u srcFullChunks = dst.width() / this->dataWidth;
for (Ncv32u j=0; j<srcFullChunks; j++)
{
memcpy(dst.ptr() + i * dst.stride() + j * this->dataWidth,
this->data.get()->ptr() + this->data.get()->stride() * srcLine,
this->dataWidth * sizeof(T));
}
Ncv32u srcLastChunk = dst.width() % this->dataWidth;
memcpy(dst.ptr() + i * dst.stride() + srcFullChunks * this->dataWidth,
this->data.get()->ptr() + this->data.get()->stride() * srcLine,
srcLastChunk * sizeof(T));
}
return true;
}
NcvBool fill(NCVVector<T> &dst)
{
ncvAssertReturn(this->isInit() &&
dst.memType() == allocatorCPU.get()->memType(), false);
if (dst.length() == 0)
{
return true;
}
Ncv32u srcLen = this->dataWidth * this->dataHeight;
Ncv32u srcFullChunks = (Ncv32u)dst.length() / srcLen;
for (Ncv32u j=0; j<srcFullChunks; j++)
{
memcpy(dst.ptr() + j * srcLen, this->data.get()->ptr(), srcLen * sizeof(T));
}
Ncv32u srcLastChunk = dst.length() % srcLen;
memcpy(dst.ptr() + srcFullChunks * srcLen, this->data.get()->ptr(), srcLastChunk * sizeof(T));
return true;
}
~NCVTestSourceProvider()
{
data.reset();
allocatorCPU.reset();
}
private:
NcvBool isInit(void)
{
return this->bInit;
}
NcvBool bInit;
std::auto_ptr< INCVMemAllocator > allocatorCPU;
std::auto_ptr< NCVMatrixAlloc<T> > data;
Ncv32u dataWidth;
Ncv32u dataHeight;
};
#endif // _ncvtestsourceprovider_hpp_
tests/gpu/src/nvidia/TestCompact.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestCompact.h 0 → 100644
View file @ 0747f2d8
/*
* Copyright 1993-2010 NVIDIA Corporation. All rights reserved.
*
* NVIDIA Corporation and its licensors retain all intellectual
* property and proprietary rights in and to this software and
* related documentation and any modifications thereto.
* Any use, reproduction, disclosure, or distribution of this
* software and related documentation without an express license
* agreement from NVIDIA Corporation is strictly prohibited.
*/
#ifndef _testhypothesescompact_h_
#define _testhypothesescompact_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
class TestCompact : public NCVTestProvider
{
public:
TestCompact(std::string testName, NCVTestSourceProvider<Ncv32u> &src,
Ncv32u length, Ncv32u badElem, Ncv32u badElemPercentage);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
TestCompact(const TestCompact&);
TestCompact& operator=(const TestCompact&);
NCVTestSourceProvider<Ncv32u> &src;
Ncv32u length;
Ncv32u badElem;
Ncv32u badElemPercentage;
};
#endif // _testhypothesescompact_h_
tests/gpu/src/nvidia/TestDrawRects.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestDrawRects.h 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestHaarCascadeApplication.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestHaarCascadeApplication.h 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestHaarCascadeLoader.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestHaarCascadeLoader.h 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestHypothesesFilter.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestHypothesesFilter.h 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestHypothesesGrow.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestHypothesesGrow.h 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestIntegralImage.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestIntegralImage.h 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestIntegralImageSquared.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestIntegralImageSquared.h 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestRectStdDev.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestRectStdDev.h 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestResize.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestResize.h 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestTranspose.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/TestTranspose.h 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia/main_nvidia.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
tests/gpu/src/nvidia_tests.cpp 0 → 100644
View file @ 0747f2d8
This diff is collapsed.
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