提交 59ed7d06 编写于 作者: A Alexander Alekhin

Merge pull request #4072 from vladimir-dudnik:opencl-opencv-interop

......@@ -71,9 +71,28 @@ using namespace cv::ocl;
//! @{
// TODO static functions in the Context class
//! @brief Creates OpenCL context from D3D11 device
//
//! @param pD3D11Device - pointer to D3D11 device
//! @return Returns reference to OpenCL Context
CV_EXPORTS Context& initializeContextFromD3D11Device(ID3D11Device* pD3D11Device);
//! @brief Creates OpenCL context from D3D10 device
//
//! @param pD3D10Device - pointer to D3D10 device
//! @return Returns reference to OpenCL Context
CV_EXPORTS Context& initializeContextFromD3D10Device(ID3D10Device* pD3D10Device);
//! @brief Creates OpenCL context from Direct3DDevice9Ex device
//
//! @param pDirect3DDevice9Ex - pointer to Direct3DDevice9Ex device
//! @return Returns reference to OpenCL Context
CV_EXPORTS Context& initializeContextFromDirect3DDevice9Ex(IDirect3DDevice9Ex* pDirect3DDevice9Ex);
//! @brief Creates OpenCL context from Direct3DDevice9 device
//
//! @param pDirect3DDevice9 - pointer to Direct3Device9 device
//! @return Returns reference to OpenCL Context
CV_EXPORTS Context& initializeContextFromDirect3DDevice9(IDirect3DDevice9* pDirect3DDevice9);
//! @}
......@@ -83,19 +102,70 @@ CV_EXPORTS Context& initializeContextFromDirect3DDevice9(IDirect3DDevice9* pDire
//! @addtogroup core_directx
//! @{
//! @brief Converts InputArray to ID3D11Texture2D
//
//! @note Note: function does memory copy from src to
//! pD3D11Texture2D
//
//! @param src - source InputArray
//! @param pD3D11Texture2D - destination D3D11 texture
CV_EXPORTS void convertToD3D11Texture2D(InputArray src, ID3D11Texture2D* pD3D11Texture2D);
//! @brief Converts ID3D11Texture2D to OutputArray
//
//! @note Note: function does memory copy from pD3D11Texture2D
//! to dst
//
//! @param pD3D11Texture2D - source D3D11 texture
//! @param dst - destination OutputArray
CV_EXPORTS void convertFromD3D11Texture2D(ID3D11Texture2D* pD3D11Texture2D, OutputArray dst);
//! @brief Converts InputArray to ID3D10Texture2D
//
//! @note Note: function does memory copy from src to
//! pD3D10Texture2D
//
//! @param src - source InputArray
//! @param pD3D10Texture2D - destination D3D10 texture
CV_EXPORTS void convertToD3D10Texture2D(InputArray src, ID3D10Texture2D* pD3D10Texture2D);
//! @brief Converts ID3D10Texture2D to OutputArray
//
//! @note Note: function does memory copy from pD3D10Texture2D
//! to dst
//
//! @param pD3D10Texture2D - source D3D10 texture
//! @param dst - destination OutputArray
CV_EXPORTS void convertFromD3D10Texture2D(ID3D10Texture2D* pD3D10Texture2D, OutputArray dst);
//! @brief Converts InputArray to IDirect3DSurface9
//
//! @note Note: function does memory copy from src to
//! pDirect3DSurface9
//
//! @param src - source InputArray
//! @param pDirect3DSurface9 - destination D3D10 texture
//! @param surfaceSharedHandle - shared handle
CV_EXPORTS void convertToDirect3DSurface9(InputArray src, IDirect3DSurface9* pDirect3DSurface9, void* surfaceSharedHandle = NULL);
//! @brief Converts IDirect3DSurface9 to OutputArray
//
//! @note Note: function does memory copy from pDirect3DSurface9
//! to dst
//
//! @param pDirect3DSurface9 - source D3D10 texture
//! @param dst - destination OutputArray
//! @param surfaceSharedHandle - shared handle
CV_EXPORTS void convertFromDirect3DSurface9(IDirect3DSurface9* pDirect3DSurface9, OutputArray dst, void* surfaceSharedHandle = NULL);
// Get OpenCV type from DirectX type, return -1 if there is no equivalent
//! @brief Get OpenCV type from DirectX type
//! @param iDXGI_FORMAT - enum DXGI_FORMAT for D3D10/D3D11
//! @return OpenCV type or -1 if there is no equivalent
CV_EXPORTS int getTypeFromDXGI_FORMAT(const int iDXGI_FORMAT); // enum DXGI_FORMAT for D3D10/D3D11
// Get OpenCV type from DirectX type, return -1 if there is no equivalent
//! @brief Get OpenCV type from DirectX type
//! @param iD3DFORMAT - enum D3DTYPE for D3D9
//! @return OpenCV type or -1 if there is no equivalent
CV_EXPORTS int getTypeFromD3DFORMAT(const int iD3DFORMAT); // enum D3DTYPE for D3D9
//! @}
......
......@@ -276,6 +276,58 @@ protected:
Impl* p;
};
/*
//! @brief Attaches OpenCL context to OpenCV
//
//! @note Note:
// OpenCV will check if available OpenCL platform has platformName name,
// then assign context to OpenCV and call clRetainContext function.
// The deviceID device will be used as target device and new command queue
// will be created.
//
// Params:
//! @param platformName - name of OpenCL platform to attach,
//! this string is used to check if platform is available
//! to OpenCV at runtime
//! @param platfromID - ID of platform attached context was created for
//! @param context - OpenCL context to be attached to OpenCV
//! @param deviceID - ID of device, must be created from attached context
*/
CV_EXPORTS void attachContext(const String& platformName, void* platformID, void* context, void* deviceID);
/*
//! @brief Convert OpenCL buffer to UMat
//
//! @note Note:
// OpenCL buffer (cl_mem_buffer) should contain 2D image data, compatible with OpenCV.
// Memory content is not copied from clBuffer to UMat. Instead, buffer handle assigned
// to UMat and clRetainMemObject is called.
//
// Params:
//! @param cl_mem_buffer - source clBuffer handle
//! @param step - num of bytes in single row
//! @param rows - number of rows
//! @param cols - number of cols
//! @param type - OpenCV type of image
//! @param dst - destination UMat
*/
CV_EXPORTS void convertFromBuffer(void* cl_mem_buffer, size_t step, int rows, int cols, int type, UMat& dst);
/*
//! @brief Convert OpenCL image2d_t to UMat
//
//! @note Note:
// OpenCL image2d_t (cl_mem_image), should be compatible with OpenCV
// UMat formats.
// Memory content is copied from image to UMat with
// clEnqueueCopyImageToBuffer function.
//
// Params:
//! @param cl_mem_image - source image2d_t handle
//! @param dst - destination UMat
*/
CV_EXPORTS void convertFromImage(void* cl_mem_image, UMat& dst);
// TODO Move to internal header
void initializeContextFromHandle(Context& ctx, void* platform, void* context, void* device);
......
......@@ -858,9 +858,9 @@ OCL_FUNC_P(cl_context, clCreateContext,
OCL_FUNC(cl_int, clReleaseContext, (cl_context context), (context))
/*
OCL_FUNC(cl_int, clRetainContext, (cl_context context), (context))
OCL_FUNC(cl_int, clRetainContext, (cl_context context), (context))
/*
OCL_FUNC_P(cl_context, clCreateContextFromType,
(const cl_context_properties * properties,
cl_device_type device_type,
......@@ -945,7 +945,6 @@ OCL_FUNC(cl_int, clGetSupportedImageFormats,
(context, flags, image_type, num_entries, image_formats, num_image_formats))
/*
OCL_FUNC(cl_int, clGetMemObjectInfo,
(cl_mem memobj,
cl_mem_info param_name,
......@@ -962,6 +961,7 @@ OCL_FUNC(cl_int, clGetImageInfo,
size_t * param_value_size_ret),
(image, param_name, param_value_size, param_value, param_value_size_ret))
/*
OCL_FUNC(cl_int, clCreateKernelsInProgram,
(cl_program program,
cl_uint num_kernels,
......@@ -1038,20 +1038,20 @@ OCL_FUNC(cl_int, clEnqueueCopyImage,
cl_event * event),
(command_queue, src_image, dst_image, src_origin, dst_origin,
region, num_events_in_wait_list, event_wait_list, event))
*/
OCL_FUNC(cl_int, clEnqueueCopyImageToBuffer,
(cl_command_queue command_queue,
cl_mem src_image,
cl_mem dst_buffer,
const size_t * src_origin[3],
const size_t * region[3],
const size_t * src_origin,
const size_t * region,
size_t dst_offset,
cl_uint num_events_in_wait_list,
const cl_event * event_wait_list,
cl_event * event),
(command_queue, src_image, dst_buffer, src_origin, region, dst_offset,
num_events_in_wait_list, event_wait_list, event))
*/
OCL_FUNC(cl_int, clEnqueueCopyBufferToImage,
(cl_command_queue command_queue,
......@@ -1100,10 +1100,10 @@ OCL_FUNC(cl_int, clGetKernelInfo,
size_t * param_value_size_ret),
(kernel, param_name, param_value_size, param_value, param_value_size_ret))
OCL_FUNC(cl_int, clRetainMemObject, (cl_mem memobj), (memobj))
*/
OCL_FUNC(cl_int, clRetainMemObject, (cl_mem memobj), (memobj))
OCL_FUNC(cl_int, clReleaseMemObject, (cl_mem memobj), (memobj))
......@@ -1348,7 +1348,7 @@ OCL_FUNC(cl_int, clReleaseEvent, (cl_event event), (event))
#define CL_VERSION_1_2
#endif
#endif
#endif // HAVE_OPENCL
#ifdef _DEBUG
#define CV_OclDbgAssert CV_DbgAssert
......@@ -2925,6 +2925,83 @@ CV_EXPORTS bool useSVM(UMatUsageFlags usageFlags)
#endif // HAVE_OPENCL_SVM
static void get_platform_name(cl_platform_id id, String& name)
{
// get platform name string length
size_t sz = 0;
if (CL_SUCCESS != clGetPlatformInfo(id, CL_PLATFORM_NAME, 0, 0, &sz))
CV_ErrorNoReturn(cv::Error::OpenCLApiCallError, "clGetPlatformInfo failed!");
// get platform name string
AutoBuffer<char> buf(sz + 1);
if (CL_SUCCESS != clGetPlatformInfo(id, CL_PLATFORM_NAME, sz, buf, 0))
CV_ErrorNoReturn(cv::Error::OpenCLApiCallError, "clGetPlatformInfo failed!");
// just in case, ensure trailing zero for ASCIIZ string
buf[sz] = 0;
name = (const char*)buf;
}
/*
// Attaches OpenCL context to OpenCV
*/
void attachContext(const String& platformName, void* platformID, void* context, void* deviceID)
{
cl_uint cnt = 0;
if(CL_SUCCESS != clGetPlatformIDs(0, 0, &cnt))
CV_ErrorNoReturn(cv::Error::OpenCLApiCallError, "clGetPlatformIDs failed!");
if (cnt == 0)
CV_ErrorNoReturn(cv::Error::OpenCLApiCallError, "no OpenCL platform available!");
std::vector<cl_platform_id> platforms(cnt);
if(CL_SUCCESS != clGetPlatformIDs(cnt, &platforms[0], 0))
CV_ErrorNoReturn(cv::Error::OpenCLApiCallError, "clGetPlatformIDs failed!");
bool platformAvailable = false;
// check if external platformName contained in list of available platforms in OpenCV
for (unsigned int i = 0; i < cnt; i++)
{
String availablePlatformName;
get_platform_name(platforms[i], availablePlatformName);
// external platform is found in the list of available platforms
if (platformName == availablePlatformName)
{
platformAvailable = true;
break;
}
}
if (!platformAvailable)
CV_ErrorNoReturn(cv::Error::OpenCLApiCallError, "No matched platforms available!");
// check if platformID corresponds to platformName
String actualPlatformName;
get_platform_name((cl_platform_id)platformID, actualPlatformName);
if (platformName != actualPlatformName)
CV_ErrorNoReturn(cv::Error::OpenCLApiCallError, "No matched platforms available!");
// do not initialize OpenCL context
Context ctx = Context::getDefault(false);
// attach supplied context to OpenCV
initializeContextFromHandle(ctx, platformID, context, deviceID);
if(CL_SUCCESS != clRetainContext((cl_context)context))
CV_ErrorNoReturn(cv::Error::OpenCLApiCallError, "clRetainContext failed!");
// clear command queue, if any
getCoreTlsData().get()->oclQueue.finish();
Queue q;
getCoreTlsData().get()->oclQueue = q;
return;
} // attachContext()
void initializeContextFromHandle(Context& ctx, void* platform, void* _context, void* _device)
{
......@@ -3150,10 +3227,10 @@ struct Kernel::Impl
bool haveTempDstUMats;
};
}}
}} // namespace cv::ocl
extern "C" {
extern "C"
{
static void CL_CALLBACK oclCleanupCallback(cl_event, cl_int, void *p)
{
((cv::ocl::Kernel::Impl*)p)->finit();
......@@ -5166,6 +5243,167 @@ MatAllocator* getOpenCLAllocator()
return allocator;
}
}} // namespace cv::ocl
namespace cv {
// three funcs below are implemented in umatrix.cpp
void setSize( UMat& m, int _dims, const int* _sz, const size_t* _steps,
bool autoSteps = false );
void updateContinuityFlag(UMat& m);
void finalizeHdr(UMat& m);
} // namespace cv
namespace cv { namespace ocl {
/*
// Convert OpenCL buffer memory to UMat
*/
void convertFromBuffer(void* cl_mem_buffer, size_t step, int rows, int cols, int type, UMat& dst)
{
int d = 2;
int sizes[] = { rows, cols };
CV_Assert(0 <= d && d <= CV_MAX_DIM);
dst.release();
dst.flags = (type & Mat::TYPE_MASK) | Mat::MAGIC_VAL;
dst.usageFlags = USAGE_DEFAULT;
setSize(dst, d, sizes, 0, true);
dst.offset = 0;
cl_mem memobj = (cl_mem)cl_mem_buffer;
cl_mem_object_type mem_type = 0;
CV_Assert(clGetMemObjectInfo(memobj, CL_MEM_TYPE, sizeof(cl_mem_object_type), &mem_type, 0) == CL_SUCCESS);
CV_Assert(CL_MEM_OBJECT_BUFFER == mem_type);
size_t total = 0;
CV_Assert(clGetMemObjectInfo(memobj, CL_MEM_SIZE, sizeof(size_t), &total, 0) == CL_SUCCESS);
CV_Assert(clRetainMemObject(memobj) == CL_SUCCESS);
CV_Assert((int)step >= cols * CV_ELEM_SIZE(type));
CV_Assert(total >= rows * step);
// attach clBuffer to UMatData
dst.u = new UMatData(getOpenCLAllocator());
dst.u->data = 0;
dst.u->allocatorFlags_ = 0; // not allocated from any OpenCV buffer pool
dst.u->flags = 0;
dst.u->handle = cl_mem_buffer;
dst.u->origdata = 0;
dst.u->prevAllocator = 0;
dst.u->size = total;
finalizeHdr(dst);
dst.addref();
return;
} // convertFromBuffer()
/*
// Convert OpenCL image2d_t memory to UMat
*/
void convertFromImage(void* cl_mem_image, UMat& dst)
{
cl_mem clImage = (cl_mem)cl_mem_image;
cl_mem_object_type mem_type = 0;
CV_Assert(clGetMemObjectInfo(clImage, CL_MEM_TYPE, sizeof(cl_mem_object_type), &mem_type, 0) == CL_SUCCESS);
CV_Assert(CL_MEM_OBJECT_IMAGE2D == mem_type);
cl_image_format fmt = { 0, 0 };
CV_Assert(clGetImageInfo(clImage, CL_IMAGE_FORMAT, sizeof(cl_image_format), &fmt, 0) == CL_SUCCESS);
int depth = CV_8U;
switch (fmt.image_channel_data_type)
{
case CL_UNORM_INT8:
case CL_UNSIGNED_INT8:
depth = CV_8U;
break;
case CL_SNORM_INT8:
case CL_SIGNED_INT8:
depth = CV_8S;
break;
case CL_UNORM_INT16:
case CL_UNSIGNED_INT16:
depth = CV_16U;
break;
case CL_SNORM_INT16:
case CL_SIGNED_INT16:
depth = CV_16S;
break;
case CL_SIGNED_INT32:
depth = CV_32S;
break;
case CL_FLOAT:
depth = CV_32F;
break;
default:
CV_Error(cv::Error::OpenCLApiCallError, "Not supported image_channel_data_type");
}
int type = CV_8UC1;
switch (fmt.image_channel_order)
{
case CL_R:
type = CV_MAKE_TYPE(depth, 1);
break;
case CL_RGBA:
case CL_BGRA:
case CL_ARGB:
type = CV_MAKE_TYPE(depth, 4);
break;
default:
CV_Error(cv::Error::OpenCLApiCallError, "Not supported image_channel_order");
break;
}
size_t step = 0;
CV_Assert(clGetImageInfo(clImage, CL_IMAGE_ROW_PITCH, sizeof(size_t), &step, 0) == CL_SUCCESS);
size_t w = 0;
CV_Assert(clGetImageInfo(clImage, CL_IMAGE_WIDTH, sizeof(size_t), &w, 0) == CL_SUCCESS);
size_t h = 0;
CV_Assert(clGetImageInfo(clImage, CL_IMAGE_HEIGHT, sizeof(size_t), &h, 0) == CL_SUCCESS);
dst.create((int)h, (int)w, type);
cl_mem clBuffer = (cl_mem)dst.handle(ACCESS_READ);
cl_command_queue q = (cl_command_queue)Queue::getDefault().ptr();
size_t offset = 0;
size_t src_origin[3] = { 0, 0, 0 };
size_t region[3] = { w, h, 1 };
CV_Assert(clEnqueueCopyImageToBuffer(q, clImage, clBuffer, src_origin, region, offset, 0, NULL, NULL) == CL_SUCCESS);
CV_Assert(clFinish(q) == CL_SUCCESS);
return;
} // convertFromImage()
///////////////////////////////////////////// Utility functions /////////////////////////////////////////////////
static void getDevices(std::vector<cl_device_id>& devices, cl_platform_id platform)
......
......@@ -46,6 +46,13 @@
namespace cv {
// forward decls, implementation is below in this file
void setSize(UMat& m, int _dims, const int* _sz, const size_t* _steps,
bool autoSteps = false);
void updateContinuityFlag(UMat& m);
void finalizeHdr(UMat& m);
// it should be a prime number for the best hash function
enum { UMAT_NLOCKS = 31 };
static Mutex umatLocks[UMAT_NLOCKS];
......@@ -123,8 +130,8 @@ void swap( UMat& a, UMat& b )
}
static inline void setSize( UMat& m, int _dims, const int* _sz,
const size_t* _steps, bool autoSteps=false )
void setSize( UMat& m, int _dims, const int* _sz,
const size_t* _steps, bool autoSteps )
{
CV_Assert( 0 <= _dims && _dims <= CV_MAX_DIM );
if( m.dims != _dims )
......@@ -176,7 +183,8 @@ static inline void setSize( UMat& m, int _dims, const int* _sz,
}
}
static void updateContinuityFlag(UMat& m)
void updateContinuityFlag(UMat& m)
{
int i, j;
for( i = 0; i < m.dims; i++ )
......@@ -199,7 +207,7 @@ static void updateContinuityFlag(UMat& m)
}
static void finalizeHdr(UMat& m)
void finalizeHdr(UMat& m)
{
updateContinuityFlag(m);
int d = m.dims;
......@@ -207,6 +215,7 @@ static void finalizeHdr(UMat& m)
m.rows = m.cols = -1;
}
UMat Mat::getUMat(int accessFlags, UMatUsageFlags usageFlags) const
{
UMat hdr;
......
......@@ -66,6 +66,8 @@ endif()
add_subdirectory(cpp)
# FIXIT: can't use cvconfig.h in samples: add_subdirectory(gpu)
add_subdirectory(opencl)
if(WIN32)
add_subdirectory(directx)
endif()
......
# cmake 3.1 needed for find_package(OpenCL)
if(CMAKE_VERSION VERSION_LESS "3.1")
message(STATUS "OpenCL samples require CMakes 3.1+")
return()
endif()
set(
OPENCV_OPENCL_SAMPLES_REQUIRED_DEPS
opencv_core
opencv_imgproc
opencv_video
opencv_imgcodecs
opencv_videoio
opencv_highgui)
ocv_check_dependencies(${OPENCV_OPENCL_SAMPLES_REQUIRED_DEPS})
if(BUILD_EXAMPLES AND OCV_DEPENDENCIES_FOUND)
find_package(OpenCL 1.2 REQUIRED)
set(project "opencl")
string(TOUPPER "${project}" project_upper)
project("${project}_samples")
ocv_include_modules_recurse(${OPENCV_OPENCL_SAMPLES_REQUIRED_DEPS})
include_directories(${OpenCL_INCLUDE_DIR})
# ---------------------------------------------
# Define executable targets
# ---------------------------------------------
MACRO(OPENCV_DEFINE_OPENCL_EXAMPLE name srcs)
set(the_target "example_${project}_${name}")
add_executable(${the_target} ${srcs})
ocv_target_link_libraries(
${the_target}
${OPENCV_LINKER_LIBS}
${OPENCV_OPENCL_SAMPLES_REQUIRED_DEPS}
${OpenCL_LIBRARY})
set_target_properties(${the_target} PROPERTIES
OUTPUT_NAME "${project}-example-${name}"
PROJECT_LABEL "(EXAMPLE_${project_upper}) ${name}")
if(ENABLE_SOLUTION_FOLDERS)
set_target_properties(${the_target} PROPERTIES FOLDER "samples//${project}")
endif()
if(WIN32)
if(MSVC AND NOT BUILD_SHARED_LIBS)
set_target_properties(${the_target} PROPERTIES LINK_FLAGS "/NODEFAULTLIB:atlthunk.lib /NODEFAULTLIB:atlsd.lib /DEBUG")
endif()
install(TARGETS ${the_target} RUNTIME DESTINATION "${OPENCV_SAMPLES_BIN_INSTALL_PATH}/${project}" COMPONENT main)
endif()
ENDMACRO()
file(GLOB all_samples RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} *.cpp)
foreach(sample_filename ${all_samples})
get_filename_component(sample ${sample_filename} NAME_WE)
file(GLOB sample_srcs RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${sample}.*)
OPENCV_DEFINE_OPENCL_EXAMPLE(${sample} ${sample_srcs})
endforeach()
endif()
此差异已折叠。
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册