added Stream support to ogl::Buffer

modules/core/include/opencv2/core/opengl.hpp

@@ -40,8 +40,12 @@
 //
 //M*/
 
-#ifndef __OPENCV_OPENGL_INTEROP_HPP__
-#define __OPENCV_OPENGL_INTEROP_HPP__
+#ifndef __OPENCV_CORE_OPENGL_HPP__
+#define __OPENCV_CORE_OPENGL_HPP__
+
+#ifndef __cplusplus
+#  error opengl.hpp header must be compiled as C++
+#endif
 
 #include "opencv2/core.hpp"
 
@@ -84,7 +88,7 @@ public:
 
     //! create buffer
     void create(int arows, int acols, int atype, Target target = ARRAY_BUFFER, bool autoRelease = false);
-    void create(Size asize, int atype, Target target = ARRAY_BUFFER, bool autoRelease = false) { create(asize.height, asize.width, atype, target, autoRelease); }
+    void create(Size asize, int atype, Target target = ARRAY_BUFFER, bool autoRelease = false);
 
     //! release memory and delete buffer object
     void release();
@@ -92,11 +96,15 @@ public:
     //! set auto release mode (if true, release will be called in object's destructor)
     void setAutoRelease(bool flag);
 
-    //! copy from host/device memory
+    //! copy from host/device memory (blocking)
     void copyFrom(InputArray arr, Target target = ARRAY_BUFFER, bool autoRelease = false);
+    //! copy from device memory (non blocking)
+    void copyFrom(InputArray arr, gpu::Stream& stream, Target target = ARRAY_BUFFER, bool autoRelease = false);
 
-    //! copy to host/device memory
-    void copyTo(OutputArray arr, Target target = ARRAY_BUFFER, bool autoRelease = false) const;
+    //! copy to host/device memory (blocking)
+    void copyTo(OutputArray arr) const;
+    //! copy to device memory (non blocking)
+    void copyTo(OutputArray arr, gpu::Stream& stream) const;
 
     //! create copy of current buffer
     Buffer clone(Target target = ARRAY_BUFFER, bool autoRelease = false) const;
@@ -111,21 +119,26 @@ public:
     Mat mapHost(Access access);
     void unmapHost();
 
-    //! map to device memory
+    //! map to device memory (blocking)
     gpu::GpuMat mapDevice();
     void unmapDevice();
 
-    int rows() const { return rows_; }
-    int cols() const { return cols_; }
-    Size size() const { return Size(cols_, rows_); }
-    bool empty() const { return rows_ == 0 || cols_ == 0; }
+    //! map to device memory (non blocking)
+    gpu::GpuMat mapDevice(gpu::Stream& stream);
+    void unmapDevice(gpu::Stream& stream);
 
-    int type() const { return type_; }
-    int depth() const { return CV_MAT_DEPTH(type_); }
-    int channels() const { return CV_MAT_CN(type_); }
-    int elemSize() const { return CV_ELEM_SIZE(type_); }
-    int elemSize1() const { return CV_ELEM_SIZE1(type_); }
+    int rows() const;
+    int cols() const;
+    Size size() const;
+    bool empty() const;
 
+    int type() const;
+    int depth() const;
+    int channels() const;
+    int elemSize() const;
+    int elemSize1() const;
+
+    //! get OpenGL opject id
     unsigned int bufId() const;
 
     class Impl;
@@ -165,7 +178,7 @@ public:
 
     //! create texture
     void create(int arows, int acols, Format aformat, bool autoRelease = false);
-    void create(Size asize, Format aformat, bool autoRelease = false) { create(asize.height, asize.width, aformat, autoRelease); }
+    void create(Size asize, Format aformat, bool autoRelease = false);
 
     //! release memory and delete texture object
     void release();
@@ -182,13 +195,14 @@ public:
     //! bind texture to current active texture unit for GL_TEXTURE_2D target
     void bind() const;
 
-    int rows() const { return rows_; }
-    int cols() const { return cols_; }
-    Size size() const { return Size(cols_, rows_); }
-    bool empty() const { return rows_ == 0 || cols_ == 0; }
+    int rows() const;
+    int cols() const;
+    Size size() const;
+    bool empty() const;
 
-    Format format() const { return format_; }
+    Format format() const;
 
+    //! get OpenGL opject id
     unsigned int texId() const;
 
     class Impl;
@@ -224,8 +238,8 @@ public:
 
     void bind() const;
 
-    int size() const { return size_; }
-    bool empty() const { return size_ == 0; }
+    int size() const;
+    bool empty() const;
 
 private:
     int size_;
@@ -260,14 +274,14 @@ enum {
 CV_EXPORTS void render(const Arrays& arr, int mode = POINTS, Scalar color = Scalar::all(255));
 CV_EXPORTS void render(const Arrays& arr, InputArray indices, int mode = POINTS, Scalar color = Scalar::all(255));
 
-}} // namespace cv::gl
+}} // namespace cv::ogl
 
 namespace cv { namespace gpu {
 
 //! set a CUDA device to use OpenGL interoperability
 CV_EXPORTS void setGlDevice(int device = 0);
 
-}} // cv::gpu
+}}
 
 namespace cv {
 
@@ -276,4 +290,149 @@ template <> CV_EXPORTS void Ptr<cv::ogl::Texture2D::Impl>::delete_obj();
 
 }
 
-#endif // __OPENCV_OPENGL_INTEROP_HPP__
+////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////
+
+inline
+cv::ogl::Buffer::Buffer(int arows, int acols, int atype, Target target, bool autoRelease) : rows_(0), cols_(0), type_(0)
+{
+    create(arows, acols, atype, target, autoRelease);
+}
+
+inline
+cv::ogl::Buffer::Buffer(Size asize, int atype, Target target, bool autoRelease) : rows_(0), cols_(0), type_(0)
+{
+    create(asize, atype, target, autoRelease);
+}
+
+inline
+void cv::ogl::Buffer::create(Size asize, int atype, Target target, bool autoRelease)
+{
+    create(asize.height, asize.width, atype, target, autoRelease);
+}
+
+inline
+int cv::ogl::Buffer::rows() const
+{
+    return rows_;
+}
+
+inline
+int cv::ogl::Buffer::cols() const
+{
+    return cols_;
+}
+
+inline
+cv::Size cv::ogl::Buffer::size() const
+{
+    return Size(cols_, rows_);
+}
+
+inline
+bool cv::ogl::Buffer::empty() const
+{
+    return rows_ == 0 || cols_ == 0;
+}
+
+inline
+int cv::ogl::Buffer::type() const
+{
+    return type_;
+}
+
+inline
+int cv::ogl::Buffer::depth() const
+{
+    return CV_MAT_DEPTH(type_);
+}
+
+inline
+int cv::ogl::Buffer::channels() const
+{
+    return CV_MAT_CN(type_);
+}
+
+inline
+int cv::ogl::Buffer::elemSize() const
+{
+    return CV_ELEM_SIZE(type_);
+}
+
+inline
+int cv::ogl::Buffer::elemSize1() const
+{
+    return CV_ELEM_SIZE1(type_);
+}
+
+///////
+
+inline
+cv::ogl::Texture2D::Texture2D(int arows, int acols, Format aformat, bool autoRelease) : rows_(0), cols_(0), format_(NONE)
+{
+    create(arows, acols, aformat, autoRelease);
+}
+
+inline
+cv::ogl::Texture2D::Texture2D(Size asize, Format aformat, bool autoRelease) : rows_(0), cols_(0), format_(NONE)
+{
+    create(asize, aformat, autoRelease);
+}
+
+inline
+void cv::ogl::Texture2D::create(Size asize, Format aformat, bool autoRelease)
+{
+    create(asize.height, asize.width, aformat, autoRelease);
+}
+
+inline
+int cv::ogl::Texture2D::rows() const
+{
+    return rows_;
+}
+
+inline
+int cv::ogl::Texture2D::cols() const
+{
+    return cols_;
+}
+
+inline
+cv::Size cv::ogl::Texture2D::size() const
+{
+    return Size(cols_, rows_);
+}
+
+inline
+bool cv::ogl::Texture2D::empty() const
+{
+    return rows_ == 0 || cols_ == 0;
+}
+
+inline
+cv::ogl::Texture2D::Format cv::ogl::Texture2D::format() const
+{
+    return format_;
+}
+
+///////
+
+inline
+cv::ogl::Arrays::Arrays() : size_(0)
+{
+}
+
+inline
+int cv::ogl::Arrays::size() const
+{
+    return size_;
+}
+
+inline
+bool cv::ogl::Arrays::empty() const
+{
+    return size_ == 0;
+}
+
+#endif /* __OPENCV_CORE_OPENGL_HPP__ */

modules/core/src/matrix.cpp

@@ -41,8 +41,6 @@
 //M*/
 
 #include "precomp.hpp"
-#include "opencv2/core/gpu.hpp"
-#include "opencv2/core/opengl.hpp"
 
 /****************************************************************************************\
 *                           [scaled] Identity matrix initialization                      *

modules/core/src/opengl.cpp

@@ -55,19 +55,19 @@ using namespace cv::gpu;
 namespace
 {
     #ifndef HAVE_OPENGL
-        void throw_no_ogl() { CV_Error(CV_OpenGlNotSupported, "The library is compiled without OpenGL support"); }
+        inline void throw_no_ogl() { CV_Error(cv::Error::OpenGlNotSupported, "The library is compiled without OpenGL support"); }
     #else
-        void throw_no_ogl() { CV_Error(CV_OpenGlApiCallError, "OpenGL context doesn't exist"); }
+        inline void throw_no_ogl() { CV_Error(cv::Error::OpenGlApiCallError, "OpenGL context doesn't exist"); }
     #endif
 
-bool checkError(const char* file, const int line, const char* func = 0)
-{
-#ifndef HAVE_OPENGL
+    bool checkError(const char* file, const int line, const char* func = 0)
+    {
+    #ifndef HAVE_OPENGL
         (void) file;
         (void) line;
         (void) func;
         return true;
-#else
+    #else
         GLenum err = gl::GetError();
 
         if (err != gl::NO_ERROR_)
@@ -102,15 +102,14 @@ bool checkError(const char* file, const int line, const char* func = 0)
         }
 
         return true;
-#endif
-}
+    #endif
+    }
 
-#if defined(__GNUC__)
+    #if defined(__GNUC__)
         #define CV_CheckGlError() CV_DbgAssert( (checkError(__FILE__, __LINE__, __func__)) )
-#else
+    #else
         #define CV_CheckGlError() CV_DbgAssert( (checkError(__FILE__, __LINE__)) )
-#endif
-
+    #endif
 } // namespace
 
 #ifdef HAVE_OPENGL
@@ -129,7 +128,7 @@ void cv::gpu::setGlDevice(int device)
     (void) device;
     throw_no_ogl();
 #else
-    #if !defined(HAVE_CUDA) || defined(CUDA_DISABLER)
+    #ifndef HAVE_CUDA
         (void) device;
         throw_no_cuda();
     #else
@@ -141,7 +140,7 @@ void cv::gpu::setGlDevice(int device)
 ////////////////////////////////////////////////////////////////////////
 // CudaResource
 
-#if defined(HAVE_OPENGL) && defined(HAVE_CUDA) && !defined(CUDA_DISABLER)
+#if defined(HAVE_OPENGL) && defined(HAVE_CUDA)
 
 namespace
 {
@@ -353,12 +352,13 @@ const Ptr<cv::ogl::Buffer::Impl>& cv::ogl::Buffer::Impl::empty()
     return p;
 }
 
-cv::ogl::Buffer::Impl::Impl() : bufId_(0), autoRelease_(true)
+cv::ogl::Buffer::Impl::Impl() : bufId_(0), autoRelease_(false)
 {
 }
 
 cv::ogl::Buffer::Impl::Impl(GLuint abufId, bool autoRelease) : bufId_(abufId), autoRelease_(autoRelease)
 {
+    CV_Assert( gl::IsBuffer(abufId) == gl::TRUE_ );
 }
 
 cv::ogl::Buffer::Impl::Impl(GLsizeiptr size, const GLvoid* data, GLenum target, bool autoRelease) : bufId_(0), autoRelease_(autoRelease)
@@ -437,29 +437,31 @@ void cv::ogl::Buffer::Impl::unmapHost()
 }
 
 #ifdef HAVE_CUDA
-    void cv::ogl::Buffer::Impl::copyFrom(const void* src, size_t spitch, size_t width, size_t height, cudaStream_t stream)
-    {
+
+void cv::ogl::Buffer::Impl::copyFrom(const void* src, size_t spitch, size_t width, size_t height, cudaStream_t stream)
+{
     cudaResource_.registerBuffer(bufId_);
     cudaResource_.copyFrom(src, spitch, width, height, stream);
-    }
+}
 
-    void cv::ogl::Buffer::Impl::copyTo(void* dst, size_t dpitch, size_t width, size_t height, cudaStream_t stream) const
-    {
+void cv::ogl::Buffer::Impl::copyTo(void* dst, size_t dpitch, size_t width, size_t height, cudaStream_t stream) const
+{
     cudaResource_.registerBuffer(bufId_);
     cudaResource_.copyTo(dst, dpitch, width, height, stream);
-    }
+}
 
-    void* cv::ogl::Buffer::Impl::mapDevice(cudaStream_t stream)
-    {
+void* cv::ogl::Buffer::Impl::mapDevice(cudaStream_t stream)
+{
     cudaResource_.registerBuffer(bufId_);
     return cudaResource_.map(stream);
-    }
+}
 
-    void cv::ogl::Buffer::Impl::unmapDevice(cudaStream_t stream)
-    {
+void cv::ogl::Buffer::Impl::unmapDevice(cudaStream_t stream)
+{
     cudaResource_.unmap(stream);
-    }
-#endif
+}
+
+#endif // HAVE_CUDA
 
 #endif // HAVE_OPENGL
 
@@ -505,16 +507,6 @@ cv::ogl::Buffer::Buffer(Size asize, int atype, unsigned int abufId, bool autoRel
 #endif
 }
 
-cv::ogl::Buffer::Buffer(int arows, int acols, int atype, Target target, bool autoRelease) : rows_(0), cols_(0), type_(0)
-{
-    create(arows, acols, atype, target, autoRelease);
-}
-
-cv::ogl::Buffer::Buffer(Size asize, int atype, Target target, bool autoRelease) : rows_(0), cols_(0), type_(0)
-{
-    create(asize, atype, target, autoRelease);
-}
-
 cv::ogl::Buffer::Buffer(InputArray arr, Target target, bool autoRelease) : rows_(0), cols_(0), type_(0)
 {
 #ifndef HAVE_OPENGL
@@ -528,16 +520,9 @@ cv::ogl::Buffer::Buffer(InputArray arr, Target target, bool autoRelease) : rows_
     switch (kind)
     {
     case _InputArray::OPENGL_BUFFER:
-        {
-            copyFrom(arr, target, autoRelease);
-            break;
-        }
-
     case _InputArray::GPU_MAT:
-        {
         copyFrom(arr, target, autoRelease);
         break;
-        }
 
     default:
         {
@@ -622,7 +607,7 @@ void cv::ogl::Buffer::copyFrom(InputArray arr, Target target, bool autoRelease)
 
     case _InputArray::GPU_MAT:
         {
-            #if !defined HAVE_CUDA || defined(CUDA_DISABLER)
+            #ifndef HAVE_CUDA
                 throw_no_cuda();
             #else
                 GpuMat dmat = arr.getGpuMat();
@@ -642,13 +627,36 @@ void cv::ogl::Buffer::copyFrom(InputArray arr, Target target, bool autoRelease)
 #endif
 }
 
-void cv::ogl::Buffer::copyTo(OutputArray arr, Target target, bool autoRelease) const
+void cv::ogl::Buffer::copyFrom(InputArray arr, gpu::Stream& stream, Target target, bool autoRelease)
 {
 #ifndef HAVE_OPENGL
     (void) arr;
+    (void) stream;
     (void) target;
     (void) autoRelease;
     throw_no_ogl();
+#else
+    #ifndef HAVE_CUDA
+        (void) arr;
+        (void) stream;
+        (void) target;
+        (void) autoRelease;
+        throw_no_cuda();
+    #else
+        GpuMat dmat = arr.getGpuMat();
+
+        create(dmat.size(), dmat.type(), target, autoRelease);
+
+        impl_->copyFrom(dmat.data, dmat.step, dmat.cols * dmat.elemSize(), dmat.rows, gpu::StreamAccessor::getStream(stream));
+    #endif
+#endif
+}
+
+void cv::ogl::Buffer::copyTo(OutputArray arr) const
+{
+#ifndef HAVE_OPENGL
+    (void) arr;
+    throw_no_ogl();
 #else
     const int kind = arr.kind();
 
@@ -656,13 +664,13 @@ void cv::ogl::Buffer::copyTo(OutputArray arr, Target target, bool autoRelease) c
     {
     case _InputArray::OPENGL_BUFFER:
         {
-            arr.getOGlBufferRef().copyFrom(*this, target, autoRelease);
+            arr.getOGlBufferRef().copyFrom(*this);
             break;
         }
 
     case _InputArray::GPU_MAT:
         {
-            #if !defined HAVE_CUDA || defined(CUDA_DISABLER)
+            #ifndef HAVE_CUDA
                 throw_no_cuda();
             #else
                 GpuMat& dmat = arr.getGpuMatRef();
@@ -684,6 +692,25 @@ void cv::ogl::Buffer::copyTo(OutputArray arr, Target target, bool autoRelease) c
 #endif
 }
 
+void cv::ogl::Buffer::copyTo(OutputArray arr, gpu::Stream& stream) const
+{
+#ifndef HAVE_OPENGL
+    (void) arr;
+    (void) stream;
+    throw_no_ogl();
+#else
+    #ifndef HAVE_CUDA
+        (void) arr;
+        (void) stream;
+        throw_no_cuda();
+    #else
+        arr.create(rows_, cols_, type_);
+        GpuMat dmat = arr.getGpuMat();
+        impl_->copyTo(dmat.data, dmat.step, dmat.cols * dmat.elemSize(), dmat.rows, gpu::StreamAccessor::getStream(stream));
+    #endif
+#endif
+}
+
 cv::ogl::Buffer cv::ogl::Buffer::clone(Target target, bool autoRelease) const
 {
 #ifndef HAVE_OPENGL
@@ -745,7 +772,7 @@ GpuMat cv::ogl::Buffer::mapDevice()
     throw_no_ogl();
     return GpuMat();
 #else
-    #if !defined HAVE_CUDA || defined(CUDA_DISABLER)
+    #ifndef HAVE_CUDA
         throw_no_cuda();
         return GpuMat();
     #else
@@ -759,7 +786,7 @@ void cv::ogl::Buffer::unmapDevice()
 #ifndef HAVE_OPENGL
     throw_no_ogl();
 #else
-    #if !defined HAVE_CUDA || defined(CUDA_DISABLER)
+    #ifndef HAVE_CUDA
         throw_no_cuda();
     #else
         impl_->unmapDevice();
@@ -767,6 +794,38 @@ void cv::ogl::Buffer::unmapDevice()
 #endif
 }
 
+gpu::GpuMat cv::ogl::Buffer::mapDevice(gpu::Stream& stream)
+{
+#ifndef HAVE_OPENGL
+    (void) stream;
+    throw_no_ogl();
+    return GpuMat();
+#else
+    #ifndef HAVE_CUDA
+        (void) stream;
+        throw_no_cuda();
+        return GpuMat();
+    #else
+        return GpuMat(rows_, cols_, type_, impl_->mapDevice(gpu::StreamAccessor::getStream(stream)));
+    #endif
+#endif
+}
+
+void cv::ogl::Buffer::unmapDevice(gpu::Stream& stream)
+{
+#ifndef HAVE_OPENGL
+    (void) stream;
+    throw_no_ogl();
+#else
+    #ifndef HAVE_CUDA
+        (void) stream;
+        throw_no_cuda();
+    #else
+        impl_->unmapDevice(gpu::StreamAccessor::getStream(stream));
+    #endif
+#endif
+}
+
 unsigned int cv::ogl::Buffer::bufId() const
 {
 #ifndef HAVE_OPENGL
@@ -824,12 +883,13 @@ const Ptr<cv::ogl::Texture2D::Impl> cv::ogl::Texture2D::Impl::empty()
     return p;
 }
 
-cv::ogl::Texture2D::Impl::Impl() : texId_(0), autoRelease_(true)
+cv::ogl::Texture2D::Impl::Impl() : texId_(0), autoRelease_(false)
 {
 }
 
 cv::ogl::Texture2D::Impl::Impl(GLuint atexId, bool autoRelease) : texId_(atexId), autoRelease_(autoRelease)
 {
+    CV_Assert( gl::IsTexture(atexId) == gl::TRUE_ );
 }
 
 cv::ogl::Texture2D::Impl::Impl(GLint internalFormat, GLsizei width, GLsizei height, GLenum format, GLenum type, const GLvoid* pixels, bool autoRelease) : texId_(0), autoRelease_(autoRelease)
@@ -935,16 +995,6 @@ cv::ogl::Texture2D::Texture2D(Size asize, Format aformat, unsigned int atexId, b
 #endif
 }
 
-cv::ogl::Texture2D::Texture2D(int arows, int acols, Format aformat, bool autoRelease) : rows_(0), cols_(0), format_(NONE)
-{
-    create(arows, acols, aformat, autoRelease);
-}
-
-cv::ogl::Texture2D::Texture2D(Size asize, Format aformat, bool autoRelease) : rows_(0), cols_(0), format_(NONE)
-{
-    create(asize, aformat, autoRelease);
-}
-
 cv::ogl::Texture2D::Texture2D(InputArray arr, bool autoRelease) : rows_(0), cols_(0), format_(NONE)
 {
 #ifndef HAVE_OPENGL
@@ -985,7 +1035,7 @@ cv::ogl::Texture2D::Texture2D(InputArray arr, bool autoRelease) : rows_(0), cols
 
     case _InputArray::GPU_MAT:
         {
-            #if !defined HAVE_CUDA || defined(CUDA_DISABLER)
+            #ifndef HAVE_CUDA
                 throw_no_cuda();
             #else
                 GpuMat dmat = arr.getGpuMat();
@@ -1098,7 +1148,7 @@ void cv::ogl::Texture2D::copyFrom(InputArray arr, bool autoRelease)
 
     case _InputArray::GPU_MAT:
         {
-            #if !defined HAVE_CUDA || defined(CUDA_DISABLER)
+            #ifndef HAVE_CUDA
                 throw_no_cuda();
             #else
                 GpuMat dmat = arr.getGpuMat();
@@ -1149,7 +1199,7 @@ void cv::ogl::Texture2D::copyTo(OutputArray arr, int ddepth, bool autoRelease) c
 
     case _InputArray::GPU_MAT:
         {
-            #if !defined HAVE_CUDA || defined(CUDA_DISABLER)
+            #ifndef HAVE_CUDA
                 throw_no_cuda();
             #else
                 ogl::Buffer buf(rows_, cols_, CV_MAKE_TYPE(ddepth, cn), ogl::Buffer::PIXEL_PACK_BUFFER);
@@ -1201,10 +1251,6 @@ template <> void cv::Ptr<cv::ogl::Texture2D::Impl>::delete_obj()
 ////////////////////////////////////////////////////////////////////////
 // ogl::Arrays
 
-cv::ogl::Arrays::Arrays() : size_(0)
-{
-}
-
 void cv::ogl::Arrays::setVertexArray(InputArray vertex)
 {
     const int cn = vertex.channels();

modules/gpu/test/test_opengl.cpp

@@ -198,7 +198,8 @@ GPU_TEST_P(Buffer, CopyToBuffer)
     cv::ogl::Buffer buf(gold, cv::ogl::Buffer::ARRAY_BUFFER, true);
 
     cv::ogl::Buffer dst;
-    buf.copyTo(dst, cv::ogl::Buffer::ARRAY_BUFFER, true);
+    buf.copyTo(dst);
+    dst.setAutoRelease(true);
 
     EXPECT_NE(buf.bufId(), dst.bufId());