背景：我使用OpenCV从磁盘读取图像，使用CUDA将其传递给GPU，现在，我正在尝试让OpenGL渲染图像。
我这里没有使用GLUT，因为我编译了我的代码并获得了32位Windows来创建一个新窗口，我将在其中渲染图像。现在，当我简单地将flipped.data传递给glTexImage2D()函数时，我翻转了OpenCV图像并获得了OpenGL来很好地渲染图像。但是，当我使用CUDA + OpenGL时，不会渲染相同的图像。
我的实际图像比当前图像大。我正在使用OpenGL像素缓冲对象和OpenGL纹理来渲染图像。利用纹理允许我指定要显示的图像部分。我的灰度图像的尺寸为w1024 x h256，其深度为8位（unsigned char / GL_UNSIGNED_BYTE）。
问题：我无法弄清楚代码中出了什么问题。我尝试仔细遵循CUDA C编程指南，并使用PBO和纹理以及实际输入数据注册/映射CUDA资源。由于我的输入图像数据来自OpenCV，我只是将flipped的数据复制到设备指针dev_inp中。我（正确？）也使用dev_inp将cudaGraphicsResourceGetMappedPointer()映射到CUDA资源。然而，窗口没有任何显示，并保持黑色。没有视口更改，我在glBegin().. glEnd()指定的坐标是正确的，因为它们正确地将flipped的数据映射到纹理。
我在这里错过了其他什么吗？我是否错误地将CUDA资源映射到PBO或设备指针？
OpenGL + CUDA互操作部分：这部分只是我代码中的CUDA + OpenGL互操作。函数DrawOpenGLScene()是从WindProc()方法调用的。
void DrawOpenGLScene()
{
initCUDADevice();
Mat image, flipped;
image = imread("K:/Ultrasound experiment images/PA_160.png", CV_LOAD_IMAGE_GRAYSCALE); // Read the file from disk
if(!image.data) // Check for invalid input
{
cout << "Could not open or find the image" << std::endl ;
}
cv::flip(image, flipped, 0);
imshow("flip", image); // displays output
//cout << "depth: " << flipped.depth() << endl;
// ===================================================================================
// opengl setup
// first, the context was created
// now, clear the window with the rendering context
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
glLoadIdentity();
// ====================================================================================
// generate the pixel buffer object (PBO)
// Generate a buffer ID called a PBO (Pixel Buffer Object)
glGenBuffers(1, &pbo);
// Make this the current UNPACK buffer (OpenGL is state-based)
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, pbo);
// Allocate data for the buffer. 4-channel 8-bit image
glBufferData(GL_PIXEL_UNPACK_BUFFER, sizeof(unsigned char) * flipped.rows * flipped.cols, NULL, GL_DYNAMIC_COPY);
//gpuErrchk(cudaGLRegisterBufferObject( pbo ));
gpuErrchk(cudaGraphicsGLRegisterBuffer(&cuda_resource, pbo, cudaGraphicsMapFlagsNone));
// ====================================================================================
// create the texture object
// enable 2D texturing
glEnable(GL_TEXTURE_2D);
// bind the texture
glGenTextures(1, &tex);
glBindTexture(GL_TEXTURE_2D, tex);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
//glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
// put flipped.data at the end, and it'll work for normal texturing
glTexImage2D(GL_TEXTURE_2D, 0, GL_LUMINANCE, image.cols, image.rows, 0, GL_LUMINANCE, GL_UNSIGNED_BYTE, NULL);
// put tex at the end, and it'll work for normal texturing
glBindTexture(GL_TEXTURE_2D, 0);
// ====================================================================================
// copy data from openCV
unsigned char *dev_inp;
gpuErrchk( cudaMalloc((void**)&dev_inp, sizeof(unsigned char)*flipped.rows*flipped.cols) );
//cudaGLMapBufferObject((void**)dev_inp, pbo);
gpuErrchk( cudaGraphicsMapResources(1, &cuda_resource, 0) );
size_t size; // = sizeof(unsigned char)*flipped.rows*flipped.cols;
gpuErrchk( cudaGraphicsResourceGetMappedPointer((void **)&dev_inp, &size, cuda_resource) );
gpuErrchk( cudaMemcpy(dev_inp, flipped.data, sizeof(unsigned char)*flipped.rows*flipped.cols, cudaMemcpyHostToDevice) );
//cudaGLUnmapBufferObject(pbo);
gpuErrchk( cudaGraphicsUnmapResources(1, &cuda_resource, 0) );
// ====================================================================================
// bind pbo and texture to render data now
glBindBuffer( GL_PIXEL_UNPACK_BUFFER, pbo);
glBindTexture(GL_TEXTURE_2D, tex);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, flipped.cols, flipped.rows, GL_LUMINANCE, GL_UNSIGNED_BYTE, NULL);
glBegin(GL_QUADS);
glTexCoord2f(0.0f, 0.0f); glVertex3f(-1.0f, -1.0f, 1.0f); // Bottom Left Of The Texture and Quad
glTexCoord2f(1.0f, 0.0f); glVertex3f( 1.0f, -1.0f, 1.0f); // Bottom Right Of The Texture and Quad
glTexCoord2f(1.0f, 1.0f); glVertex3f( 1.0f, 1.0f, 1.0f); // Top Right Of The Texture and Quad
glTexCoord2f(0.0f, 1.0f); glVertex3f(-1.0f, 1.0f, 1.0f); // Top Left Of The Texture and Quad
glEnd();
glFlush(); // force rendering to happen
//glBindTexture(GL_TEXTURE_2D, 0);
}
整个代码：
LRESULT CALLBACK WndProc(HWND, UINT, WPARAM, LPARAM);
void DrawOpenGLScene(void);
HGLRC SetUpOpenGLContext(HWND hWnd);
GLuint tex;
GLuint pbo;
struct cudaGraphicsResource *cuda_resource;
int WINAPI WinMain (HINSTANCE hInstance, HINSTANCE hPrevInstance,
LPSTR lpszCmdLine, int nCmdShow)
{
static char szClassName[] = "Myclass";
static char szTitle[]="A Simple Win32 API OpenGL Program";
WNDCLASS wc;
MSG msg;
HWND hWnd;
wc.style = CS_HREDRAW | CS_VREDRAW;
wc.lpfnWndProc = (WNDPROC)WndProc;
wc.cbClsExtra = 0;
wc.cbWndExtra = 0;
wc.hInstance = hInstance;
wc.hIcon = NULL;
wc.hCursor = LoadCursor(NULL, IDC_ARROW);
wc.hbrBackground = (HBRUSH)GetStockObject (BLACK_BRUSH);
wc.lpszMenuName = NULL;
wc.lpszClassName = szClassName;
if (!RegisterClass (&wc))
return 0;
hWnd = CreateWindow(szClassName, szTitle,
WS_OVERLAPPEDWINDOW |
// NEED THESE for OpenGL calls to work!
WS_CLIPCHILDREN | WS_CLIPSIBLINGS,
0, 0, 1024, 256,
NULL, NULL, hInstance, NULL);
ShowWindow(hWnd, nCmdShow);
UpdateWindow( hWnd );
while (GetMessage(&msg, NULL, 0, 0))
{
TranslateMessage( &msg );
DispatchMessage( &msg );
}
return(msg.wParam);
}
LRESULT CALLBACK WndProc( HWND hWnd, UINT msg,
WPARAM wParam, LPARAM lParam )
{
HDC hDC;
static HGLRC hRC; // Note this is STATIC!
PAINTSTRUCT ps;
switch (msg)
{
case WM_CREATE:
// Select a pixel format and create a rendering context
hRC = SetUpOpenGLContext(hWnd);
break;
case WM_PAINT:
// Draw the scene
// Get a DC, make RC current & associate it with this DC
hDC = BeginPaint(hWnd, &ps);
wglMakeCurrent(hDC, hRC);
DrawOpenGLScene(); // Draw
// We're done with the RC, so deselect it
wglMakeCurrent(NULL, NULL);
EndPaint(hWnd, &ps);
break;
case WM_DESTROY:
//cudaGLUnregisterBufferObject(pbo);
cudaGraphicsUnregisterResource(cuda_resource);
// Clean up and terminate
wglDeleteContext(hRC);
PostQuitMessage(0);
break;
default:
return DefWindowProc(hWnd, msg, wParam, lParam);
}
return (0);
}
//*******************************************************
// SetUpOpenGL sets the pixel format and a rendering
// context then returns the RC
//*******************************************************
HGLRC SetUpOpenGLContext(HWND hWnd)
{
static PIXELFORMATDESCRIPTOR pfd = {
sizeof (PIXELFORMATDESCRIPTOR), // strcut size
1, // Version number
PFD_DRAW_TO_WINDOW | // Flags, draw to a window,
PFD_SUPPORT_OPENGL, // use OpenGL
PFD_TYPE_RGBA, // RGBA pixel values
24, // 24-bit color
0, 0, 0, // RGB bits & shift sizes.
0, 0, 0, // Don't care about them
0, 0, // No alpha buffer info
0, 0, 0, 0, 0, // No accumulation buffer
32, // 32-bit depth buffer
0, // No stencil buffer
0, // No auxiliary buffers
PFD_MAIN_PLANE, // Layer type
0, // Reserved (must be 0)
0, // No layer mask
0, // No visible mask
0 // No damage mask
};
int nMyPixelFormatID;
HDC hDC;
HGLRC hRC;
hDC = GetDC(hWnd);
nMyPixelFormatID = ChoosePixelFormat(hDC, &pfd);
SetPixelFormat(hDC, nMyPixelFormatID, &pfd);
hRC = wglCreateContext(hDC);
ReleaseDC(hWnd, hDC);
return hRC;
}
//***********************************************************
// initCUDADevice uses CUDA commands to initiate the CUDA
// enabled graphics card. This is prior to resource mapping,
// and rendering.
//***********************************************************
void initCUDADevice() {
gpuErrchk(cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() ));
}
//********************************************************
// DrawOpenGLScene uses OpenGL commands to draw the scene
// This is where we put the OpenGL drawing commands
//********************************************************
void DrawOpenGLScene()
{
initCUDADevice();
Mat image, flipped;
image = imread("K:/Ultrasound experiment images/PA_160.png", CV_LOAD_IMAGE_GRAYSCALE); // Read the file from disk
if(!image.data) // Check for invalid input
{
cout << "Could not open or find the image" << std::endl ;
}
cv::flip(image, flipped, 0);
imshow("flip", image); // displays output
//cout << "depth: " << flipped.depth() << endl;
// ===================================================================================
// opengl setup
// first, the context was created
// now, clear the window with the rendering context
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
glLoadIdentity();
// ====================================================================================
// generate the pixel buffer object (PBO)
// Generate a buffer ID called a PBO (Pixel Buffer Object)
glGenBuffers(1, &pbo);
// Make this the current UNPACK buffer (OpenGL is state-based)
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, pbo);
// Allocate data for the buffer. 4-channel 8-bit image
glBufferData(GL_PIXEL_UNPACK_BUFFER, sizeof(unsigned char) * flipped.rows * flipped.cols, NULL, GL_DYNAMIC_COPY);
//gpuErrchk(cudaGLRegisterBufferObject( pbo ));
gpuErrchk(cudaGraphicsGLRegisterBuffer(&cuda_resource, pbo, cudaGraphicsMapFlagsNone));
// ====================================================================================
// create the texture object
// enable 2D texturing
glEnable(GL_TEXTURE_2D);
// bind the texture
glGenTextures(1, &tex);
glBindTexture(GL_TEXTURE_2D, tex);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
//glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
// put flipped.data at the end, and it'll work for normal texturing
glTexImage2D(GL_TEXTURE_2D, 0, GL_LUMINANCE, image.cols, image.rows, 0, GL_LUMINANCE, GL_UNSIGNED_BYTE, NULL);
// put tex at the end, and it'll work for normal texturing
glBindTexture(GL_TEXTURE_2D, 0);
// ====================================================================================
// copy data from openCV
unsigned char *dev_inp;
gpuErrchk( cudaMalloc((void**)&dev_inp, sizeof(unsigned char)*flipped.rows*flipped.cols) );
//cudaGLMapBufferObject((void**)dev_inp, pbo);
gpuErrchk( cudaGraphicsMapResources(1, &cuda_resource, 0) );
size_t size; // = sizeof(unsigned char)*flipped.rows*flipped.cols;
gpuErrchk( cudaGraphicsResourceGetMappedPointer((void **)&dev_inp, &size, cuda_resource) );
gpuErrchk( cudaMemcpy(dev_inp, flipped.data, sizeof(unsigned char)*flipped.rows*flipped.cols, cudaMemcpyHostToDevice) );
//cudaGLUnmapBufferObject(pbo);
gpuErrchk( cudaGraphicsUnmapResources(1, &cuda_resource, 0) );
// ====================================================================================
// bind pbo and texture to render data now
glBindBuffer( GL_PIXEL_UNPACK_BUFFER, pbo);
glBindTexture(GL_TEXTURE_2D, tex);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, flipped.cols, flipped.rows, GL_LUMINANCE, GL_UNSIGNED_BYTE, NULL);
glBegin(GL_QUADS);
glTexCoord2f(0.0f, 0.0f); glVertex3f(-1.0f, -1.0f, 1.0f); // Bottom Left Of The Texture and Quad
glTexCoord2f(1.0f, 0.0f); glVertex3f( 1.0f, -1.0f, 1.0f); // Bottom Right Of The Texture and Quad
glTexCoord2f(1.0f, 1.0f); glVertex3f( 1.0f, 1.0f, 1.0f); // Top Right Of The Texture and Quad
glTexCoord2f(0.0f, 1.0f); glVertex3f(-1.0f, 1.0f, 1.0f); // Top Left Of The Texture and Quad
glEnd();
glFlush(); // force rendering to happen
//glBindTexture(GL_TEXTURE_2D, 0);
}
答案 0 :(得分：0)
如果其他人遇到同样的问题，这个线程可以帮助他们。
我通过更改DrawOpenGLScene()中的几个电话解决了我的问题。
事实证明cudaGraphicsResourceGetMappedPointer()返回指向OpenGL PBO并从OpenGL PBO派生的指针，并将该指针放在dev_inp中。它会根据之前建立的size = sizeof(unsigned char) * flipped.rows * flipped.cols和dev_inp调用为glBufferData()内部分配cudaGraphicsGLRegisterBuffer()个内存。
完成此操作后，我之前使用cudaMalloc()分配的内存现在不再存在，因为调用cudaGraphicsResourceGetMappedPointer()会将指针放在dev_inp中。删除cudaMalloc()和cudaFree()允许程序按原计划运行。
为了释放内存，应该解除分配PBO，因为OpenGL是内存的“所有者”，而CUDA只是共享对OpenGL拥有的内存的访问权限。
修改后的DrawOpenGLScene()例程粘贴在下面：
#define GET_PROC_ADDRESS( str ) wglGetProcAddress( str )
PFNGLBINDBUFFERARBPROC glBindBuffer = NULL;
PFNGLDELETEBUFFERSARBPROC glDeleteBuffers = NULL;
PFNGLGENBUFFERSARBPROC glGenBuffers = NULL;
PFNGLBUFFERDATAARBPROC glBufferData = NULL;
void initCUDADevice() {
gpuErrchk(cudaGLSetGLDevice( cutGetMaxGflopsDeviceId() ));
}
//********************************************************
// DrawOpenGLScene uses OpenGL commands to draw the scene
// This is where we put the OpenGL drawing commands
//********************************************************
void DrawOpenGLScene()
{
// Clear Color and Depth Buffers
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
// Reset transformations
glLoadIdentity();
// ====================================================================================
// initiate GPU by setting it correctly
initCUDADevice();
// ====================================================================================
// read the image that needs to be textured
Mat image, flipped;
image = imread("K:/OCT experiment images/PA_175.png", CV_LOAD_IMAGE_GRAYSCALE); // Read the file from disk
if(!image.data) // Check for invalid input
{
cout << "Could not open or find the image" << std::endl ;
}
cv::flip(image, flipped, 0);
imshow("OpenCV - image", image); // displays output
// ====================================================================================
// allocate the PBO, texture, and CUDA resource
glBindBuffer = (PFNGLBINDBUFFERARBPROC)GET_PROC_ADDRESS("glBindBuffer");
glDeleteBuffers = (PFNGLDELETEBUFFERSARBPROC)GET_PROC_ADDRESS("glDeleteBuffers");
glGenBuffers = (PFNGLGENBUFFERSARBPROC)GET_PROC_ADDRESS("glGenBuffers");
glBufferData = (PFNGLBUFFERDATAARBPROC)GET_PROC_ADDRESS("glBufferData");
// ====================================================================================
// generate the pixel buffer object (PBO)
// Generate a buffer ID called a PBO (Pixel Buffer Object)
glGenBuffers(1, &pbo);
// Make this the current UNPACK buffer (OpenGL is state-based)
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, pbo);
// Allocate data for the buffer. 4-channel 8-bit image
glBufferData(GL_PIXEL_UNPACK_BUFFER, sizeof(unsigned char) * flipped.rows * flipped.cols, NULL, GL_STREAM_DRAW);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
gpuErrchk(cudaGraphicsGLRegisterBuffer(&cuda_resource, pbo, cudaGraphicsMapFlagsNone));
// ====================================================================================
// create the texture object
// enable 2D texturing
glEnable(GL_TEXTURE_2D);
// generate and bind the texture
glGenTextures(1, &tex);
glBindTexture(GL_TEXTURE_2D, tex);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
// put flipped.data at the end for cpu rendering
glTexImage2D(GL_TEXTURE_2D, 0, GL_LUMINANCE, image.cols, image.rows, 0, GL_LUMINANCE, GL_UNSIGNED_BYTE, 0 );
// put tex at the end for cpu rendering
glBindTexture(GL_TEXTURE_2D, 0);
// ====================================================================================
// copy OpenCV flipped image data into the device pointer
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
unsigned char *dev_inp;
//gpuErrchk( cudaMalloc((void**)&dev_inp, sizeof(unsigned char)*flipped.rows*flipped.cols) );
gpuErrchk( cudaGraphicsMapResources(1, &cuda_resource, 0) );
size_t size;
gpuErrchk( cudaGraphicsResourceGetMappedPointer((void **)&dev_inp, &size, cuda_resource) );
gpuErrchk( cudaMemcpy(dev_inp, flipped.data, sizeof(unsigned char)*flipped.rows*flipped.cols, cudaMemcpyHostToDevice) );
gpuErrchk( cudaGraphicsUnmapResources(1, &cuda_resource, 0) );
// ====================================================================================
// bind pbo and texture to render data now
glBindBuffer( GL_PIXEL_UNPACK_BUFFER, pbo);
//
glBindTexture(GL_TEXTURE_2D, tex);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, flipped.cols, flipped.rows, GL_LUMINANCE, GL_UNSIGNED_BYTE, NULL);
gpuErrchk( cudaGraphicsUnregisterResource(cuda_resource));
gpuErrchk( cudaThreadSynchronize());
//gpuErrchk(cudaFree(dev_inp));
// ====================================================================================
// map the texture coords to the vertex coords
glBegin(GL_QUADS);
// Front Face
glTexCoord2f(0.0f, 0.0f); glVertex3f(-1.0f, -1.0f, 1.0f); // Bottom Left Of The Texture and Quad
glTexCoord2f(1.0f, 0.0f); glVertex3f( 1.0f, -1.0f, 1.0f); // Bottom Right Of The Texture and Quad
glTexCoord2f(1.0f, 1.0f); glVertex3f( 1.0f, 1.0f, 1.0f); // Top Right Of The Texture and Quad
glTexCoord2f(0.0f, 1.0f); glVertex3f(-1.0f, 1.0f, 1.0f); // Top Left Of The Texture and Quad
glEnd();
glFlush(); // force rendering
glDisable(GL_TEXTURE_2D);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
glDeleteBuffers(1, &pbo);