Accessing GpuMat on another GPU

See also questions close to this topic

• Const struct initialization with custom elements in C++

  I am porting some C code to C++ and I am trying to initialize a struct with some values. I want the struct to be stored in flash (const) and not in RAM, and its values are typedef'd elements.

  Originally, I had it like this:

  typedef struct
  {
      typeA_t elementA;
      typeB_t elementB;
      uint8_t elementC;
  } structTypeA_t;
  

  And to instantiate them in flash, I simply did the following:

  const structTypeA_t sA = {
      .elementA = ONE,
      .elementB = TWO,
      .elementC = 3
  };
  

  I know that this type of initializing is not allowed in C++. How can I achieve it in C++?
• Snake game bugged, weird output

  I'm a beginner programmer so PLEASE bear with me. I have a problem trying to make a snake game in the c++ console debugger. To avoid blinking i use a double buffer and then output the characters in said buffer to a console screen buffer. Problem is.. the output is weird and i can't figure out the problem. Please help (I know this is gonna get so much hate but never hurts to ask people that know better than you right?)

  I've already setting break points everywhere (I'm in visual studio) but i still can't figure out the problem.

  void draw(HANDLE hconsole, Snake& pl, wchar_t* buf, COORD dwSize) {
  
      int xpos = pl.getxpos() + 1;
      int ypos = pl.getypos() + 1;
      DWORD dwBytesWritten = 0;
  
      for (int i = 0; i < nFieldHeight; ++i) {
          for (int j = 0; j < nFieldWidth; ++j) {
              if (i == 0 || i == nFieldHeight - 1) buf[i + j] = L'#';
              else if (j == 1 || i == nFieldWidth - 1) buf[i + j] = L'#';
              if (i == ypos && j == xpos) {
                  for (int k = 0; k < pl.getlen(); ++k)
                      buf[i + j - k] = L'*';
              }
              else buf[i + j] = L' ';
          }
          buf[nFieldWidth * i] = L'\n';
      }
  
      WriteConsoleOutputCharacter(hconsole, buf, dwSize.X * dwSize.Y, { 0, 0 }, &dwBytesWritten);
  
      clear(buf, dwSize);
  }
  

  There should be a rectangle of # characters and the snake in the middle, but i get like [?] characters even with an L prefix before any of the characters i write to the buffer.
• How to block message killfocus for other window?

  Can you help me how i can block message WM_KILLFOCUS for other window? I would like deceive other window, that he have always focus.
• Image Detection & Recognition by drawing a specific zone like snipping tool in real-time screen capture with OpenCV?

  I'm kinda new to OpenCV (Python), and after looking for many hours I don't seem to find what I'm looking for in terms of informations on how to do a specific task I want to implement in my project.

  On Windows when you use the snipping tool to take a screenshot, it makes like an overlay zone that you control with your mouse, then it takes the screenshot of the zone you defined and displays it onto another window.

  1. Detection part: I need to do the same thing but this time on real-time screen capture from my computer (define a zone to capture with my mouse), and I also need to display this isolated zone onto another window like the snipping tool program, but this time on real-time.

  2. Recognition part: Then I need the isolated box zone to be compared on real-time with a set of pre defined images and trying to recognize to which image that corresponds, to finally tell and put a label "this is a cat, dog, bike, car ..." just under the isolated zone on the second window.

  To limit errors, the screen captured & thus the isolated zone will always show known things (if we use the previous example above with (cat, dog, bike, car) we won't show birds or planes for ex, only these 4 categories kind of images.

  And to make it simple for anyone to use it, I've thought about a small GUI with a simple button "set new zone", once clicked, you make the zone with your mouse, then the 2nd window pops up with the isolated zone & the description label

  I already managed to get my screen captured on real time with OpenCV and it's working fine but for the rest, I'm pretty much stuck right now on my problem so any help, ideas and suggestions would be really apreciated

  thanks
• How to add an image over another image using x,y coordinates?

  I want to add image 'abc.jpg' on xyz.jpg using openCV and python. I have got the coordinates x,y on which I have to add the image and also resized my 'abc.jpg' so that it will fit on the image. Now how can I add it?
• Want to translate this Matlab code into OpenCV code

  I want to translate some Matlab codes into OpenCV code. Which functions I need to use.

  For element-wise multiplication, I don't know which function to use. For example,

  cv::Mat A; cv::Mat B; cv::Mat C;
  C = A + B;
  //or
  cv::add(A, B, C);
  
  //Matlab Code 
  //G and b are constant 
  //Rb, CRb and Rb_final are the images
  
  Rb = G(CRb.Rb+b); min3 = min(min(Rb)); // minimun in the Rb image max3 = max(max(Rb)); // maximum in the Rb image
  
  Rb_final = uint8(255*(Rb-min3)/(max3-min3)); // This function is to scale the image into 0 ~ 255 using maximum and minimum value in the image. And then convert the image into 8-bit unsigned image.
  
• How to pass the structure to cuda？

  I want to pass the structure to cuda device,and my structure contains pointer,the pointer is a structure.

  there is visual studio 2017 community and cuda 10.0

  there is the structure

  typedef struct mod2entry { 
  int row; 
  int col; 
  struct mod2entry *right, *left, *up, *down; 
  float pr, lr; 
  } mod2entry; 
  
  typedef struct mod2block { 
  struct mod2block *next; 
  mod2entry entry[Mod2sparse_block]; 
  }mod2block; 
  
  typedef struct { 
  int n_rows; 
  int n_cols; 
  mod2entry *rows; 
  mod2entry *cols; 
  mod2entry *next_free; 
  mod2block *blocks; 
  } mod2sparse; 
  

  in the host

  mod2sparse *H,*d_H;
  

  H is in the host memory and d_H is in the device memory. how to pass H to d_H.
• In cuda, how to dynamically specify the gpu device id?

  I want to implement dynamic binding. I use LD_PRELOAD to hijack the cudaSetDevice function. After selecting the appropriate gpu, I modify the device id to implement dynamic binding, but there are some problems.

  When I used the hook to hijack the cuda code, the hijacking was successful. However, if you modify the device id (the parameter of cudaSetDevice), a "Segmentation fault" error will occur.

  // cuda 9.0 cuda runtime api   
  typedef int(*cuda_set_device_fp)(int);
  
  // define dynamic library same name function
  int cudaSetDevice(int device)
  {
    static void *handle = NULL;
    static cuda_set_device_fp orig_cuda_set_device = NULL;
  
    if( !handle )
    {
      handle = dlopen("libcuda.so", RTLD_LAZY);
      orig_cuda_set_device = (cuda_set_device_fp)dlsym(handle, "cudaSetDevice");
    }
    device = 1;
    printf("oops!!! hack function invoked. device = %d\n", device);
    return orig_cuda_set_device (device);
  }
  

  Does this mean using the method hook of LD_PRELOAD, can't modify the parameters of the function?

  The hijacking succeeds, mapping the user to the gpu task on device 0 and re-mapping to device 1.
• How do I copy the cuda input array into the shared array?

  I'm trying to copy a cuda input array into a shared memory array. The first n values copy into the shared array perfectly but after that there are some pretty weird patterns happening. Can anyone find what I'm doing wrong/what's happening here?

  I've tried using s_x[tid] = x[tid], s_x[row]= x[row], and the other two combinations as well. I know they wouldn't logically make sense but gotta try everything.

  Here is the kernel function I wrote:

      __global__ void mv_cuda_shared(float* y, float* A, float* x, int n)
      {
          extern __shared__ float s_x[];
          int row = blockIdx.x * blockDim.x + threadIdx.x;
          int tid = threadIdx.x;
          s_x[row] = x[row];
          __syncthreads();
          if(tid == 0 && row == 0)
                  for(int i = 0; i < n; i++){printf("s_x[%d] = %10.6f x = %10.6f\n", i, s_x[i], x[i]);}
          float temp = 0;
          if(row < n)
          {
              for(int k = 0; k < n; k++)
              {
                  temp += (A[row*n + k] * s_x[k]);
              }
          }
          y[row] = temp;
      }
  

  and here is where I call it in the main method (omitting all the cuda mallocs and copying):

      mv_cuda_shared<<<(N/BLOCK_SIZE), BLOCK_SIZE, N*sizeof(float)>>>(d_y, d_A, d_x, N);
  

  So I expect s_x to match x. But, instead the first N (n=N) elements are copied correctly but then the rest seem to follow a pattern. For example the 4 entries after s_x[N-1] is say, 1. Then the 4 after that are like, .24566.... (I'll check and the number isn't even in the original array). So anywho, can anyone see what I've done wrong?