Lesson10 - 2D Convolution - Conclusions

Available Implementations:

Using Global Memory:

Using Texture Memory:

Using Shared Memory:

  • 1 thread associated to 2 pixels with overlap
  • 1 thread associated to 2 pixels without overlap
  • 1 thread, 1 pixel, n elements shared memory:
  • incremental, processes and loads it

References - CUDA:

Giving Feedback:

Please fill out here your feedback forms. Each student should give feedback on two papers from colleagues

Template for Feedback: Formulário

Performance comparison

Implementations are available in desempenho convolução.