Notes

CUDA notes

Filtering notes

Programming assignment

Extend the code in rgb2gray.cu to perform Sobel edge detection on the Jetson TK1 GPU. Your code should be able to process a 640 x 480 grayscale image in less than 1/30th of a second.

Turn in a report including the following elements.

1. Your code

2. A screen shot showing the edge detector working

3. Discuss your strategy for utilizing the GPU. Include the following elements. a. In your discussion explain how you divided the work among blocks and threads (course vs. fine-grained parallelism).

   b. How many kernels did you write to compute the Sobel edge image?

   c. What grid size (number of blocks) and block size (number of threads) did you use for each kernel launch.

   d. Explain how you used shared memory to coordinate threads within a block.

   e. What efforts were made to coalesce accesses to global memory.