lesson3 some important things in CUDA

40

For a 128*128 image, there are different strategies to create specified number of blocks and threads to compute value in each pixel in parallel. 

For example, we could allocate 128 blocks and 128 threads in each block. A block stands for each line in the image, however, a thread for one pixel in each line.

Otherwise, using 2D blocks and threads, a (8,8) 2-dimension blocks and (16,16) 2-dimension threads can be utilized.

41

threadIdx is the index of thread in block.

blockDim denotes how many threads in each block

blockIdx is the index of block in grid.

gridDim expresses how many blocks in the grid

62