In this section, it shows several different environments and strategies for exclusive scan:
When there are only limited number of cores (like two), it is better to divide the array to cores and do sequential scan respectively and then merge.
When the array is small (like 32 elements), we could simply use SIMD instead of the sophisticated work-efficient algorithm.
When the array is larger, we could use a multi-thread SIMD implementation.
I realized that there is no silver bullet in parallel computing, and we need to design the algorithm with respect to the scale of the problem and the computing resource we have.
In this section, it shows several different environments and strategies for exclusive scan:
I realized that there is no silver bullet in parallel computing, and we need to design the algorithm with respect to the scale of the problem and the computing resource we have.