This slide may not be covered in the lecture, but I find it interesting. Specifying specific dependencies looks like more natural, and the programmers could have a better description of the workflow. When the granularity is finer, there are less unnecessary constraints, then the program could have better parallelism.
I also considered specifying specific dependencies for the grid solver problem. But I came up with something very similar to the barrier version.
zale
Functional programming can also be a way to specify dependencies in the program, since many functional languages require side-effects to be made explicit. However, being significantly more abstract than C/C++, they are certainly much harder to optimize correctly if performance is the main issue.
This slide may not be covered in the lecture, but I find it interesting. Specifying specific dependencies looks like more natural, and the programmers could have a better description of the workflow. When the granularity is finer, there are less unnecessary constraints, then the program could have better parallelism.
I also considered specifying specific dependencies for the grid solver problem. But I came up with something very similar to the barrier version.
Functional programming can also be a way to specify dependencies in the program, since many functional languages require side-effects to be made explicit. However, being significantly more abstract than C/C++, they are certainly much harder to optimize correctly if performance is the main issue.