Is it right to assume that when we say 'Decomposition need not happen statically,' we are talking about work division using pthreads?
And, when we say that 'New tasks can be identified as program executes,' we are referring to ISPC tasks, as an example?
thunder
Please correct me if I'm wrong. For the first question, I think "statically" here is opposed to the concept "dynamically". And the point for the sentence is that the decomposition of all the work might not be determined at the beginning of the program, but changing through the execution. For the programs in our assignment, all the decompositions are statical. For the example of a dynamically decomposed parallel task, I can only think of a quick sort problem in which the decomposition can be finalized only after the partition is done. For the second question, I think the task here is just a general idea and ISPC task can be one example of it.
muchanon
I think that when the notes say 'Decomposition need not happen statically' and 'New tasks can be identified as program executes', the notes are just trying to say the same thing. I don't believe it is exclusively pthreads or ISPC tasks. Both of them are just trying to say that you can use dynamic allocation, regardless of your parallel implementation.
Is it right to assume that when we say 'Decomposition need not happen statically,' we are talking about work division using pthreads? And, when we say that 'New tasks can be identified as program executes,' we are referring to ISPC tasks, as an example?
Please correct me if I'm wrong. For the first question, I think "statically" here is opposed to the concept "dynamically". And the point for the sentence is that the decomposition of all the work might not be determined at the beginning of the program, but changing through the execution. For the programs in our assignment, all the decompositions are statical. For the example of a dynamically decomposed parallel task, I can only think of a quick sort problem in which the decomposition can be finalized only after the partition is done. For the second question, I think the task here is just a general idea and ISPC task can be one example of it.
I think that when the notes say 'Decomposition need not happen statically' and 'New tasks can be identified as program executes', the notes are just trying to say the same thing. I don't believe it is exclusively pthreads or ISPC tasks. Both of them are just trying to say that you can use dynamic allocation, regardless of your parallel implementation.