Previous | Next --- Slide 37 of 69
Back to Lecture Thumbnails

I am a bit confused. I thought data parallel requires the instruction streams to be independent, then why do we need any communication in the first place?


The communication could include where to get the data from. For example, in our common example C[i] = A[i] + B[i], each operation needs to know the location of the arrays A, B, and C.