In the getting stated with TVM tutorial for vector addition there is a step where you split the tensor C along it’s first axis as seen below. What does factor mean and what is the significance of it being 64 here?
I’ve searched google, the forums, and the API but I’m still new to all of this so I apologize if this has been answered somewhere!
Splitting a loop with a given factor will break a loop nest of extent (number of iterations) N into N/factor pieces with each piece having extent factor. This is a classical way to partition work on a GPU for locality as in the intent is to give assign each block of threads 64 (continuous) indices. The choice of 64 here is fairly arbitrary, but it is usually related to hardware properties (e.g., the ideal number of threads that should be allocated per block for this computation).
