Clarification: Imagine we want to execute the logic described by the pseudocode on the right for each element of a large array of numbers, and we want to do so as efficiently as possible on our fictitious processor which is capable of eight-wide SIMD execution. In the code, assume that there is an input array A, and that x is initialized to A[i]. Also assume there's an output array B and that B[i] = refl. The code segment on this slide should be changed to something more intuitive (I apologize).
Now consider the fundamental problem at hand. We have to perform a computation that logically may take a different sequence of control for each element in the input array, but we have a processor that operates by executing one instruction on eight execution units at once. We don't know what sequence of control each iteration requires at compile time, because the answer is dependent on the value of A[i]. How do we map this program onto this execution resource (given the constraints of the resource)?
Clarification: Imagine we want to execute the logic described by the pseudocode on the right for each element of a large array of numbers, and we want to do so as efficiently as possible on our fictitious processor which is capable of eight-wide SIMD execution. In the code, assume that there is an input array
A, and thatxis initialized toA[i]. Also assume there's an output arrayBand thatB[i] = refl. The code segment on this slide should be changed to something more intuitive (I apologize).Now consider the fundamental problem at hand. We have to perform a computation that logically may take a different sequence of control for each element in the input array, but we have a processor that operates by executing one instruction on eight execution units at once. We don't know what sequence of control each iteration requires at compile time, because the answer is dependent on the value of
A[i]. How do we map this program onto this execution resource (given the constraints of the resource)?This comment was marked helpful 0 times.