Video / Imaging DesignWire

A Generic Instruction Stream
This problem can be solved by realizing that even though different operations are to be performed, the SIMD processing model can still take place in two or more compute units by converting the equations such as (3) and (6) to a more generalized generic instruction stream that carries both equations within it but calls up local coefficients stored in each compute unit to produce its localized solution.

For example, equations (3) and (6) for P₀ can be generalized as follow,

for ap = 1, equation (3) can be rewritten as

(7)

and for ap = 0, equation (6) can be rewritten as

(8)

Equation (7) can then be generalized to:

(9)

and equation (8) can be generalized to:

(10)

It can be seen that equation (9) and equation (10) are in the same form, except that equation (10) for P₀ and ap = 0 has no p₂ or q₀ term. The generic instruction stream can be represented as

(11)

where all the terms in both equations (9) and (10) are represented, p_0, p₁, p_2, q_0, q₁, but with different coefficients K₀ - K₄.  The specific coefficient set is selected according to ap and aq and when applied, each of the compute units generates its localized solution to the generic instruction stream.

NEXT: Dual Compute Units

Page 3: next page

Pages: 1 2 3 4 5