Speedupoverall = 1/((1 – .7) + (.7/40)
Solving the above equation gives a Speedupoverall of 3.149
To find what percentage of vectorization would need to be achieved to match the hardware
doubling method simply solve for Fractionenhanced as follows:
Fractionenhanced = speedupoverall * speedupenhanced - speedupenhanced/ speedupoverall * speedupenhanced–
speedupoverall
So,
Fractionenhanced = (3.14 * 20 – 20)/(3.14 * 20 – 3.14) = .718
Therefore, a percentage vectorization of around 1% would match the hardware doubling
technique. I would recommend using the compiler to crew to gain the extra 1% vectorization