Question

We have an application that was profiled and it was discovered that the do_work function accounted for 50% of the execution time. If do_work was implemented in hardware, what is the largest speedup one could expect for the co-designed system?

Answer 1

Hello,
Please find the answer attached as under. Please give a thumbs up rating if you find the answer useful! Have a rocking day ahead!

First of all, you have not mentioned the specifications of the hardware and the function do_work. Without a proper spec of these two things, I can only help you with certain hints, on how functions implemented in hardware may perform:

1. If the function is a purely linear function, that is, the output is related to the input only by a gain, the function can be realized by a resistive network, in which case, the speed up will be enormous, since resistances do not offer any delays.

2. If the function is a linear function, but with an intercept, then also, it can be realized using resistive networks with voltage sources. In this case also, the speed up is huge because of negligible delay.

3. If the function is non-linear, and/or contains additive/subtractive terms, then we need to make use of op-amps. Here also, the delays are in the order of nano-seconds.

4. If the function contains terms which are to be integrated and/or differentiated, then we need to make use of capacitors and/or inductors, which may delay things a bit. Depending on the value of the capacitor and inductor, the time constants might vary from milli-seconds to the order of seconds, which may introduce more delay into the system. Thus, these functions are better realised using software.

Having said all of the above, you could try to re-post the same question giving more details on the function and hardware, so that experts can help you with more precise answers.

*************************************************************************************

PS: Please do not forget the thumbs up!