02Verification

Find the parameter a generator loop ignores

Problem

Intended behavior. Gains drives two 8-bit outputs from one Scala method, gain(in, doublings, bias), which builds in * 2^doublings + bias from a chain of adders. x4p3 is gain(in, 2, 3), that is 4 * in + 3, and x2p1 is gain(in, 1, 1), that is 2 * in + 1. Both wrap modulo 256.

Current behavior. The starter compiles and runs. x2p1 is right for every input, but x4p3 is wrong for the nonzero inputs in the public run.

Your task. Find out why one call of the same method builds the wrong circuit, and fix gain itself. Keep the ports and both calls, with their arguments, unchanged.

Read and predict

gain prints one ELABORATE line for each doubling adder it builds and one more for the bias adder. Those lines count the adder expressions the generator builds, before any later optimization.

Before running, write down how many doubling lines each call should print, and both outputs for inputs 1 and 5. Then run the starter and compare, first the ELABORATE lines and then the SIMULATE lines.

Diagnose

The SIMULATE lines tell you which output is wrong; the ELABORATE lines tell you why. Count the doubling lines for each call and compare the count with each argument of that call. A loop that builds the wrong number of adders is reading the wrong Scala value, and a call whose arguments happen to be equal hides the mistake.

Nothing done at simulation time can repair this. The loop has finished before the first poke.

Completion check

x4p3 equals 4 * in + 3 and x2p1 equals 2 * in + 1, both modulo 256, for every 8-bit input.
Each call prints as many doubling lines as its doublings argument.
A fixed loop count that suits gain(in, 2, 3) breaks gain(in, 1, 1). The fix has to work for any arguments.

Hints

Hint 1

Put the number of doubling lines for gain(2, 3) next to its two arguments. Which argument does the count match?

Hint 2

Look at the upper bound of the for range. A generator parameter that the loop never reads can change freely without changing the circuit.

Concept

In a generator, Scala parameters are the only inputs that shape the hardware. When editing a parameter changes nothing, or changes the wrong thing, trace where each loop bound and each Scala if gets its value. Prints during elaboration are the quickest way to see what the generator actually built.

Variation

Add a third output driven by gain(io.in, 0, 7). With the fixed gain, predict its ELABORATE lines and its value for input 10. Then work out what the starter's version would have printed and computed for the same call.