Improve performance in espresso_exprs

[danjujan]

In case espresso_exprs has only one expression as argument, the output vector is always [1]. This change results in a significant speedup for expressions with a large cover, as we don't need to compare the cover with itself.

[cjdrake]

Thanks for the pull request. I haven't had a chance to review it yet.

Please help me understand the background by providing an example of how this improves performance.

[danjujan]

Sure. When espresso_exprs is called with only one argument, fscover is the same as f.cover. The outvec loop checks if fcube is in fscube. But since they are the same, we will always append 1 to the outvec. So we can skip the outvec loop in this case, which saves us len(fscover) * len(f.cover)operations. Basically, the function boils down to this:

def __espresso_expr(dnf: Expression) -> Expression:
    support = dnf.support
    inputs = sorted(support)

    ninputs = len(inputs)
    noutputs = 1

    invec = [0] * ninputs
    cover = set()
    for cube in dnf.cover:
        for i, var in enumerate(inputs):
            if ~var in cube:
                invec[i] = 1
            elif var in cube:
                invec[i] = 2
            else:
                invec[i] = 3
        cover.add((tuple(invec), (1,)))

    set_config(**CONFIG)

    cover = espresso(ninputs, noutputs, cover, intype=FTYPE)
    return _cover2exprs(inputs, noutputs, cover)[0]

Before this change, one call to espresso_exprs took about 30 seconds for an expression with 22 variables and a cover of about 22000 elements. Now it takes only a second. The fixed size lists just avoid allocating memory for every append in the loop.

[cjdrake]

LGTM