Um, that's not what the paper does at all. There's exactly one invocation of gcc in the whole paper, and its only point is to enable the (human) developer to see what the calling convention for a function should look like. A couple of sentences after the one you quote we get: "Let's compile it using gcc [...] Generating this file from Scheme is straightforward." and then comes the very first compiler in the incrementally-improving series, which accepts a small integer and outputs assembler code for a function that returns that integer as its value.