Maybe. If you're just loading constants onto the stack, it seems like it would be pretty forgiving of optimizations and such, and even work across architectures

Maybe. If you're just loading constants onto the stack, it seems like it would be pretty forgiving of optimizations and such, and even work across architectures (assuming the same endian-ness, at least). I'd definitely look for init code for some calculation (as it appears the author has done here), rather than actual computation code, since, as you point out, optimizations could have a huge impact on the instructions emitted for those.