gofmt works on the AST. When you look at https://golang.org/pkg/go/ast you'll see that every node has functions Pos() and End() which return the token.Pos of the beginning and end respectively. These are essentially offsets in the source file and as such know nothing about line numbers/breaks.
But when combined with a token.Fileset such a token.Pos can be converted into a token.Position which includes the line number. gofmt does that in the function printer.go:lineFor().
The actual insertion of linebreaks is done in nodes.go:linebreak(). The first argument to linebreak() is a line number obtained by calling the aforementioned lineFor() on the respective token.Pos. The function computes the difference between this line number and the line number of the last token that was printed (tracked in the pos field of struct printer). This tells it if the token to be printed now is on the same line in the input file as the previous token. If it isn't, that means the programmer included one or more line breaks in the original source and linebreak() will output at most 1 empty line. While it could preserve all input line breaks, gofmt's policy is to compress series of empty lines down to only 1 empty line.
If the reason you're asking this question is that you want to customize gofmt, take a look at https://github.com/mbenkmann/goformat
