The bytecode is specifically designed so that it maps directly to the output.

  // The third phase operates primarily using a bytecode interpreter; it
  // generates a program in "FizzBuzz bytecode", for which each byte of
  // bytecode generates one byte of output. The bytecode language is
  // designed so that it can be interpreted using SIMD instructions; 32
  // bytes of bytecode can be loaded from memory, interpreted, and have
  // its output stored back into memory using just four machine
  // instructions.

  // The bytecode format is very simple (in order to allow it to be
  // interpreted in just a couple of machine instructions):
  // - A negative byte represents a literal character (e.g. to produce
  //   a literal 'F', you use the bytecode -'F', i.e. -70 = 0xba)
  // - A byte 0..7 represents the hundreds..billions digit of the line
  //   number respectively, and asserts that the hundreds digit of the
  //   line number is even
  // - A byte 8..15 represents the hundreds..billions digit of the line
  //   number respectively, and asserts that the hundreds digit of the
  //   line number is odd

  // %ymm2 holds the bytecode for outputting the hundreds and more
  // significant digits of a line number. The most significant digits of
  // this can be obtained by converting LINENO_TOP from high-decimal to
  // the corresponding bytecode, which is accomplished by subtracting
  // from 198 (i.e. 256 - 10 - '0'). The constant parts of LINENO_TOP
  // are 198 minus the bytecode for outputting the hundreds to billions
  // digit of a number; this makes it possible for a single endian
  // shuffle to deal with all 16 of the mid and high digits at once.