Jump table optimization in the Jason library?

preciz · October 29, 2019, 11:20am

michalmuskala/jason/blob/master/lib/decoder.ex#L36


@moduledoc false


import Bitwise


alias Jason.{DecodeError, Codegen}


import Codegen, only: [bytecase: 2, bytecase: 3]


# @compile :native


# We use integers instead of atoms to take advantage of the jump table
# optimization
@terminate 0
@array     1
@key       2
@object    3


def parse(data, opts) when is_binary(data) do
  key_decode = key_decode_function(opts)
  string_decode = string_decode_function(opts)
  try do

Could somebody elaborate on this optimization and how to achieve it in Elixir?
(in the context of Elixir, not simple mentions of assembly)

dom · October 29, 2019, 11:39am

Check out “select_val optimizations” in The BEAM Book. Pattern matching on a variable when all cases are integers close to each other makes it possible to use a jump table (jump_on_val) rather than linear or binary search.