[erlang-questions] Pattern matching of literal small/big/bignum integers and binary search

[Edmond Begumisa]

Thank you Björn,

This answers all my questions precisely, especially revealing how bignum  
literals are treated differently w.r.t binary search.

And erts_debug:df/1 is pure gold. Far more useful than erlc -S. Even made  
it easy to locate the relevant instruction implementation in beam_emu.c

Much appreciated.

- Edmond -


On Mon, 17 Jun 2019 15:45:24 +1000, Björn Gustavsson <bjorn@REDACTED>  
wrote:

> On Sun, Jun 16, 2019 at 10:28 AM Edmond Begumisa
> <ebegumisa@REDACTED> wrote:
>>
>> Hello all,
>>
>> I have some questions about how pattern matches on integer literals are
>> evaluated by the emulator.
>>
>> As an example, say I have the following function which expects an  
>> integer
>> that's a power of 2 and returns corresponding position of the lone "on"
>> bit...
>>
>>
>>          single_cardinalty_to_pos(2#1) -> 1;
>>          single_cardinalty_to_pos(2#10) -> 2;
>>          single_cardinalty_to_pos(2#100) -> 3;
>>          ...
>>          single_cardinalty_to_pos(2#100...0) -> 128.
>
> The easiest way to determine what the loader will do with the code is
> to disassemble the loaded code by calling erts_debug:df/1. If your
> module is called 'foo', erts_debug:df('foo') will disassemble the
> loaded coded for the module into the file foo.dis.
>
> In the case of your example, the loader will split up the select_val
> instruction into multiple instructions. There will be a
> select_val_bins instruction that will do a binary search of all
> integers that fit into one word. If that instruction fails, there will
> be one i_is_ne_exact_literal instruction for comparing each of the
> bignum values. Here are the three last instructions from your example
> (matching 2^126, 2^127, 2^128):
>
> 0000000018E150B8: i_is_ne_exact_literal_fxc f(0000000018E15160) x(0)
> 85070591730234615865843651857942052864
> 0000000018E150D0: i_is_ne_exact_literal_fxc f(0000000018E15150) x(0)
> 42535295865117307932921825928971026432
> 0000000018E150E8: i_is_ne_exact_literal_fxc f(0000000018E15140) x(0)
> 21267647932558653966460912964485513216
> .
> .
> .
> 0000000018E15160: move_return_c 126
> 0000000018E15170: move_return_c 127
> 0000000018E15180: move_return_c 128
>
> /Björn
>
> P.S. In OTP 22 documentation, there is internal documentation for the
> beam_makeops script and the loader:
>
> http://erlang.org/doc/apps/erts/beam_makeops.html
>


-- 
Using Opera's mail client: http://www.opera.com/mail/