Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > comp.lang.c++ > #82812

Re: Can anyone improve this ?

From Tim Rentsch <tr.17687@z991.linuxsc.com>
Newsgroups comp.lang.c++
Subject Re: Can anyone improve this ?
Date 2022-01-17 16:16 -0800
Organization A noiseless patient Spider
Message-ID <86wnixrj8t.fsf@linuxsc.com> (permalink)
References <srjdtm$d9c$1@dont-email.me> <86ee56rzbz.fsf@linuxsc.com> <ss4ksr$d9d$1@dont-email.me>

Show all headers | View raw


"Alf P. Steinbach" <alf.p.steinbach@gmail.com> writes:

> On 17 Jan 2022 19:28, Tim Rentsch wrote:
>
>> Bonita Montero <Bonita.Montero@gmail.com> writes:
>>
>>> I had an idea how to write a fast UTF-8 strlen, here it is:
>>>
>>> size_t utf8Strlen( char const *str )
>>> {
>>>      struct encode_t { size_t lenIncr, strIncr; };
>>>      static encode_t const encodes[] =
>>>      {
>>>          { 1, 1 },
>>>          { 0, 0 },
>>>          { 1, 2 },
>>>          { 1, 3 },
>>>          { 1, 4 },
>>>          { 0, 0 },
>>>          { 0, 0 },
>>>          { 0, 0 },
>>>          { 0, 0 }
>>>      };
>>>      size_t len = 0;
>>>      for( unsigned char c; (c = *str); )
>>>      {
>>>          encode_t const &enc =
>>>             encodes[(size_t)countl_zero<unsigned char>( ~c )];
>>>          if( !enc.lenIncr ) [[unlikely]]
>>>              return -1;
>>>          len += enc.lenIncr;
>>>          for( char const *cpEnd = str + enc.strIncr; ++str != cpEnd; )
>>>              if( ((unsigned char)*str & 0x0C0) != 0x080 ) [[unlikely]]
>>>                  return -1;
>>>      }
>>>      return len;
>>> }
>>>
>>> Has anyone further ideas to improve this ?
>>
>> Just for fun -
>>
>> size_t
>> utf8_units_two( const char *s ){
>>      size_t r = -1;
>>      unsigned char c;
>>
>>    next:
>>      switch(  r++,  c = *s++,  c >> 3  ){
>> 	cases(16,17,18,19,20,21,22,23,31):			return  -1;
>>
>> 	cases(30):		if(  c = *s++,  c >> 6 != 2  )	return  -1;
>> 	cases(28,29):		if(  c = *s++,  c >> 6 != 2  )	return  -1;
>> 	cases(24,25,26,27):	if(  c = *s++,  c >> 6 != 2  )	return  -1;
>> 	cases(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15):		goto  next;
>>
>> 	cases(0):		if(  c != 0  )			goto  next;
>>      }
>>
>>      return  r;
>> }
>
> That's very ugly.

Beauty is in the eye of the beholder.  Personally I think it's
rather pretty.

>> (Note:  the cases() macro produces one 'case X:' for each
>> argument X except the last one where the ':' is omitted,
>> written using the standard C preprocessor, and left as an
>> exercise for any ambitious readers.)
>
> Exercise for you:  test that unrevealed macro with Visual C++.

I don't have access to a Visual C++ system.  The macro I wrote
compiles under gcc and g++ as C99, C11, C++11, C++14, and C++17,
using a -pedantic-errors option.  (Using clang and clang++ gives
the same results.)

> (Making it work also with Visual C++ is doable.)

Does this need more than just writing the definition in
conforming C11 or C++11?

>> Incidentally, this code provides the only example I remember
>> where using 'goto' is pretty much unavoidable, in that any
>> re-writing without 'goto' seems awkward or inferior in some
>> other way.  I guess I should say, at least not that I could
>> find, maybe someone else can do better.
>
> The `for(;;)`, `continue;` and `break;` constructs come to mind, so
> you're right, someone else can do better.

It is of course possible to write the function without using any
'goto's.  Whether such a writing is an improvement is another
matter.  To my eye the goto version expresses the structure more
directly and in a form that is easier to comprehend.  I did try
actually writing a version with a continue-able loop around the
switch(), but it looked clumsy and not as easy to follow as the
goto version.  If you have an example to post I definitely would
like to see it.

> At a guess the code counts the number of Unicode code points?

Yes, sorry, I thought that was apparent from the earlier
context.

> Does it return -1 also for "too long" UTF-8 sequence?

The -1 return values is for character sequences that are not
syntactically well-formed for UTF-8.

>> This code also has the interesting property that along the main
>> code path there are no conditional branches (as compiled by gcc
>> under -O2).
>
> Smart compiler removed all the `if`'s?
>
> I guess you're flame-baiting a little just for fun, but I chose to
> take it seriously.

I'm not flame-baiting at all.  By "main code path" I mean that
part of the code that deals with normal ASCII characters (and not
counting the terminating null character).  The switch() statement
generates an unconditional indirect branch, indexing a table of
32 targets, and the 15 targets for normal ASCII characters simply
go around the loop again, with no tests.  I expect you can see
that based on the line with 15 cases, which just does a 'goto'
to start the loop again.

Back to comp.lang.c++ | Previous | NextPrevious in thread | Next in thread | Find similar | Unroll thread


Thread

Can anyone improve this ? Bonita Montero <Bonita.Montero@gmail.com> - 2022-01-11 09:08 +0100
  Re: Can anyone improve this ? Juha Nieminen <nospam@thanks.invalid> - 2022-01-11 09:50 +0000
    Re: Can anyone improve this ? Juha Nieminen <nospam@thanks.invalid> - 2022-01-11 11:07 +0000
    Re: Can anyone improve this ? Ben Bacarisse <ben.usenet@bsb.me.uk> - 2022-01-11 11:08 +0000
    Re: Can anyone improve this ? Bonita Montero <Bonita.Montero@gmail.com> - 2022-01-11 18:28 +0100
      Re: Can anyone improve this ? Christian Gollwitzer <auriocus@gmx.de> - 2022-01-11 18:41 +0100
        Re: Can anyone improve this ? Bonita Montero <Bonita.Montero@gmail.com> - 2022-01-11 19:36 +0100
      Re: Can anyone improve this ? Öö Tiib <ootiib@hot.ee> - 2022-01-11 21:54 -0800
  Re: Can anyone improve this ? Bonita Montero <Bonita.Montero@gmail.com> - 2022-01-11 21:51 +0100
    Re: Can anyone improve this ? Bonita Montero <Bonita.Montero@gmail.com> - 2022-01-12 14:19 +0100
    Re: Can anyone improve this ? Bonita Montero <Bonita.Montero@gmail.com> - 2022-01-12 14:20 +0100
      Re: Can anyone improve this ? Ben Bacarisse <ben.usenet@bsb.me.uk> - 2022-01-12 14:14 +0000
        Re: Can anyone improve this ? Bonita Montero <Bonita.Montero@gmail.com> - 2022-01-12 17:20 +0100
          Re: Can anyone improve this ? "james...@alumni.caltech.edu" <jameskuyper@alumni.caltech.edu> - 2022-01-12 08:38 -0800
          Re: Can anyone improve this ? Ben Bacarisse <ben.usenet@bsb.me.uk> - 2022-01-12 16:55 +0000
        Re: Can anyone improve this ? Tim Rentsch <tr.17687@z991.linuxsc.com> - 2022-01-16 12:09 -0800
        Re: Can anyone improve this ? Tim Rentsch <tr.17687@z991.linuxsc.com> - 2022-01-16 12:12 -0800
          Re: Can anyone improve this ? Öö Tiib <ootiib@hot.ee> - 2022-01-16 13:00 -0800
            Re: Can anyone improve this ? Tim Rentsch <tr.17687@z991.linuxsc.com> - 2022-01-17 10:13 -0800
  Re: Can anyone improve this ? Tim Rentsch <tr.17687@z991.linuxsc.com> - 2022-01-17 10:28 -0800
    Re: Can anyone improve this ? "Alf P. Steinbach" <alf.p.steinbach@gmail.com> - 2022-01-17 21:51 +0100
      Re: Can anyone improve this ? Tim Rentsch <tr.17687@z991.linuxsc.com> - 2022-01-17 16:16 -0800
      Re: Can anyone improve this ? Tim Rentsch <tr.17687@z991.linuxsc.com> - 2022-01-17 22:43 -0800
        Re: Can anyone improve this ? "Alf P. Steinbach" <alf.p.steinbach@gmail.com> - 2022-01-18 12:09 +0100
    Re: Can anyone improve this ? Bonita Montero <Bonita.Montero@gmail.com> - 2022-01-18 09:18 +0100

csiph-web