Re: GNU Guile branch, master, updated. v2.1.0-158-g3c12fc3

Re: GNU Guile branch, master, updated. v2.1.0-158-g3c12fc3

[Mark H Weaver]

Hi Andy,

> commit 24ea9f9c3abb8d9398df4810b815075593ba67c8
> Author: Andy Wingo <wingo-e+AXbWqSrlAAvxtiuMwx3w@public.gmane.org>
> Date:   Tue Mar 6 22:21:39 2012 +0100
>
>     ports.c: inline get_codepoint
>     
>     * libguile/ports.c (get_codepoint): Add inline keyword.  It showed up
>       high in benchmarks, and it's static, so it's probably important to
>       inline.

It's generally frowned upon nowadays to explicitly ask the C compiler to
inline functions, as modern compilers usually do a better job of
deciding when to inline than we do.  It's not a very small function, and
now there will be two copies of it, maybe more in the future.  The
overhead of calling a static function is extremely low.  Were you able
to measure how much this helps performance?

    Thanks,
      Mark

Re: GNU Guile branch, master, updated. v2.1.0-158-g3c12fc3

[Andy Wingo]

Hi,

On Thu 08 Mar 2012 03:37, Mark H Weaver <mhw@netris.org> writes:

>> commit 24ea9f9c3abb8d9398df4810b815075593ba67c8
>> Author: Andy Wingo <wingo-e+AXbWqSrlAAvxtiuMwx3w@public.gmane.org>
>> Date:   Tue Mar 6 22:21:39 2012 +0100
>>
>>     ports.c: inline get_codepoint
>
> It's generally frowned upon nowadays to explicitly ask the C compiler to
> inline functions, as modern compilers usually do a better job of
> deciding when to inline than we do.

This was in response to benchmarking.  In my experience the compiler
usually does a good job but sometimes needs help.  In this case it
helped my benchmark by a percent or so without increasing code size.

Regards,

Andy
-- 
http://wingolog.org/

Re: GNU Guile branch, master, updated. v2.1.0-158-g3c12fc3

[David Kastrup]

Andy Wingo <wingo@pobox.com> writes:

> Hi,
>
> On Thu 08 Mar 2012 03:37, Mark H Weaver <mhw@netris.org> writes:
>
>>> commit 24ea9f9c3abb8d9398df4810b815075593ba67c8
>>> Author: Andy Wingo <wingo@pobox.com>
>>> Date:   Tue Mar 6 22:21:39 2012 +0100
>>>
>>>     ports.c: inline get_codepoint
>>
>> It's generally frowned upon nowadays to explicitly ask the C compiler to
>> inline functions, as modern compilers usually do a better job of
>> deciding when to inline than we do.
>
> This was in response to benchmarking.  In my experience the compiler
> usually does a good job but sometimes needs help.  In this case it
> helped my benchmark by a percent or so without increasing code size.

With your compiler, the current compiler version, and on a particular
platform.

-- 
David Kastrup