PATCH: Add bits/byteswap-16.h and use GCC builtin 32/64-bit bswap

[Andreas Jaeger]

On 04/24/2012 11:31 AM, Chung-Lin Tang wrote:
> On 2012/4/7 03:40 AM, H.J. Lu wrote:
>> 2012-04-06  H.J. Lu  <hongjiu.lu@intel.com>
>>
>> 	* bits/byteswap.h (__bswap_16): Removed.
>> 	Include <bits/byteswap-16.h> to get __bswap_16.
>> 	(__bswap_32): Use __builtin_bswap32 for GCC >= 4.2.
>> 	(__bswap_64): Use __builtin_bswap64 for GCC >= 4.2.
>> 	* bits/byteswap-16.h: New file.
> 
> For SH, this triggered a build fail, due to those inline C functions
> being added to a assembly file (the SH strlen.S use of endian.h) after a
> complex include sequence.
> 
> Of course, there's nothing wrong with those __bswap* functions. It seems
> that string/endian.h includes a bunch of stuff not usable anyways for
> assembler, so here's trivial patch to correct this.
> 
> Thanks,
> Chung-Lin
> 
> 2012-04-24  Chung-Lin Tang  <cltang@codesourcery.com>
> 
> 	* string/endian.h: Add !__ASSEMBLER__ condition for including
> 	conversion interfaces.
> 

coming back to an old email and reversing my opinion...

Having seen this also mentioned on libc-ports, I think we should add the
assembler protection as Chung-Lin proposes in his patch. I'd like
somebody else to comment on this as well...

References:
http://sourceware.org/ml/libc-ports/2012-05/msg00016.html
http://sourceware.org/ml/libc-alpha/2012-04/msg01141.html

Andreas

> diff --git a/string/endian.h b/string/endian.h
> index 0c293f6..51084f1 100644
> --- a/string/endian.h
> +++ b/string/endian.h
> @@ -55,7 +55,7 @@
>  #endif
>  
>  
> -#ifdef __USE_BSD
> +#if defined __USE_BSD && !defined __ASSEMBLER__
>  /* Conversion interfaces.  */
>  # include <bits/byteswap.h>


-- 
 Andreas Jaeger aj@{suse.com,opensuse.org} Twitter/Identica: jaegerandi
  SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
   GF: Jeff Hawn,Jennifer Guild,Felix Imendörffer,HRB16746 (AG Nürnberg)
    GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126