Dual Mode Binaries

Dual Mode Binaries

[Brian D. McGrew]

Good morning!

I've googled all I can think of and perhaps I'm not searching the correct
keywords, or it can't be done...

I'm looking to create a dual-mode binary for 32-bit and 64-bit Linux in the
same executable...  Similar to the way Apple puts out dual mode binaries for
PPC and Intel.

Can this be done with gcc-4.4.2???  If so, where can I find the howto's?

-b

Re: Dual Mode Binaries

[Joel Dice]

On Sat, 5 Dec 2009, Brian D. McGrew wrote:

> Good morning!
>
> I've googled all I can think of and perhaps I'm not searching the correct
> keywords, or it can't be done...
>
> I'm looking to create a dual-mode binary for 32-bit and 64-bit Linux in the
> same executable...  Similar to the way Apple puts out dual mode binaries for
> PPC and Intel.
>
> Can this be done with gcc-4.4.2???  If so, where can I find the howto's?

This may be what you're looking for:

http://icculus.org/fatelf/

Re: Dual Mode Binaries

[Kalle Olavi Niemitalo]

Joel Dice <dicej@mailsnare.net> writes:

> This may be what you're looking for:
>
> http://icculus.org/fatelf/

Those patches were not accepted into Linux and glibc,
and the author gave up.

http://osnews.com/story.php/22446/Ryan_Gordon_Halts_FatELF_Project

Re: Dual Mode Binaries

[Tim Prince]

Kalle Olavi Niemitalo wrote:
> Joel Dice <dicej@mailsnare.net> writes:
> 
>> This may be what you're looking for:
>>
>> http://icculus.org/fatelf/
> 
> Those patches were not accepted into Linux and glibc,
> and the author gave up.
> 
> http://osnews.com/story.php/22446/Ryan_Gordon_Halts_FatELF_Project
The incentive for combining 32- and 64-bit binaries is mighty slim; in 
most cases where 32-bit is satisfactory, the 32-bit code runs as well or 
better on the 64-bit OS, while the larger size binary detracts from any 
advantage.
gcc has survived without providing much support for multiple paths, e.g. 
  non-SSE, SSE3, ... where the incentive might have been greater.

Re: Dual Mode Binaries

[Kalle Olavi Niemitalo]

Tim Prince <n8tm@aol.com> writes:

> gcc has survived without providing much support for multiple paths,
> e.g. non-SSE, SSE3, ... where the incentive might have been greater.

For those, the upcoming ifunc attribute should be a better fit,
as it allows compiling multiple variants of only the real
hotspots and sharing all the rest.  The downside is that ifunc
requires changes in the source code, rather than only compiler
options.