"Brandon Van Every" <[EMAIL PROTECTED]> writes:
> On 9/13/07, Goswin von Brederlow <[EMAIL PROTECTED]> wrote:
>>
>> For dynamic code (in C/C++/obj-c/fortran/anythign with C linkage) you
>> must use -fPIC or variant thereof because otherwise your jump labels
>> will be too small and linking will fail sometimes on x86, nearly
>> always on amd64, ppc, always on ia64, hppa, ...
>>
>> And if you build static code with -fPIC then you get the reverse on
>> some archs and loose, some claim up to 30%, speed on x86 and amd64.
>
> Great. I just took a stab at redrafting the new FAQ material for
> clarity, not accuracy. So now I will ask, is what's described
> actually correct, at least for Linux and GCC?
> http://www.cmake.org/Wiki/CMake_FAQ#Library_questions
> If it's all YMMV for different Unixes or compilers, then we need
> strong warnings about trying to apply this trick cross-platform. As
> well as performance consequences you mention.
What your hack builds there is the *.so and *_pic.a. You build a
static library suitable for use in a dynamic library. Afaik no common
architecture has problems with using -fPIC for a static lib but as
mentioned you loose the speed gain associated with static code.
As to functionality I think it should work on other unix systems too
and other compilers as well provided the two options you used exist.
Don't ask me about windows dlls though.
MfG
Goswin
_______________________________________________
CMake mailing list
[email protected]
http://www.cmake.org/mailman/listinfo/cmake
