On Tue, 6 Oct 2015 20:45:18 -0400
William Harrington <[email protected]> wrote:

> Another option for GMP to build a generic library is to use --build during
> configure. Distros will build GMP for a generic AMD64 or i386 build.


That's really good to know and probably the best way to go.


On Mon, 05 Oct 2015 14:04:42 -0400
Wayne Sallee <[email protected]> wrote:

> The one that I ran the test in is AMD.

AMD is often a tad slow to add Intel's newest commands. The BMI2 stuff is
not even in the Piledriver processors, it won't be until the Excavator
series that AMD will support BMI2:

https://en.wikipedia.org/wiki/Advanced_Bit_Manipulation

It can be understandable though - a lot of the newer commands don't add
as much performance as Intel suggests they do. For example, in 2007 Linus
Torvalds went on a rant about the lack of virtues of cmov:

http://yarchive.net/comp/linux/cmov.html

The BMI2 stuff might be an exception here when very large numbers need to
be manipulated (as libgmp does). This might explain why libgmp runs
noticably faster on Intel's stuff:

https://gmplib.org/pi-with-gmp.html

A 2.9 GHz Intel Haswell running more than 25% faster than a 4 GHz
AMD Piledriver? Sheezz. I'm still an AMD fan though. Note how it was the
inverse case in the GMP 4.2/4.3 days.

> One thing that I am wondering about; what about all of the programs that
> I have compiled with the previous install of gmp? I am wondering if I
> might run into problems with them working on a different computer.

You should be OK unless libgmp was statically compiled into something.
Note the standard LFS build of libgmp even disables the static ability:

http://www.linuxfromscratch.org/lfs/view/development/chapter06/gmp.html

Issues like this are the entire point of dynamic libraries - the library
can be updated without having to recompile everything. Furthermore, I
don't think many applications will link with gmp - it's gcc itself
that is using it. On my machine even bc does not link to gmp.

Gmp is a math library that supports the handling of numbers of arbitrary
precision and size. It's often used with cryptography applications because
they work with such large primes. I wonder why gcc needs it (and only
recently it seems this became true).



    Cheers,

    Mike Shell

 
-- 
http://lists.linuxfromscratch.org/listinfo/blfs-support
FAQ: http://www.linuxfromscratch.org/blfs/faq.html
Unsubscribe: See the above information page

Reply via email to