Multiarch crazyiness

Callum Lerwick seg at haxxed.com
Tue Oct 23 05:24:44 UTC 2007


On Mon, 2007-10-22 at 12:20 -0400, Jesse Keating wrote:
> On those arches, running 64bit code comes at a penalty.  In fact, even
> on x86_64 there is a penalty.  Larger memory overhead.  However on
> x86_64 the extra CPU registers gained by running in x86_64 mode
> outweigh the memory overhead.

Not to mention with x86_64, you can assume the presence of cmov, MMX and
SSE2. Whereas on i386, we're still for the most part compiling for 386s.

In fact I've written patches for OpenJPEG that are specifically designed
to take advantage of gcc's ability to perform if conversions with cmovs,
to eliminate branching. Fedora's i386 misses out on this. :)

Passing function arguments, including structures, in registers is part
of the x86_64 ABI.

x86_64 also has an instruction pointer relative addressing mode,
allowing it to support position independent code without a speed hit.

<3 <3 <3 x86_64
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://listman.redhat.com/archives/fedora-devel-list/attachments/20071023/aab4ce96/attachment.sig>


More information about the fedora-devel-list mailing list