svn commit: r229415 - head/lib/libvgl
Dimitry Andric
dim at FreeBSD.org
Tue Jan 10 08:02:14 UTC 2012
On 2012-01-10 05:48, Eitan Adler wrote:
> On Mon, Jan 9, 2012 at 11:02 PM, John-Mark Gurney<jmg at funkthat.com> wrote:
>> Last time I did benchmarking on this ~10 years ago, it was slower to
>> use xor instead of using assignments...
>
> I'd be shocked if the compiler didn't optimize this away to an XCHG.
Even with our ancient gcc 4.2 (on i386):
VGLLine:
[...]
movl 12(%ebp), %edx
movl 16(%ebp), %eax
movl $1, -56(%ebp)
xchgl %eax, %edx
movl %edx, 12(%ebp)
movl 24(%ebp), %edx
movl %eax, 16(%ebp)
movl %ecx, %eax
xchgl %edx, %esi
movl %edx, 24(%ebp)
movl -84(%ebp), %edx
xchgl %eax, %edx
movl %eax, -80(%ebp)
movl %edx, -84(%ebp)
With clang, the corresponding local variables seem to have been
optimized away entirely (or rather, moved around).
More information about the svn-src-all
mailing list