Re: [PATCH net-next] net: Implement fast csum_partial for x86_64

On Mon, Jan 4, 2016 at 3:52 PM, Eric Dumazet [off-list ref] wrote:

On Mon, 2016-01-04 at 15:34 -0800, Tom Herbert wrote:

quoted

On Mon, Jan 4, 2016 at 2:36 PM, Eric Dumazet [off-list ref] wrote:

quoted

On Sun, 2016-01-03 at 15:22 -0800, Tom Herbert wrote:
\...

quoted

+402: /* Length 2, align is 1, 3, or 5 */
+     movb    (%rdi), %al
+     movb    1(%rdi), %ah

Looks like a movw (%rdi),%ax

Wouldn't that be an unaligned access?

x86 does not care. (This is why we have
CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)

I bet it is faster using a single instruction.

Okay, I'll re-implement without worrying about alignment. If it's not
an issue at all (even for eight bytes) then that will be a speedup.

Thanks,
Tom