The OR seems to be correct. %r9d previously contains 1 for a non-Intel
CPU. The next line isolates the XOP bit but *also* preserves that 1.
This figures in later in in line 121 to avoid tagging as Intel. (Most
codepaths in the AMD-specific branch go straight to .Lgeneric, but some
go to .Lintel. This probably could be done more clearly as there's some
complicated code that's subtly different...)

Change-Id: Ie7b9566074498333f26416e46478e4f4df68194f
Reviewed-on: https://boringssl-review.googlesource.com/5433
Reviewed-by: Adam Langley <agl@google.com>