KVM: emulate: avoid per-byte copying in instruction fetches

We do not need a memory copying loop anymore in insn_fetch; we
can use a byte-aligned pointer to access instruction fields directly
from the fetch_cache.  This eliminates 50-150 cycles (corresponding to
a 5-10% improvement in performance) from each instruction.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

diff --cc arch/x86/kvm/emulate.c
Simple merge