- Dup shuffles broadcast a single lane from a source
  operand to all lanes of the destination. Conceptually
  similar to a splat, they require special handling since
  the splatted value must be extracted from a source. The
  32x4 case is already well handled (pshufd) but 16x8 and
  8x16 currently generate the general shuffle code sequence.
- Adds IA32S16x8Dup, IA32S8x16Dup  opcodes.

Bug: v8:6020
Change-Id: Ia4f044aa7e25cae30e8f9007c2488db738ca6cfc
Reviewed-on: https://chromium-review.googlesource.com/1128513Reviewed-by: Jing Bao <jing.bao@intel.com>
Commit-Queue: Bill Budge <bbudge@chromium.org>
Cr-Commit-Position: refs/heads/master@{#54362}