This eliminates one instruction for following pattern:
  movl rX, rY       // TruncateInt64ToInt32
  subl rX, imm32
==>
  leal rX, [rY - imm32]

R=bmeurer@chromium.org

Change-Id: I4164e1407f5953302051e905555da14d3ca6680a
Reviewed-on: https://chromium-review.googlesource.com/1046381Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Commit-Queue: Kanghua Yu <kanghua.yu@intel.com>
Cr-Commit-Position: refs/heads/master@{#53021}