This optimization greatly enhances efficiency of creating new instructions by

masking and shifting operands directly into their place in the instruction,
instead of the old-fashioned way of ORing in each bit separately.

llvm-svn: 7179