Request for reviews (XS): 6708714: Optimize long LShift on 32-bits x86

[Vladimir Kozlov]

http://webrev.invokedynamic.info/kvn/6708714/index.html

Fixed 6708714: Optimize long LShift on 32-bits x86

For small (1-3 bits) left long shifts in 32-bits VM use sets of
add+addc instructions instead of shld+shl on AMD cpus.
It is AMD cpus specific optimization.

No regression in refWorkload.

Reviewed by:
Fix verified (y/n): y, generated code and micro benchmark

Other testing:
JPRT, refWorkload