RFR: Use R15-relative bad-mask in interpreter and C1 on x86_64

[Rickard Bäckman]

The Interpreter and C1 have so far used a test reg, [rip+offset] to
execute the ZGC barrier.
On some machines it seems the distance between the CodeCache and global
statics are larger than what is possible to represent with a 32-bit
displacement, causing crashes.

This patch modifies the barrier used by C1 and the Interpreter to
instead use what C2 has been using by default. mov reg, [r15+offset],
which is the thread local copy of the mask.

Here is the webrev:
http://cr.openjdk.java.net/~rbackman/zgc_r15_barrier/

Thanks
/R