If you look closely at the disassembly for G1 post-barrier in JDK-8130918, then an interesting thing pops out on a hot path in the barrier:
  http://cr.openjdk.java.net/~shade/8130918/g1.perfasm

 ; load receiver address, unpack to %r10
 lea (%r12,%r13,8),%r10

 ; take the value address and unpack it to %r11
 mov %rbx,%r11
 shl $0x3,%r11

 ; xor them
 xor %r10,%r11

 ; shift them by region size
 shr $0x15,%r11

 ; test for 0, if so, this is an intra-region store, bail
 test %r11,%r11
 je 0x00007f82d0af4460

But, we might as well operate on the raw pointers, if we know the compressed oops mode we are operating at.
In this case, with zero-based compressed oops with 3-bit shift, we might just emit:

 ; receiver compressed oop is already at %r13
 ; value compressed oop is already at %rbx
 ; xor them
 xor %r13,%rbx

 ; shift them by region size, plus compressed oops adjustment
 shr $0x12,%r11

 ; test for 0, if so, this is an intra-region store, bail
 test %r11,%r11
 je 0x00007f82d0af4460

Since this code executes for each reference store, the performance improvement may be significant.