Enable vectorizedHashCode intrinsic for OffHeap on POWER

[luke-li-2003]

Calculate the address for the array data in OffHeap mode correctly for vectorizedHashCode

[luke-li-2003]

FYI @zl-wang

[luke-li-2003]

Performance numbers:

load	gencon	offheap	offheap with changes
int[2]	471	245	400
3	375	210	333
4	259	190	240
16	158	65	161

[zl-wang]

LGTM

[zl-wang]

almost identical code between gencon and offheap ... it is somewhat unexpected that int[2] & int[3] performance are not closer to gencon, isn't it? should be more than 90% of the gencon corresponding numbers.

[luke-li-2003]

I did some more runs and the numbers were the same, so they should be accurate. My guess is that the extra load required is significant enough on the short arrays.

[zl-wang]

Jenkins test sanity aix,plinux jdk8,jdk21

[zl-wang]

in particular, for int[3], how long is the sequence for gencon and off-heap respectively? we can figure out if the load vs. addi (that is the only difference) can explain it.

[luke-li-2003]

It turns out the old numbers for int[3] were wrong. All the new tests I run have 375 for gencon and 333 for offheap. The other lengths' throughputs are still consistent.

[zl-wang]

better. please update the performance numbers.

[luke-li-2003]

Numbers has been updated here: #21856 (comment)