AArch64: Regressions in ArrayCopyUnalignedDst.testByte/testChar for 65-78 bytes when UseSIMDForMemoryOps is on

Filed on behalf of Evgeny Astigeevich, eastig@amazon.com.

JDK-8256488 fixed most of regressions but ArrayCopyUnalignedDst.testByte/testChar for 65-78 bytes still have them:

Benchmark length Old Score New Score Units Diff
ArrayCopyUnalignedDst.testByte 65 6.739 7.957 ns/op 18.07%
ArrayCopyUnalignedDst.testByte 66 6.725 7.935 ns/op 17.99%
ArrayCopyUnalignedDst.testByte 67 6.714 7.853 ns/op 16.96%
ArrayCopyUnalignedDst.testByte 68 6.671 8.242 ns/op 23.55%
ArrayCopyUnalignedDst.testByte 69 6.714 8.219 ns/op 22.42%
ArrayCopyUnalignedDst.testByte 70 6.707 8.228 ns/op 22.68%
ArrayCopyUnalignedDst.testByte 71 6.716 7.839 ns/op 16.72%
ArrayCopyUnalignedDst.testByte 72 6.667 8.312 ns/op 24.67%
ArrayCopyUnalignedDst.testByte 73 6.684 8.216 ns/op 22.92%
ArrayCopyUnalignedDst.testByte 74 6.69 8.245 ns/op 23.24%
ArrayCopyUnalignedDst.testByte 75 6.677 7.93 ns/op 18.77%
ArrayCopyUnalignedDst.testByte 76 6.668 8.316 ns/op 24.72%
ArrayCopyUnalignedDst.testByte 77 6.684 8.23 ns/op 23.13%
ArrayCopyUnalignedDst.testByte 78 6.684 8.242 ns/op 23.31%
ArrayCopyUnalignedDst.testChar 33 7.147 7.754 ns/op 8.49%
ArrayCopyUnalignedDst.testChar 34 7.003 8.126 ns/op 16.04%
ArrayCopyUnalignedDst.testChar 35 7.102 7.687 ns/op 8.24%
ArrayCopyUnalignedDst.testChar 36 7.012 8.153 ns/op 16.27%
ArrayCopyUnalignedDst.testChar 37 7.027 7.749 ns/op 10.27%
ArrayCopyUnalignedDst.testChar 38 7.013 8.216 ns/op 17.15%