Poor ByteBuffer performance on SPARC

Unaligned byte access conditions resulting in (unnecessarily) poor code on the SPARC architecture. Poor results are demonstrated in e.g., java.nio.ByteBuffers.testByteBufferBulkGet10, where the code executed on SPARC uses a simplistic load-byte, store-byte approach (compared to the merged load-quad used on x86).

Code generation on SPARC could be improved by using the new (S5) support for misaligned loads/stores and improved byte copy layout for inline intrinsics.