Use FFM API instead of Unsafe

[wendigo]

Raw version - will add some cleanup/refactors

[wendigo]

cc @dain

[wendigo]

MemoryCopyBenchmark.copy      ARRAY_COPY          0  thrpt   10  474491681.129 ±  3265306.927  ops/s
MemoryCopyBenchmark.copy      ARRAY_COPY         32  thrpt   10  285042279.681 ±  2462686.492  ops/s
MemoryCopyBenchmark.copy      ARRAY_COPY        128  thrpt   10  159996540.251 ± 15054813.264  ops/s
MemoryCopyBenchmark.copy      ARRAY_COPY        512  thrpt   10   72404594.618 ±   489402.830  ops/s
MemoryCopyBenchmark.copy      ARRAY_COPY       1024  thrpt   10   47357057.119 ±  1201945.213  ops/s
MemoryCopyBenchmark.copy      ARRAY_COPY    1048576  thrpt   10      41088.703 ±     4961.142  ops/s
MemoryCopyBenchmark.copy      ARRAY_COPY  134217728  thrpt   10        312.411 ±       12.321  ops/s
MemoryCopyBenchmark.copy           SLICE          0  thrpt   10  295805908.140 ±  2085091.805  ops/s
MemoryCopyBenchmark.copy           SLICE         32  thrpt   10  189399971.406 ±  3461827.652  ops/s
MemoryCopyBenchmark.copy           SLICE        128  thrpt   10  140171047.804 ±   977496.348  ops/s
MemoryCopyBenchmark.copy           SLICE        512  thrpt   10   76250723.985 ±  2047258.969  ops/s
MemoryCopyBenchmark.copy           SLICE       1024  thrpt   10   49895982.908 ±   543046.942  ops/s
MemoryCopyBenchmark.copy           SLICE    1048576  thrpt   10      55195.630 ±     2027.515  ops/s
MemoryCopyBenchmark.copy           SLICE  134217728  thrpt   10        226.097 ±        4.955  ops/s
MemoryCopyBenchmark.copy     CUSTOM_LOOP          0  thrpt   10  541104779.380 ±  4493537.557  ops/s
MemoryCopyBenchmark.copy     CUSTOM_LOOP         32  thrpt   10  265335751.007 ±  2960040.019  ops/s
MemoryCopyBenchmark.copy     CUSTOM_LOOP        128  thrpt   10  127626464.503 ± 15330290.165  ops/s
MemoryCopyBenchmark.copy     CUSTOM_LOOP        512  thrpt   10   52954182.844 ±  1601716.854  ops/s
MemoryCopyBenchmark.copy     CUSTOM_LOOP       1024  thrpt   10   31604216.249 ±   248654.945  ops/s
MemoryCopyBenchmark.copy     CUSTOM_LOOP    1048576  thrpt   10      26121.004 ±     2754.518  ops/s
MemoryCopyBenchmark.copy     CUSTOM_LOOP  134217728  thrpt   10        198.818 ±        3.548  ops/s
MemoryCopyBenchmark.copy          UNSAFE          0  thrpt   10  331847053.122 ±   539142.963  ops/s
MemoryCopyBenchmark.copy          UNSAFE         32  thrpt   10  187917069.206 ±  7990464.460  ops/s
MemoryCopyBenchmark.copy          UNSAFE        128  thrpt   10  133080275.413 ±   589945.484  ops/s
MemoryCopyBenchmark.copy          UNSAFE        512  thrpt   10   76809746.683 ±   565967.987  ops/s
MemoryCopyBenchmark.copy          UNSAFE       1024  thrpt   10   49797502.548 ±   480430.134  ops/s
MemoryCopyBenchmark.copy          UNSAFE    1048576  thrpt   10      52984.040 ±     3870.650  ops/s
MemoryCopyBenchmark.copy          UNSAFE  134217728  thrpt   10        221.023 ±       22.489  ops/s
MemoryCopyBenchmark.copy             FFM          0  thrpt   10  311388867.857 ±  1783603.147  ops/s
MemoryCopyBenchmark.copy             FFM         32  thrpt   10  212324332.813 ±  1894856.373  ops/s
MemoryCopyBenchmark.copy             FFM        128  thrpt   10  149091866.397 ± 17204132.219  ops/s
MemoryCopyBenchmark.copy             FFM        512  thrpt   10   72916935.083 ±   482069.949  ops/s
MemoryCopyBenchmark.copy             FFM       1024  thrpt   10   47854814.809 ±   350433.072  ops/s
MemoryCopyBenchmark.copy             FFM    1048576  thrpt   10      42325.853 ±     2703.085  ops/s
MemoryCopyBenchmark.copy             FFM  134217728  thrpt   10        310.401 ±       19.358  ops/s

[wendigo]

I got it working but I'm worried about performance. @dain can you take a look?