Encoder compress performance

[Slixe]

Hey there!

I don't know if I should do a PR or not because it's kinda weird:
https://github.com/BurntSushi/rust-snappy/blob/master/src/compress.rs#L514
could be simply replaced by a table.fill(0); and give some performance boost in some cases
but also being slower on others benchs.

Here is an idea of results with this change (few examples, not all):

snap/compress/zflat02_jpg/123093                                                                            
                        time:   [3.4687 µs 3.4811 µs 3.4970 µs]
                        thrpt:  [32.782 GiB/s 32.931 GiB/s 33.050 GiB/s]
                 change:
                        time:   [-12.804% -11.879% -11.104%] (p = 0.00 < 0.05)
                        thrpt:  [+12.491% +13.480% +14.684%]
                        Performance has improved.
Found 2 outliers among 50 measurements (4.00%)
  1 (2.00%) high mild
  1 (2.00%) high severe
...
snap/compress/zflat08_txt3/426754                                                                            
                        time:   [826.13 µs 829.73 µs 834.85 µs]
                        thrpt:  [487.49 MiB/s 490.50 MiB/s 492.64 MiB/s]
                 change:
                        time:   [+3.8076% +4.4894% +5.2353%] (p = 0.00 < 0.05)
                        thrpt:  [-4.9749% -4.2965% -3.6679%]
                        Performance has regressed.

but there is side effects on decompress?

snap/decompress/uflat03_jpg_200/200                                                                           
                        time:   [114.52 ns 115.63 ns 116.80 ns]
                        thrpt:  [1.5947 GiB/s 1.6108 GiB/s 1.6265 GiB/s]
                 change:
                        time:   [-4.5188% -3.5631% -2.6538%] (p = 0.00 < 0.05)
                        thrpt:  [+2.7262% +3.6947% +4.7326%]
                        Performance has improved.

snap/decompress/uflat04_pdf/102400                                                                            
                        time:   [3.0975 µs 3.1150 µs 3.1403 µs]
                        thrpt:  [30.369 GiB/s 30.615 GiB/s 30.789 GiB/s]
                 change:
                        time:   [-8.6944% -7.3807% -6.3959%] (p = 0.00 < 0.05)
                        thrpt:  [+6.8329% +7.9688% +9.5223%]
                        Performance has improved.
Found 3 outliers among 50 measurements (6.00%)
  3 (6.00%) high severe
...
snap/decompress/uflat10_pb/118588                                                                            
                        time:   [15.782 µs 15.806 µs 15.833 µs]
                        thrpt:  [6.9757 GiB/s 6.9875 GiB/s 6.9982 GiB/s]
                 change:
                        time:   [+3.0507% +3.3915% +3.6961%] (p = 0.00 < 0.05)
                        thrpt:  [-3.5643% -3.2803% -2.9604%]
                        Performance has regressed.

Not sure what I miss, and if its really worth it to make such change

[BurntSushi]

I would be surprised if your change resulted in a change to the codegen. That's what you should check.

I wrote this crate a very long time ago. Routines like <&[T]>::fill(...) didn't exist. I probably would have used it if it did.