[Zlib-devel] Google's version of zlib code

[glennrp at comcast.net]

----- Frédéric Kayser <f.kayser at free.fr> wrote:
> Hi Glenn,
> could you rerun it but force the filter 0 (none) this time? I'm no sure bzip2 and xz really benefit from the filters.

      Method           time   IDAT bytes
    original files      n/a    5019516

    pngcrush -brute   310.8    4281038
    gzip -9n            5.6    4339189
    zopfli --zlib     596.7    4035633
    bzip2 -9            3.8    4015342
    xz                 14.6    3813468

    pngcrush default   24.5    4299905
    gzip -9n            5.5    4349009
    zopfli --zlib     590.4    4047719
    bzip2 -9            3.6    4022529
    xz                 14.5    3820192

    pngcrush none      18.6    5136096
    zopfli --zlib      98.9    4864938
    gzip -n9            2.8    5132487
    bzip2 -9            3.8    4614708
    xz                 12.8    4138560

    pngcrush adaptive  17.4    4763290
    zopfli --zlib      99.0    5153680
    gzip -n9            2.8    4752190
    bzip2 -9            2.6    5492706
    xz                 15.2    4231288

Evidently the filters make the files more compressible for
all methods.  The more highly-compressible ones take
much longer with zopfli than the less-compressible ones.

xz is clearly the winner in all of these cases.

Glenn