[Zlib-devel] infnew-5 available for testing

[Greg Roelofs]

Chris Anderson wrote:

> Just did a quick minigzip run with 5 (icc):

> $ time bin/minigzip -d < test.gz > /dev/null

> real    0m12.998s
> user    0m12.418s
> sys     0m0.408s

I did a similar test with a 313 MB archive on a Linux box, gcc 2.95.3,
1.400 GHz Athlon XP:

meltdown:/util/compression/zlib-1.1.4-infnew 226> time ./minigzip-static.stock -d < /backups/quantum-TM3200A-archive.tgz  > /dev/null
14.760u 1.000s 0:19.88 79.2%    0+0k 0+0io 115pf+0w
14.940u 1.110s 0:19.86 80.8%    0+0k 0+0io 115pf+0w
14.580u 1.110s 0:19.69 79.6%    0+0k 0+0io 115pf+0w
14.720u 1.040s 0:19.40 81.2%    0+0k 0+0io 115pf+0w
14.630u 0.920s 0:19.64 79.1%    0+0k 0+0io 115pf+0w

meltdown:/util/compression/zlib-1.1.4-infnew 231> time ./minigzip-static -d < /backups/quantum-TM3200A-archive.tgz > /dev/null
13.830u 1.060s 0:19.06 78.1%    0+0k 0+0io 112pf+0w
13.770u 1.300s 0:19.00 79.3%    0+0k 0+0io 112pf+0w
13.680u 1.250s 0:19.08 78.2%    0+0k 0+0io 112pf+0w
14.030u 0.950s 0:18.84 79.5%    0+0k 0+0io 112pf+0w
13.980u 1.000s 0:19.00 78.8%    0+0k 0+0io 112pf+0w

meltdown:/util/compression/zlib-1.1.4-infnew 244> time ./ung-static < /backups/quantum-TM3200A-archive.tgz > /dev/null
13.220u 1.160s 0:18.33 78.4%    0+0k 0+0io 99pf+0w
13.280u 1.110s 0:18.26 78.8%    0+0k 0+0io 99pf+0w
13.480u 1.020s 0:18.14 79.9%    0+0k 0+0io 99pf+0w
13.350u 1.030s 0:18.27 78.7%    0+0k 0+0io 99pf+0w
13.280u 1.020s 0:18.25 78.3%    0+0k 0+0io 99pf+0w

16% reduction in page faults for ung.c, 7.3% faster (wallclock).

This version of gcc has significantly poorer code-generation than the
3.x series, but I haven't had time to compile a new version yet.

Greg