Hi.

This patch builds bzip2 with profile guided optimization, the --combine 
option, and -fomit-frame-pointer:
http://www.linuxfromscratch.org/~robert/new/patches/bzip2-1.0.5-faster.diff

The non-shared bzip2 program decompresses 7% faster for me, without accounting 
for -fomit-frame-pointer. The libbz2.so library runs 4.5% faster for me. The 
profiling optimization works a lot better when the all the code is built 
together, so the non-shared bzip2 program is suggested. So:
cp -v bzip2 /bin/bzip2 # Not bzip2-shared

I also tried other options like -fwhole-program, -O3, and -finline-functions, 
and they made bzip2 and the library slower. My tests were only with 
decompression, of a gcc tarball, but compression should also be faster.

There are also other bzip2 packages that do parallel/multithread compression, 
but not parallel decompression.

robert

Attachment: pgpuvuWU0zIr0.pgp
Description: PGP signature

-- 
http://linuxfromscratch.org/mailman/listinfo/lfs-chat
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page

Reply via email to