|
From: | Antonio Diaz Diaz |
Subject: | Re: [Lzip-bug] Question: parallel lzip? |
Date: | Thu, 25 Jun 2009 16:39:39 +0200 |
User-agent: | Mozilla/5.0 (X11; U; Linux i586; en-US; rv:1.7.11) Gecko/20050905 |
John Reiser wrote:
Even before any code might appear, I am interested in hearing *where* parallelism might be applied. Decompression seems to be inherently serial
See the code of lzip 1.7. I have removed prev_byte and now copy_block can be executed in parallel with decoding of the next distance-length pair. Not sure if it will be worth the added complexity.
Actual encoding also seems to be serial by nature. Only the finding and the ranking of matches with previous substrings appears to be fruitful for parallelism. This activity ought to be similar across all LZ schemes, including Mark Adler's "pigz" code for gzip: http://www.zlib.net/pigz/
Yes, match finding is my primary target. Thanks for the link. I'll take a look at it as soon as I can.
Regards, Antonio.
[Prev in Thread] | Current Thread | [Next in Thread] |