Gif Decompression optimization using larger suffix chunks.
Reported by
tusharkh...@gmail.com,
Jul 2
|
|
Issue descriptionStore LZW dictionary suffixes in larger chunks To write out an LZW dictionary entry each byte requires two table lookups, one of which must complete before the next lookup can begin. This change makes it possible to write up to eight bytes per pair of lookups. Testing with image_decode_bench show performance improvement by 20-25% on elm and kevin. There is an overall improvement in decoding performance. There are some regressions (mostly on little core) when decoding time is very less. Apparently, Benchmarking showed image_decode_bench ran always on one of the big cores. Link to the results of benchmarking done using image_decode_bench: https://docs.google.com/spreadsheets/d/1oh9NSztquw9j6916WxSjdood1iW39T_uvAL1sKIq-TM/edit#gid=0
,
Sep 26
If I click the link, it says the file is deleted.
,
Sep 26
@scroggo: updated benchmark data can be found here: https://docs.google.com/spreadsheets/d/17-jWvxu6CA8B-4Esu3iRlAAQ3wqM0YGAoJ0pnu6FfVU/edit?usp=sharing Includes ARM (big.LITTLE) and x86 (courtesy of Richard Townsend@arm). |
|
►
Sign in to add a comment |
|
Comment 1 by cavalcantii@chromium.org
, Jul 2Owner: cavalcantii@chromium.org
Status: Started (was: Unconfirmed)