Faster JBIG2 bitmap decoding #5114
Conversation
| @@ -159,6 +219,13 @@ var Jbig2Image = (function Jbig2ImageClosure() { | |||
| error('JBIG2 error: MMR encoding is not supported'); | |||
| } | |||
|
|
|||
| // Use optimized version for the most common case | |||
| if (templateIndex === 0 && !useskip && !prediction && at.length === 4 && | |||
There was a problem hiding this comment.
Isn't useskip not declared until line 229?
There was a problem hiding this comment.
Ouch. Thanks, I'll fix that.
|
Fixed unnecessary newlines. |
|
I noticed that an extra for loop for the case of i < 2 in my previous commit was not necessary. This commit does the same but is 20 lines shorter. |
|
Each time I commit the code I think it is as simple it can get, and nothing can be improved. |
|
/botio test |
From: Bot.io (Windows)ReceivedCommand cmd_test from @timvandermeij received. Current queue size: 0 Live output at: http://107.22.172.223:8877/da966027eba9712/output.txt |
From: Bot.io (Linux)ReceivedCommand cmd_test from @timvandermeij received. Current queue size: 0 Live output at: http://107.21.233.14:8877/dbe0a25ad6f39a6/output.txt |
From: Bot.io (Windows)SuccessFull output at http://107.22.172.223:8877/da966027eba9712/output.txt Total script time: 19.62 mins
|
From: Bot.io (Linux)SuccessFull output at http://107.21.233.14:8877/dbe0a25ad6f39a6/output.txt Total script time: 22.65 mins
|
|
/botio-linux preview |
From: Bot.io (Linux)ReceivedCommand cmd_preview from @timvandermeij received. Current queue size: 0 Live output at: http://107.21.233.14:8877/94cea4f3c3450e3/output.txt |
From: Bot.io (Linux)SuccessFull output at http://107.21.233.14:8877/94cea4f3c3450e3/output.txt Total script time: 0.76 mins Published
|
|
BTW: There's a FF regression in nightly 34 (Linux only): The university logo in scanned_example_1.pdf turns black. It's not caused by this PR. |
|
@fkaelberer Please open a separate issue for that. Thank you! |
Faster JBIG2 bitmap decoding
|
Looks good thanks |
|
Reported upstream: https://bugzilla.mozilla.org/show_bug.cgi?id=1050788 |
While debugging the JBIG2 code I noticed that jbig2dec and xpdf both have special code paths for each of the four standard templates, for performance reasons. And indeed, the code can be simplified a lot if the context template is known.
In this PR, I wrote the decoding procedure from scratch for the case of the most common standard template and the
atpixels at the standard location (which is the case for most jbig2 documents I have). In this case, each pixel in the main loop can be decoded without looping or branching, and it is much easier to understand what's happening.The decoding time of #3816 and #2909 and others is reduced by ~25%.