Update SampledImageReader.java #122
Conversation
optimized from1Bit(): - Fixed an uncaught abort on failure if subsampling is enabled and the row was being skipped. - Optimized bit to byte expansion by explicitly starting at startx instead of looping over bytes, using the integer sign to extract bits instead of a mask, and not calculating subsampling unless necessary.
|
Do you have a file that shows the speed increase / how did you test? I tried with the file from https://issues.apache.org/jira/browse/PDFBOX-2123 and didn't really notice a difference with PDFDebugger (but it's difficult to make benchmarks). |
|
Yeah as I wrote in the initial comment, it's not really slow. I simply timed the method itself in the sourcecode with System nanoseconds and for the big mask it went down from 55 to something 25ish. Once you go through a few pages of that Alfa catalog PDF you can see consisten times. As I said it wasn't really slow. It was just very wasteful in the original version with not calculating start offsets and all. And the error if readLen!=rowLen was not caught with subsampling if the line was left out. |
|
Fixed in https://issues.apache.org/jira/browse/PDFBOX-5224, thank you! |
…122 git-svn-id: https://svn.apache.org/repos/asf/pdfbox/branches/2.0@1891007 13f79535-47bb-0310-9956-ffa450edef68
2 participants
This fixes an issue with a missed logged error + loop breaking if an erroneous read falls onto a skipped row.
It also improves the speed at least by factor twice (due to sign test instead of bit masking and boolean flag for subsampling).
It will be even faster with regions, as there was a poor design choice of full inner loop computations even if x was below startX.
(The method wasn't very slow to begin with, though.)