Faster JPX decoding #4264

fkaelberer · 2014-02-08T19:23:17Z

This PR reduces decoding time of JPX images by some 10 to 25 percent, and so improves #2790 and #2103 a little bit.

Roughly 50 % of JPX decoding time is spend for (arithmetic) decoding of the byte stream, and the remaining 50 % is (mostly) spend on the wavelet transform. This PR concerns wavelet transform.
During the wavelet transform, rows and columns of the images are processed individually, and the bottleneck is mainly memory copy performance to access rows and cols.

This PR increases memory performance by

using Array.set() when reading rows and
reading multiple columns at a time to avoid cache misses:

      // Accesses to the items array can take long, because it may not fit into
      // CPU cache and has to be fetched from main memory. Since subsequent
      // accesses to the items array are not local when reading columns, we
      // have a cache miss every time. To reduce cache misses, get up to
      // 'numBuffers' items at a time and store them into the individual
      // buffers. The colBuffers should be small enough to fit into CPU cache.

Using different settings, I measured the following timings. Numbers are averaged over the first 4 pages of #2790, but are similar for other JPX-heavy pdfs.

5201ms --> 3827ms (-26%) Firefox Nightly 30a1 Linux (in a VM)
4681ms --> 3951ms (-16%) Firefox Nightly 30a1 Win7-64
6704ms --> 5649ms (-16%) Chrome 34 dev Linux (in a VM)
5538ms --> 4797ms (-13%) Chrome 34 dev Win7-64

Snuffleupagus · 2014-02-08T19:35:54Z

/botio test

pdfjsbot · 2014-02-08T19:35:56Z

From: Bot.io (Linux)

Received

Command cmd_test from @Snuffleupagus received. Current queue size: 0

Live output at: http://107.21.233.14:8877/1a2e69cdc7f0d44/output.txt

pdfjsbot · 2014-02-08T19:35:56Z

From: Bot.io (Windows)

Received

Command cmd_test from @Snuffleupagus received. Current queue size: 0

Live output at: http://107.22.172.223:8877/259980401037aca/output.txt

pdfjsbot · 2014-02-08T19:57:15Z

From: Bot.io (Linux)

Failed

Full output at http://107.21.233.14:8877/1a2e69cdc7f0d44/output.txt

Total script time: 21.33 mins

Font tests: Passed
Unit tests: Passed
Regression tests: FAILED

Image differences available at: http://107.21.233.14:8877/1a2e69cdc7f0d44/reftest-analyzer.xhtml#web=eq.log

Snuffleupagus · 2014-02-08T19:59:33Z

Since Chrome timed out, let's re-run the test.

/botio-linux test

pdfjsbot · 2014-02-08T19:59:37Z

From: Bot.io (Linux)

Received

Command cmd_test from @Snuffleupagus received. Current queue size: 0

Live output at: http://107.21.233.14:8877/bac70aa27711e01/output.txt

pdfjsbot · 2014-02-08T20:14:09Z

From: Bot.io (Windows)

Success

Full output at http://107.22.172.223:8877/259980401037aca/output.txt

Total script time: 38.20 mins

Font tests: Passed
Unit tests: Passed
Regression tests: Passed

pdfjsbot · 2014-02-08T20:23:23Z

From: Bot.io (Linux)

Success

Full output at http://107.21.233.14:8877/bac70aa27711e01/output.txt

Total script time: 23.76 mins

Font tests: Passed
Unit tests: Passed
Regression tests: Passed

timvandermeij · 2014-02-08T20:34:04Z

/botio-linux preview

pdfjsbot · 2014-02-08T20:34:05Z

From: Bot.io (Linux)

Received

Command cmd_preview from @timvandermeij received. Current queue size: 0

Live output at: http://107.21.233.14:8877/a196da3bbb51d1d/output.txt

pdfjsbot · 2014-02-08T20:34:19Z

From: Bot.io (Linux)

Success

Full output at http://107.21.233.14:8877/a196da3bbb51d1d/output.txt

Total script time: 0.23 mins

Published

Viewer: http://107.21.233.14:8877/a196da3bbb51d1d/web/viewer.html
B2G Viewer: http://107.21.233.14:8877/a196da3bbb51d1d/extensions/b2g/content/web/viewer.html
Extension: http://107.21.233.14:8877/a196da3bbb51d1d/extensions/firefox/pdf.js.xpi
Extension (AMO): http://107.21.233.14:8877/a196da3bbb51d1d/extensions/firefox/pdf.js.amo.xpi

timvandermeij · 2014-02-08T20:44:28Z

@fkaelberer It is indeed loading faster, but I notice the page loading indicators stop spinning when the pages are loading. This isn't the case with the current viewer. Can you reproduce that?

fkaelberer · 2014-02-08T21:29:33Z

No... at least not on Firefox / Windows / pdf.js.xpi

timvandermeij · 2014-02-09T12:18:08Z

@fkaelberer I see that I should have been more specific, so I'm not sure we were talking about the same issue. I meant that the loading indicator shows up, but doesn't spin when loading pages from the bird book PDF from #2790. The default (Tracemonkey) PDF for example is fine, only PDFs with JPX images seem to be affected.

Steps to reproduce for me are:

Download the bird book PDF from Slow rendering of the books from Archive.org #2790.
Open it with http://107.21.233.14:8877/a196da3bbb51d1d/web/viewer.html. Notice the page loaders not spinning when a page is loading.
Open the same file with http://mozilla.github.io/pdf.js/web/viewer.html. Notice the page loaders spinning when a page is loading.

My configuration: Windows 7 x64, Firefox Nightly 30.0a1 (HWA on, 64 bit) and the viewer from the generated preview above.

Snuffleupagus · 2014-02-09T12:41:37Z

@timvandermeij Following your STR, I'm not able to reproduce that issue. The spinners work fine even with this patch applied.

Configuration: Windows 7 (64-bit), Firefox Nightly 30 (buildID: 20140208030207) with HWA on.

timvandermeij · 2014-02-09T15:45:39Z

Hm, I cleared the cache again and now it's working. That's strange because I have cache disabled by default, but I'm happy that it's not an issue caused by this patch!

yurydelendik · 2014-02-10T18:47:42Z

src/core/jpx.js

-        for (var v = 0; v < height; v++, k += width, l++)
-          buffer[l] = items[k];
+        // if we ran out of buffers, copy several image columns at once
+        if (currentBuffer === buffersCached) {


To improve readability, can you introduce e.g. buffersToProcess that will be set to buffersCached, reduced after var buffer = colBuffers[currentBuffer]; and be used detect when to flush or populate the buffer? (As alternative currentBuffer can be used but in reverse order)

yurydelendik · 2014-02-10T18:49:36Z

Looks good with the changes above.

fkaelberer · 2014-02-12T15:53:18Z

All right, I'll do that in the next few days.

fkaelberer · 2014-02-24T18:06:44Z

Ok, there you go. I made currentBuffer run in reversed order, so the conditions to fill and flush the buffers are indeed much easier to read. Also, buffersCached is not needed anymore.

timvandermeij · 2014-02-24T19:00:52Z

/botio-linux preview

pdfjsbot · 2014-02-24T19:00:53Z

From: Bot.io (Linux)

Received

Command cmd_preview from @timvandermeij received. Current queue size: 0

Live output at: http://107.21.233.14:8877/df95c2df5e844f8/output.txt

pdfjsbot · 2014-02-24T19:01:06Z

From: Bot.io (Linux)

Success

Full output at http://107.21.233.14:8877/df95c2df5e844f8/output.txt

Total script time: 0.22 mins

Published

Viewer: http://107.21.233.14:8877/df95c2df5e844f8/web/viewer.html
B2G Viewer: http://107.21.233.14:8877/df95c2df5e844f8/extensions/b2g/content/web/viewer.html
Extension: http://107.21.233.14:8877/df95c2df5e844f8/extensions/firefox/pdf.js.xpi
Extension (AMO): http://107.21.233.14:8877/df95c2df5e844f8/extensions/firefox/pdf.js.amo.xpi

timvandermeij · 2014-02-24T19:04:11Z

/botio test

pdfjsbot · 2014-02-24T19:04:12Z

From: Bot.io (Linux)

Received

Command cmd_test from @timvandermeij received. Current queue size: 0

Live output at: http://107.21.233.14:8877/40ad25e4b57ae91/output.txt

pdfjsbot · 2014-02-24T19:04:12Z

From: Bot.io (Windows)

Received

Command cmd_test from @timvandermeij received. Current queue size: 0

Live output at: http://107.22.172.223:8877/15a3c3c609fe2db/output.txt

pdfjsbot · 2014-02-24T19:27:49Z

From: Bot.io (Linux)

Failed

Full output at http://107.21.233.14:8877/40ad25e4b57ae91/output.txt

Total script time: 23.62 mins

Font tests: Passed
Unit tests: Passed
Regression tests: FAILED

Image differences available at: http://107.21.233.14:8877/40ad25e4b57ae91/reftest-analyzer.xhtml#web=eq.log

pdfjsbot · 2014-02-24T19:39:27Z

From: Bot.io (Windows)

Success

Full output at http://107.22.172.223:8877/15a3c3c609fe2db/output.txt

Total script time: 35.24 mins

Font tests: Passed
Unit tests: Passed
Regression tests: Passed

Faster JPX decoding

yurydelendik · 2014-02-24T19:54:03Z

Thank you for the patch.

yurydelendik reviewed Feb 10, 2014
View reviewed changes

Faster JPX decoding

5f8546c

yurydelendik added a commit that referenced this pull request Feb 24, 2014

Merge pull request #4264 from fkaelberer/FasterJPXdecoding

5de3e55

Faster JPX decoding

yurydelendik merged commit 5de3e55 into mozilla:master Feb 24, 2014

fkaelberer deleted the FasterJPXdecoding branch February 26, 2014 08:28

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Faster JPX decoding #4264

Faster JPX decoding #4264

fkaelberer commented Feb 8, 2014

Snuffleupagus commented Feb 8, 2014

pdfjsbot commented Feb 8, 2014

pdfjsbot commented Feb 8, 2014

pdfjsbot commented Feb 8, 2014

Snuffleupagus commented Feb 8, 2014

pdfjsbot commented Feb 8, 2014

pdfjsbot commented Feb 8, 2014

pdfjsbot commented Feb 8, 2014

timvandermeij commented Feb 8, 2014

pdfjsbot commented Feb 8, 2014

pdfjsbot commented Feb 8, 2014

timvandermeij commented Feb 8, 2014

fkaelberer commented Feb 8, 2014

timvandermeij commented Feb 9, 2014

Snuffleupagus commented Feb 9, 2014

timvandermeij commented Feb 9, 2014

yurydelendik Feb 10, 2014

yurydelendik commented Feb 10, 2014

fkaelberer commented Feb 12, 2014

fkaelberer commented Feb 24, 2014

timvandermeij commented Feb 24, 2014

pdfjsbot commented Feb 24, 2014

pdfjsbot commented Feb 24, 2014

timvandermeij commented Feb 24, 2014

pdfjsbot commented Feb 24, 2014

pdfjsbot commented Feb 24, 2014

pdfjsbot commented Feb 24, 2014

pdfjsbot commented Feb 24, 2014

yurydelendik commented Feb 24, 2014

Faster JPX decoding #4264

Faster JPX decoding #4264

Conversation

fkaelberer commented Feb 8, 2014

Snuffleupagus commented Feb 8, 2014

pdfjsbot commented Feb 8, 2014

From: Bot.io (Linux)

Received

pdfjsbot commented Feb 8, 2014

From: Bot.io (Windows)

Received

pdfjsbot commented Feb 8, 2014

From: Bot.io (Linux)

Failed

Snuffleupagus commented Feb 8, 2014

pdfjsbot commented Feb 8, 2014

From: Bot.io (Linux)

Received

pdfjsbot commented Feb 8, 2014

From: Bot.io (Windows)

Success

pdfjsbot commented Feb 8, 2014

From: Bot.io (Linux)

Success

timvandermeij commented Feb 8, 2014

pdfjsbot commented Feb 8, 2014

From: Bot.io (Linux)

Received

pdfjsbot commented Feb 8, 2014

From: Bot.io (Linux)

Success

Published

timvandermeij commented Feb 8, 2014

fkaelberer commented Feb 8, 2014

timvandermeij commented Feb 9, 2014

Snuffleupagus commented Feb 9, 2014

timvandermeij commented Feb 9, 2014

yurydelendik Feb 10, 2014

Choose a reason for hiding this comment

yurydelendik commented Feb 10, 2014

fkaelberer commented Feb 12, 2014

fkaelberer commented Feb 24, 2014

timvandermeij commented Feb 24, 2014

pdfjsbot commented Feb 24, 2014

From: Bot.io (Linux)

Received

pdfjsbot commented Feb 24, 2014

From: Bot.io (Linux)

Success

Published

timvandermeij commented Feb 24, 2014

pdfjsbot commented Feb 24, 2014

From: Bot.io (Linux)

Received

pdfjsbot commented Feb 24, 2014

From: Bot.io (Windows)

Received

pdfjsbot commented Feb 24, 2014

From: Bot.io (Linux)

Failed

pdfjsbot commented Feb 24, 2014

From: Bot.io (Windows)

Success

yurydelendik commented Feb 24, 2014