Avoid out of range array access in JBIG2 decoder

[fkaelberer]

Fixes #5663. Chrome rendering times are down from 4 seconds to <1.5s.

Benchmark with document from #2909:

browser	Count	Baseline(ms)	Current(ms)	+/-	%	Result(P<.05)
chrome	180	3215	997	-2218	-69.00	faster
firefoxNightly	180	1156	1139	-18	-1.53

[Snuffleupagus]

/botio test

[pdfjsbot]

From: Bot.io (Windows)

---

Received

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

Live output at: http://107.22.172.223:8877/5ada5a015667842/output.txt

[pdfjsbot]

From: Bot.io (Linux)

---

Received

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

Live output at: http://107.21.233.14:8877/fb7bce13911d1bd/output.txt

[pdfjsbot]

From: Bot.io (Windows)

---

Failed

Full output at http://107.22.172.223:8877/5ada5a015667842/output.txt

Total script time: 3.84 mins

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

Image differences available at: http://107.22.172.223:8877/5ada5a015667842/reftest-analyzer.html#web=eq.log

[pdfjsbot]

From: Bot.io (Linux)

---

Success

Full output at http://107.21.233.14:8877/fb7bce13911d1bd/output.txt

Total script time: 23.03 mins

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

[yurydelendik]

/botio-windows test

[pdfjsbot]

From: Bot.io (Windows)

---

Received

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

Live output at: http://107.22.172.223:8877/768c15980bd259d/output.txt

[pdfjsbot]

From: Bot.io (Windows)

---

Success

Full output at http://107.22.172.223:8877/768c15980bd259d/output.txt

Total script time: 17.93 mins

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

[timvandermeij]

/botio-linux preview

[pdfjsbot]

From: Bot.io (Linux)

---

Received

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

Live output at: http://107.21.233.14:8877/cdce1ac6b2db209/output.txt

[pdfjsbot]

From: Bot.io (Linux)

---

Success

Full output at http://107.21.233.14:8877/cdce1ac6b2db209/output.txt

Total script time: 0.78 mins

Published

• Viewer: http://107.21.233.14:8877/cdce1ac6b2db209/web/viewer.html
• B2G Viewer: http://107.21.233.14:8877/cdce1ac6b2db209/extensions/b2g/content/web/viewer.html
• Extension: http://107.21.233.14:8877/cdce1ac6b2db209/extensions/firefox/pdf.js.xpi

[timvandermeij]

Really nice, thank you!

[CodingFabian]

I was just looking at the code, and I wondered if you have tested unrolling that magic or yourself

contextLabel = ((contextLabel & OLD_PIXEL_MASK) << 1) |
                         (j + 3 < width ? row2[j + 3] << 11 : 0) |
                       (j + 4 < width ? row1[j + 4] << 4 : 0) | pixel;

could be

contextLabel = (contextLabel & OLD_PIXEL_MASK) << 1;
if (contextLabel) {
  continue;
} 
if (j + 3 < width) {
 contextLabel = row2[j + 3] << 11;
 if (contextLabel) continue;
}
if (j + 4 < width) {
 contextLabel = row1[j + 4] << 4;
 if (contextLabel) continue;
}
contextLabel = pixel;

I dont have the setup handy, maybe if thats interesting for you, try it out and let me know. I know the code looks way uglier than before, but maybe its even worth.

[fkaelberer]

@CodingFabian Before I committed this fix, I have tested similar code to avoid the branching in the j + x < width branching in the loop, but the performance gain was below 1 or 2% for ~15 additional lines of code. But I preferred simplicity/readability over the marginal speed gain. And don't confuse | with ||: you can't continue early if contextLabel is != 0 :-]