Increases chunk size in SMask composition to 1M pixels

[fkaelberer]

This PR increases the chunk size for SMask composition from 64k to 1M pixels to speed up #5515.
Chunking avoids the duplication of large buffers when ctx.getImageData() is called in favor for the duplication of many small buffers. But since calling ctx.[get|put]ImageData() is soo slow, too small chunks slow down SMask processing significantly.
The 1M chunk size makes decoding #5515 several times faster (unless the zoom level is very small). Issue #2648 renders also noticeably quicker.

I was surprised so see that this PR also reduces peak memory consumption:

Chunk Size	Peak Mem
64k	>2000M
1M	~820M
8M	~920M

(using win7 x64, pdf from issue 5515 at zoom=200, http://www.partnerplast.com/images/seismikk/2844_PPMSG_Seismic_REV_012014_PREVEIW.pdf#zoom=200&page=10)

[Snuffleupagus]

/botio test

[pdfjsbot]

From: Bot.io (Windows)

---

Received

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

Live output at: http://107.22.172.223:8877/158e2816d990bed/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/bb05148fdbf247d/output.txt

[pdfjsbot]

From: Bot.io (Windows)

---

Success

Full output at http://107.22.172.223:8877/158e2816d990bed/output.txt

Total script time: 20.69 mins

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

[pdfjsbot]

From: Bot.io (Linux)

---

Success

Full output at http://107.21.233.14:8877/bb05148fdbf247d/output.txt

Total script time: 23.88 mins

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

[CodingFabian]

this might be pure guessing, but afaik this causes the memory to be used on the graphics card instead of main memory.
I stumbled across the pixel limitations a few times and always wondered what the real purpose is. because mask and layer are in memory fully. so what the chunking does is that the get pixel data regions are only created for a smaller subpart, which would affect peak memory rather than total memory use.
64k is pretty low, i think 1M is much more reasonable.
+1 for the change.

[CodingFabian]

Its a 4 times improvement for referenced doc on page 10 for me.
before:
Page: 10
Page Request 15ms
Rendering 4689ms
Overall 4704ms

after:
Page: 10
Page Request 20ms
Rendering 1114ms
Overall 1134ms

[yurydelendik]

Really good. Thanks!