Use Workers to do tile decompression and unpremultiply #9208
Conversation
|
Thanks for opening this pull request! Things that will help get your PR across the finish line:
|
Cwiiis
force-pushed
the
messing-with-workers
branch
3 times, most recently
from
3430ea8
to
f6b9b8e
Compare
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| break; | ||
|
|
||
| default: | ||
| console.error('Unrecognised preprocessor message', e); |
Check warning
Code scanning / CodeQL
Log injection Medium
Cwiiis
force-pushed
the
messing-with-workers
branch
from
f6b9b8e
to
c6f8206
Compare
|
|
||
| console.log('MessagePreProcessor initialised', self.fzstd); | ||
|
|
||
| function onMessage(e) { |
Check warning
Code scanning / CodeQL
Missing origin verification in `postMessage` handler Medium
Cwiiis
force-pushed
the
messing-with-workers
branch
4 times, most recently
from
672568b
to
823be3a
Compare
caolanm
force-pushed
the
messing-with-workers
branch
from
823be3a
to
55f1a41
Compare
| lastMask = data[rleMaskOffset++]; | ||
| } | ||
| if (!(lastMask & bitToCheck)) { | ||
| this.lastPixel.set(data.subarray(pixSrc, pixSrc+4)) |
Check notice
Code scanning / CodeQL
Semicolon insertion Note
| break; | ||
|
|
||
| default: | ||
| console.error('Unrecognised preprocessor message', e); |
Check warning
Code scanning / CodeQL
Log injection Medium
| var keyframeBufferSize = 0; | ||
| var keyframeBuffer; | ||
|
|
||
| function onMessage(e) { |
Check warning
Code scanning / CodeQL
Missing origin verification in `postMessage` handler Medium
Pre-allocate decompression buffers and use a Worker to decompress and unpremultiply tile data asynchronously while avoiding memory allocation/ deallocation thrashing. Signed-off-by: Chris Lord <[email protected]> Change-Id: I2282e3df8bf40bdc3a2381880d46df33f2451758
caolanm
force-pushed
the
messing-with-workers
branch
from
55f1a41
to
89d69cd
Compare
|
|
||
| if ('undefined' === typeof window) { | ||
| importScripts('../../node_modules/fzstd/umd/index.js'); | ||
| importScripts('../layer/tile/CanvasTileUtils.js'); |
There was a problem hiding this comment.
I suspect that this has not been tested with a bundlified product build (?) or do we distribute CanvasTileUtils.js on its own ?
| if (alpha === 0) { | ||
| data.fill(0, i, i + 3); | ||
| } else if (alpha !== 255) { | ||
| data[i] = Math.ceil((data[i] * 255) / alpha); |
There was a problem hiding this comment.
We should probably use a 64k byte lookup table here as some of the core LibreOffice code does =)
| lastMask = data[rleMaskOffset++]; | ||
| } | ||
| if (!(lastMask & bitToCheck)) { | ||
| this.lastPixel.set(data.subarray(pixSrc, pixSrc + 4)); |
There was a problem hiding this comment.
I thought we decided that subarray was unfeasibly slow on firefox, and we didn't want to over-use it ?
| @@ -459,7 +484,14 @@ app.definitions.Socket = L.Class.extend({ | |||
| // process so - slurp and then emit at idle - its faster to delay! | |||
| _slurpMessage: function(e) { | |||
| this._extractTextImg(e); | |||
| if (e.image && e.image.rawData && this._worker) { | |||
| this._queueWorkerTileProcessing(e); | |||
There was a problem hiding this comment.
Am I reading this right that we take 'tile:' messages only - and we potentially then emit them out of order with subsequent 'delta:' and 'update:' messages which could arrive before the tiles are de-compressed by the thread - that is likely to lead to corrupted pixels.
Hmm.
|
I'm going to close this as I have a newer, better version of the patch in a different branch. I'll likely need advice resolving some of these comments - I'll try to flag these issues explicitly. |
Summary
Two patches; the first fixes a bug where if we receive new keyframe tiledata and need to ensure the canvas, we end up decompressing/unpremultiplying/putting the old data unnecessarily (I think). The second uses a worker to offload the majority of tile image data decompression and unpremultiplying from the main thread and improves response during large update operations (such as scrolling through a document).
The code is not properly formatted yet, I'm opening this PR now and will update it with fixes once I figure out what those fixes are...
Checklist
make checkmake runand manually verified that everything looks okay