faster decoding

[jerch]

The PR introduces a webassembly decoder for faster decoding. Speedup is 3-4x.

Also fixes #17.

TODO:

• reconcile decoder interface
  • remove BE handling (needed to avoid second wasm build)
  • remove old decoder #23 - postponed
• more test cases
• more prebundles
• set up gh actions as CI
• fix example scripts
• polish docs

[jerch]

@PerBothner Just finished the rework of the decoder details and some of the bundling aspects. It is still not fully finished for release (tests missing, old decoder to be removed, docs cleanup). In case you want to test things early, you can do that as follows:

git clone git@github.com:jerch/node-sixel.git
cd node-sixel
git fetch origin pull/20/head:faster_decode
git checkout faster_decode
npm install

(This will take some time to install emscripten.)

If all went well, you'll find under /dist several browser bundles. I guess you are only interested in the decoder part, thus the decode.umd.js (umd build) or decode.esm.js (es6 module) should do.

Note that the decoder interface changed alot (mostly simplified, for now check in Decoder.ts).

On my machine these are the benchmark results:

   Context "./lib/index.benchmark.js"
      Context "testimage"
         Context "pixel transfer"
            Case "toPixelData - with fillColor" : 20 runs - average runtime: 1.48 ms
            Case "toPixelData - without fillColor" : 20 runs - average runtime: 0.87 ms
         Context "decode (DefaultDecoder)"
            Case "decode" : 20 runs - average runtime: 3.83 ms
            Case "decodeString" : 20 runs - average runtime: 4.11 ms
            Case "decode + pixel transfer" : 20 runs - average runtime: 3.09 ms
         Context "decode (WasmDecoder)"
            Case "decode" : 20 runs - average runtime: 0.76 ms
            Case "decodeString" : 20 runs - average runtime: 1.48 ms
         Context "encode"
            Case "sixelEncode" : 20 runs - average runtime: 21.57 ms
      Context "decode - testfiles (DefaultDecoder)"
         Case "test1_clean.sixel" : 20 runs - average runtime: 16.20 ms
         Case "test1_clean.sixel" : 20 runs - average throughput: 38.57 MB/s
         Case "test2_clean.sixel" : 20 runs - average runtime: 6.49 ms
         Case "test2_clean.sixel" : 20 runs - average throughput: 48.75 MB/s
         Case "sampsa_reencoded_clean.six" : 20 runs - average runtime: 15.76 ms
         Case "sampsa_reencoded_clean.six" : 20 runs - average throughput: 40.98 MB/s
         Case "FullHD 12bit noise" : 20 runs - average runtime: 224.61 ms
         Case "FullHD 12bit noise" : 20 runs - average throughput: 69.03 MB/s
      Context "decode - testfiles (WasmDecoder)"
         Case "test1_clean.sixel" : 20 runs - average runtime: 3.89 ms
         Case "test1_clean.sixel" : 20 runs - average throughput: 152.63 MB/s
         Case "test2_clean.sixel" : 20 runs - average runtime: 1.91 ms
         Case "test2_clean.sixel" : 20 runs - average throughput: 165.01 MB/s
         Case "sampsa_reencoded_clean.six" : 20 runs - average runtime: 4.47 ms
         Case "sampsa_reencoded_clean.six" : 20 runs - average throughput: 146.42 MB/s
         Case "FullHD 12bit noise" : 20 runs - average runtime: 48.53 ms
         Case "FullHD 12bit noise" : 20 runs - average throughput: 319.51 MB/s

(DefaultDecoder is the old one, WasmDecoder the new one.)
Speedup is around 3-4 times now. 😸