Javascript version

[wshager]

It would be interesting to see if a javascript version of this library could be created using SIMD.js.

https://developer.mozilla.org/nl/docs/Web/JavaScript/Reference/Global_Objects/SIMD
https://github.com/tc39/ecmascript_simd
https://github.com/jonbrennecke/node-simd

[lemire]

Yes, this looks possible now.

[wshager]

The Emscripten compilation of this lib also look promising, but I don't know what a dynamic call from javascript would look like. Emscripten generates a lot of constants and heap allocations.

[lemire]

@wshager

I am somewhat pessimistic regarding any approach that does not translate into SIMD instructions at the machine level.

[wshager]

Please note that SIMD is currently in the crosswalk V8 fork:
https://github.com/crosswalk-project/v8-crosswalk

[wshager]

Hmm perhaps you are right, nevermind. I can't find it in the repo.
It may be too premature, except maybe for Firefox Nightly.

[wshager]

Continuing from #10, the performance of the first example compiled to javascript with Emscripten performs well, but performance of other 2 tests is extremely poor. It seems SIMD.js is nowhere near being compatible with this project.

[lemire]

@wshager

Can you tell us which examples fare well and which ones fare badly?

[wshager]

When I compiled previously with my smmintrin shim the first two test would run, but simple_demo would hang, currently the second example didn't finish in any acceptable time, only the first compress_decompress example finishes in ff nightly.

[lemire]

What happens after this last commit b7e1ecc

?

[wshager]

@lemire yes, that script does finish, in about 8 seconds in FF nightly on my i7-3612QM (8GB).

[lemire]

Ok. Can you share the resulting output (just so we can document it).

[wshager]

@lemire sorry, I did, but I replied to the commit. Here goes:
http://lagua.nl/uploadfiles/example.html
http://lagua.nl/uploadfiles/example.uncompressed.html

[wshager]

example.js.txt
example.uncompressed.js.txt

[lemire]

@wshager

Thanks, but I am inviting you to publish the performance results you got.

My concern right now is whether SIMD instructions are used at all. On my laptop, the script completes, but very slowly and with speeds that indicate that SIMD instructions are not used.