Hey, it looks like in the past month or so, the various blockers to running
SSE1 with Emscripten have been disappearing, so thanks to Dan Gohman,
Ningxin Hu and others who've spent time adding this support to Emscripten!

Here's an interesting chart showing a microbenchmark of synthetic SSE1
performance when compiling natively and in Firefox Nightly:

https://dl.dropboxusercontent.com/u/40949268/emcc/results_sse1_2015_03_18.html

We are definitely getting in the ballpark in terms of performance and SSE1
support is becoming more complete and you can see the advantages SIMD has
over scalar execution. Note though that the JS SIMD spec is not
standardized yet, and you will need experimental Firefox Nightly and Chrome
dev channels to test these features on.

The numbers in the above graph can have some amount of noise in them so
don't take the specific numbers too seriously. If you want to run the
benchmark yourself, try the latest Emscripten incoming branch and type

    python tests/benchmark_sse1.py

in the Emscripten root directory.

In the coming months, the SIMD support will be gradually extended to cover
SSE2 as well, which will expand the number of SIMD codebases that can be
ported with Emscripten. If you have your own codebases that utilize the
SSE1 instruction set, please do give those a test, and report back on the
bugs and performance issues that you might find.

Cheers,
   Jukka

-- 
You received this message because you are subscribed to the Google Groups 
"emscripten-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to