Icculus: EmScripten Audio Conversion Performance In The Web Browser

[liam]

Um did you read the article?
native: 6.5 seconds
Under Chrome without SIMD.js it took 18 seconds
With Firefox, also without SIMD.js, it took about 11 seconds
With Firefox Nightly using SIMD.js and EmScripten building their SSE2 code yielded a time of about 15 seconds.

1.3 times? ok. Nice job cherry picking one synthetic result, where it doesn't suck.

As others have said, we don't know how well the simd.js was employed. Since it was slower and this test maps well to simd it looks something is amiss.
The references I cited were composed of many more tests than a single example of unknown code quality.
However, please take a look at the article Michael released today that includes the pentium and i7700k.
Feel free to share more thorough benchmarks in that thread.

[caligula]

In general? Show me the numbers.
What I've found is the following:

• the i7700k ~ 2.6 faster than the pentium - http://cpu.userbenchmark.com/Compare...00K/3539vs3647
• native clang is about 1.3 times faster than wasm - https://hacks.mozilla.org/2016/10/we...owser-preview/

Um did you read the article?
native: 6.5 seconds
Under Chrome without SIMD.js it took 18 seconds
With Firefox, also without SIMD.js, it took about 11 seconds
With Firefox Nightly using SIMD.js and EmScripten building their SSE2 code yielded a time of about 15 seconds.

1.3 times? ok. Nice job cherry picking one synthetic result, where it doesn't suck.

[liam]

Nobody really wants native code programs these days. Native performance scares them. If you compare those numbers to the OpenCL benchmarks with Intel CPUs, emscripten / asm.js is like buying a $350 CPU (i7 7700k) and executing programs like you had a $60 CPU (Pentium G4400). What a great way to waste power.

In general? Show me the numbers.
What I've found is the following:

• the i7700k ~ 2.6 faster than the pentium - http://cpu.userbenchmark.com/Compare...00K/3539vs3647
• native clang is about 1.3 times faster than wasm - https://hacks.mozilla.org/2016/10/we...owser-preview/

[hansg]

It would be interesting to see a comparison with the original native code, instead of just various emscripten versions.

[ldo17]

There are some secrets of Nature that Man was never meant to tamper with.. .

[caligula]

I agree that it is interesting. It's just the
"Take a C/C++ program and turn it into javascript to run in a browser. What could possibly go wrong?" that scares me.

Nobody really wants native code programs these days. Native performance scares them. If you compare those numbers to the OpenCL benchmarks with Intel CPUs, emscripten / asm.js is like buying a $350 CPU (i7 7700k) and executing programs like you had a $60 CPU (Pentium G4400). What a great way to waste power.

[marco-c]

It's not fully ready yet, but Emscripten already has experimental support for WebAssembly target, and Rust input. See:
* https://users.rust-lang.org/t/compil...mscripten/7627
* https://blog.rust-lang.org/2016/12/22/Rust-1.14.html

Yes, what I was trying to say is that the result of Icculus' benchmarks are not about pure JavaScript, but of something really close to WebAssembly.

[shmerl]

Emscripten converts to asm.js, which is a subset of JavaScript pretty similar to WebAssembly (infact, WebAssembly was inspired by asm.js).

It's not fully ready yet, but Emscripten already has experimental support for WebAssembly target, and Rust input. See:
* https://users.rust-lang.org/t/compil...mscripten/7627
* https://blog.rust-lang.org/2016/12/22/Rust-1.14.html

[marco-c]

Speaking of WebAssembly, I'd much rather see this test when converted to that rather than JavaScript.

Emscripten converts to asm.js, which is a subset of JavaScript pretty similar to WebAssembly (infact, WebAssembly was inspired by asm.js).

I agree that it is interesting. It's just the
"Take a C/C++ program and turn it into javascript to run in a browser. What could possibly go wrong?" that scares me.

Well, it's already been done with several large projects and it's worked pretty well (see Unity, Unreal Engine, etc.).

[starshipeleven]

Why? It's an interesting project. Should help for example compiling Rust into WebAssembly.

I agree that it is interesting. It's just the
"Take a C/C++ program and turn it into javascript to run in a browser. What could possibly go wrong?" that scares me.