SIMD optimizations

[karlzwinnen]

As Liquidfun uses SIMD for its particle system, i was wondering if there are any plans to optimize the box2d part as well?

It seems there already is a fork of box2d available which makes use of SIMD (and also enables this for SIMD.js through emscripten): http://peterjensen.github.io/html5-box2d/
The source seems to be hosted here: https://github.com/huningxin/box2d.js/tree/sse

Would it be difficult to merge this back in the box2d core liquidfun uses?

[jsanmiya]

Hey Karizwinnen,

That's a cool fork of box2d. It looks like they optimized a few key
functions with intrinsics
https://github.com/huningxin/box2d.js/blob/sse/Box2D_v2.2.1/Box2D/Dynamics/Contacts/b2ContactSolver.cpp#L674,
so yes, it seems possible to pull that in, although we'd have to ensure it
works for all platforms, not just Intel processors.

That being said, I think it's nice to stick as close as possible to
official Box2D releases (currently we're based on the 2.3.0 release of
Box2D). This way, people can use LiquidFun as a drop-in replacement.
Pulling in code from other places increases the chance that we'll introduce
a deviation.

One thing to note: their SIMD optimizations use SSE
https://en.wikipedia.org/wiki/Streaming_SIMD_Extensions, so they're for
Intel and AMD chips. LiquidFun's SIMD assembly code is for NEON
http://www.arm.com/products/processors/technologies/neon.php, so for ARM
chips. So they're optimizations for different platforms. Still useful, but
slightly different.

On Wed, Oct 28, 2015 at 8:30 AM, karlzwinnen notifications@github.com
wrote:

As Liquidfun uses SIMD for its particle system, i was wondering if there
are any plans to optimize the box2d part as well?

It seems there already is a fork of box2d available which makes use of
SIMD (and also enables this for SIMD.js through emscripten):
http://peterjensen.github.io/html5-box2d/
The source seems to be hosted here:
https://github.com/huningxin/box2d.js/tree/sse

Would it be difficult to merge this back in the box2d core liquidfun uses?

—
Reply to this email directly or view it on GitHub
#58.