[sldev] Optimization target: Avatar skinning, LLViewerJointMesh, matrix multiply

James Cook james at lindenlab.com
Sat Apr 14 14:39:43 PDT 2007


Sadly, Second Life currently supports processors that don't have SSE. 
So I need to be able to have code chosen at runtime based on whether or 
not we detect SSE.  I don't know how to use the compiler's 
auto-vectorizer in this situation.  I suspect I need to set up a 
function pointer at runtime and call a special SSE version.

I wrote an SSE version on my Mac this morning and it was slower than the 
compiler generated version.  :-(

James

Callum Lerwick wrote:
> On Sat, 2007-04-14 at 08:17 -0700, James Cook wrote:
>> My next step is to try to speed up this math using SSE instructions.
>> References I found online suggest SSE can double the speed of Vector3
>> Matrix4 multiplies.  If anyone here has experience with SSE, I could
>> sure use some advice.
>>
>> I think the matrix lerp() math can also be done with SSE.
> 
> Use the auto-vectorizer, Luke.
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Click here to unsubscribe or manage your list subscription:
> /index.html


More information about the SLDev mailing list