"PhysX hobbled on CPU by x87 code"

[Scali]

Better still, why not you and Scali form an email together and send it to him addressing the most important ingredient in all this, which is..... MAKE YOURSELF CLEAR!! AND UNDERSTOOD!!!

I did, but not getting anywhere.
His last answer was this:
"Scali,

I've been pretty clear, so I'll re-iterate myself here:

This is a fine discussion for the forums at RWT, that's exactly why they exist. Our forums are filled with people who know much more about assembly, C/C++, optimization, etc. than either of us - there's always someone more knowledgeable out there.

If you want to have this discussion there, that's fine then feel free to post something, but I'm not really interested in continuing this discussion via email.

David"

I don't see his point. Various people have now confirmed that his SSE claims are completely wrong. nVidia has also set him straight on the multithreading...
Why would we need to discuss this on his forum? He needs (yes NEEDS, his integrity as a journalist is at stake here) to amend his article, and correct the misinformation. Us posting at his forum isn't going to help any.
Clearly he is reluctant, as I already said. He just doesn't seem good-willing.

 

[Scali]

I'll have to add... At least Andrew Lauritzen bothered to look into Bullet some more himself, and publish his results at Beyond3D, so Andrew has redeemed some of his integrity:
http://74.200.65.90/showpost.php?p=1451158&postcount=136

They pretty much match with mine and Schmide's.
He also agrees with nVidia's response on the matter, so indirectly he disagrees with Kanter's claims (he says he doesn't think nVidia is intentionally hurting CPU performance).

DKanter is an active member on that forum... I wonder if he's going to ask Andrew to post it on the RWT forum aswell, before he does anything with the information.

 

[aphorism]

i find it odd that bullet's use of SSE intrinsics is slower than MSVC's generated SSE code.

 

[Scali]

i find it odd that bullet's use of SSE intrinsics is slower than MSVC's generated SSE code.

It isn't. Where did you get that idea?
From Andrew Lauritzen's figures:
MSVC generating SSE2 code, intrinsics disabled:
/arch:SSE2
Results for 3000 fall: 18.083166
Results for 1000 stack: 12.192101
Results for 136 ragdolls: 9.896542
Results for 1000 convex: 15.049071
Results for prim-trimesh: 8.446912
Results for convex-trimesh: 13.625685
Results for raytests: 18.630299

MSVC generating SSE2 code, intrinsics enabled:
/arch:SSE2 + USE_SSE
Results for 3000 fall: 17.373240
Results for 1000 stack: 11.402561
Results for 136 ragdolls: 9.099781
Results for 1000 convex: 14.487597
Results for prim-trimesh: 8.132652
Results for convex-trimesh: 13.579372
Results for raytests: 18.672155

These results are times in ms, so lower is better.