🔒 Locked

d3dx9math - fast enough!?

Started by gamelife Jan 10, 2003 at 5:33 AM 4 replies 450+ views

Original Post

Jan 10, 2003 at 5:33 AM

Is it fast enough? AFAIK, using AMD''s 3DNow or Pentium''s SSE can boost the speed of vector/matrix operations *N* times faster. I am curious.. does MS optimized functions like D3DXMatrixMultiply by 3DNow/SSE?

S1CA

Jan 10, 2003 at 7:04 AM

Yes. Actually engineers from AMD and Intel wrote & optimised the code for Microsoft specifically for D3D and D3DX. It''s known as the PSGP, "processor specific geometry pipeline/path"

If you step through the call to D3DXMatrixMultiply you''ll see it using whatever SIMD is present. If you look at the symbols in the .lib you''ll see things like sse_D3DXMatrixMultiply, x3d_D3DXMatrixMultiply etc

--
Simon O''Connor
Creative Asylum Ltd
www.creative-asylum.com

Joe Forhens

Jan 10, 2003 at 1:38 PM

Sweet to hear

tok_junior

Jan 10, 2003 at 6:14 PM

Yup.
I actually tried to write faster code myself, and didn''t manage to in most cases, this code is ultra optimized

--
MFC is sorta like the swedish police... It''''s full of crap, and nothing can communicate with anything else.

gamelife

Jan 10, 2003 at 7:48 PM

Nice. It''s always better just rely on Microsoft...

One more question: suppose I want to multiply an array of one million D3DXVECTOR4 (or D3DXMATRIX) by one *single* D3DXMATRIX, do I need to call D3DXVec4Transform (or D3DXMatrixMultiply) 1 million times?
I cannot find D3DX functions for batch transforms. Some articles from google presented batch multiplication functions that can be optimized further and faster.

xyzzy00

Jan 10, 2003 at 7:50 PM

D3DXVec4TransformArray was added in D3DX9.

xyzzy

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.