Re: vector & matrix operations

the GPU runs shaders. nothing else.

readbacks from vertex operations are not permitted.

the vertex operations output goes to the pixel shader. with SM 3.0 you can
ping-pong between vs and ps, but you cant get it back to the CPU.

this is a hw architecture isssue.

as far as SIMD optimizing D3DX math routines, suggest it to
directx@xxxxxxxxxxxxx, its a good suggestion.

"DPR" <temp@xxxxxxx> wrote in message
news:eV7QrbfNFHA.1300@xxxxxxxxxxxxxxxxxxxxxxx
> Thx for your overview ...
>
>> thats a lot of overhead for an add, and thats before the cost of the
>> readback and readbacks are very expensive.
>
> Well it is worth the effort if you want to use DirectX/the GPU to help the
> CPU with a lot of complex matrix/vector calculations. I.e. if it takes a
> relative long time before you do have data for the GPU to render and
> display - so the GPU has to wait a long time until it is able to process
> new
> data.
> That's what I was looking for: using the GPU for some parallelized
> calculations, that is to say load sharing for the mentioned calculations.
>
> BTW: after looking into "d3dx9math.inl" I was really surprised that none
> of
> the functions is optimized for SIMD operations. Compared to my existing
> SIMD
> optimized algos the DirectX ops took at least twice the time ...
>
> Regards,
>
> --
>
> DPR
>
>


.

Loading