ok i was trying to see how accurate this can be done using tex2Dproj. first, i looked at two sources: ShaderTech link Deferred Shading demo I tried the sample code from the second source, but couldnt get it work though. What happened was that i could get screen space texcoords with a slight distortion, about 2-3 pixels being offsetted along the x axis. What I did was to render a quad with screen space coords. And then a second pass tries to uses tex2Dproj to lookup the earlier texture to see the lookup was done correctly or not. I use a modified DX texture tool to read back the pixel values (like Image Debugger) to check if the 2nd pass was performing the lookup accurately. Unfortunately, it didnt. :(

D3DXMATRIXA16 matPostProjRescale;
D3DXMATRIXA16 translation, scaling;
D3DXMatrixIdentity(&matPostProjRescale);
D3DXMatrixIdentity(&translation);
D3DXMatrixIdentity(&scaling);
D3DXMatrixTranslation(&translation, 1.0f + 1.0f / 512.0f, -1.0f - 1.0f / 512.0f, 0.0f);
D3DXMatrixScaling(&scaling, 0.5f, -0.5f, 1.0f);
D3DXMatrixMultiply(&matPostProjRescale,&translation,&scaling);

anyone tried this way of getting screen space texcoords before? if u do, do you get extremely accurate results or also have slight distortions... then again, I might as well get an X1K card and use VPOS instead. And I hope ATI supports VPOS... phew! Edwin

After you have applied this matrix there is a final step to get screen coordinates. X,Y and Z need to be divided by W. If you are sure that W is always 1 you don’t need to do this.

ok i believe the method from the website and yours are essentially the same. But i believe the problem could lie with ATI 24bit precision as i realized that during readback, I have trailing errors with accumulate... sigh

thx again!
Edwin