I have 3DLabs WildCat II 5110 card, and I used glConvolutionFilter2DEXT to apply a 3x3 filter on a 24-bit 1600x1200 BMP image and glDrawPixels. But it takes me around 2 seconds.
Does anybody can help me to improve the performance?
Use a fragemnt program for your calculation - but I´m not sure if your HW uspports it.
If so - render a full screen quad with texture coordinates in an orthogonal projection and then do the convolution of your texture in a fragment program with GLSL.