Prev: Business Analyst Online Training @ CVK OLT starts on Feb, 1st 8:30 PM EST
Next: Intel documentation mistake for single floating point precision !
From: Skybuck Flying on 31 Jan 2010 04:26 Remaining question is: What does the hardware do when it encounters an inprecise 3D texture coordinate ? For example instead of: 0, 0, 0.265 it encounters: 0, 0, 0.2649999856948853 (53/200 in 32 bit floating point) The hardware "knows" the texture is 500x400x200 If the hardware is inprecise it will calculate a lineair offset using 32 bit floating points, the end result will be incorrect: 500 * 400 * 200 * 0.2649999856948853 = 10599999.427795412 The correct result is: 500 * 400 * 200 * 0.265 = 10600000 So I guess the internal precision of the hardware can be tested by creating such a texture map and setting offset/voxel 10599999 to red and 10600000 to green ;) :) offset 10600000 = voxel 0, 0, 53 offset 10599999 = voxel 499, 399, 52 This will be a most interesting test ! ;) Bye, Skybuck. |