At 11:30 PM 11/10/04 -0700, SherLok Merfy wrote:
Increasing the precision (mathtolerance=/-1) seems to result in an out-zoom. . . .
Yes. The program will not calculate at a magnitude beyond the magnitude at which the designated precision fails. Decreasing the precision permits greater magnitudes to be reached, albeit at the cost of some image degradation.
. . . Do you mean that chip-math is correct when the arbitrary precision is supposed to be more accurate? . . .
Both math routines are accurate, but they work at different magnitude ranges, and arbitrary is much slower. I believe arbitrary precision is available for only the mandel, julia, manzpower, and julzpower formulas. I include a mathtolerance entry in the parameter files of images that have a magnitude very near the limit of precision. I do this to assure that no unintended outzoom happens when the file is run. I also add a mathtolerance entry to 'mandel' images with a magnitude very near the point where arbitrary precision math kicks in. I do this to assure that the image uses the fast chip math when it renders.
. . . Back when I was telling FracTint that I had an ATI Mach32 (It's a 64, now. I gave the MicroChannel machine away.), I had trouble using the arbitrary precision at all.
Maybe you were trying to use it with a formula that it does not work with. As far as I know, arbitrary math works with only the four formulas I mentioned above. Jim M.