DirectCompute Double precision Mandelbrot demo and more..

In addition to first demo using double precision on GL 4.0 here now on DirectCompute:
THIS DEMO NEEDS DX JUNE 2010 RUNTIMES
so update if needed

this test on AMD shows a ATI DirectCompute DPFP bug.. it shows incorrect rendering..

Also note I learned DirectCompute doesn't admit division with doubles so I have to change /2 with *0.5.
Nvidia Fermi works OK!
DirectCompute Double precision Mandelbrot (includes source based almost 100% on Voxilla demo):

use test.bat app starts at big zoom so it shows DP in action.. if you exit with esc then shows same rendering at SPFP.. note with mouse you can zoom in out.. 

bat calls mandel.exe 0 for SP or mandel.exe 1 for DP..

Also note I expected better perf for AMD than Nvidia but two work very slow i.e. Nvidia runs at full speed (i.e. capped 8x vs Teslas) but AMD has perf issues as it should run at least 3-4x vs Nvidia Fermi..
Also has vector mode running somewhat faster than scalar shader (but not much could run up to 4x faster if compiler didn't extract perf of scalar code but runs not much faster compared to SP where vector code outperforms scalar code by a higher amount).. fermi perf is unaffected by using vector code..

Correct behavior:

Double precision (on GTX 470)

See full Window

Single precision

On AMD 5850 DP renders as (i will post image soon):

Related also I patched Nvidia Physx Demo to work on AMD changing GLSL code using Cg non standard functions.. it exhibits some OpenGL bugs.

Instructions:

Download Nvidia Physx Demo here ((select FLUIDS: TECHNOLOGY DEMO)

and use this exectuable for running on AMD cards (extract on demo dir).
It shows artifacts on AMD card not on rendering but on desktop outside of program window..
On AMD 5850 DP bad renders as (i will post image soon):