CUDA and Raymarching
category: code [glöplog]
i agree.. both major companies have to do a lot more fixes to their OpenCL compilers before one can actually call them usable (especially for "real" programs, not just toy stuff)..
Again I'd opt for DirectCompute / DX Compute shaders. Yes it's slightly more limited, but reasonably stable and "just there".
The straight CS5.0 port of BurjBabil (no multi pixel optimizations) was 8 times faster than the original opengl version. However the compile time was several minutes - ie the general problem of avoiding unrolling/inlining.
The straight CS5.0 port of BurjBabil (no multi pixel optimizations) was 8 times faster than the original opengl version. However the compile time was several minutes - ie the general problem of avoiding unrolling/inlining.
