I am seeing rather large discrepancies between the slice render and the volumetric renderer for the smoke_test.fds, and when enabling the GPU with adaptive integration it seems to be broken on my end (latest drivers using a RTX 2080 Super on Windows 11).

This is the slice renderer (0.125 m cells):

Which is in line with the verification test:

This is the volumetric renderer without GPU with adaptive integration on (does not seem to do anything when toggling):

And this is the volumetric renderer using the GPU with adaptive integration:

And this is without adaptive integration and GPU:

I assume the difference/bug is in the step length done in the ray marching process since it changes when unticking the adaptive integration.

Also, from my quick view (I only looked quickly at shaders.c and the SetVolSmokeShaders section) it does not look like Smokeview uses this approach to step length:

https://github.com/cgyurgyik/fast-voxel-traversal-algorithm/blob/master/overview/FastVoxelTraversalOverview.md

But this approach actually seems to be very suitable for traversing the mesh data (voxels), enabling use of all of the available data and reducing banding due to semi-set step length being used now (with the downside of being potentially slower due to it being impossible to do loop unrolling by the shader compiler, but it is done in a while-loop now anyway and with the default way Smokeview renders currently with only one frame once the camera is still I think speed is not an issue, ).

I also have issues with adding lights, it either crashes or doesn't show anything (it does render out the position but no in-scattering is being shown on screen), but since this is not a documented feature I assume it is a WIP.