Global Illumination is Super Slow

[yvt]

Global Illumination (r_radiosity) works perfectly and fast on MacBook Air (Mid 2012, Intel HD Graphics 4000), but not on others.

Radiosity rendering is done as following:

1. Computation is done in background by CPU, for 512x512x64 grid points.
2. The computation result is streamed into GPU using glTexSubImage3D (each call transfers 16x16x16 block)
3. Shader (Shadow/MapRadiosity.fs) samples this and calculates the final value for every visible solid pixels.

The problem is that the step 2 sometimes takes too much time. Even only one glTexSubImage3D call took 300ms on some computers.

Using PBO had no effects.

[dany-on-demand]

Possibly a driver issue? Maybe intel GPU forums will have info on slow glTexSubImage3D on windows.

[learn-more]

Could be a memory alignment issue, did you try other blocksizes?

[feikname]

@yvt can this be closed?

EDIT: Accidentally hit "Close and comment" button