LuxCoreRender Forums

LuxCoreRender 2.2 beta 1, build 004d2aa

Today, I noticed there can be a big difference in color and brightness if PGIC is enabled compared to disabled.
Please look at the image, they have both the same fixed tone map settings (Reinard).
Why is the render darker with PGIC? What happens to the energy?
Talking about Reinard ... Did anyone else notice that preview in viewport and final render do not have the same brightness?
Why does Reinard take some dozen samples to stabilize its brightness?

Without any information about material types,light sources, cache type and/or a test scene, it is hard to say.

FarbigeWelt wrote: Tue Jun 11, 2019 4:58 pm they have both the same fixed tone map settings (Reinard).

Have you tried with linear tone mapping ? Reinard is dynamic like auto linear tone mapping.

Dade wrote: Tue Jun 11, 2019 9:49 pm Without any information about material types,light sources, cache type and/or a test scene, it is hard to say.

FarbigeWelt wrote: Tue Jun 11, 2019 4:58 pm they have both the same fixed tone map settings (Reinard).

Have you tried with linear tone mapping ? Reinard is dynamic like auto linear tone mapping.

I did not know that Reinard is dynamic too. Best is in this case to re-render with linear but without auto brightness.

The materials are a mix of glass 95% and matte 5%.
There are three area lights and the sun with sky.

Dade wrote: Tue Jun 11, 2019 9:49 pm Without any information about material types,light sources, cache type and/or a test scene, it is hard to say.

Have you tried with linear tone mapping ?

Please have a look at the attached scene. Meanwhile yes, please have a look at the two pictures. Objects are darker with PGIC, indirect plus caustic cache, than without. Sky background seems to have the same brightness. Do you have any ideas why PGIC is darker?

Have you tried increasing the depth of PGI?

On phone now, but IIRC that was a property. Your problem looks like glass with a too small path depth

The problem is present only in GPU rendering, you get the same brightness on CPU. I will check what is going wrong.

It is a damn AMD compiler bug, the same code works fine with NVIDIA GPU. I have tracked down the problem and the AMD code just doesn't write a value to the requested location; I will try to find a workaround but it may be not even possible.

Dade wrote: Thu Jun 13, 2019 11:24 am It is a damn AMD compiler bug, the same code works fine with NVIDIA GPU. I have tracked down the problem and the AMD code just doesn't write a value to the requested location; I will try to find a workaround but it may be not even possible.

Why AMD driver seem to encounter so much little issue while being opencl only compute oriented while nvidia do both cuda/opencl and seem to work seamlessly ?

The Nvidia compiler has its fair shair of problems as well ...

Dade wrote: Thu Jun 13, 2019 11:24 am I will try to find a workaround but it may be not even possible.

I have found a ways to workaround the problem and now the scene should work fine on AMD GPUs too.