WIP: Merge noise-info features from VGA internal branch #54
Conversation
Original commit: Author: Saku Haikio <[email protected]> Date: Tue Nov 7 12:11:45 2023 +0200 add material-id, metallic feature and roughness feature
Original commit: commit 3e3e786c09fe1dc5c178b47931051b88c01ad8a8 Author: Saku Haikio <[email protected]> Date: Thu Jan 25 11:34:23 2024 +0200 add pdf features
Material and instance-id gbuffers are required only when the options require them, and the code to read them is injected with preprocessor macros. Therefore performance should be unaffected when they are not used.
|
I added metallic, roughness and instance-id options to BMFR, these should hopefully work correctly. The --noise-data-str parameter should probably be separated to at least use a different name for BMFR, as the gbuffer options aren't related to the noise data renderer. There seems to be some work in progress stuff in the noise data renderer from Saku to allow a maximum of 8 noise data features instead of 4 (using 2 buffers), I could have a look to see if this could be integrated with BMFR. I think some additional testing is needed to make sure the noise data stuff works correctly with BMFR (mainly that the buffers have the expected data in them when used with BMFR). |
|
@sasamiyatu thanks, the BMFR output with metallic, roughness and instance-id looks reasonable, so it probably works correctly, but I didn't verify the contents of the buffers. I agree that --noise-data-str should rather be for example --bmfr.buffers (similar to how SVGF and ReSTIR specify their own parameters). If you can take a look at the integration of max 8 noise data features, that would be great. |
This is a work in progress. Do not merge this yet.
Requesting review to check that the buffers output by the various noise-data options are what are expected. Some may have changed unintentionally, as I had to change the computations somewhat to fit the current internal structure of the path tracing loop.