Proposed SIG-Graphics-Audio meeting agenda for 2022-07-06

[thefranke]

Meeting Details

• Date/Time: July 6, 2022 @ 6:00pm UTC / 10:00am PT
• Location: Discord SIG-Graphics-Audio Voice Room
• Moderator: @thefranke
• Note Taker @thefranke

The SIG-Graphics-Audio Meetings repo contains the history past calls, including a link to the agenda, recording, notes, and resources.

SIG Updates

See previous meeting notes here: #50

Meeting Agenda

• @mekhontsev will present their work on ShaderCanvas.
• Discuss Proposed RFC Feature Material Canvas #51 in the next meeting (just dropped at the start, too long to go through) and let @gadams3 fill us in on the details
• Discuss possible solutions to make the RT code aware of materials. Proposals include custom shaders that will act as drop-in for a material that can be sampled by the raytracing code, proxy geometry with vertex colors or rendering out albedo textures that can be sampled (which however implies diffuse-only).
• Pull in @dmcdiarmid-ly for a discussion on hybrid rendering. It is currently unclear when the RT acceleration structures are built. Does the user need to have a GI component in the scene, or what triggers this?

Outcomes from Discussion topics

Discuss outcomes from agenda

Action Items

Open Discussion Items

List any additional items below!

[jeremyong-az]

For material-aware RT shaders, I believe #53 coupled with ongoing bindless and material canvas work is a viable path forward. There are a few problems to address here:

• Resource binding: Attempting to load a resource (e.g. texture) directly in shader code is an immediate non-starter if trying to use material code in a DXR shader. An abstraction is needed to substitute a request for a texture with a load from a bindless table or SRG depending on the context.
• Shader evaluation: Materials that deform geometry would need to rely on intersection shaders in the the general case. Exposing the routines that modify vertex positions/tangent spaces to deform geo in a normal hit shader could work also, but this is tricky and I would suggest handling only static/skinned geometry initially
• BRDF parameter evaluation: the work proposed in the first link above suggests that materials expose a function for parameter evaluation directly given UV inputs. Coupled with a resource binding abstraction, this code would be usable in a normal pixel shader, or a DXR shader
• We'd need a way to maintain a shader table for all materials in the scene that are accessible from a DXR shader.