[rendering] Cluster LOD: rigged emissive surfaces no longer cast ReSTIR/GI emissive light #1009

New issue

Open

opened 2026-10-24 01:50:29 +00:00 by sky_line · 0 comments

sky_line commented

2026-10-24 01:50:29 +00:00

Owner

With cluster LOD on at startup (the cluster path owns rigged meshes, and their legacy geometry is released), rigged emissive triangles drop out of the ReSTIR DI emissive light build. Glowing attachments, eyes and rave gear still glow on screen (bloom), but they no longer light their surroundings under GI / ReSTIR emissive.

Same party, ReSTIR DI emissive stats:

session	emitters	emissive triangles live	c_rs_emissive
legacy	56-96	~12.3k	4.1 ms
CLOD turned on live (legacy triangles still recorded)	56	12.2k	4.1 ms
CLOD on at startup	45-46	~1.8k	1.6 ms

About 10.5k rigged emissive triangles are missing. This also made CLOD look ~2.5 ms better than legacy in emissive-heavy scenes (#1008).

Fix

The emissive build's triangle records for rigged meshes on the cluster path must come from the cluster data instead of the released legacy rigged buffers. Take the emissive material's triangles from the cluster pages or the current cut, skinned, at a level of detail suited to light sampling (a coarse cut is fine: emissive light is soft).
Keep the per-emitter radiance and area the same as the legacy path, so the look matches (A/B against CLOD off).
Respect the triangle cap, and #1006's incremental and amortised rebuild once it lands (this work is queued behind #1006).
Stats: rigged-cluster emissive triangles in the ReSTIR DI log line.

Refs #1008, #1006.

With cluster LOD on **at startup** (the cluster path owns rigged meshes, and their legacy geometry is released), rigged emissive triangles drop out of the ReSTIR DI emissive light build. Glowing attachments, eyes and rave gear still glow on screen (bloom), but they no longer light their surroundings under GI / ReSTIR emissive. Same party, ReSTIR DI emissive stats: | session | emitters | emissive triangles live | c_rs_emissive | |---|---|---|---| | legacy | 56-96 | ~12.3k | 4.1 ms | | CLOD turned on live (legacy triangles still recorded) | 56 | 12.2k | 4.1 ms | | **CLOD on at startup** | 45-46 | **~1.8k** | 1.6 ms | About 10.5k rigged emissive triangles are missing. This also made CLOD look ~2.5 ms better than legacy in emissive-heavy scenes (#1008). **Fix** - The emissive build's triangle records for rigged meshes on the cluster path must come from the cluster data instead of the released legacy rigged buffers. Take the emissive material's triangles from the cluster pages or the current cut, skinned, at a level of detail suited to light sampling (a coarse cut is fine: emissive light is soft). - Keep the per-emitter radiance and area the same as the legacy path, so the look matches (A/B against CLOD off). - Respect the triangle cap, and #1006's incremental and amortised rebuild once it lands (this work is queued behind #1006). - Stats: rigged-cluster emissive triangles in the ReSTIR DI log line. Refs #1008, #1006.

sky_line added the

bug

rendering

labels

2026-10-24 01:50:29 +00:00

sky_line added this to the Slipstream project

2026-10-24 01:50:29 +00:00

sky_line added the

status

in-progress

label

2026-10-24 02:06:14 +00:00