GI shade and RT reflections: hit/miss compaction and a material/lighting split #1036

New issue

Closed

opened 2026-10-25 02:12:09 +00:00 by sky_line · 1 comment

sky_line commented

2026-10-25 02:12:09 +00:00

Owner

Have the trace stage sort texels into hit and miss lists, and run a small miss kernel and a hit kernel over them (indirect dispatch).
Split the hit work into a material stage that writes a compact packed surface record and a lighting stage that shades from it. Apply the same structure to the RT reflections.

Done when: each new kernel is well under the current register counts (target ~128), the image is unchanged, and GI shade + reflection GPU timers drop at the same scene.

Follow-up to #1031. After the GI split, the GI SHADE stage is the heaviest at 212 registers (1.15 ms on an RTX 5090). It runs the miss path (sky + probe far field), material evaluation (rtSurfaceEval, ~40-float surface), the weather coat and the lighting (rtShadeHit, including a shadow ray) in one kernel. The RT reflections (~1.36 ms) use the same rtSurfaceEval / rtShadeHit. 1. Have the trace stage sort texels into hit and miss lists, and run a small miss kernel and a hit kernel over them (indirect dispatch). 2. Split the hit work into a material stage that writes a compact packed surface record and a lighting stage that shades from it. Apply the same structure to the RT reflections. **Done when:** each new kernel is well under the current register counts (target ~128), the image is unchanged, and GI shade + reflection GPU timers drop at the same scene.

sky_line added the

perf

rendering

labels

2026-10-25 02:12:09 +00:00

sky_line added this to the Slipstream project

2026-10-25 02:12:09 +00:00

sky_line referenced this issue

2026-10-25 03:01:22 +00:00

Avatar very noisy for a moment when it goes from occluded to visible #1038

sky_line referenced this issue

2026-10-25 03:19:37 +00:00