Slim per-instance record for the world vertex shaders #1033

New issue

Closed

opened 2026-10-25 01:02:05 +00:00 by sky_line · 1 comment

sky_line commented

2026-10-25 01:02:05 +00:00

Owner

Have the cull pass write a compact vertex-shader instance stream (3x4 model matrix plus the ids and flags the VS actually reads, ~64 bytes) and read that in clean_draw.vert / clean_draw_rigged.vert instead.

Done when: identical image (including motion vectors, attachments and spin), with lower G-buffer / prepass timers.

Profiled with Nsight Graphics GPU Trace + shader profiler on an RTX 5090 (full GI scene, ~61 FPS, GPU ~16.3 ms/frame). Vertex shading is ~21% of GPU shader samples. The hottest vertex lines (clean_draw.vert, first use of `inst` / `pinst`) stall 97-99% on memory: every vertex loads the full 224-byte Instance record, and the motion prepass also loads the previous frame's record. Have the cull pass write a compact vertex-shader instance stream (3x4 model matrix plus the ids and flags the VS actually reads, ~64 bytes) and read that in clean_draw.vert / clean_draw_rigged.vert instead. **Done when:** identical image (including motion vectors, attachments and spin), with lower G-buffer / prepass timers.

sky_line added the

perf

rendering

labels

2026-10-25 01:02:05 +00:00

sky_line added this to the Slipstream project

2026-10-25 01:02:05 +00:00

sky_line added the

status

testing

label

2026-10-25 01:19:01 +00:00