workshopsoft/ludic
2
0
Fork 0
Code Issues Pull requests Projects Releases 26 Packages Wiki Activity Actions

fix(render3d): the Vulkan renderer stops leaking as it loads and draws

Browse source 
- every texture upload malloc'd a CPU copy of its pixels and never freed it (236 MB over the
  valley's boot, and again for every model loaded later): the pixels are converted straight
  into the mapped staging buffer
- the per-level and per-layer views gvk_view_of made outlived their texture, and on MoltenVK a
  view keeps its Metal texture alive: a released texture takes its views with it, and the cache
  is keyed by numbers instead of a string built on every call
- the Vulkan structs filled for a draw, pass, barrier, descriptor set, buffer, allocation or
  upload (about sixty call sites) come from a reused 1 MB scratch ring (gvk_tmp)
- the descriptor-set cache and the retired buffers are emptied in place, not replaced; the grass
  cull's dispatch arguments are made once
- macOS drains an autorelease pool each frame (Vk.frame_pool, lvk_frame_pool in vk_mac.ll)
- R3D_VK_PROF reports Vulkan objects made and destroyed by kind, and every cache's length
- examples/rendering/smooth presents through render3d, so it runs on Vulkan too

The full valley on headless Vulkan loads to 2.18 GB and holds (it passed 8 GB while loading
before). ludic-dev test 305/305, selfhost-test 33/33.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Orkun ÇAKILKAYA 2026-09-27 23:03:38 +03:00
parent 9539f7bcec
commit 80764f85c1
11 changed files with 284 additions and 113 deletions
Download patch file Download diff file Expand all files Collapse all files

19
packages/ludic.render3d/grass_gpu.ludic
View file

@ -102,9 +102,14 @@ function gg_cull_frame(render3d_st: mut Render3dState) -> void {
gg_tiles(render3d_st, 4, 0.0, 16.0)
gg_tiles(render3d_st, 8, 16.0, 40.0)
gg_tiles(render3d_st, 16, 40.0, render3d_st.grass_radius)
let rb = words(4)
# made once: a words() a frame is a calloc nothing frees
if render3d_st.gg_rb == null {
render3d_st.gg_rb = words(4); render3d_st.gg_cb = words(1); render3d_st.gg_bufs = words(3)
render3d_st.gg_texs = words(2); render3d_st.gg_pr = words(48)
}
let rb = render3d_st.gg_rb
rb[0] = GG_BANDS; rb[1] = 0; rb[2] = 0; rb[3] = 0
let cb = words(1)
let cb = render3d_st.gg_cb
cb[0] = render3d_st.gg_cmds
gpu_dispatch(render3d_st, render3d_st.gg_reset, data_of(rb), 16, cb, 1)
if render3d_st.gg_n == 0 { return }
@ -112,16 +117,16 @@ function gg_cull_frame(render3d_st: mut Render3dState) -> void {
let tb = render3d_st.gg_tiles_buf[render3d_st.r3d_test_frame % 2]
gpu_buffer_upload(render3d_st, tb, render3d_st.gg_n * 16, data_of(render3d_st.gg_tv), GPU_DYNAMIC)
let pr = gg_params(render3d_st)
let bufs = words(3)
let bufs = render3d_st.gg_bufs
bufs[0] = tb; bufs[1] = render3d_st.gg_out; bufs[2] = render3d_st.gg_cmds
let texs = words(2)
let texs = render3d_st.gg_texs
texs[0] = render3d_st.ter_height_tex; texs[1] = render3d_st.ter_ortho_tex
gpu_dispatch_tex(render3d_st, render3d_st.gg_cull, data_of(pr), 192, bufs, texs, (render3d_st.gg_max + 63) / 64, render3d_st.gg_n)
}
# grass_cull.comp's Params, std140: 12 vec4s
function gg_params(render3d_st: Render3dState) -> words {
let pr = words(48)
# grass_cull.comp's Params, std140: 12 vec4s, into the block made once in gg_cull_frame
function gg_params(render3d_st: mut Render3dState) -> words {
let pr = render3d_st.gg_pr
for i in 0 .. 48 { pr[i] = 0 }
if render3d_st.cam_planes != null { for i in 0 .. 16 { pr[i] = float_bits(render3d_st.cam_planes[i]) } }
pr[16] = float_bits(render3d_st.cam_pos[0]); pr[17] = float_bits(render3d_st.cam_pos[1]); pr[18] = float_bits(render3d_st.cam_pos[2]); pr[19] = float_bits(render3d_st.grass_radius)