Stage 0 of the roadmap asks for a memory baseline before any more
optimisation, and explicitly not from RSS: the allocator and the driver
both retain high-water memory that RSS cannot tell apart from live
resources.
`gpu-live` counts from the sizes this renderer allocated, in the
categories large enough to decide where the next fix goes:
slots total shadow targets text_atlas wire_arena block_geometry
It is partial by construction and says so, rather than presenting itself
as a total. Hatches, images, meshes and Face3D are not counted yet.
Three of the numbers needed a size that was not reachable before: the
block-geometry chunk records its bytes at build because in storage mode
the buffer belongs to the bind group, the arena exposes its allocation
rather than its occupied prefix — headroom is what a budget has to answer
for — and the text atlas reports its `R8` texel count.
Emitted only when a number moves. `prepare` runs thousands of times in a
session and these change a handful of times, so a per-frame line would
bury the log it is meant to explain.
This answers the question the previous fix left open: whether the arena's
52 MiB per slot, the 16 MiB shadow map every slot allocates whether or not
shadows are on, or the render targets dominate once a paper layout has
four or five viewports on screen.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The crash in #358 was not the out-of-memory itself. wgpu reports OOM as
recoverable and the app already installs a handler that logs and keeps
the session alive. What killed the process was what happened next:
create_buffer_init maps the new buffer unconditionally, and mapping the
invalid buffer an OOM device returns panics. That panic unwinds through
iced's main-thread redraw, trips the wgpu-hal swapchain assertion in a
destructor, and becomes a non-unwinding abort -- no chance to autosave.
Every allocation whose size scales with the drawing now goes through
src/scene/pipeline/gpu_upload.rs, which creates the buffer unmapped and
fills it with Queue::write_buffer. A failed allocation then surfaces as a
validation error on the write, which reaches install_gpu_error_handler
and degrades the frame -- exactly what that handler's comment promised:
"A bad frame must degrade, not end the session."
Converted: block-wire vertices and instances, the wire const storage
buffer, all three wire-arena allocators, face3d vertices and instances,
block-text instances and text vertices, both hatch backends, and the
silhouette buffers. mesh_gpu already allocated this way, which is why the
mesh path never produced this panic; the rest now match it.
No mapped_at_creation:true remains in the renderer. The create_buffer_init
calls left are fixed-size or bounded-small -- uniforms, the viewcube's
static geometry, per-image and per-wipeout quads -- where an allocation
failure is not a realistic outcome.
Two trade-offs worth stating:
* write_buffer stages through a copy, which mapping at creation avoided.
Chunking bounds that transient cost, and the measured block-wire
upload got faster with chunking, not slower.
* `queue` had to be threaded through the wire, face3d, text and hatch
builders and their callers. Mechanical, but it widens the diff.
This touches every upload path, so it needs a visual check against a real
drawing before it goes upstream.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
max_buffer_size is an API validation limit, not physical VRAM. NVIDIA
reports several GB of it on a 2 GB card, so sizing a chunk from it
produces buffers larger than the whole device. That is the same bug as
#203 (mesh_gpu), #358 (face3d_gpu) and the block-wire vertex buffer:
rediscovered three times because each upload path spelled the budget out
for itself.
mesh_gpu.rs was the only site that clamped the result. Move that clamp
into src/scene/pipeline/gpu_budget.rs and route every allocating path
through it, so a new upload path cannot reintroduce the mistake by
copying one of the unclamped expressions.
Five sites had no clamp at all: text_gpu (block-text instances),
wire_gpu (block-wire instances), and face3d_gpu (vertices, instances,
and upload_chunks).
Three behaviour changes worth noting:
* Silhouette chunks were capped at a private 64 MiB constant and now
use the shared 32 MiB budget.
* The storage-hatch chunk limit was the minimum of two device limits,
both of them validation figures; it is now also capped by the
budget. The `*_fits` predicates next to it are genuine "would wgpu
reject this" checks and are left alone.
* mesh_gpu's instanced path used the unclamped ceiling to decide
whether a mesh fits in a single buffer, which would still admit a
multi-GB allocation. It now uses the clamped budget, so a few more
meshes take the chunked path.
OCS_GPU_CHUNK_MIB overrides the cap. Raising it above the real VRAM is
the cheapest way to force an allocation failure and check that the
renderer degrades instead of aborting.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Sync current main and companion codec changes before preserving unsupported frame rows, rendering in the entity plane, and sharing final layout metrics with placement preview.
Share source buffers across block placements and render nested content with GPU instance transforms.\n\nChunk silhouettes and compact mesh material data so large drawings stay under device buffer limits and remain responsive.
Pin Iced 0.15 so dialogs can measure content before distributing the resolved frame across child panels. Keep the web plugin notice compact and non-resizable.\n\nCloses #582
Paper content viewports with a circle / spline / ellipse clip boundary
drew their model content to the full bounding rectangle instead of the
shape. Each paper content viewport is its own render into its own screen
rectangle, so clipping is done per viewport: the boundary polygon is
projected into that viewport's render-target NDC (paper shape mapped
through the same visible-sub-rect crop as the content, so it stays fixed
to the viewport frame) and stamped into the stencil buffer with an
even-odd Invert, which marks the interior of any shape, convex or not,
without triangulation. Every content pass then draws only where the
stencil is set, so hatch, wire, mesh, 3DFACE, text, image, wipeout and
selection are all clipped uniformly.
Because Invert turns the cleared 0x00 into 0xFF, clipped content draws
with stencil reference 0xFF (unclipped content keeps reference 0). The
shared depth buffer moves to Depth24PlusStencil8 across every pipeline.
Rectangular viewports keep clipping via their own render rectangle and
XCLIP keeps its existing CPU wire clip; neither needs the stencil path.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Command previews (MOVE, COPY, ROTATE, SCALE, STRETCH, MIRROR, ARRAY,
PASTE, INSERT) build ghost WireModels via WireModel::translated/rotated/
scaled/stretched/mirrored, but those helpers moved only `points`, never
`text_verts`, and the ghosts' glyphs were never uploaded — so text
vanished (or froze at its source) during the preview, same root cause as
the grip-drag case.
Transform text_verts alongside points in all five helpers (shared
map_text_verts, reconstructing + re-splitting the double-single world
position), and gather preview_wires' glyphs into the per-frame
preview-text buffer in the render path. This covers every command
preview automatically. The grip-drag else branch no longer sets
preview_text explicitly (the render gather handles it); the grip-slide
fast path keeps its own scene.preview_text channel (empty preview_wires,
so no double-draw).
MIRROR honours MIRRTEXT: on -> true glyph mirror via mirrored(); off ->
readable relocation via translated().
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Dragging text by a grip hid it until release. The dragged entity is
hidden from the base wire set and shown as a preview overlay, but its
SDF glyph quads (text_verts) were never uploaded for the preview, so
the text vanished mid-drag.
Add a per-frame preview_text GPU buffer (sibling to preview_wires). A
rigid whole-entity move of pure text (TEXT/MTEXT, Square insertion grip)
slides the drag-start glyphs by the drag delta instead of re-shaping the
run every move; reshape grips (MTEXT width), dimensions and wire
entities re-tessellate and feed the same buffer.
Also fold preview_text into render_signature: a pure-text slide leaves
preview_wires empty, so without it the scene-render cache froze the
dragged text at its first frame (grip moved but text stayed, then
jumped to place on release).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Wires the SDF glyph atlas into the render loop end-to-end for top-level
TEXT entities, gated by the OCS_TEXT_SDF env var so the default path is
unchanged. Verified rendering on a real drawing: glyph quads land exactly
over the existing stroke text (position, size, rotation), confirming the
atlas → shader → pipeline → screen path and the RTE / UV math.
- text::text_run_placement — extracted from to_truck (behaviour-preserved)
so the stroke and SDF-quad paths compute an identical run placement.
- sdf_atlas::text_atlas — process-wide GlyphAtlas (Mutex/OnceLock), shared
by the collector (bakes glyphs) and the GPU upload (reads texels).
- Scene::sdf_text_vertices — collects per-glyph TextVertex for the frame's
TEXT entities via layout_glyph_quads + push_glyph_vertices. Empty unless
OCS_TEXT_SDF is set.
- ViewportData.text_verts + Pipeline text_pipeline/atlas/vbuf, upload_text
(rebuilds the R8 atlas texture when it grows), and a text render pass
drawn over the wires.
Still additive over the stroke text; only top-level TEXT is wired (MTEXT,
attributes, and block-internal text are next), and the collector re-runs
per frame (vertex caching by geometry epoch is a follow-up).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Groundwork to render CAD text as per-glyph SDF quads (box-quad snap,
perf, no LOD) instead of stroke-polyline wires. CPU side fully tested;
GPU side compiles but is not yet wired into the render loop.
- scene::text::sdf_atlas — GlyphAtlas: shelf packer + R8 SDF tiles baked
per (font, char). Unified field so one shader draws both kinds: TrueType
= signed distance to the filled outline, LFF stroke = pen band. Uses only
crates already in the tree (ttf-parser/lyon via the existing Glyph).
- scene::text::glyph_quads — layout_glyph_quads: lays a run out as per-glyph
quads using the exact tessellate_text_run transform, so quads land where
the strokes would. Per-char advance (TTF shaping is a follow-up).
- shaders/text.wgsl — SDF fragment shader (smoothstep over fwidth), RTE
vertex path mirroring image.wgsl, per-vertex colour + draw-order bias.
- scene::pipeline::text_gpu — TextVertex + double-single vertex builder,
R8 atlas texture upload, bind group, and the image-cloned pipeline.
13 unit tests across the three CPU stages, all green. Modules carry
#![allow(dead_code)] until the render-loop integration lands.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Christopher Overbeck