feat: GPU text rendering for RenderDirect

[kolkov]

Text is not visible when using RenderDirect because it only renders GPU-accelerated shapes. Text is currently CPU-rasterized to an internal pixel buffer, and RenderDirect (zero-copy GPU→surface path) does not upload that buffer.

Current behavior

Method	GPU shapes	CPU text
RenderDirect	✅	❌
RenderTo	✅	✅

Workaround

Use RenderTo instead of RenderDirect:

canvas.RenderTo(dc.AsTextureDrawer())

Solution

Integrate the existing MSDF text pipeline into the unified GPU render session so DrawString goes through a GPU MSDF batch instead of CPU rasterization.

Completed infrastructure

Task	Description	Status
TASK-050 Phase 1	Text pipeline: shaper, bidi, layout, cache	✅ Done
TASK-050b	Glyph-as-path conversion	✅ Done
TASK-050c	Glyph cache	✅ Done
TASK-050f	MSDF generator (Pure Go)	✅ Done
TASK-050g	MSDF atlas packer	✅ Done
TASK-050h	MSDF WGSL shader (msdf_text.wgsl)	✅ Done
TASK-050i	Emoji support	✅ Done
TASK-050j	Subpixel positioning	✅ Done

Existing code:

• internal/gpu/text_pipeline.go — TextPipeline struct (stub RenderText)
• internal/gpu/shaders/msdf_text.wgsl — WGSL shader with fill, outline, shadow modes
• Spec: msdfgen-go-spec.md — standalone Pure Go MSDF library design

Remaining work

Wire DrawString → TextPipeline.RenderText within the GPU render session:

1. Upload MSDF atlas to GPU texture
2. Glyph layout → vertex buffer (quad per glyph)
3. Integrate text batch into unified render pass
4. Zero readbacks, resolution-independent

Related

• Reported by @amortaza in Discussion #47
• Backlog task: GG-TXT-001 (milestone v0.31.0)

[kolkov]

Update: Type Compatibility Analysis (2026-02-20)

Current State: gg/text/msdf/ ≠ msdfgen-go types

The existing gg/text/msdf/ implementation (~5,800 lines, 87% coverage) is a working prototype but has incompatible types with the planned standalone msdfgen-go/msdfgen library:

Aspect	Current gg/text/msdf/	Planned msdfgen-go/msdfgen
Input	*text.GlyphOutline (gg-coupled)	Shape (standalone)
Vector	Point{X,Y} (local)	Vec2 (reusable)
Output	MSDF{Data []byte} (RGB concrete)	Bitmap[T] (generic)
Edge types	Edge (concrete struct)	EdgeSegment (interface)
Dependencies	github.com/gogpu/gg/text	ZERO

Migration Path

Phase 1: Create msdfgen-go/msdfgen (standalone, spec's types)
  ├── Vec2, Shape, Contour, EdgeSegment (interface)
  ├── Bitmap[T] generic bitmap
  ├── SDF/MSDF/MTSDF generators
  └── v0.1.0 release

Phase 2: Create msdfgen-go/msdfgen-atlas (standalone)
  ├── Font loading, rectangle packing
  └── v0.1.0 release

Phase 3: Replace gg/text/msdf/ internals
  ├── Import msdfgen-go/msdfgen
  ├── Adapter: GlyphOutline → msdfgen.Shape
  ├── Delete ~4,000 lines (replaced by import)
  └── Keep atlas.go, shelf.go (or use msdfgen-atlas)

Phase 4: Wire into GPU render session (gg#115 core)
  ├── Upload atlas → GPU texture
  ├── DrawString → GPU text batch
  └── Zero readbacks

Algorithms are correct

The current prototype correctly implements Chlumsky's MSDF algorithm:

• Edge coloring with angle threshold
• Per-channel signed distance calculation
• Parallel row processing (4 workers)
• Median filter + error correction
• Generator pool for concurrent use

These algorithms will transfer directly to msdfgen-go — only types change.

[kolkov]

Implemented in gg v0.29.0 — MSDF text pipeline (Tier 4) renders text entirely on GPU via ggcanvas.Canvas.RenderDirect(). Resolution-independent anti-aliased text across Vulkan, DX12, and GLES backends.