EasyGPU

Status: 0.1.0 — command encoding + FakeGPU complete; real backend on the roadmap.

EasyGPU is a Python wrapper for WebGPU with a built-in FakeGPU that lets you TDD your command-encoding logic without a GPU.

What works in 0.1.0

• ✅ GPU Protocol (73 methods) + configure/get_gpu
• ✅ FakeGPU — records and asserts command encoding without a GPU, including render_pass_sequence() / compute_pass_sequence() call trees
• ✅ Core objects — Buffer, Texture/TextureView/Sampler, ShaderModule, RenderPipeline/ComputePipeline, BindGroup + PipelineLayout, QuerySet, Adapter/Device/Queue
• ✅ Complete render pipeline — depth/stencil, blend, multisample + A2C, indexed draws, viewport/scissor, MRT via multiple color targets
• ✅ Compute pipeline — storage buffers, dispatch, indirect dispatch
• ✅ Render bundles — reusable draw batches
• ✅ Indirect draws on render passes
• ✅ Query sets — timestamp/occlusion + resolve_query_set
• ✅ Buffer mapping — map_async/get_mapped_range/unmap
• ✅ Texture ops — write_texture, copy_texture_to_*, generate_mipmap
• ✅ Context managers — with encoder.begin_render_pass(...): auto-ends
• ✅ Spec-aligned enums — IDL flags for ShaderStage/BufferUsage/TextureUsage
• ✅ Zero runtime dependencies — struct.pack for data, pure-Python easygpu.matrix
• ✅ 13 examples — 11 webgpu-samples ports + 2 self-authored (triangle, alpha_to_coverage)

What does NOT work in 0.1.0

• ❌ No real GPU backend — the GPU Protocol is the deliverable; the wrappers raise GPUValidationError unless you configure() a fake or real backend
• ❌ No swapchain / window output (examples bind color attachment handle 0)
• ❌ No dynamic-offset bind groups (subset-range bindings work via BufferSlice; see two_cubes)
• ❌ No numpy integration: vertex data is raw bytes (float sequences are struct-packed); the numpy upload path is deferred
• ❌ WGSL execution (FakeGPU records calls, it does not run shaders)

Examples

All examples live in the examples/ directory of the repository. They are not shipped in the wheel — run them from a clone or from the source distribution.

Ports of webgpu-samples, each encoding one frame against FakeGPU and verified by a pytest. 11 webgpu-samples ports + 2 self-authored examples (triangle, alpha_to_coverage); some ports are simplified (wireframe, reversed_z). WGSL and data are inlined/adapted (no browser, no image assets).

Run one directly, e.g.:

uv run python examples/textured_cube.py
uv run pytest tests/test_textured_cube.py -v

Quickstart

From source (contributors)

uv sync
uv run pytest                # headless, no GPU needed
uv run python examples/triangle.py

From PyPI (users)

pip install easygpu
python -c "from easygpu.fake_gpu import FakeGPU; print(FakeGPU())"

Note: importing easygpu emits a UserWarning explaining that 0.1.0 ships no real backend. It will go away once one lands.

Roadmap

• 0.1.0 (current): command encoding + FakeGPU, 13 examples, zero runtime deps
• 0.2.0: real backend behind the GPU Protocol (wgpu-py), keeping the Protocol stable
• 0.3.0: swapchain + windowed output
• 0.4.0: dynamic-offset bind groups, GPU timing readback
• Later: more webgpu-samples ports

Project layout

• src/easygpu/ — library (Protocol seam + wrapper objects + fake_gpu.py)
• examples/ — runnable examples (11 webgpu-samples ports + 2 self-authored)
• tests/ — pytest suite for the library and every example (headless)

The layout above describes the source repository. The published wheel contains only src/easygpu/ — examples/ and tests/ live in the repository and in the source distribution (easygpu-0.1.0.tar.gz).