Add WebGPU backend for web platform#1458
Open
edwardxfshen wants to merge 52 commits into
Open
Conversation
edwardxfshen
requested review from
Hparty,
domchen,
kevingpqi123,
richardshan0614 and
shlzxjp
as code owners
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…on with 101 passing. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…reated textures have RENDER_ATTACHMENT usage in their descriptor, allowing them to be used as render pass color attachments.
…e copy rect dimensions do not exceed either source or destination texture bounds.
…t dimensions do not exceed render target dimensions.
…ual method to GPUBuffer base class and integrate async buffer mapping in SurfaceReadback::lockPixels() with polling.
Move emitChild(colorizerIndex) outside the if-else conditional to satisfy WGSL uniformity requirements. When the colorizer contains texture sampling (e.g., TextureGradientColorizer), textureSample must be in uniform control flow, not inside conditional branches.
Removed printf/console.log debug statements from SurfaceReadback, WebGPUBuffer, and WebTestRunner. Added Tint-based SPIR-V to WGSL shader conversion to replace Emscripten SPIR-V extension (which Chrome WebGPU doesn't support). Modified shader compilation pipeline to convert shaderc-generated SPIR-V to WGSL text before passing to WebGPU. Fixed test filter to properly exclude BackgroundBlurTest. Currently investigating Tint SPIR-V reader compatibility issues.
Investigated SPIR-V generation from GLSL shaders using shaderc. Identified root cause: GLSL sampler2D compiles to combined image-sampler in Vulkan SPIR-V, which WGSL/WebGPU doesn't support. Tried SPIRV-Tools optimizer split pass (failed in Emscripten) and shaderc auto-bind options (ineffective). Current blocker: Need to separate combined image-samplers into distinct texture and sampler resources. Solutions require either: (1) modifying GLSL generation to use separate samplers, (2) finding Tint's internal transform pass, or (3) post-processing SPIR-V with tools compatible with Emscripten. Asyncify buffer mapping and partial rendering pipeline confirmed working. Full test suite blocked by shader compilation failure.
…ources for WebGPU Tint compatibility.
…crash on readPixels.
Use global sequential counter for shader locations instead of per-buffer index. This aligns WebGPU pipeline descriptor with WGSL shader expectations where location indices increment globally across all vertex inputs, matching Metal and OpenGL backends. Result: Fixes 265 out of 292 failing tests (from 292 → 27 failures). Commit: $(git rev-parse --short HEAD)
Pipeline layout now derives texture/sampler binding indices from TEXTURE_BINDING_POINT_START to match WGSL shader expectations, instead of using layoutEntries.size() which fails when UBO count != 2. Also adds validation error capture for pipeline creation and WGSL binding debug logging.
1. Set depthSlice = WGPU_DEPTH_SLICE_UNDEFINED for mipmap generation render pass color attachments (was 0, which is invalid for non-3D textures). 2. Create separate single-mip-level render views for textures with mipmaps. WebGPU requires render target views to have exactly 1 mip level, while sampling views need all levels. Added webgpuRenderView() accessor. 3. Use webgpuRenderView() for render pass color attachments and resolve targets instead of the default full-mipchain textureView. Fixes 16 blank renders (from 25 down to 9) and reduces Invalid CommandBuffer errors from 11 to 1.
WebGPU requires primitive topology at pipeline creation time, but TGFX specifies it at draw time. Created dual pipeline variants (TriangleList and TriangleStrip) for each render pipeline, selecting the correct one based on the PrimitiveType parameter in draw/drawIndexed calls. Previously all draws used TriangleList, causing TriangleStrip(4 vertices) draws to only render half of each quad (1 triangle instead of 2). Result: Fixes 46 additional tests (from 291 down to 245 failures).
…n and Y-flip in rtAdjust.
…before uploading to R8 atlas texture.
…ersion to onMakeTexture and using copy composite in readImagePixels.
… for JS-side mapAsync.
…sion across configure and build phases.
…ses synchronous getBufferSubData for readback.
…lready includes encode support.
…ss without Asyncify issues.
…date_baseline entry point.
edwardxfshen
force-pushed
the
feature/partyhuang_webgpu_backend
branch
from
9a9b89c to
9569a7d
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
为 Web 平台新增 WebGPU 渲染后端，与现有 WebGL 后端并行支持。
主要改动
WebGPU 后端核心实现
像素回读（readPixels）
测试基础设施
.cache/{backend}/，各后端 MD5 完全隔离update_baseline.sh作为唯一入口，支持USE_WEBGL/USE_WEBGPUaccept_baseline.sh自动同步 version.json 到 Web 端缓存工具链
setup_emsdk.sh自动检测和版本校验（需要 emsdk 4.0.15）测试结果
WebGPU 后端通过全部测试（仅排除
FlushSemaphore，WebGPU 无信号量语义）。