Bump bgfx.cmake (TextureRegion blit/read, shader bin v12, Vulkan bindings)

[bkaradzic-microsoft]

Summary

Bumps bgfx.cmake to 9a569c775917e3dd2a8cfb8836aeddbe790a8deb (bgfx 0837ead via BabylonJS/bgfx#78 / bgfx.cmake#144) and adapts Babylon Native to the Graphics/shader packaging changes in that range.

API adaptations

• Blit / read: encoder.blit / bgfx::blit take TextureRegion pairs; bgfx::readTexture → bgfx::read(TextureRegion, data). Updated in FrameBuffer, DeviceImpl, NativeCapture, NativeEngine.
• Shader binary v12: CreateBgfxShader writes BGFX_CHUNK_MAGIC_*S_VERSION 12, emits per-uniform texture meta (AppendUniformTextureMeta), and carries D3D SRV/UAV masks.
• Canvas headers: embedded canvas shader headers bumped v11 → v12.
• Depth formats: D3D depth RT prefers D32F → D24 → D24S8 (Android depth-only stays D24S8). Backbuffer Depth32 maps to D32F (D3D has no DSV for plain D32).
• ShaderCache: CACHE_VERSION 4 → 5 (v12 binaries are not compatible with prior cache entries).

Vulkan packaging (shaderc / bgfx v12 binding model)

• Shared VS/FS layoutBinding per sampler name (no per-stage compact remap that desynced UniformStages).
• Image bindings start at 2; separate samplers at image + 16; UBOs stay 0/1.
• AppendSamplersVulkan: separate_images, strip Texture suffix, set regIndex from binding.
• Reject shaders with >16 distinct sampler textures (bgfx texture stage limit) with an explicit error.
• Packaging helpers live in ShaderCompilerVulkan.cpp / common CreateBgfxShader sampler-set path (not #if VULKAN in Common).

Tests / CI

• Stage-sampler tests: disjoint VS/FS names and partially shared names (D3D).
• ShaderCompilation.VulkanRejectsMoreThanSixteenSamplers (compiled only when GRAPHICS_API=Vulkan): unexpected success fails outside the catch path (no false pass via a thrown "…>16…" message).
• New Ubuntu_Clang_Vulkan CI job: GRAPHICS_API=Vulkan, lavapipe ICD, runs that gtest.
• Minimal Vulkan RendererType / DeviceImpl_Vulkan / ExternalTexture_Vulkan stub / Helpers.Vulkan so the API can configure and link. ExternalTexture remains unimplemented (SKIP_EXTERNAL_TEXTURE_TESTS).

Upstream deps already merged

• UWP: QI IDXGISwapChain2 before SetMaximumFrameLatency bgfx#77 (QI IDXGISwapChain2 before SetMaximumFrameLatency)
• Define local IID_IDXGISwapChain2 GUID bgfx#78 (local IID_IDXGISwapChain2 GUID; UWP does not link dxguid)
• Matching bgfx.cmake pins V8 teardown crash: order-of-destruction between JS context and Napi::Env #143 / Added minimal change to allow v8 to shut down correctly. #144

Test plan

• Existing Win32 D3D11/D3D12 CI paths
• Linux OpenGL UnitTests / validation (existing jobs)
• Ubuntu_Clang_Vulkan green (build + VulkanRejectsMoreThanSixteenSamplers under lavapipe)
• Spot-check motion blur / thin-instance scenarios if any residual bgfx behavior changes show up in playground validation

[Copilot]

🔵 Needs a closer look

It changes shader binary packaging and core GPU blit/readback paths across multiple backends, which is high-impact and needs full CI/platform validation.

Pull request overview

Updates Babylon Native’s bgfx integration to a newer bgfx.cmake/bgfx revision and adapts engine/shader tooling to match breaking API and shader-binary format changes introduced in that range.

Changes:

• Bump bgfx.cmake to 1a135dbf... (bgfx 3b85a1c) via CMakeLists.txt.
• Update runtime code to use TextureRegion-based blit/read APIs (encoder->blit(...) and bgfx::read(TextureRegion, ...)).
• Update shader binary emission to bgfx shader package v12 (including v12-required fields) and bump embedded Canvas precompiled shader headers to v12.

File summaries

File	Description
Polyfills/Canvas/Source/Shaders/spirv/vs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/spirv/vs_fspass.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/spirv/fs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/spirv/fs_gaussblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/spirv/fs_boxblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/metal/vs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/metal/vs_fspass.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/metal/fs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/metal/fs_gaussblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/metal/fs_boxblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/glsl/vs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/glsl/vs_fspass.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/glsl/fs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/glsl/fs_gaussblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/glsl/fs_boxblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/essl/vs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/essl/vs_fspass.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/essl/fs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/essl/fs_gaussblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/essl/fs_boxblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/dxil/vs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/dxil/vs_fspass.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/dxil/fs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/dxil/fs_gaussblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/dxil/fs_boxblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/dxbc/vs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/dxbc/vs_fspass.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/dxbc/fs_nanovg_fill.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/dxbc/fs_gaussblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Polyfills/Canvas/Source/Shaders/dxbc/fs_boxblur.h	Bump embedded shader binary header to v12 format (adds v12 fields).
Plugins/ShaderCompiler/Source/ShaderCompilerCommon.cpp	Emit bgfx shader binary v12 (raw SRV/UAV masks + per-uniform texture metadata).
Plugins/NativeEngine/Source/NativeEngine.cpp	Update blit + readback paths to TextureRegion APIs (copy + readTexture pipeline).
Plugins/NativeCapture/Source/NativeCapture.cpp	Update framebuffer readback blit to TextureRegion API.
Core/Graphics/Source/FrameBuffer.cpp	Update FrameBuffer::Blit to TextureRegion API and preserve “whole texture” semantics.
Core/Graphics/Source/DeviceImpl.cpp	Switch async readback to bgfx::read(TextureRegion, ...).
CMakeLists.txt	Bump bgfx.cmake dependency tag to the targeted commit.

Review details

Suppressed comments (2)

Plugins/NativeEngine/Source/NativeEngine.cpp:2375

• This ReadTexture blit fallback block is over-indented compared to the rest of the else-branch, which makes it look like it is nested differently than it actually is.

            // If the image needs to be cropped, the texture lacks the READ_BACK flag, or we are reading a
                        // specific cube-map face, blit to a temp 2D texture. bgfx::read addresses a whole mip of one
                        // slice via TextureRegion::z, but a cropped sub-rect still needs the blit path. Cube-face
                        // reads use srcZ = face index on the source region of that blit.
                        if (isCubeFace || x != 0 || y != 0 || width != mipWidth || height != mipHeight || (texture->Flags() & BGFX_TEXTURE_READ_BACK) == 0)

Plugins/ShaderCompiler/Source/ShaderCompilerCommon.cpp:312

• This BGFX_SHADER_BIN_VERSION/comment block is over-indented compared to the rest of CreateBgfxShader, which makes the shader-binary header layout harder to follow.

        // Must match BGFX_SHADER_BIN_VERSION in bgfx tools/shaderc/shaderc.cpp.
                // v12 requires raw SRV/UAV binding masks after the in/out hashes, and
                // uniform entries always carry texComponent/texDimension/texFormat (v8/v10).
                constexpr uint8_t BGFX_SHADER_BIN_VERSION{12};

• Files reviewed: 36/36 changed files
• Comments generated: 5
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[bkaradzic-microsoft]

UWP-only remaining failure was a bgfx compile error from present-pacing:

dxgi.cpp: SetMaximumFrameLatency is not a member of IDXGISwapChain1

WinRT SwapChainI is IDXGISwapChain1; the method is on IDXGISwapChain2+.

Fixed upstream:

• UWP: QI IDXGISwapChain2 before SetMaximumFrameLatency bgfx#77
• Bump bgfx for UWP IDXGISwapChain2 frame-latency fix bgfx.cmake#143 (ede5513a)

BN pin bumped in 02c0586.

[Copilot]

🟡 Changes recommended

The v12 shader package selects bgfx’s new Vulkan binding model without emitting its required shifted bindings.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Review details

• Files reviewed: 36/36 changed files
• Comments generated: 1
• Review effort level: Balanced

[bghgary]

[Reviewed by Copilot on behalf of @bghgary]

Concerns inline.

[bghgary]

[Reviewed by Copilot on behalf of @bghgary]

Concern and nits inline.

[bghgary]

[Commented by Copilot on behalf of @bghgary]

The Vulkan sampler-limit regression test merged with two gaps: no CI job configures GRAPHICS_API=Vulkan, so it is compiled out everywhere; and an unexpectedly ready effect throws "expected compile failure for >16 samplers" inside the same try, so the catch plus "16" assertion treats success as the expected rejection. Please make unexpected success fail outside the caught path and add a Vulkan CI configuration that runs the test.

[bkaradzic-microsoft]

Addressed:

1. False pass on unexpected success — VulkanRejectsMoreThanSixteenSamplers now records unexpected_success / threw / not_ready without throwing a “>16” string inside the try. Unexpected success fails with ASSERT_NE(status, "unexpected_success"); the “16” check only runs on a real rejection path (JS throw, not-ready, or native unhandled exception).

2. Vulkan CI — added Ubuntu_Clang_Vulkan (GRAPHICS_API=Vulkan, Mesa lavapipe, gtest filter ShaderCompilation.VulkanRejectsMoreThanSixteenSamplers). Also added the minimal Graphics/ExternalTexture/UnitTest Vulkan stubs needed so that configuration can configure and link (ExternalTexture still skipped).

Pushed in 79b363c.

[bkaradzic-microsoft]

#1857 was already merged when this landed on the fork branch. Follow-up is in a new PR:

#1859

That PR has:

1. Test fix so unexpected success no longer false-passes via a thrown “>16” string inside the catch path
2. Ubuntu_Clang_Vulkan CI (GRAPHICS_API=Vulkan + lavapipe) running ShaderCompilation.VulkanRejectsMoreThanSixteenSamplers
3. Minimal Vulkan Graphics/ExternalTexture/UnitTest stubs so the job can configure and link