CNA-GAP-076: WEBGPU and DIRECTX9 refuse a RenderTargetCube face in a multiple-render-target set, although both report MultipleRenderTargets

CNA snapshot 3b43bd86  ·  Known Issues › Functional gaps  ·  source links pinned to 3b43bd86

SetRenderTargets with a cube face among two or more bindings throws NotSupportedException on WEBGPU and DIRECTX9, while XNA's HiDef profile accepts it and SDL_GPU binds it since 51fb39d5a; a single cube face and sets of 2D targets work.

Identifier

CNA-GAP-076

Category

Functional gap

Subsystem

Graphics & renderers

Status

Open

Verified against

CNA 3b43bd86 (3b43bd8687ed46523d873f9bec4b32bd8481dcb5)

Evidence basis

Source-verified: read at TARGET, not executed

Tests touching this area

Yes: see Current tests

Affected contract

GraphicsDevice::SetRenderTargets with a RenderTargetBinding for a RenderTargetCube face among several bindings (HiDef), on WEBGPU and DIRECTX9

Expected behaviour

XNA 4.0's HiDef profile accepts a RenderTargetCube face in any slot of a multiple-render-target set and refuses only the sets it names (the same cube twice, different sample counts or sizes). CNA's conformance program cube_mrt_binding_test.cpp checks that each face slot receives its own pixel-shader output and that the refused sets are refused.

Actual behaviour at TARGET

WebGPURenderer.cpp's SetRenderTargets throws System::NotSupportedException (“WebGPU: a RenderTargetCube face cannot be bound in a multiple-render-target set ... Bind up to 4 RenderTarget2D targets instead”) for any cube face in a set of two or more, and DirectX9Renderer.cpp's does the same (“cube faces in a multi-target set are not implemented by this CNA renderer”), while both report GraphicsCapability::MultipleRenderTargets. A single bound cube face and sets of RenderTarget2D targets work on both. handoff.md section 5.3 records it as a capability boundary: cube_mrt_binding M1-M4 and the plural cube-face RenderTargetSemantics cases skip on the refusal there, and HEADLESS accepts such sets. SDL_GPU binds a cube face in any slot since 51fb39d5a (CNA-DEMOS DC-R1).

Source locations

• modules/renderers/webgpu/src/WebGPURenderer.cpp — SetRenderTargets: a RenderTargetCube face in an MRT set throws NotSupportedException
• modules/renderers/directx9/src/DirectX9Renderer.cpp — SetRenderTargets: cube faces in a multi-target set are not implemented
• modules/graphics/examples/conformance/cube_mrt_binding_test.cpp — M1-M4 face slots and R1-R4 refusals
• handoff.md — section 5.3: WEBGPU and DIRECTX9 refuse cube faces in a multi-render-target set; SDL_GPU binds them since 51fb39d5a
• CNA_DEMOS_REPAIR_MATRIX.json — additional_validation_findings DC-R1: SDL_GPU, FIXED_VALIDATED

Evidence

Read at 3b43bd86: both refusals and the conformance program. That the M-legs skip on the refusal on WEBGPU and DIRECTX9 is CNA's record (handoff.md); nothing was built or executed for this entry, and DIRECTX9 runs only under Wine. An earlier pin's review did not list the boundary as an entry while SDL_GPU refused it too.

Focused reproduction

No focused reproduction is known. Nothing has been invented here; the evidence above is what exists.

Current tests

cube_mrt_binding_test.cpp (registered through the parity fixture list; on SDL_GPU SdlGpu_Conformance_cube_mrt_binding) asserts the face slots where a renderer binds them and skips where the binding is refused; nothing requires WEBGPU or DIRECTX9 to bind them.

Regression test

cube_mrt_binding M1-M4 passing on WEBGPU and DIRECTX9 instead of skipping.

Blast radius

HiDef games that render into cube faces together with other targets in one pass (for example a dynamic environment map with a second output) on WEBGPU or DIRECTX9. Single cube-face rendering, 2D multiple render targets and SDL_GPU, EasyGL and SOFTWARE are unaffected.

Workaround

Render each cube face in its own pass, or bind 2D targets and copy the results into the cube.