Compat mode: make f16 and f32 rendering optional.

SenorBlanco · SenorBlanco · commit aab57bf3398e · 2024-11-20T12:19:11.000-05:00
Introduce float16-renderable and float32-renderable as optional features
in Compatibility mode, and required in Core.
diff --git a/proposals/compatibility-mode.md b/proposals/compatibility-mode.md
@@ -197,7 +197,19 @@ generate a validation error.
 other APIs only support the first vertex so only `@interpolation(flat, either)` is supported in
 compatibility mode.
 
-## 18. Introduce new `maxStorageBuffersInVertexStage` and `maxStorageTexturesInVertexStage` limits.
+## 18. Introduce new `float16-renderable` and `float32-renderable` features.
+
+When supported, `float16-renderable` allows the `RENDER_ATTACHMENT` usage on textures with format `"r16float"`, `"rg16float"`, and `"rgba16float"`.
+
+When supported, `float32-renderable` allows the `RENDER_ATTACHMENT` usage on textures with format `"r32float"`, `"rg32float"`, and `"rgba32float"`.
+
+Without support, an error will occur at texture creation time as described in section 6.1.3.
+
+Support for both features is mandatory in core WebGPU.
+
+**Justification**: OpenGL ES 3.1 does not require the relevant f16- or f32-based texture formats (`R16F`, `RG16F`, `RGBA16F`, `R32F`, `RG32F`, and `RGBA32F`) to be color-renderable. While there exist OpenGL ES extensions to enable renderability (`GL_EXT_COLOR_BUFFER_HALF_FLOAT` and `GL_EXT_COLOR_BUFFER_FLOAT`), there are a significant number of devices which lack support for these extensions.
+
+## 19. Introduce new `maxStorageBuffersInVertexStage` and `maxStorageTexturesInVertexStage` limits.
 
 If the number of shader variables of type `storage_buffer` in a vertex shader exceeds the `maxStorageBuffersInVertexStage` limit, a validation error will occur at pipeline creation time.
 
