fix(dx12): Map components properly for textureLoad from stencil (#9520)

Author: andyleiserson

CHANGELOG.md

@@ -215,6 +215,7 @@ By @beholdnec in [#8505](https://github.com/gfx-rs/wgpu/pull/8505).
 
 - Fixed use of a texture view without `TextureUsage::TEXTURE_BINDING` as a read-only depth attachment. By @andyleiserson in [#9346](https://github.com/gfx-rs/wgpu/pull/9346).
 - Fixed a `debug_assert` during stride validation for indirect multi draw. By @kristoff3r in [#9332](https://github.com/gfx-rs/wgpu/pull/9332)
+- Fixed stencil values read with `textureLoad` appearing in G instead of R. By @andyleiserson in [#9520](https://github.com/gfx-rs/wgpu/pull/9520).
 
 #### Metal
 

cts_runner/fail.lst

@@ -11,6 +11,11 @@
 //
 // The `cts_runner` integration test `lst_files_are_sorted` verifies that test selectors
 // appear in this file in order -- including ones in comments.
+
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="depth24plus-stencil8";depthReadOnly=false;stencilReadOnly=true // https://github.com/gfx-rs/wgpu/issues/8705
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="depth24plus-stencil8";depthReadOnly=true;stencilReadOnly=false // https://github.com/gfx-rs/wgpu/issues/8705
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="depth32float-stencil8";depthReadOnly=false;stencilReadOnly=true // https://github.com/gfx-rs/wgpu/issues/8705
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="depth32float-stencil8";depthReadOnly=true;stencilReadOnly=false // https://github.com/gfx-rs/wgpu/issues/8705
 webgpu:api,operation,rendering,3d_texture_slices:* // https://github.com/gfx-rs/wgpu/issues/9455
 
 webgpu:api,validation,buffer,mapping:* // crash

cts_runner/test.lst

@@ -36,6 +36,15 @@ webgpu:api,operation,compute_pipeline,overrides:*
 //FAIL: webgpu:api,operation,compute,basic:large_dispatch:*
 webgpu:api,operation,compute,basic:memcpy:*
 webgpu:api,operation,device,lost:*
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="depth16unorm";*
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="depth24plus";*
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="depth24plus-stencil8";depthReadOnly=false;stencilReadOnly=false
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="depth24plus-stencil8";depthReadOnly=true;stencilReadOnly=true
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="depth32float";*
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="depth32float-stencil8";depthReadOnly=false;stencilReadOnly=false
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="depth32float-stencil8";depthReadOnly=true;stencilReadOnly=true
+webgpu:api,operation,memory_sync,texture,readonly_depth_stencil:sampling_while_testing:format="stencil8";*
+
 webgpu:api,operation,render_pass,storeOp:*
 webgpu:api,operation,render_pipeline,overrides:*
 webgpu:api,operation,render_pipeline,pipeline_output_targets:color,component_count,blend:*

wgpu-hal/src/dx12/conv.rs

@@ -447,3 +447,21 @@ pub(crate) fn map_acceleration_structure_copy_mode(
         }
     }
 }
+
+pub(crate) const fn make_shader_component_mapping(
+    src0: Direct3D12::D3D12_SHADER_COMPONENT_MAPPING,
+    src1: Direct3D12::D3D12_SHADER_COMPONENT_MAPPING,
+    src2: Direct3D12::D3D12_SHADER_COMPONENT_MAPPING,
+    src3: Direct3D12::D3D12_SHADER_COMPONENT_MAPPING,
+) -> Direct3D12::D3D12_SHADER_COMPONENT_MAPPING {
+    const M: i32 = Direct3D12::D3D12_SHADER_COMPONENT_MAPPING_MASK as i32;
+    const S: i32 = Direct3D12::D3D12_SHADER_COMPONENT_MAPPING_SHIFT as i32;
+    Direct3D12::D3D12_SHADER_COMPONENT_MAPPING(
+        (src0.0 & M)
+            | (src1.0 & M) << S
+            | (src2.0 & M) << (S * 2)
+            | (src3.0 & M) << (S * 3)
+            | Direct3D12::D3D12_SHADER_COMPONENT_MAPPING_ALWAYS_SET_BIT_AVOIDING_ZEROMEM_MISTAKES
+                as i32,
+    )
+}

wgpu-hal/src/dx12/view.rs

@@ -41,12 +41,33 @@ fn aspects_to_plane(aspects: crate::FormatAspects) -> u32 {
     }
 }
 
+/// Shader component mapping for stencil views
+///
+/// Stencil views use `DXGI_FORMAT_X24_TYPELESS_G8_UINT` or
+/// `DXGI_FORMAT_X32_TYPELESS_G8X24_UINT`, which have the stencil value in
+/// the green component. WebGPU specifies that the stencil value be in the
+/// red component. It also specifies that the remaining components _should_
+/// be (0, 0, 1), but may have an unspecified value.
+const STENCIL_COMPONENT_MAPPING: Direct3D12::D3D12_SHADER_COMPONENT_MAPPING =
+    super::conv::make_shader_component_mapping(
+        Direct3D12::D3D12_SHADER_COMPONENT_MAPPING_FROM_MEMORY_COMPONENT_1,
+        Direct3D12::D3D12_SHADER_COMPONENT_MAPPING_FORCE_VALUE_0,
+        Direct3D12::D3D12_SHADER_COMPONENT_MAPPING_FORCE_VALUE_0,
+        Direct3D12::D3D12_SHADER_COMPONENT_MAPPING_FORCE_VALUE_1,
+    );
+
 impl ViewDescriptor {
     pub(crate) unsafe fn to_srv(&self) -> Option<Direct3D12::D3D12_SHADER_RESOURCE_VIEW_DESC> {
+        let swizzle = if self.aspects == crate::FormatAspects::STENCIL {
+            STENCIL_COMPONENT_MAPPING.0 as u32
+        } else {
+            Direct3D12::D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING
+        };
+
         let mut desc = Direct3D12::D3D12_SHADER_RESOURCE_VIEW_DESC {
             Format: self.srv_uav_format?,
             ViewDimension: Direct3D12::D3D12_SRV_DIMENSION_UNKNOWN,
-            Shader4ComponentMapping: Direct3D12::D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING,
+            Shader4ComponentMapping: swizzle,
             Anonymous: Default::default(),
         };