[dx12] use typeless formats for textures that might be viewed as srgb or non-srgb (#3555)

query `CastingFullyTypedFormatSupported` and skip typeless formats if available

CHANGELOG.md

@@ -114,6 +114,7 @@ By @teoxoy in [#3534](https://github.com/gfx-rs/wgpu/pull/3534)
 #### DX12
 
 - Fix DXC validation issues when using a custom `dxil_path`. By @Elabajaba in [#3434](https://github.com/gfx-rs/wgpu/pull/3434)
+- Use typeless formats for textures that might be viewed as srgb or non-srgb. By @teoxoy in [#3555](https://github.com/gfx-rs/wgpu/pull/3555)
 
 #### GLES
 

wgpu-hal/src/auxil/dxgi/conv.rs

@@ -163,17 +163,47 @@ pub fn map_texture_format_for_copy(
     })
 }
 
-pub fn map_texture_format_depth_stencil_typeless(
+pub fn map_texture_format_for_resource(
     format: wgt::TextureFormat,
+    usage: crate::TextureUses,
+    has_view_formats: bool,
+    casting_fully_typed_format_supported: bool,
 ) -> dxgiformat::DXGI_FORMAT {
-    match format {
-        wgt::TextureFormat::Depth16Unorm => dxgiformat::DXGI_FORMAT_R16_TYPELESS,
-        wgt::TextureFormat::Depth32Float => dxgiformat::DXGI_FORMAT_R32_TYPELESS,
-        wgt::TextureFormat::Depth32FloatStencil8 => dxgiformat::DXGI_FORMAT_R32G8X24_TYPELESS,
-        wgt::TextureFormat::Stencil8
-        | wgt::TextureFormat::Depth24Plus
-        | wgt::TextureFormat::Depth24PlusStencil8 => dxgiformat::DXGI_FORMAT_R24G8_TYPELESS,
-        _ => unreachable!(),
+    use wgt::TextureFormat as Tf;
+    use winapi::shared::dxgiformat::*;
+
+    if casting_fully_typed_format_supported {
+        map_texture_format(format)
+
+    // We might view this resource as srgb or non-srgb
+    } else if has_view_formats {
+        match format {
+            Tf::Rgba8Unorm | Tf::Rgba8UnormSrgb => DXGI_FORMAT_R8G8B8A8_TYPELESS,
+            Tf::Bgra8Unorm | Tf::Bgra8UnormSrgb => DXGI_FORMAT_B8G8R8A8_TYPELESS,
+            Tf::Bc1RgbaUnorm | Tf::Bc1RgbaUnormSrgb => DXGI_FORMAT_BC1_TYPELESS,
+            Tf::Bc2RgbaUnorm | Tf::Bc2RgbaUnormSrgb => DXGI_FORMAT_BC2_TYPELESS,
+            Tf::Bc3RgbaUnorm | Tf::Bc3RgbaUnormSrgb => DXGI_FORMAT_BC3_TYPELESS,
+            Tf::Bc7RgbaUnorm | Tf::Bc7RgbaUnormSrgb => DXGI_FORMAT_BC7_TYPELESS,
+            format => map_texture_format(format),
+        }
+
+    // We might view this resource as SRV/UAV but also as DSV
+    } else if format.is_depth_stencil_format()
+        && usage.intersects(
+            crate::TextureUses::RESOURCE
+                | crate::TextureUses::STORAGE_READ
+                | crate::TextureUses::STORAGE_READ_WRITE,
+        )
+    {
+        match format {
+            Tf::Depth16Unorm => DXGI_FORMAT_R16_TYPELESS,
+            Tf::Depth32Float => DXGI_FORMAT_R32_TYPELESS,
+            Tf::Depth32FloatStencil8 => DXGI_FORMAT_R32G8X24_TYPELESS,
+            Tf::Stencil8 | Tf::Depth24Plus | Tf::Depth24PlusStencil8 => DXGI_FORMAT_R24G8_TYPELESS,
+            _ => unreachable!(),
+        }
+    } else {
+        map_texture_format(format)
     }
 }
 

wgpu-hal/src/dx12/adapter.rs

@@ -151,8 +151,18 @@ impl super::Adapter {
             hr == 0 && features2.DepthBoundsTestSupported != 0
         };
 
-        //Note: `D3D12_FEATURE_D3D12_OPTIONS3::CastingFullyTypedFormatSupported` can be checked
-        // to know if we can skip "typeless" formats entirely.
+        let casting_fully_typed_format_supported = {
+            let mut features3: crate::dx12::types::D3D12_FEATURE_DATA_D3D12_OPTIONS3 =
+                unsafe { mem::zeroed() };
+            let hr = unsafe {
+                device.CheckFeatureSupport(
+                    21, // D3D12_FEATURE_D3D12_OPTIONS3
+                    &mut features3 as *mut _ as *mut _,
+                    mem::size_of::<crate::dx12::types::D3D12_FEATURE_DATA_D3D12_OPTIONS3>() as _,
+                )
+            };
+            hr == 0 && features3.CastingFullyTypedFormatSupported != 0
+        };
 
         let private_caps = super::PrivateCapabilities {
             instance_flags,
@@ -166,6 +176,7 @@ impl super::Adapter {
                 super::MemoryArchitecture::NonUnified
             },
             heap_create_not_zeroed: false, //TODO: winapi support for Options7
+            casting_fully_typed_format_supported,
         };
 
         // Theoretically vram limited, but in practice 2^20 is the limit

wgpu-hal/src/dx12/device.rs

@@ -406,20 +406,12 @@ impl crate::Device<super::Api> for super::Device {
             Height: desc.size.height,
             DepthOrArraySize: desc.size.depth_or_array_layers as u16,
             MipLevels: desc.mip_level_count as u16,
-            Format: if crate::FormatAspects::from(desc.format).contains(crate::FormatAspects::COLOR)
-                || !desc.usage.intersects(
-                    crate::TextureUses::RESOURCE
-                        | crate::TextureUses::STORAGE_READ
-                        | crate::TextureUses::STORAGE_READ_WRITE,
-                ) {
-                auxil::dxgi::conv::map_texture_format(desc.format)
-            } else {
-                // This branch is needed if it's a depth texture, and it's ever needed to be viewed as SRV or UAV,
-                // because then we'd create a non-depth format view of it.
-                // Note: we can skip this branch if
-                // `D3D12_FEATURE_D3D12_OPTIONS3::CastingFullyTypedFormatSupported`
-                auxil::dxgi::conv::map_texture_format_depth_stencil_typeless(desc.format)
-            },
+            Format: auxil::dxgi::conv::map_texture_format_for_resource(
+                desc.format,
+                desc.usage,
+                !desc.view_formats.is_empty(),
+                self.private_caps.casting_fully_typed_format_supported,
+            ),
             SampleDesc: dxgitype::DXGI_SAMPLE_DESC {
                 Count: desc.sample_count,
                 Quality: 0,

wgpu-hal/src/dx12/mod.rs

@@ -41,6 +41,7 @@ mod device;
 mod instance;
 mod shader_compilation;
 mod suballocation;
+mod types;
 mod view;
 
 use crate::auxil::{self, dxgi::result::HResult as _};
@@ -174,6 +175,7 @@ struct PrivateCapabilities {
     memory_architecture: MemoryArchitecture,
     #[allow(unused)] // TODO: Exists until windows-rs is standard, then it can probably be removed?
     heap_create_not_zeroed: bool,
+    casting_fully_typed_format_supported: bool,
 }
 
 #[derive(Default)]

wgpu-hal/src/dx12/types.rs

@@ -0,0 +1,34 @@
+#![allow(non_camel_case_types)]
+#![allow(non_snake_case)]
+
+winapi::ENUM! {
+    enum D3D12_VIEW_INSTANCING_TIER {
+        D3D12_VIEW_INSTANCING_TIER_NOT_SUPPORTED  = 0,
+        D3D12_VIEW_INSTANCING_TIER_1 = 1,
+        D3D12_VIEW_INSTANCING_TIER_2 = 2,
+        D3D12_VIEW_INSTANCING_TIER_3 = 3,
+    }
+}
+
+winapi::ENUM! {
+    enum D3D12_COMMAND_LIST_SUPPORT_FLAGS {
+        D3D12_COMMAND_LIST_SUPPORT_FLAG_NONE = 0,
+        // D3D12_COMMAND_LIST_SUPPORT_FLAG_DIRECT,
+        // D3D12_COMMAND_LIST_SUPPORT_FLAG_BUNDLE,
+        // D3D12_COMMAND_LIST_SUPPORT_FLAG_COMPUTE,
+        // D3D12_COMMAND_LIST_SUPPORT_FLAG_COPY,
+        // D3D12_COMMAND_LIST_SUPPORT_FLAG_VIDEO_DECODE,
+        // D3D12_COMMAND_LIST_SUPPORT_FLAG_VIDEO_PROCESS,
+        // D3D12_COMMAND_LIST_SUPPORT_FLAG_VIDEO_ENCODE,
+    }
+}
+
+winapi::STRUCT! {
+    struct D3D12_FEATURE_DATA_D3D12_OPTIONS3 {
+        CopyQueueTimestampQueriesSupported: winapi::shared::minwindef::BOOL,
+        CastingFullyTypedFormatSupported: winapi::shared::minwindef::BOOL,
+        WriteBufferImmediateSupportFlags: D3D12_COMMAND_LIST_SUPPORT_FLAGS,
+        ViewInstancingTier: D3D12_VIEW_INSTANCING_TIER,
+        BarycentricsSupported: winapi::shared::minwindef::BOOL,
+    }
+}