D3D11: Fix UAV creation failure for P8/format-convert textures

Textures created for GPU palette expansion and format conversion were
using DXGI_FORMAT_R8G8B8A8_UNORM, which does not allow R32_UINT UAV
reinterpretation. This caused CreateUnorderedAccessView to fail,
falling back to the slow CPU path on every upload.

Fix: create these textures as R8G8B8A8_TYPELESS (same typeless family
as R32_UINT), and resolve to R8G8B8A8_UNORM when creating SRVs for
sampling in HostSync and the blit path.

Author: PatrickvL

src/core/hle/D3D8/Rendering/Backend/Backend_D3D11.cpp

@@ -786,7 +786,7 @@ HRESULT CxbxD3D11Blt(
 	// Create temporary SRV for source
 	ID3D11ShaderResourceView* pSRV = nullptr;
 	D3D11_SHADER_RESOURCE_VIEW_DESC srvDesc = {};
-	srvDesc.Format = srcDesc.Format;
+	srvDesc.Format = (srcDesc.Format == DXGI_FORMAT_R8G8B8A8_TYPELESS) ? DXGI_FORMAT_R8G8B8A8_UNORM : srcDesc.Format;
 	srvDesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE2D;
 	srvDesc.Texture2D.MipLevels = 1;
 	HRESULT hr = g_pD3DDevice->CreateShaderResourceView(pSrc, &srvDesc, &pSRV);

src/core/hle/D3D8/Rendering/HostResourceCreate.cpp

@@ -324,16 +324,16 @@ static HRESULT CreateGpuPixelContainerResource(
 				desc.BindFlags = D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_UNORDERED_ACCESS;
 				desc.CPUAccessFlags = 0;
 			} else if (bSwizzled && X_Format == xbox::X_D3DFMT_P8) {
-				// P8 textures use DEFAULT + UAV for GPU palette expand CS
-				// Use R8G8B8A8_UNORM (not B8G8R8A8) for R32_UINT UAV compatibility
-				desc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
+				// P8 textures use DEFAULT + UAV for GPU palette expand CS.
+				// Typeless allows R32_UINT UAV reinterpretation; SRV uses R8G8B8A8_UNORM.
+				desc.Format = DXGI_FORMAT_R8G8B8A8_TYPELESS;
 				desc.Usage = D3D11_USAGE_DEFAULT;
 				desc.BindFlags = D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_UNORDERED_ACCESS;
 				desc.CPUAccessFlags = 0;
 			} else if (bConvertTextureFormat && CxbxGetFormatConvertType(X_Format) != 0) {
-				// Format-convertible textures use DEFAULT + UAV for GPU CS format conversion
-				// Output is always R8G8B8A8_UNORM (CS writes packed RGBA via R32_UINT UAV)
-				desc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
+				// Format-convertible textures use DEFAULT + UAV for GPU CS format conversion.
+				// Typeless allows R32_UINT UAV reinterpretation; SRV uses R8G8B8A8_UNORM.
+				desc.Format = DXGI_FORMAT_R8G8B8A8_TYPELESS;
 				desc.Usage = D3D11_USAGE_DEFAULT;
 				desc.BindFlags = D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_UNORDERED_ACCESS;
 				desc.CPUAccessFlags = 0;

src/core/hle/D3D8/Rendering/HostSync.cpp

@@ -344,7 +344,12 @@ static void CxbxBindTextureSRV(int stage, ID3D11Resource* pHostBaseTexture, bool
 	case D3D11_RESOURCE_DIMENSION_TEXTURE2D: {
 		D3D11_TEXTURE2D_DESC texDesc = {};
 		((ID3D11Texture2D*)pHostBaseTexture)->GetDesc(&texDesc);
-		srvDesc.Format = IsDepthFormat(texDesc.Format) ? GetDepthSRVFormat(texDesc.Format) : texDesc.Format;
+		if (IsDepthFormat(texDesc.Format))
+			srvDesc.Format = GetDepthSRVFormat(texDesc.Format);
+		else if (texDesc.Format == DXGI_FORMAT_R8G8B8A8_TYPELESS)
+			srvDesc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
+		else
+			srvDesc.Format = texDesc.Format;
 		if (texDesc.ArraySize == 6) {
 			srvDesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURECUBE;
 			srvDesc.TextureCube.MipLevels = texDesc.MipLevels;