[d3d9] Implement YV12 video format

Needed for #1726 otherwise it will upload dump that upload garbage in a YUV2 texture.

src/d3d9/d3d9_common_texture.cpp

@@ -189,7 +189,7 @@ namespace dxvk {
 
     const uint32_t planeCount = m_mapping.ConversionFormatInfo.PlaneCount;
 
-    return planeCount
+    return std::min(planeCount, 2u)
          * formatInfo.elementSize
          * blockCount.width
          * blockCount.height

src/d3d9/d3d9_format.cpp

@@ -411,6 +411,15 @@ namespace dxvk {
         { D3D9ConversionFormat_NV12, 2u, VK_FORMAT_B8G8R8A8_UNORM }
       };
 
+      case D3D9Format::YV12: return {
+        VK_FORMAT_R8_UNORM,
+        VK_FORMAT_UNDEFINED,
+        VK_IMAGE_ASPECT_COLOR_BIT,
+        { VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY,
+          VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY },
+        { D3D9ConversionFormat_YV12, 3u, VK_FORMAT_B8G8R8A8_UNORM }
+      };
+
       case D3D9Format::RAWZ: return {}; // Unsupported
 
       default:

src/d3d9/d3d9_format.h

@@ -135,6 +135,7 @@ namespace dxvk {
     D3D9ConversionFormat_X8L8V8U8,
     D3D9ConversionFormat_A2W10V10U10,
     D3D9ConversionFormat_NV12,
+    D3D9ConversionFormat_YV12,
     D3D9ConversionFormat_Count
   };
 

src/d3d9/d3d9_format_helpers.cpp

@@ -5,6 +5,7 @@
 #include <d3d9_convert_x8l8v8u8.h>
 #include <d3d9_convert_a2w10v10u10.h>
 #include <d3d9_convert_nv12.h>
+#include <d3d9_convert_yv12.h>
 
 namespace dxvk {
 
@@ -40,6 +41,10 @@ namespace dxvk {
         ConvertGenericFormat(conversionFormat, dstImage, dstSubresource, srcBuffer, VK_FORMAT_R16_UINT, 0, { 2u, 1u });
         break;
 
+      case D3D9ConversionFormat_YV12:
+        ConvertGenericFormat(conversionFormat, dstImage, dstSubresource, srcBuffer, VK_FORMAT_R8_UINT, 0, { 1u, 1u });
+        break;
+
       case D3D9ConversionFormat_L6V5U5:
         ConvertGenericFormat(conversionFormat, dstImage, dstSubresource, srcBuffer, VK_FORMAT_R16_UINT, 0, { 1u, 1u });
         break;
@@ -115,6 +120,7 @@ namespace dxvk {
     m_shaders[D3D9ConversionFormat_X8L8V8U8] = InitShader(d3d9_convert_x8l8v8u8);
     m_shaders[D3D9ConversionFormat_A2W10V10U10] = InitShader(d3d9_convert_a2w10v10u10);
     m_shaders[D3D9ConversionFormat_NV12] = InitShader(d3d9_convert_nv12);
+    m_shaders[D3D9ConversionFormat_YV12] = InitShader(d3d9_convert_yv12);
   }
 
 

src/d3d9/meson.build

@@ -7,7 +7,8 @@ d3d9_shaders = files([
   'shaders/d3d9_convert_l6v5u5.comp',
   'shaders/d3d9_convert_x8l8v8u8.comp',
   'shaders/d3d9_convert_a2w10v10u10.comp',
-  'shaders/d3d9_convert_nv12.comp'
+  'shaders/d3d9_convert_nv12.comp',
+  'shaders/d3d9_convert_yv12.comp'
 ])
 
 d3d9_src = [

src/d3d9/shaders/d3d9_convert_yv12.comp

@@ -0,0 +1,58 @@
+#version 450
+#extension GL_GOOGLE_include_directive : enable
+
+#include "d3d9_convert_common.h"
+
+layout(
+  local_size_x = 8,
+  local_size_y = 8,
+  local_size_z = 1) in;
+
+layout(binding = 0)
+writeonly uniform image2D dst;
+
+layout(binding = 1) uniform usamplerBuffer src;
+
+layout(push_constant)
+uniform u_info_t {
+  uvec2 extent;
+} u_info;
+
+// Format is:
+// YYYYYYYY...
+// VVVV...
+// UUUU...
+
+float fetchUnorm(usamplerBuffer source, uint offset) {
+  return unpackUnorm(texelFetch(src, int(offset)).r);
+}
+
+void main() {
+  ivec3 thread_id = ivec3(gl_GlobalInvocationID);
+
+  if (all(lessThan(thread_id.xy, u_info.extent))) {
+    uvec2 pitch = uvec2(u_info.extent.x, u_info.extent.y);
+
+    uint offset = thread_id.x
+                + thread_id.y * pitch.x;
+
+    // Fetch a Y, luminance sample.
+    float y = fetchUnorm(src, offset) - (16 / 255.0);        
+
+    // Go into the second plane to get a V, chroma sample
+    offset = (thread_id.x / 2)
+           + (thread_id.y / 2) * (pitch.x / 2)
+           + pitch.x * pitch.y;
+
+    float v = fetchUnorm(src, offset) - (128 / 255.0);
+
+    // Go into the third plane to get a U, chroma sample
+    offset += (pitch.x / 2) * (pitch.y / 2);
+    float u = fetchUnorm(src, offset) - (128 / 255.0);
+
+    // TODO: Is this the right color space?
+    vec4 color = convertBT_709(vec3(y, u, v));
+
+    imageStore(dst, thread_id.xy, color);
+  }
+}