fix(win/video): don't offload chroma subsampling math to texture sampler when downscaling

src/platform/windows/display_vram.cpp

@@ -107,6 +107,10 @@
   blob_t convert_yuv420_packed_uv_type0_ps_linear_hlsl;
   blob_t convert_yuv420_packed_uv_type0_ps_perceptual_quantizer_hlsl;
   blob_t convert_yuv420_packed_uv_type0_vs_hlsl;
+  blob_t convert_yuv420_packed_uv_type0s_ps_hlsl;
+  blob_t convert_yuv420_packed_uv_type0s_ps_linear_hlsl;
+  blob_t convert_yuv420_packed_uv_type0s_ps_perceptual_quantizer_hlsl;
+  blob_t convert_yuv420_packed_uv_type0s_vs_hlsl;
   blob_t convert_yuv420_planar_y_ps_hlsl;
   blob_t convert_yuv420_planar_y_ps_linear_hlsl;
   blob_t convert_yuv420_planar_y_ps_perceptual_quantizer_hlsl;
@@ -488,6 +492,110 @@
       frame_texture->AddRef();
       output_texture.reset(frame_texture);
 
+      HRESULT status = S_OK;
+
+#define create_vertex_shader_helper(x, y)                                                                    \
+  if (FAILED(status = device->CreateVertexShader(x->GetBufferPointer(), x->GetBufferSize(), nullptr, &y))) { \
+    BOOST_LOG(error) << "Failed to create vertex shader " << #x << ": " << util::log_hex(status);            \
+    return -1;                                                                                               \
+  }
+#define create_pixel_shader_helper(x, y)                                                                    \
+  if (FAILED(status = device->CreatePixelShader(x->GetBufferPointer(), x->GetBufferSize(), nullptr, &y))) { \
+    BOOST_LOG(error) << "Failed to create pixel shader " << #x << ": " << util::log_hex(status);            \
+    return -1;                                                                                              \
+  }
+
+      const bool downscaling = display->width > width || display->height > height;
+
+      switch (format) {
+        case DXGI_FORMAT_NV12:
+          // Semi-planar 8-bit YUV 4:2:0
+          create_vertex_shader_helper(convert_yuv420_planar_y_vs_hlsl, convert_Y_or_YUV_vs);
+          create_pixel_shader_helper(convert_yuv420_planar_y_ps_hlsl, convert_Y_or_YUV_ps);
+          create_pixel_shader_helper(convert_yuv420_planar_y_ps_linear_hlsl, convert_Y_or_YUV_fp16_ps);
+          if (downscaling) {
+            create_vertex_shader_helper(convert_yuv420_packed_uv_type0s_vs_hlsl, convert_UV_vs);
+            create_pixel_shader_helper(convert_yuv420_packed_uv_type0s_ps_hlsl, convert_UV_ps);
+            create_pixel_shader_helper(convert_yuv420_packed_uv_type0s_ps_linear_hlsl, convert_UV_fp16_ps);
+          }
+          else {
+            create_vertex_shader_helper(convert_yuv420_packed_uv_type0_vs_hlsl, convert_UV_vs);
+            create_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_hlsl, convert_UV_ps);
+            create_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_linear_hlsl, convert_UV_fp16_ps);
+          }
+          break;
+
+        case DXGI_FORMAT_P010:
+          // Semi-planar 16-bit YUV 4:2:0, 10 most significant bits store the value
+          create_vertex_shader_helper(convert_yuv420_planar_y_vs_hlsl, convert_Y_or_YUV_vs);
+          create_pixel_shader_helper(convert_yuv420_planar_y_ps_hlsl, convert_Y_or_YUV_ps);
+          if (display->is_hdr()) {
+            create_pixel_shader_helper(convert_yuv420_planar_y_ps_perceptual_quantizer_hlsl, convert_Y_or_YUV_fp16_ps);
+          }
+          else {
+            create_pixel_shader_helper(convert_yuv420_planar_y_ps_linear_hlsl, convert_Y_or_YUV_fp16_ps);
+          }
+          if (downscaling) {
+            create_vertex_shader_helper(convert_yuv420_packed_uv_type0s_vs_hlsl, convert_UV_vs);
+            create_pixel_shader_helper(convert_yuv420_packed_uv_type0s_ps_hlsl, convert_UV_ps);
+            if (display->is_hdr()) {
+              create_pixel_shader_helper(convert_yuv420_packed_uv_type0s_ps_perceptual_quantizer_hlsl, convert_UV_fp16_ps);
+            }
+            else {
+              create_pixel_shader_helper(convert_yuv420_packed_uv_type0s_ps_linear_hlsl, convert_UV_fp16_ps);
+            }
+          }
+          else {
+            create_vertex_shader_helper(convert_yuv420_packed_uv_type0_vs_hlsl, convert_UV_vs);
+            create_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_hlsl, convert_UV_ps);
+            if (display->is_hdr()) {
+              create_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_perceptual_quantizer_hlsl, convert_UV_fp16_ps);
+            }
+            else {
+              create_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_linear_hlsl, convert_UV_fp16_ps);
+            }
+          }
+          break;
+
+        case DXGI_FORMAT_R16_UINT:
+          // Planar 16-bit YUV 4:4:4, 10 most significant bits store the value
+          create_vertex_shader_helper(convert_yuv444_planar_vs_hlsl, convert_Y_or_YUV_vs);
+          create_pixel_shader_helper(convert_yuv444_planar_ps_hlsl, convert_Y_or_YUV_ps);
+          if (display->is_hdr()) {
+            create_pixel_shader_helper(convert_yuv444_planar_ps_perceptual_quantizer_hlsl, convert_Y_or_YUV_fp16_ps);
+          }
+          else {
+            create_pixel_shader_helper(convert_yuv444_planar_ps_linear_hlsl, convert_Y_or_YUV_fp16_ps);
+          }
+          break;
+
+        case DXGI_FORMAT_AYUV:
+          // Packed 8-bit YUV 4:4:4
+          create_vertex_shader_helper(convert_yuv444_packed_vs_hlsl, convert_Y_or_YUV_vs);
+          create_pixel_shader_helper(convert_yuv444_packed_ayuv_ps_hlsl, convert_Y_or_YUV_ps);
+          create_pixel_shader_helper(convert_yuv444_packed_ayuv_ps_linear_hlsl, convert_Y_or_YUV_fp16_ps);
+          break;
+
+        case DXGI_FORMAT_Y410:
+          // Packed 10-bit YUV 4:4:4
+          create_vertex_shader_helper(convert_yuv444_packed_vs_hlsl, convert_Y_or_YUV_vs);
+          create_pixel_shader_helper(convert_yuv444_packed_y410_ps_hlsl, convert_Y_or_YUV_ps);
+          if (display->is_hdr()) {
+            create_pixel_shader_helper(convert_yuv444_packed_y410_ps_perceptual_quantizer_hlsl, convert_Y_or_YUV_fp16_ps);
+          }
+          else {
+            create_pixel_shader_helper(convert_yuv444_packed_y410_ps_linear_hlsl, convert_Y_or_YUV_fp16_ps);
+          }
+          break;
+
+        default:
+          BOOST_LOG(error) << "Unable to create shaders because of the unrecognized surface format";
+          return -1;
+      }
+
+#undef create_vertex_shader_helper
+#undef create_pixel_shader_helper
+
       auto out_width = width;
       auto out_height = height;
 
@@ -676,83 +784,6 @@
         BOOST_LOG(warning) << "Failed to increase encoding GPU thread priority. Please run application as administrator for optimal performance.";
       }
 
-#define create_vertex_shader_helper(x, y)                                                                    \
-  if (FAILED(status = device->CreateVertexShader(x->GetBufferPointer(), x->GetBufferSize(), nullptr, &y))) { \
-    BOOST_LOG(error) << "Failed to create vertex shader " << #x << ": " << util::log_hex(status);            \
-    return -1;                                                                                               \
-  }
-#define create_pixel_shader_helper(x, y)                                                                    \
-  if (FAILED(status = device->CreatePixelShader(x->GetBufferPointer(), x->GetBufferSize(), nullptr, &y))) { \
-    BOOST_LOG(error) << "Failed to create pixel shader " << #x << ": " << util::log_hex(status);            \
-    return -1;                                                                                              \
-  }
-
-      switch (format) {
-        case DXGI_FORMAT_NV12:
-          // Semi-planar 8-bit YUV 4:2:0
-          create_vertex_shader_helper(convert_yuv420_planar_y_vs_hlsl, convert_Y_or_YUV_vs);
-          create_pixel_shader_helper(convert_yuv420_planar_y_ps_hlsl, convert_Y_or_YUV_ps);
-          create_pixel_shader_helper(convert_yuv420_planar_y_ps_linear_hlsl, convert_Y_or_YUV_fp16_ps);
-          create_vertex_shader_helper(convert_yuv420_packed_uv_type0_vs_hlsl, convert_UV_vs);
-          create_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_hlsl, convert_UV_ps);
-          create_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_linear_hlsl, convert_UV_fp16_ps);
-          break;
-
-        case DXGI_FORMAT_P010:
-          // Semi-planar 16-bit YUV 4:2:0, 10 most significant bits store the value
-          create_vertex_shader_helper(convert_yuv420_planar_y_vs_hlsl, convert_Y_or_YUV_vs);
-          create_pixel_shader_helper(convert_yuv420_planar_y_ps_hlsl, convert_Y_or_YUV_ps);
-          create_vertex_shader_helper(convert_yuv420_packed_uv_type0_vs_hlsl, convert_UV_vs);
-          create_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_hlsl, convert_UV_ps);
-          if (display->is_hdr()) {
-            create_pixel_shader_helper(convert_yuv420_planar_y_ps_perceptual_quantizer_hlsl, convert_Y_or_YUV_fp16_ps);
-            create_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_perceptual_quantizer_hlsl, convert_UV_fp16_ps);
-          }
-          else {
-            create_pixel_shader_helper(convert_yuv420_planar_y_ps_linear_hlsl, convert_Y_or_YUV_fp16_ps);
-            create_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_linear_hlsl, convert_UV_fp16_ps);
-          }
-          break;
-
-        case DXGI_FORMAT_R16_UINT:
-          // Planar 16-bit YUV 4:4:4, 10 most significant bits store the value
-          create_vertex_shader_helper(convert_yuv444_planar_vs_hlsl, convert_Y_or_YUV_vs);
-          create_pixel_shader_helper(convert_yuv444_planar_ps_hlsl, convert_Y_or_YUV_ps);
-          if (display->is_hdr()) {
-            create_pixel_shader_helper(convert_yuv444_planar_ps_perceptual_quantizer_hlsl, convert_Y_or_YUV_fp16_ps);
-          }
-          else {
-            create_pixel_shader_helper(convert_yuv444_planar_ps_linear_hlsl, convert_Y_or_YUV_fp16_ps);
-          }
-          break;
-
-        case DXGI_FORMAT_AYUV:
-          // Packed 8-bit YUV 4:4:4
-          create_vertex_shader_helper(convert_yuv444_packed_vs_hlsl, convert_Y_or_YUV_vs);
-          create_pixel_shader_helper(convert_yuv444_packed_ayuv_ps_hlsl, convert_Y_or_YUV_ps);
-          create_pixel_shader_helper(convert_yuv444_packed_ayuv_ps_linear_hlsl, convert_Y_or_YUV_fp16_ps);
-          break;
-
-        case DXGI_FORMAT_Y410:
-          // Packed 10-bit YUV 4:4:4
-          create_vertex_shader_helper(convert_yuv444_packed_vs_hlsl, convert_Y_or_YUV_vs);
-          create_pixel_shader_helper(convert_yuv444_packed_y410_ps_hlsl, convert_Y_or_YUV_ps);
-          if (display->is_hdr()) {
-            create_pixel_shader_helper(convert_yuv444_packed_y410_ps_perceptual_quantizer_hlsl, convert_Y_or_YUV_fp16_ps);
-          }
-          else {
-            create_pixel_shader_helper(convert_yuv444_packed_y410_ps_linear_hlsl, convert_Y_or_YUV_fp16_ps);
-          }
-          break;
-
-        default:
-          BOOST_LOG(error) << "Unable to create shaders because of the unrecognized surface format";
-          return -1;
-      }
-
-#undef create_vertex_shader_helper
-#undef create_pixel_shader_helper
-
       auto default_color_vectors = ::video::color_vectors_from_colorspace(::video::colorspace_e::rec601, false);
       if (!default_color_vectors) {
         BOOST_LOG(error) << "Missing color vectors for Rec. 601"sv;
@@ -1923,6 +1954,10 @@
     compile_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_linear);
     compile_pixel_shader_helper(convert_yuv420_packed_uv_type0_ps_perceptual_quantizer);
     compile_vertex_shader_helper(convert_yuv420_packed_uv_type0_vs);
+    compile_pixel_shader_helper(convert_yuv420_packed_uv_type0s_ps);
+    compile_pixel_shader_helper(convert_yuv420_packed_uv_type0s_ps_linear);
+    compile_pixel_shader_helper(convert_yuv420_packed_uv_type0s_ps_perceptual_quantizer);
+    compile_vertex_shader_helper(convert_yuv420_packed_uv_type0s_vs);
     compile_pixel_shader_helper(convert_yuv420_planar_y_ps);
     compile_pixel_shader_helper(convert_yuv420_planar_y_ps_linear);
     compile_pixel_shader_helper(convert_yuv420_planar_y_ps_perceptual_quantizer);

src_assets/windows/assets/shaders/directx/convert_yuv420_packed_uv_type0_vs.hlsl

@@ -11,5 +11,5 @@ cbuffer rotate_texture_steps_cbuffer : register(b1) {
 
 vertex_t main_vs(uint vertex_id : SV_VertexID)
 {
-    return generate_fullscreen_triangle_vertex(vertex_id, subsample_offset.x, rotate_texture_steps);
+    return generate_fullscreen_triangle_vertex(vertex_id, subsample_offset, rotate_texture_steps);
 }

src_assets/windows/assets/shaders/directx/convert_yuv420_packed_uv_type0s_ps.hlsl

@@ -0,0 +1,5 @@
+#include "include/convert_base.hlsl"
+
+#define LEFT_SUBSAMPLING_SCALE
+
+#include "include/convert_yuv420_packed_uv_ps_base.hlsl"

src_assets/windows/assets/shaders/directx/convert_yuv420_packed_uv_type0s_ps_linear.hlsl

@@ -0,0 +1,5 @@
+#include "include/convert_linear_base.hlsl"
+
+#define LEFT_SUBSAMPLING_SCALE
+
+#include "include/convert_yuv420_packed_uv_ps_base.hlsl"

src_assets/windows/assets/shaders/directx/convert_yuv420_packed_uv_type0s_ps_perceptual_quantizer.hlsl

@@ -0,0 +1,5 @@
+#include "include/convert_perceptual_quantizer_base.hlsl"
+
+#define LEFT_SUBSAMPLING_SCALE
+
+#include "include/convert_yuv420_packed_uv_ps_base.hlsl"

src_assets/windows/assets/shaders/directx/convert_yuv420_packed_uv_type0s_vs.hlsl

@@ -0,0 +1,15 @@
+cbuffer subsample_offset_cbuffer : register(b0) {
+    float2 subsample_offset;
+};
+
+cbuffer rotate_texture_steps_cbuffer : register(b1) {
+    int rotate_texture_steps;
+};
+
+#define LEFT_SUBSAMPLING_SCALE
+#include "include/base_vs.hlsl"
+
+vertex_t main_vs(uint vertex_id : SV_VertexID)
+{
+    return generate_fullscreen_triangle_vertex(vertex_id, subsample_offset, rotate_texture_steps);
+}

src_assets/windows/assets/shaders/directx/convert_yuv420_planar_y_vs.hlsl

@@ -6,5 +6,5 @@ cbuffer rotate_texture_steps_cbuffer : register(b1) {
 
 vertex_t main_vs(uint vertex_id : SV_VertexID)
 {
-    return generate_fullscreen_triangle_vertex(vertex_id, rotate_texture_steps);
+    return generate_fullscreen_triangle_vertex(vertex_id, float2(0, 0), rotate_texture_steps);
 }

src_assets/windows/assets/shaders/directx/convert_yuv444_packed_vs.hlsl

@@ -6,5 +6,5 @@ cbuffer rotate_texture_steps_cbuffer : register(b1) {
 
 vertex_t main_vs(uint vertex_id : SV_VertexID)
 {
-    return generate_fullscreen_triangle_vertex(vertex_id, rotate_texture_steps);
+    return generate_fullscreen_triangle_vertex(vertex_id, float2(0, 0), rotate_texture_steps);
 }

src_assets/windows/assets/shaders/directx/convert_yuv444_planar_vs.hlsl

@@ -15,7 +15,7 @@ cbuffer color_matrix_cbuffer : register(b3) {
 
 vertex_t main_vs(uint vertex_id : SV_VertexID)
 {
-    vertex_t output = generate_fullscreen_triangle_vertex(vertex_id % 3, rotate_texture_steps);
+    vertex_t output = generate_fullscreen_triangle_vertex(vertex_id % 3, float2(0, 0), rotate_texture_steps);
 
     output.viewport = vertex_id / 3;
 

src_assets/windows/assets/shaders/directx/cursor_vs.hlsl

@@ -6,5 +6,5 @@ cbuffer rotate_texture_steps_cbuffer : register(b2) {
 
 vertex_t main_vs(uint vertex_id : SV_VertexID)
 {
-    return generate_fullscreen_triangle_vertex(vertex_id, rotate_texture_steps);
+    return generate_fullscreen_triangle_vertex(vertex_id, float2(0, 0), rotate_texture_steps);
 }

src_assets/windows/assets/shaders/directx/include/base_vs.hlsl

@@ -1,12 +1,6 @@
 #include "include/base_vs_types.hlsl"
 
-#if defined(LEFT_SUBSAMPLING)
-vertex_t generate_fullscreen_triangle_vertex(uint vertex_id, float subsample_offset, int rotate_texture_steps)
-#elif defined(TOPLEFT_SUBSAMPLING)
 vertex_t generate_fullscreen_triangle_vertex(uint vertex_id, float2 subsample_offset, int rotate_texture_steps)
-#else
-vertex_t generate_fullscreen_triangle_vertex(uint vertex_id, int rotate_texture_steps)
-#endif
 {
     vertex_t output;
     float2 tex_coord;
@@ -30,11 +24,24 @@ vertex_t generate_fullscreen_triangle_vertex(uint vertex_id, int rotate_texture_
                                      sin(rotation_radians), cos(rotation_radians) };
         float2 rotation_center = { 0.5, 0.5 };
         tex_coord = round(rotation_center + mul(rotation_matrix, tex_coord - rotation_center));
+
+        if (rotate_texture_steps % 2) {
+            subsample_offset.xy = subsample_offset.yx;
+        }
     }
 
 #if defined(LEFT_SUBSAMPLING)
-    output.tex_right_left_center = float3(tex_coord.x, tex_coord.x - subsample_offset, tex_coord.y);
-#elif defined (TOPLEFT_SUBSAMPLING)
+    output.tex_right_left_center = float3(tex_coord.x, tex_coord.x - subsample_offset.x, tex_coord.y);
+#elif defined(LEFT_SUBSAMPLING_SCALE)
+    float2 halfsample_offset = subsample_offset / 2;
+    float3 right_center_left = float3(tex_coord.x + halfsample_offset.x,
+                                      tex_coord.x - halfsample_offset.x,
+                                      tex_coord.x - 3 * halfsample_offset.x);
+    float2 top_bottom = float2(tex_coord.y - halfsample_offset.y,
+                               tex_coord.y + halfsample_offset.y);
+    output.tex_right_center_left_top = float4(right_center_left, top_bottom.x);
+    output.tex_right_center_left_bottom = float4(right_center_left, top_bottom.y);
+#elif defined(TOPLEFT_SUBSAMPLING)
     output.tex_right_left_top = float3(tex_coord.x, tex_coord.x - subsample_offset.x, tex_coord.y - subsample_offset.y);
     output.tex_right_left_bottom = float3(tex_coord.x, tex_coord.x - subsample_offset.x, tex_coord.y);
 #else

src_assets/windows/assets/shaders/directx/include/base_vs_types.hlsl

@@ -3,9 +3,12 @@ struct vertex_t
     float4 viewpoint_pos : SV_Position;
 #if defined(LEFT_SUBSAMPLING)
     float3 tex_right_left_center : TEXCOORD;
-#elif defined (TOPLEFT_SUBSAMPLING)
-    float3 tex_right_left_top : TEXCOORD;
-    float3 tex_right_left_bottom : TEXCOORD;
+#elif defined(LEFT_SUBSAMPLING_SCALE)
+    float4 tex_right_center_left_top : TEXCOORD0;
+    float4 tex_right_center_left_bottom : TEXCOORD1;
+#elif defined(TOPLEFT_SUBSAMPLING)
+    float3 tex_right_left_top : TEXCOORD0;
+    float3 tex_right_left_bottom : TEXCOORD1;
 #else
     float2 tex_coord : TEXCOORD;
 #endif

src_assets/windows/assets/shaders/directx/include/convert_yuv420_packed_uv_ps_base.hlsl

@@ -17,6 +17,15 @@ float2 main_ps(vertex_t input) : SV_Target
     float3 rgb_left = image.Sample(def_sampler, input.tex_right_left_center.xz).rgb;
     float3 rgb_right = image.Sample(def_sampler, input.tex_right_left_center.yz).rgb;
     float3 rgb = CONVERT_FUNCTION((rgb_left + rgb_right) * 0.5);
+#elif defined(LEFT_SUBSAMPLING_SCALE)
+    float3 rgb = image.Sample(def_sampler, input.tex_right_center_left_top.yw).rgb; // top-center
+    rgb += image.Sample(def_sampler, input.tex_right_center_left_bottom.yw).rgb; // bottom-center
+    rgb *= 2;
+    rgb += image.Sample(def_sampler, input.tex_right_center_left_top.xw).rgb; // top-right
+    rgb += image.Sample(def_sampler, input.tex_right_center_left_top.zw).rgb; // top-left
+    rgb += image.Sample(def_sampler, input.tex_right_center_left_bottom.xw).rgb; // bottom-right
+    rgb += image.Sample(def_sampler, input.tex_right_center_left_bottom.zw).rgb; // bottom-left
+    rgb = CONVERT_FUNCTION(rgb * (1./8));
 #elif defined(TOPLEFT_SUBSAMPLING)
     float3 rgb_top_left = image.Sample(def_sampler, input.tex_right_left_top.xz).rgb;
     float3 rgb_top_right = image.Sample(def_sampler, input.tex_right_left_top.yz).rgb;