Use optimized pixel shaders for the normal SDR and HDR colorspaces

app/shaders/build_hlsl.bat

@@ -1,4 +1,6 @@
 fxc /T vs_4_0_level_9_3 /Fo d3d11_vertex.fxc d3d11_vertex.hlsl
 
 fxc /T ps_4_0_level_9_3 /Fo d3d11_overlay_pixel.fxc d3d11_overlay_pixel.hlsl
-fxc /T ps_4_0_level_9_3 /Fo d3d11_video_pixel.fxc d3d11_video_pixel.hlsl
+fxc /T ps_4_0_level_9_3 /Fo d3d11_video_pixel.fxc d3d11_video_pixel.hlsl
+fxc /T ps_4_0_level_9_3 /Fo d3d11_bt601lim_pixel.fxc d3d11_bt601lim_pixel.hlsl
+fxc /T ps_4_0_level_9_3 /Fo d3d11_bt2020lim_pixel.fxc d3d11_bt2020lim_pixel.hlsl

app/shaders/d3d11_bt2020lim_pixel.hlsl

@@ -0,0 +1,35 @@
+Texture2D<min16float> luminancePlane : register(t0);
+Texture2D<min16float2> chrominancePlane : register(t1);
+SamplerState theSampler : register(s0);
+
+static const min16float3x3 cscMatrix =
+{
+    1.1644, 1.1644, 1.1644,
+    0.0, -0.1874, 2.1418,
+    1.6781, -0.6505, 0.0,
+};
+
+static const min16float3 offsets =
+{
+    16.0 / 255.0, 128.0 / 255.0, 128.0 / 255.0
+};
+
+struct ShaderInput
+{
+    min16float4 pos : SV_POSITION;
+    min16float2 tex : TEXCOORD0;
+};
+
+min16float4 main(ShaderInput input) : SV_TARGET
+{
+    min16float3 yuv = min16float3(luminancePlane.Sample(theSampler, input.tex),
+                                  chrominancePlane.Sample(theSampler, input.tex));
+
+    // Subtract the YUV offset for limited vs full range
+    yuv -= offsets;
+
+    // Multiply by the conversion matrix for this colorspace
+    yuv = mul(yuv, cscMatrix);
+
+    return min16float4(yuv, 1.0);
+}

app/shaders/d3d11_bt601lim_pixel.hlsl

@@ -0,0 +1,35 @@
+Texture2D<min16float> luminancePlane : register(t0);
+Texture2D<min16float2> chrominancePlane : register(t1);
+SamplerState theSampler : register(s0);
+
+static const min16float3x3 cscMatrix =
+{
+    1.1644, 1.1644, 1.1644,
+    0.0, -0.3917, 2.0172,
+    1.5960, -0.8129, 0.0,
+};
+
+static const min16float3 offsets =
+{
+    16.0 / 255.0, 128.0 / 255.0, 128.0 / 255.0
+};
+
+struct ShaderInput
+{
+    min16float4 pos : SV_POSITION;
+    min16float2 tex : TEXCOORD0;
+};
+
+min16float4 main(ShaderInput input) : SV_TARGET
+{
+    min16float3 yuv = min16float3(luminancePlane.Sample(theSampler, input.tex),
+                                  chrominancePlane.Sample(theSampler, input.tex));
+
+    // Subtract the YUV offset for limited vs full range
+    yuv -= offsets;
+
+    // Multiply by the conversion matrix for this colorspace
+    yuv = mul(yuv, cscMatrix);
+
+    return min16float4(yuv, 1.0);
+}

app/shaders/d3d11_video_pixel.hlsl

@@ -25,5 +25,5 @@ min16float4 main(ShaderInput input) : SV_TARGET
     // Multiply by the conversion matrix for this colorspace
     yuv = mul(yuv, cscMatrix);
 
-    return min16float4(saturate(yuv), 1.0);
+    return min16float4(yuv, 1.0);
 }