Add CopyToArray method for IPalette.

By allowing a palette to be copied to an array, a speedup can be gained in HardwarePalette since palettes can be block copied using native magic rather than having to copy them item by item. We transpose the buffer in HardwarePalette in order to allow a contiguous block copy into this buffer, transposing again in the shader to keep everything correct.

OpenRA.Game/Graphics/HardwarePalette.cs

@@ -23,7 +23,7 @@ namespace OpenRA.Graphics
 		readonly Dictionary<string, MutablePalette> modifiablePalettes = new Dictionary<string, MutablePalette>();
 		readonly IReadOnlyDictionary<string, MutablePalette> readOnlyModifiablePalettes;
 		readonly Dictionary<string, int> indices = new Dictionary<string, int>();
-		readonly uint[,] buffer = new uint[Palette.Size, MaxPalettes];
+		readonly uint[,] buffer = new uint[MaxPalettes, Palette.Size];
 
 		public HardwarePalette()
 		{
@@ -85,8 +85,7 @@ namespace OpenRA.Graphics
 
 		void CopyPaletteToBuffer(int index, IPalette p)
 		{
-			for (var i = 0; i < Palette.Size; i++)
-				buffer[i, index] = p[i];
+			p.CopyToArray(buffer, index * Palette.Size);
 		}
 
 		void CopyModifiablePalettesToBuffer()
@@ -109,8 +108,7 @@ namespace OpenRA.Graphics
 			{
 				var originalPalette = palettes[kvp.Key];
 				var modifiedPalette = kvp.Value;
-				for (var i = 0; i < Palette.Size; i++)
-					modifiedPalette[i] = originalPalette[i];
+				modifiedPalette.SetFromPalette(originalPalette);
 			}
 		}
 	}

OpenRA.Game/Graphics/Palette.cs

@@ -8,6 +8,7 @@
  */
 #endregion
 
+using System;
 using System.Collections.Generic;
 using System.Drawing;
 using System.Drawing.Imaging;
@@ -17,7 +18,11 @@ using System.Runtime.InteropServices;
 
 namespace OpenRA.Graphics
 {
-	public interface IPalette { uint this[int index] { get; } }
+	public interface IPalette
+	{
+		uint this[int index] { get; }
+		void CopyToArray(Array destination, int destinationOffset);
+	}
 	public interface IPaletteRemap { Color GetRemappedColor(Color original, int index); }
 
 	public static class Palette
@@ -52,8 +57,7 @@ namespace OpenRA.Graphics
 			var data = b.LockBits(new Rectangle(0, 0, b.Width, b.Height),
 								  ImageLockMode.WriteOnly, PixelFormat.Format32bppArgb);
 			var temp = new uint[Palette.Size];
-			for (int i = 0; i < temp.Length; i++)
-				temp[i] = palette[i];
+			palette.CopyToArray(temp, 0);
 			Marshal.Copy((int[])(object)temp, 0, data.Scan0, Size);
 			b.UnlockBits(data);
 			return b;
@@ -71,6 +75,10 @@ namespace OpenRA.Graphics
 			IPalette palette;
 			public ReadOnlyPalette(IPalette palette) { this.palette = palette; }
 			public uint this[int index] { get { return palette[index]; } }
+			public void CopyToArray(Array destination, int destinationOffset)
+			{
+				palette.CopyToArray(destination, destinationOffset);
+			}
 		}
 	}
 
@@ -83,6 +91,11 @@ namespace OpenRA.Graphics
 			get { return colors[index]; }
 		}
 
+		public void CopyToArray(Array destination, int destinationOffset)
+		{
+			Buffer.BlockCopy(colors, 0, destination, destinationOffset * 4, Palette.Size * 4);
+		}
+
 		public ImmutablePalette(string filename, int[] remap)
 		{
 			using (var s = File.OpenRead(filename))
@@ -141,10 +154,14 @@ namespace OpenRA.Graphics
 			set { colors[index] = value; }
 		}
 
+		public void CopyToArray(Array destination, int destinationOffset)
+		{
+			Buffer.BlockCopy(colors, 0, destination, destinationOffset * 4, Palette.Size * 4);
+		}
+
 		public MutablePalette(IPalette p)
 		{
-			for (int i = 0; i < Palette.Size; i++)
-				this[i] = p[i];
+			SetFromPalette(p);
 		}
 
 		public void SetColor(int index, Color color)
@@ -152,6 +169,11 @@ namespace OpenRA.Graphics
 			colors[index] = (uint)color.ToArgb();
 		}
 
+		public void SetFromPalette(IPalette p)
+		{
+			p.CopyToArray(colors, 0);
+		}
+
 		public void ApplyRemap(IPaletteRemap r)
 		{
 			for (var i = 0; i < Palette.Size; i++)

glsl/shp.frag

@@ -3,6 +3,6 @@ uniform sampler2D DiffuseTexture, Palette;
 void main()
 {
 	vec4 x = texture2D(DiffuseTexture, gl_TexCoord[0].st);
-	vec2 p = vec2(gl_TexCoord[0].p, dot(x, gl_TexCoord[1]));
+	vec2 p = vec2(dot(x, gl_TexCoord[1]), gl_TexCoord[0].p);
 	gl_FragColor = texture2D(Palette, p);
 }

glsl/vxl.frag

@@ -7,11 +7,11 @@ uniform vec3 AmbientLight, DiffuseLight;
 void main()
 {
 	vec4 x = texture2D(DiffuseTexture, gl_TexCoord[0].st);
-	vec4 color = texture2D(Palette, vec2(PaletteRows.x, dot(x, gl_TexCoord[1])));
+	vec4 color = texture2D(Palette, vec2(dot(x, gl_TexCoord[1]), PaletteRows.x));
 	if (color.a < 0.01)
 		discard;
 
-	vec4 normal = (2.0 * texture2D(Palette, vec2(PaletteRows.y, dot(x, gl_TexCoord[2]))) - 1.0);
+	vec4 normal = (2.0 * texture2D(Palette, vec2(dot(x, gl_TexCoord[2]), PaletteRows.y)) - 1.0);
 	vec3 intensity = AmbientLight + DiffuseLight * max(dot(normal, LightDirection), 0.0);
 	gl_FragColor = vec4(intensity * color.rgb, color.a);
 }