Speed up Util.FastCopyIntoChannel.

The assets for the Tiberian Dawn HD mod are much larger than assets for the default mods, causing a lot of load time to be spent in Util.FastCopyIntoChannel.

We can provide a special case for the SpriteFrameType.Bgra32 format, which is the same format as the destination buffer. In this scenario we can just perform memory copies between the source and destination. Additionally, whilst the default mods require all their assets to get their alpha premultiplied, many of the Tiberian Dawn assets are already premultiplied. Being able to skip this step for these assets saves us having to interpret the bytes into colors and back again.

For the default mods, there isn't a noticeable timing difference. For Tiberian Dawn HD or other mods with modern assets sizes, a large speedup is achieved.

OpenRA.Game/Graphics/SpriteLoader.cs

@@ -22,20 +22,30 @@ namespace OpenRA.Graphics
 	/// </summary>
 	public enum SpriteFrameType
 	{
-		// 8 bit index into an external palette
+		/// <summary>
+		/// 8 bit index into an external palette.
+		/// </summary>
 		Indexed8,
 
-		// 32 bit color such as returned by Color.ToArgb() or the bmp file format
+		/// <summary>
-		// (remember that little-endian systems place the little bits in the first byte!)
+		/// 32 bit color such as returned by Color.ToArgb() or the bmp file format
+		/// (remember that little-endian systems place the little bits in the first byte).
+		/// </summary>
 		Bgra32,
 
-		// Like BGRA, but without an alpha channel
+		/// <summary>
+		/// Like BGRA, but without an alpha channel.
+		/// </summary>
 		Bgr24,
 
-		// 32 bit color in big-endian format, like png
+		/// <summary>
+		/// 32 bit color in big-endian format, like png.
+		/// </summary>
 		Rgba32,
 
-		// Like RGBA, but without an alpha channel
+		/// <summary>
+		/// Like RGBA, but without an alpha channel.
+		/// </summary>
 		Rgb24
 	}
 

OpenRA.Game/Graphics/Util.cs

@@ -10,6 +10,7 @@
 #endregion
 
 using System;
+using System.Runtime.InteropServices;
 using OpenRA.FileFormats;
 using OpenRA.Primitives;
 
@@ -106,67 +107,21 @@ namespace OpenRA.Graphics
 		public static void FastCopyIntoChannel(Sprite dest, byte[] src, SpriteFrameType srcType, bool premultiplied = false)
 		{
 			var destData = dest.Sheet.GetData();
+			var stride = dest.Sheet.Size.Width;
+			var x = dest.Bounds.Left;
+			var y = dest.Bounds.Top;
 			var width = dest.Bounds.Width;
 			var height = dest.Bounds.Height;
 
 			if (dest.Channel == TextureChannel.RGBA)
 			{
-				var destStride = dest.Sheet.Size.Width;
+				CopyIntoRgba(src, srcType, premultiplied, destData, x, y, width, height, stride);
-				unsafe
-				{
-					// Cast the data to an int array so we can copy the src data directly
-					fixed (byte* bd = &destData[0])
-					{
-						var data = (uint*)bd;
-						var x = dest.Bounds.Left;
-						var y = dest.Bounds.Top;
-
-						var k = 0;
-						for (var j = 0; j < height; j++)
-						{
-							for (var i = 0; i < width; i++)
-							{
-								byte r, g, b, a;
-								switch (srcType)
-								{
-									case SpriteFrameType.Bgra32:
-									case SpriteFrameType.Bgr24:
-									{
-										b = src[k++];
-										g = src[k++];
-										r = src[k++];
-										a = srcType == SpriteFrameType.Bgra32 ? src[k++] : (byte)255;
-										break;
-									}
-
-									case SpriteFrameType.Rgba32:
-									case SpriteFrameType.Rgb24:
-									{
-										r = src[k++];
-										g = src[k++];
-										b = src[k++];
-										a = srcType == SpriteFrameType.Rgba32 ? src[k++] : (byte)255;
-										break;
-									}
-
-									default:
-										throw new InvalidOperationException($"Unknown SpriteFrameType {srcType}");
-								}
-
-								var cc = Color.FromArgb(a, r, g, b);
-								if (premultiplied)
-									data[(y + j) * destStride + x + i] = cc.ToArgb();
-								else
-									data[(y + j) * destStride + x + i] = PremultiplyAlpha(cc).ToArgb();
-							}
-						}
-					}
-				}
 			}
 			else
 			{
-				var destStride = dest.Sheet.Size.Width * 4;
+				// Copy into single channel of destination.
-				var destOffset = destStride * dest.Bounds.Top + dest.Bounds.Left * 4 + ChannelMasks[(int)dest.Channel];
+				var destStride = stride * 4;
+				var destOffset = destStride * y + x * 4 + ChannelMasks[(int)dest.Channel];
 				var destSkip = destStride - 4 * width;
 
 				var srcOffset = 0;
@@ -183,56 +138,119 @@ namespace OpenRA.Graphics
 			}
 		}
 
+		static void CopyIntoRgba(
+			byte[] src, SpriteFrameType srcType, bool premultiplied, byte[] dest, int x, int y, int width, int height, int stride)
+		{
+			var si = 0;
+			var di = y * stride + x;
+			var d = MemoryMarshal.Cast<byte, uint>(dest);
+
+			// SpriteFrameType.Brga32 is a common source format, and it matches the destination format.
+			// Provide a fast past that just performs memory copies.
+			if (srcType == SpriteFrameType.Bgra32)
+			{
+				var s = MemoryMarshal.Cast<byte, uint>(src);
+				for (var h = 0; h < height; h++)
+				{
+					s[si..(si + width)].CopyTo(d[di..(di + width)]);
+
+					if (!premultiplied)
+					{
+						for (var w = 0; w < width; w++)
+						{
+							d[di] = PremultiplyAlpha(Color.FromArgb(d[di])).ToArgb();
+							di++;
+						}
+
+						di -= width;
+					}
+
+					si += width;
+					di += stride;
+				}
+
+				return;
+			}
+
+			for (var h = 0; h < height; h++)
+			{
+				for (var w = 0; w < width; w++)
+				{
+					byte r, g, b, a;
+					switch (srcType)
+					{
+						case SpriteFrameType.Bgra32:
+						case SpriteFrameType.Bgr24:
+							b = src[si++];
+							g = src[si++];
+							r = src[si++];
+							a = srcType == SpriteFrameType.Bgra32 ? src[si++] : byte.MaxValue;
+							break;
+
+						case SpriteFrameType.Rgba32:
+						case SpriteFrameType.Rgb24:
+							r = src[si++];
+							g = src[si++];
+							b = src[si++];
+							a = srcType == SpriteFrameType.Rgba32 ? src[si++] : byte.MaxValue;
+							break;
+
+						default:
+							throw new InvalidOperationException($"Unknown SpriteFrameType {srcType}");
+					}
+
+					var c = Color.FromArgb(a, r, g, b);
+					if (!premultiplied)
+						c = PremultiplyAlpha(c);
+					d[di++] = c.ToArgb();
+				}
+
+				di += stride - width;
+			}
+		}
+
 		public static void FastCopyIntoSprite(Sprite dest, Png src)
 		{
 			var destData = dest.Sheet.GetData();
-			var destStride = dest.Sheet.Size.Width;
+			var stride = dest.Sheet.Size.Width;
+			var x = dest.Bounds.Left;
+			var y = dest.Bounds.Top;
 			var width = dest.Bounds.Width;
 			var height = dest.Bounds.Height;
 
-			unsafe
+			var si = 0;
+			var di = y * stride + x;
+			var d = MemoryMarshal.Cast<byte, uint>(destData);
+
+			for (var h = 0; h < height; h++)
 			{
-				// Cast the data to an int array so we can copy the src data directly
+				for (var w = 0; w < width; w++)
-				fixed (byte* bd = &destData[0])
 				{
-					var data = (uint*)bd;
+					Color c;
-					var x = dest.Bounds.Left;
+					switch (src.Type)
-					var y = dest.Bounds.Top;
-
-					var k = 0;
-					for (var j = 0; j < height; j++)
 					{
-						for (var i = 0; i < width; i++)
+						case SpriteFrameType.Indexed8:
-						{
+							c = src.Palette[src.Data[si++]];
-							Color cc;
+							break;
-							switch (src.Type)
-							{
-								case SpriteFrameType.Indexed8:
-								{
-									cc = src.Palette[src.Data[k++]];
-									break;
-								}
 
-								case SpriteFrameType.Rgba32:
+						case SpriteFrameType.Rgba32:
-								case SpriteFrameType.Rgb24:
+						case SpriteFrameType.Rgb24:
-								{
+							var r = src.Data[si++];
-									var r = src.Data[k++];
+							var g = src.Data[si++];
-									var g = src.Data[k++];
+							var b = src.Data[si++];
-									var b = src.Data[k++];
+							var a = src.Type == SpriteFrameType.Rgba32 ? src.Data[si++] : byte.MaxValue;
-									var a = src.Type == SpriteFrameType.Rgba32 ? src.Data[k++] : (byte)255;
+							c = Color.FromArgb(a, r, g, b);
-									cc = Color.FromArgb(a, r, g, b);
+							break;
-									break;
-								}
 
-								// Pngs don't support BGR[A], so no need to include them here
+						// PNGs don't support BGR[A], so no need to include them here
-								default:
+						default:
-									throw new InvalidOperationException($"Unknown SpriteFrameType {src.Type}");
+							throw new InvalidOperationException($"Unknown SpriteFrameType {src.Type}");
-							}
-
-							data[(y + j) * destStride + x + i] = PremultiplyAlpha(cc).ToArgb();
-						}
 					}
+
+					d[di++] = PremultiplyAlpha(c).ToArgb();
 				}
+
+				di += stride - width;
 			}
 		}