Remove more allocations and add tasks.

src/ImageSharp/Formats/Webp/Lossless/BackwardReferenceEncoder.cs

@@ -52,6 +52,8 @@ internal static class BackwardReferenceEncoder
         Vp8LHashChain? hashChainBox = null;
         Vp8LStreaks stats = new();
         Vp8LBitEntropy bitsEntropy = new();
+
+        ColorCache[] colorCache = new ColorCache[WebpConstants.MaxColorCacheBits + 1];
         for (int lz77Type = 1; lz77TypesToTry > 0; lz77TypesToTry &= ~lz77Type, lz77Type <<= 1)
         {
             int cacheBitsTmp = cacheBitsInitial;
@@ -76,7 +78,7 @@ internal static class BackwardReferenceEncoder
             }
 
             // Next, try with a color cache and update the references.
-            cacheBitsTmp = CalculateBestCacheSize(memoryAllocator, bgra, quality, worst, cacheBitsTmp);
+            cacheBitsTmp = CalculateBestCacheSize(memoryAllocator, colorCache, bgra, quality, worst, cacheBitsTmp);
             if (cacheBitsTmp > 0)
             {
                 BackwardRefsWithLocalCache(bgra, cacheBitsTmp, worst);
@@ -123,6 +125,7 @@ internal static class BackwardReferenceEncoder
     /// <returns>Best cache size.</returns>
     private static int CalculateBestCacheSize(
         MemoryAllocator memoryAllocator,
+        Span<ColorCache> colorCache,
         ReadOnlySpan<uint> bgra,
         uint quality,
         Vp8LBackwardRefs refs,
@@ -138,12 +141,8 @@ internal static class BackwardReferenceEncoder
         double entropyMin = MaxEntropy;
         int pos = 0;
 
-        // TODO: Pass from outer loop and clear.
+        using Vp8LHistogramSet histos = new(memoryAllocator, colorCache.Length, 0);
-        ColorCache[] colorCache = new ColorCache[WebpConstants.MaxColorCacheBits + 1];
+        for (int i = 0; i < colorCache.Length; i++)
-
-        // TODO: Use fixed size.
-        using Vp8LHistogramSet histos = new(memoryAllocator, WebpConstants.MaxColorCacheBits + 1, 0);
-        for (int i = 0; i <= WebpConstants.MaxColorCacheBits; i++)
         {
             histos[i].PaletteCodeBits = i;
             colorCache[i] = new ColorCache(i);
@@ -448,12 +447,12 @@ internal static class BackwardReferenceEncoder
         int ix = useColorCache ? colorCache!.Contains(color) : -1;
         if (ix >= 0)
         {
-            double mul0 = 0.68;
+            const double mul0 = 0.68;
             costVal += costModel.GetCacheCost((uint)ix) * mul0;
         }
         else
         {
-            double mul1 = 0.82;
+            const double mul1 = 0.82;
             if (useColorCache)
             {
                 colorCache!.Insert(color);
@@ -700,10 +699,8 @@ internal static class BackwardReferenceEncoder
                             bestLength = MaxLength;
                             break;
                         }
-                        else
+
-                        {
+                        bestLength = currLength;
-                            bestLength = currLength;
-                        }
                     }
                 }
             }

src/ImageSharp/Formats/Webp/Lossless/HistogramEncoder.cs

@@ -2,6 +2,7 @@
 // Licensed under the Six Labors Split License.
 #nullable disable
 
+using System.Buffers;
 using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Memory;
 
@@ -45,9 +46,9 @@ internal static class HistogramEncoder
         int imageHistoRawSize = histoXSize * histoYSize;
         const int entropyCombineNumBins = BinSize;
 
-        // TODO: Allocations!
+        using IMemoryOwner<ushort> tmp = memoryAllocator.Allocate<ushort>(imageHistoRawSize * 2, AllocationOptions.Clean);
-        ushort[] mapTmp = new ushort[imageHistoRawSize];
+        Span<ushort> mapTmp = tmp.Slice(0, imageHistoRawSize);
-        ushort[] clusterMappings = new ushort[imageHistoRawSize];
+        Span<ushort> clusterMappings = tmp.Slice(imageHistoRawSize, imageHistoRawSize);
 
         using Vp8LHistogramSet origHisto = new(memoryAllocator, imageHistoRawSize, cacheBits);
 
@@ -60,13 +61,12 @@ internal static class HistogramEncoder
         bool entropyCombine = numUsed > entropyCombineNumBins * 2 && quality < 100;
         if (entropyCombine)
         {
-            ushort[] binMap = mapTmp;
             int numClusters = numUsed;
             double combineCostFactor = GetCombineCostFactor(imageHistoRawSize, quality);
-            HistogramAnalyzeEntropyBin(imageHisto, binMap);
+            HistogramAnalyzeEntropyBin(imageHisto, mapTmp);
 
             // Collapse histograms with similar entropy.
-            HistogramCombineEntropyBin(imageHisto, histogramSymbols, clusterMappings, tmpHisto, binMap, entropyCombineNumBins, combineCostFactor);
+            HistogramCombineEntropyBin(imageHisto, histogramSymbols, clusterMappings, tmpHisto, mapTmp, entropyCombineNumBins, combineCostFactor);
 
             OptimizeHistogramSymbols(clusterMappings, numClusters, mapTmp, histogramSymbols);
         }
@@ -128,7 +128,7 @@ internal static class HistogramEncoder
     /// Partition histograms to different entropy bins for three dominant (literal,
     /// red and blue) symbol costs and compute the histogram aggregate bitCost.
     /// </summary>
-    private static void HistogramAnalyzeEntropyBin(Vp8LHistogramSet histograms, ushort[] binMap)
+    private static void HistogramAnalyzeEntropyBin(Vp8LHistogramSet histograms, Span<ushort> binMap)
     {
         int histoSize = histograms.Count;
         DominantCostRange costRange = new();
@@ -198,9 +198,9 @@ internal static class HistogramEncoder
     private static void HistogramCombineEntropyBin(
         Vp8LHistogramSet histograms,
         Span<ushort> clusters,
-        ushort[] clusterMappings,
+        Span<ushort> clusterMappings,
         Vp8LHistogram curCombo,
-        ushort[] binMap,
+        ReadOnlySpan<ushort> binMap,
         int numBins,
         double combineCostFactor)
     {
@@ -276,7 +276,7 @@ internal static class HistogramEncoder
     /// Given a Histogram set, the mapping of clusters 'clusterMapping' and the
     /// current assignment of the cells in 'symbols', merge the clusters and assign the smallest possible clusters values.
     /// </summary>
-    private static void OptimizeHistogramSymbols(ushort[] clusterMappings, int numClusters, ushort[] clusterMappingsTmp, Span<ushort> symbols)
+    private static void OptimizeHistogramSymbols(Span<ushort> clusterMappings, int numClusters, Span<ushort> clusterMappingsTmp, Span<ushort> symbols)
     {
         bool doContinue = true;
 
@@ -303,7 +303,7 @@ internal static class HistogramEncoder
 
         // Create a mapping from a cluster id to its minimal version.
         int clusterMax = 0;
-        clusterMappingsTmp.AsSpan().Clear();
+        clusterMappingsTmp.Clear();
 
         // Re-map the ids.
         for (int i = 0; i < symbols.Length; i++)
@@ -515,7 +515,6 @@ internal static class HistogramEncoder
             histograms.DisposeAt(idx2);
 
             // Remove pairs intersecting the just combined best pair.
-            // TODO: Reversing this will avoid the need to remove from the end of the list.
             for (int i = 0; i < histoPriorityList.Count;)
             {
                 HistogramPair p = histoPriorityList[i];

src/ImageSharp/Formats/Webp/Lossless/Vp8LEncoder.cs

@@ -885,7 +885,7 @@ internal class Vp8LEncoder : IDisposable
 
     private void StoreFullHuffmanCode(Span<HuffmanTree> huffTree, HuffmanTreeToken[] tokens, HuffmanTreeCode tree)
     {
-        // TODO: Allocations.
+        // TODO: Allocations. This method is called in a loop.
         int i;
         byte[] codeLengthBitDepth = new byte[WebpConstants.CodeLengthCodes];
         short[] codeLengthBitDepthSymbols = new short[WebpConstants.CodeLengthCodes];
@@ -1628,6 +1628,7 @@ internal class Vp8LEncoder : IDisposable
             }
         }
 
+        // TODO: Allocations.
         int end = 5 * histogramImage.Count;
         for (int i = 0; i < end; i++)
         {
@@ -1641,9 +1642,9 @@ internal class Vp8LEncoder : IDisposable
         }
 
         // Create Huffman trees.
-        // TODO: Allocations. Size here has a max and can be sliced.
+        // TODO: Allocations.
         bool[] bufRle = new bool[maxNumSymbols];
-        Span<HuffmanTree> huffTree = stackalloc HuffmanTree[3 * maxNumSymbols];
+        HuffmanTree[] huffTree = new HuffmanTree[3 * maxNumSymbols];
 
         for (int i = 0; i < histogramImage.Count; i++)
         {