optimize sketch reset (#273)

* unroll

* cleanup

* rename

* rename

Author: bitfaster

BitFaster.Caching.Benchmarks/Lfu/SketchReset.cs

@@ -0,0 +1,79 @@
+
+using BenchmarkDotNet.Attributes;
+using BenchmarkDotNet.Jobs;
+
+namespace BitFaster.Caching.Benchmarks.Lfu
+{
+    [SimpleJob(RuntimeMoniker.Net60)]
+    public class SketchReset
+    {
+        static long ResetMask = 0x7777777777777777L;
+        static long OneMask = 0x1111111111111111L;
+
+        long[] table;
+
+        [Params(4, 128, 8192, 1048576)]
+        public int Size { get; set; }
+
+        [GlobalSetup]
+        public void Setup()
+        {
+            table = new long[Size];
+        }
+
+        [Benchmark(Baseline = true)]
+        public int Reset1()
+        {
+            int count = 0;
+            for (int i = 0; i < table.Length; i++)
+            {
+                count += BitOps.BitCount(table[i] & OneMask);
+                table[i] = (long)((ulong)table[i] >> 1) & ResetMask;
+            }
+
+            return count;
+        }
+
+        [Benchmark()]
+        public int Reset2()
+        {
+            int count0 = 0;
+            int count1 = 0;
+
+            for (int i = 0; i < table.Length; i += 2)
+            {
+                count0 += BitOps.BitCount(table[i] & OneMask);
+                count1 += BitOps.BitCount(table[i + 1] & OneMask);
+
+                table[i] = (long)((ulong)table[i] >> 1) & ResetMask;
+                table[i + 1] = (long)((ulong)table[i + 1] >> 1) & ResetMask;
+            }
+
+            return count0 + count1;
+        }
+
+        [Benchmark()]
+        public int Reset4()
+        {
+            int count0 = 0;
+            int count1 = 0;
+            int count2 = 0;
+            int count3 = 0;
+
+            for (int i = 0; i < table.Length; i += 4)
+            {
+                count0 += BitOps.BitCount(table[i] & OneMask);
+                count1 += BitOps.BitCount(table[i + 1] & OneMask);
+                count2 += BitOps.BitCount(table[i + 2] & OneMask);
+                count3 += BitOps.BitCount(table[i + 3] & OneMask);
+
+                table[i] = (long)((ulong)table[i] >> 1) & ResetMask;
+                table[i + 1] = (long)((ulong)table[i + 1] >> 1) & ResetMask;
+                table[i + 2] = (long)((ulong)table[i + 2] >> 1) & ResetMask;
+                table[i + 3] = (long)((ulong)table[i + 3] >> 1) & ResetMask;
+            }
+
+            return (count0 + count1) + (count2 + count3);
+        }
+    }
+}

BitFaster.Caching/Lfu/CmSketch.cs

@@ -158,19 +158,37 @@ private bool IncrementAt(int i, int j)
 
         private void Reset()
         {
-            int count = 0;
-            for (int i = 0; i < table.Length; i++)
+            // unroll, almost 2x faster
+            int count0 = 0;
+            int count1 = 0;
+            int count2 = 0;
+            int count3 = 0;
+
+            for (int i = 0; i < table.Length; i += 4)
             {
-                count += BitOps.BitCount(table[i] & OneMask);
+                count0 += BitOps.BitCount(table[i] & OneMask);
+                count1 += BitOps.BitCount(table[i + 1] & OneMask);
+                count2 += BitOps.BitCount(table[i + 2] & OneMask);
+                count3 += BitOps.BitCount(table[i + 3] & OneMask);
+
                 table[i] = (long)((ulong)table[i] >> 1) & ResetMask;
+                table[i + 1] = (long)((ulong)table[i + 1] >> 1) & ResetMask;
+                table[i + 2] = (long)((ulong)table[i + 2] >> 1) & ResetMask;
+                table[i + 3] = (long)((ulong)table[i + 3] >> 1) & ResetMask;
             }
-            size = (size - (count >> 2)) >> 1;
+
+            count0 = (count0 + count1) + (count2 + count3);
+
+            size = (size - (count0 >> 2)) >> 1;
         }
 
         private void EnsureCapacity(long maximumSize)
         {
             int maximum = (int)Math.Min(maximumSize, int.MaxValue >> 1);
 
+            // clamp to 4 as min size
+            maximum = Math.Max(4, maximum);
+
             table = new long[(maximum == 0) ? 1 : BitOps.CeilingPowerOfTwo(maximum)];
             tableMask = Math.Max(0, table.Length - 1);
             sampleSize = (maximumSize == 0) ? 10 : (10 * maximum);