From cbbe45d39ca74bf2971b62dc57a56d37b4aa9f2f Mon Sep 17 00:00:00 2001
From: WinCPP <mandards@gmail.com>
Date: Tue, 18 Apr 2017 08:11:36 +0530
Subject: [PATCH] Issue #17118 Loop unrolling in Span.CopyTo slow path

---
 src/System.Memory/src/System/SpanHelpers.cs | 39 ++++++++++++++-------
 1 file changed, 26 insertions(+), 13 deletions(-)

diff --git a/src/System.Memory/src/System/SpanHelpers.cs b/src/System.Memory/src/System/SpanHelpers.cs
index 1767772d96ff..42c341b78c83 100644
--- a/src/System.Memory/src/System/SpanHelpers.cs
+++ b/src/System.Memory/src/System/SpanHelpers.cs
@@ -61,22 +61,35 @@ ref Unsafe.Add(ref srcBytes, (IntPtr)index),
             }
             else
             {
-                if (srcGreaterThanDst)
+                int direction = srcGreaterThanDst ? 1 : -1;
+                int runCount = srcGreaterThanDst ? 0 : srcLength - 1;
+
+                int loopCount = 0;
+                for (; loopCount < (srcLength & ~7); loopCount += 8)
                 {
-                    // Source address greater than or equal to destination address. Can do normal copy.
-                    for (int i = 0; i < srcLength; i++)
-                    {
-                        Unsafe.Add<T>(ref dst, i) = Unsafe.Add<T>(ref src, i);
-                    }
+                    Unsafe.Add<T>(ref dst, runCount + direction * 0) = Unsafe.Add<T>(ref src, runCount + direction * 0);
+                    Unsafe.Add<T>(ref dst, runCount + direction * 1) = Unsafe.Add<T>(ref src, runCount + direction * 1);
+                    Unsafe.Add<T>(ref dst, runCount + direction * 2) = Unsafe.Add<T>(ref src, runCount + direction * 2);
+                    Unsafe.Add<T>(ref dst, runCount + direction * 3) = Unsafe.Add<T>(ref src, runCount + direction * 3);
+                    Unsafe.Add<T>(ref dst, runCount + direction * 4) = Unsafe.Add<T>(ref src, runCount + direction * 4);
+                    Unsafe.Add<T>(ref dst, runCount + direction * 5) = Unsafe.Add<T>(ref src, runCount + direction * 5);
+                    Unsafe.Add<T>(ref dst, runCount + direction * 6) = Unsafe.Add<T>(ref src, runCount + direction * 6);
+                    Unsafe.Add<T>(ref dst, runCount + direction * 7) = Unsafe.Add<T>(ref src, runCount + direction * 7);
+                    runCount += direction * 8;
                 }
-                else
+                if (loopCount < (srcLength & ~3))
+                {
+                    Unsafe.Add<T>(ref dst, runCount + direction * 0) = Unsafe.Add<T>(ref src, runCount + direction * 0);
+                    Unsafe.Add<T>(ref dst, runCount + direction * 1) = Unsafe.Add<T>(ref src, runCount + direction * 1);
+                    Unsafe.Add<T>(ref dst, runCount + direction * 2) = Unsafe.Add<T>(ref src, runCount + direction * 2);
+                    Unsafe.Add<T>(ref dst, runCount + direction * 3) = Unsafe.Add<T>(ref src, runCount + direction * 3);
+                    runCount += direction * 4;
+                    loopCount += 4;
+                }
+                for (; loopCount < srcLength; ++loopCount)
                 {
-                    // Source address less than destination address. Must do backward copy.
-                    int i = srcLength;
-                    while (i-- != 0)
-                    {
-                        Unsafe.Add<T>(ref dst, i) = Unsafe.Add<T>(ref src, i);
-                    }
+                    Unsafe.Add<T>(ref dst, runCount) = Unsafe.Add<T>(ref src, runCount);
+                    runCount += direction;
                 }
             }
         }