Use ArrayPool in FileStream.CopyToAsync

Rather than allocating a new buffer in CopyToAsync, take one from the shared array pool.  This drastically reduces the memory allocation from a FileStream.CopyToAsync operation, and helps to improve throughput as a result.  In one microbenchmark that repeatedly copied an 80K file with CopyToAsync's default args, throughput increased by 20% and gen0 GCs decreased by 100x.

Author: stephentoub

src/System.IO.FileSystem/src/System/IO/Win32FileStream.cs

@@ -2,6 +2,7 @@
 // The .NET Foundation licenses this file to you under the MIT license.
 // See the LICENSE file in the project root for more information.
 
+using System.Buffers;
 using System.Diagnostics;
 using System.Diagnostics.Contracts;
 using System.Runtime.InteropServices;
@@ -1775,13 +1776,13 @@ private async Task AsyncModeCopyToAsync(Stream destination, int bufferSize, Canc
                 readAwaitable._position = _pos;
             }
 
-            // Create the buffer to use for the copy operation, as the base CopyToAsync does. We don't try to use
+            // Get the buffer to use for the copy operation, as the base CopyToAsync does. We don't try to use
             // _buffer here, even if it's not null, as concurrent operations are allowed, and another operation may
             // actually be using the buffer already. Plus, it'll be rare for _buffer to be non-null, as typically
             // CopyToAsync is used as the only operation performed on the stream, and the buffer is lazily initialized.
             // Further, typically the CopyToAsync buffer size will be larger than that used by the FileStream, such that
-            // we'd likely be unable to use it anyway.  A better option than using _buffer would be a future pooling solution.
-            byte[] copyBuffer = new byte[bufferSize];
+            // we'd likely be unable to use it anyway.  Instead, we rent the buffer from a pool.
+            byte[] copyBuffer = ArrayPool<byte>.Shared.Rent(bufferSize);
 
             // Allocate an Overlapped we can use repeatedly for all operations
             var awaitableOverlapped = new PreAllocatedOverlapped(AsyncCopyToAwaitable.s_callback, readAwaitable, copyBuffer);
@@ -1914,6 +1915,7 @@ private async Task AsyncModeCopyToAsync(Stream destination, int bufferSize, Canc
                 // Cleanup from the whole copy operation
                 cancellationReg.Dispose();
                 awaitableOverlapped.Dispose();
+                ArrayPool<byte>.Shared.Return(copyBuffer, clearArray: true);
 
                 // Make sure the stream's current position reflects where we ended up
                 if (!_handle.IsClosed && _parent.CanSeek)

src/System.IO.FileSystem/src/netcore50/project.json

@@ -1,6 +1,7 @@
 {
   "dependencies": {
     "Microsoft.NETCore.Platforms": "1.0.1",
+    "System.Buffers": "4.0.0",
     "System.Collections": "4.0.10",
     "System.Diagnostics.Contracts": "4.0.0",
     "System.Diagnostics.Debug": "4.0.10",

src/System.IO.FileSystem/src/project.json

@@ -3,6 +3,7 @@
     "netstandard1.3": {
       "dependencies": {
         "Microsoft.NETCore.Platforms": "1.0.1",
+        "System.Buffers": "4.0.0",
         "System.Collections": "4.0.10",
         "System.Diagnostics.Contracts": "4.0.0",
         "System.Diagnostics.Debug": "4.0.10",

src/System.IO.FileSystem/src/win/project.json

@@ -3,6 +3,7 @@
     "netstandard1.3": {
       "dependencies": {
         "Microsoft.NETCore.Platforms": "1.0.1",
+        "System.Buffers": "4.0.0",
         "System.Collections": "4.0.10",
         "System.Diagnostics.Contracts": "4.0.0",
         "System.Diagnostics.Debug": "4.0.10",