buffer: speed up swap16/32, add swap64

* Speed up buffer.swap16 and swap32 by using builtins. Up to ~6x gain.
  Drop transition point between JS and C++ implementations accordingly.
  Amount of performance improvement not only depends on buffer size but
  also memory alignment.
* Fix tests: C++ impl tests were testing 0-filled buffers so were
  always passing.
* Add similar buffer.swap64 method.
* Make buffer-swap benchmark mirror JS impl.

doc/api/buffer.markdown has an entry of "added: REPLACEME" that should
be changed to the correct release number before tagged.

Because node is currently using a very old version of cpplint.py it
doesn't know that std::swap() has moved from <algorithm> to <utility> in
c++11. So until cpplint.py is updated simply NOLINT the line.
Technically it should be NOLINT(build/include_what_you_use), but that
puts the line over 80 characters causing another lint error.

PR-URL: https://github.com/nodejs/node/pull/7157
Reviewed-By: Trevor Norris <trev.norris@gmail.com>
Reviewed-By: James M Snell <jasnell@gmail.com>

5 files changed, 351 insertions, 111 deletions

diff --git a/benchmark/buffers/buffer-swap.js b/benchmark/buffers/buffer-swap.js
index 4ff9a23d41..0793c77c8f 100644
--- a/benchmark/buffers/buffer-swap.js
+++ b/benchmark/buffers/buffer-swap.js
@@ -1,61 +1,90 @@
 'use strict';
 
 const common = require('../common.js');
+const v8 = require('v8');
 
 const bench = common.createBenchmark(main, {
-  method: ['swap16', 'swap32', 'htons', 'htonl'],
-  len: [4, 64, 512, 768, 1024, 1536, 2056, 4096, 8192],
-  n: [1e6]
+  aligned: ['true', 'false'],
+  method: ['swap16', 'swap32', 'swap64'/*, 'htons', 'htonl', 'htonll'*/],
+  len: [8, 64, 128, 256, 512, 768, 1024, 1536, 2056, 4096, 8192],
+  n: [5e7]
 });
 
 // The htons and htonl methods below are used to benchmark the
 // performance difference between doing the byteswap in pure
 // javascript regardless of Buffer size as opposed to dropping
-// down to the native layer for larger Buffer sizes.
+// down to the native layer for larger Buffer sizes. Commented
+// out by default because they are slow for big buffers. If
+// re-evaluating the crossover point, uncomment those methods
+// and comment out their implementations in lib/buffer.js so
+// C++ version will always be used.
+
+function swap(b, n, m) {
+  const i = b[n];
+  b[n] = b[m];
+  b[m] = i;
+}
 
 Buffer.prototype.htons = function htons() {
   if (this.length % 2 !== 0)
     throw new RangeError();
-  for (var i = 0, n = 0; i < this.length; i += 2) {
-    n = this[i];
-    this[i] = this[i + 1];
-    this[i + 1] = n;
+  for (var i = 0; i < this.length; i += 2) {
+    swap(this, i, i + 1);
   }
   return this;
 };
 
 Buffer.prototype.htonl = function htonl() {
-  if (this.length % 2 !== 0)
+  if (this.length % 4 !== 0)
+    throw new RangeError();
+  for (var i = 0; i < this.length; i += 4) {
+    swap(this, i, i + 3);
+    swap(this, i + 1, i + 2);
+  }
+  return this;
+};
+
+Buffer.prototype.htonll = function htonl() {
+  if (this.length % 8 !== 0)
     throw new RangeError();
-  for (var i = 0, n = 0; i < this.length; i += 4) {
-    n = this[i];
-    this[i] = this[i + 3];
-    this[i + 3] = n;
-    n = this[i + 1];
-    this[i + 1] = this[i + 2];
-    this[i + 2] = n;
+  for (var i = 0; i < this.length; i += 8) {
+    swap(this, i, i + 7);
+    swap(this, i + 1, i + 6);
+    swap(this, i + 2, i + 5);
+    swap(this, i + 3, i + 4);
   }
   return this;
 };
 
-function createBuffer(len) {
+function createBuffer(len, aligned) {
+  len += aligned ? 0 : 1;
   const buf = Buffer.allocUnsafe(len);
   for (var i = 1; i <= len; i++)
     buf[i - 1] = i;
-  return buf;
+  return aligned ? buf : buf.slice(1);
 }
 
-function bufferSwap(n, buf, method) {
-  for (var i = 1; i <= n; i++)
-    buf[method]();
+function genMethod(method) {
+  const fnString =
+      'return function ' + method + '(n, buf) {' +
+      '  for (var i = 0; i <= n; i++)' +
+      '    buf.' + method + '();' +
+      '}';
+  return (new Function(fnString))();
 }
 
 function main(conf) {
   const method = conf.method;
   const len = conf.len | 0;
   const n = conf.n | 0;
-  const buf = createBuffer(len);
+  const aligned = conf.aligned || 'true';
+  const buf = createBuffer(len, aligned === 'true');
+  const bufferSwap = genMethod(method);
+
+  v8.setFlagsFromString('--allow_natives_syntax');
+  eval('%OptimizeFunctionOnNextCall(bufferSwap)');
+
   bench.start();
-  bufferSwap(n, buf, method);
+  bufferSwap(n, buf);
   bench.end(n);
 }
diff --git a/doc/api/buffer.md b/doc/api/buffer.md
index 66832d1edd..3102348525 100644
--- a/doc/api/buffer.md
+++ b/doc/api/buffer.md
@@ -1470,10 +1470,10 @@ calls can be chained.
 ```js
 const buf = Buffer.from([0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8]);
 console.log(buf);
-  // Prints Buffer(0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8)
+  // Prints <Buffer 01 02 03 04 05 06 07 08>
 buf.swap16();
 console.log(buf);
-  // Prints Buffer(0x2, 0x1, 0x4, 0x3, 0x6, 0x5, 0x8, 0x7)
+  // Prints <Buffer 02 01 04 03 06 05 08 07>
 ```
 
 ### buf.swap32()
@@ -1491,12 +1491,36 @@ calls can be chained.
 ```js
 const buf = Buffer.from([0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8]);
 console.log(buf);
-  // Prints Buffer(0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8)
+  // Prints <Buffer 01 02 03 04 05 06 07 08>
 buf.swap32();
 console.log(buf);
-  // Prints Buffer(0x4, 0x3, 0x2, 0x1, 0x8, 0x7, 0x6, 0x5)
+  // Prints <Buffer 04 03 02 01 08 07 06 05>
 ```
 
+### buf.swap64()
+<!-- YAML
+added: REPLACEME
+-->
+
+* Return: {Buffer}
+
+Interprets the `Buffer` as an array of 64-bit numbers and swaps
+the byte-order *in-place*. Throws a `RangeError` if the `Buffer` length is
+not a multiple of 64 bits. The method returns a reference to the Buffer, so
+calls can be chained.
+
+```js
+const buf = Buffer.from([0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8]);
+console.log(buf);
+  // Prints <Buffer 01 02 03 04 05 06 07 08>
+buf.swap64();
+console.log(buf);
+  // Prints <Buffer 08 07 06 05 04 03 02 01>
+```
+
+Note that JavaScript cannot encode 64-bit integers. This method is intended
+for working with 64-bit floats.
+
 ### buf.toString([encoding[, start[, end]]])
 
 * `encoding` {String} Default: `'utf8'`
diff --git a/lib/buffer.js b/lib/buffer.js
index 2bdc28e0be..21f7d2f21d 100644
--- a/lib/buffer.js
+++ b/lib/buffer.js
@@ -25,47 +25,6 @@ var poolSize, poolOffset, allocPool;
 
 binding.setupBufferJS(Buffer.prototype, bindingObj);
 
-const swap16n = binding.swap16;
-const swap32n = binding.swap32;
-
-function swap(b, n, m) {
-  const i = b[n];
-  b[n] = b[m];
-  b[m] = i;
-}
-
-Buffer.prototype.swap16 = function swap16() {
-  // For Buffer.length < 512, it's generally faster to
-  // do the swap in javascript. For larger buffers,
-  // dropping down to the native code is faster.
-  const len = this.length;
-  if (len % 2 !== 0)
-    throw new RangeError('Buffer size must be a multiple of 16-bits');
-  if (len < 512) {
-    for (var i = 0; i < len; i += 2)
-      swap(this, i, i + 1);
-    return this;
-  }
-  return swap16n.apply(this);
-};
-
-Buffer.prototype.swap32 = function swap32() {
-  // For Buffer.length < 1024, it's generally faster to
-  // do the swap in javascript. For larger buffers,
-  // dropping down to the native code is faster.
-  const len = this.length;
-  if (len % 4 !== 0)
-    throw new RangeError('Buffer size must be a multiple of 32-bits');
-  if (len < 1024) {
-    for (var i = 0; i < len; i += 4) {
-      swap(this, i, i + 3);
-      swap(this, i + 1, i + 2);
-    }
-    return this;
-  }
-  return swap32n.apply(this);
-};
-
 // |binding.zeroFill| can be undefined when running inside an isolate where we
 // do not own the ArrayBuffer allocator.  Zero fill is always on in that case.
 const zeroFill = bindingObj.zeroFill || [0];
@@ -1303,3 +1262,67 @@ Buffer.prototype.writeDoubleBE = function writeDoubleBE(val, offset, noAssert) {
     binding.writeDoubleBE(this, val, offset, true);
   return offset + 8;
 };
+
+const swap16n = binding.swap16;
+const swap32n = binding.swap32;
+const swap64n = binding.swap64;
+
+function swap(b, n, m) {
+  const i = b[n];
+  b[n] = b[m];
+  b[m] = i;
+}
+
+
+Buffer.prototype.swap16 = function swap16() {
+  // For Buffer.length < 128, it's generally faster to
+  // do the swap in javascript. For larger buffers,
+  // dropping down to the native code is faster.
+  const len = this.length;
+  if (len % 2 !== 0)
+    throw new RangeError('Buffer size must be a multiple of 16-bits');
+  if (len < 128) {
+    for (var i = 0; i < len; i += 2)
+      swap(this, i, i + 1);
+    return this;
+  }
+  return swap16n(this);
+};
+
+
+Buffer.prototype.swap32 = function swap32() {
+  // For Buffer.length < 192, it's generally faster to
+  // do the swap in javascript. For larger buffers,
+  // dropping down to the native code is faster.
+  const len = this.length;
+  if (len % 4 !== 0)
+    throw new RangeError('Buffer size must be a multiple of 32-bits');
+  if (len < 192) {
+    for (var i = 0; i < len; i += 4) {
+      swap(this, i, i + 3);
+      swap(this, i + 1, i + 2);
+    }
+    return this;
+  }
+  return swap32n(this);
+};
+
+
+Buffer.prototype.swap64 = function swap64() {
+  // For Buffer.length < 192, it's generally faster to
+  // do the swap in javascript. For larger buffers,
+  // dropping down to the native code is faster.
+  const len = this.length;
+  if (len % 8 !== 0)
+    throw new RangeError('Buffer size must be a multiple of 64-bits');
+  if (len < 192) {
+    for (var i = 0; i < len; i += 8) {
+      swap(this, i, i + 7);
+      swap(this, i + 1, i + 6);
+      swap(this, i + 2, i + 5);
+      swap(this, i + 3, i + 4);
+    }
+    return this;
+  }
+  return swap64n(this);
+};
diff --git a/src/node_buffer.cc b/src/node_buffer.cc
index 2472c0bb82..25a14f5c24 100644
--- a/src/node_buffer.cc
+++ b/src/node_buffer.cc
@@ -12,6 +12,7 @@
 
 #include <string.h>
 #include <limits.h>
+#include <utility>
 
 #define BUFFER_ID 0xB0E4
 
@@ -51,12 +52,37 @@
 #define BUFFER_MALLOC(length)                                               \
   zero_fill_all_buffers ? calloc(length, 1) : malloc(length)
 
-#define SWAP_BYTES(arr, a, b)                                               \
-  do {                                                                      \
-    const uint8_t lo = arr[a];                                              \
-    arr[a] = arr[b];                                                        \
-    arr[b] = lo;                                                            \
-  } while (0)
+#if defined(__GNUC__) || defined(__clang__)
+#define BSWAP_INTRINSIC_2(x) __builtin_bswap16(x)
+#define BSWAP_INTRINSIC_4(x) __builtin_bswap32(x)
+#define BSWAP_INTRINSIC_8(x) __builtin_bswap64(x)
+#elif defined(__linux__)
+#include <byteswap.h>
+#define BSWAP_INTRINSIC_2(x) bswap_16(x)
+#define BSWAP_INTRINSIC_4(x) bswap_32(x)
+#define BSWAP_INTRINSIC_8(x) bswap_64(x)
+#elif defined(_MSC_VER)
+#include <intrin.h>
+#define BSWAP_INTRINSIC_2(x) _byteswap_ushort(x);
+#define BSWAP_INTRINSIC_4(x) _byteswap_ulong(x);
+#define BSWAP_INTRINSIC_8(x) _byteswap_uint64(x);
+#else
+#define BSWAP_INTRINSIC_2(x) ((x) << 8) | ((x) >> 8)
+#define BSWAP_INTRINSIC_4(x)                                                  \
+  (((x) & 0xFF) << 24) |                                                      \
+  (((x) & 0xFF00) << 8) |                                                     \
+  (((x) >> 8) & 0xFF00) |                                                     \
+  (((x) >> 24) & 0xFF)
+#define BSWAP_INTRINSIC_8(x)                                                  \
+  (((x) & 0xFF00000000000000ull) >> 56) |                                     \
+  (((x) & 0x00FF000000000000ull) >> 40) |                                     \
+  (((x) & 0x0000FF0000000000ull) >> 24) |                                     \
+  (((x) & 0x000000FF00000000ull) >> 8) |                                      \
+  (((x) & 0x00000000FF000000ull) << 8) |                                      \
+  (((x) & 0x0000000000FF0000ull) << 24) |                                     \
+  (((x) & 0x000000000000FF00ull) << 40) |                                     \
+  (((x) & 0x00000000000000FFull) << 56)
+#endif
 
 namespace node {
 
@@ -1150,29 +1176,88 @@ void IndexOfNumber(const FunctionCallbackInfo<Value>& args) {
                                 : -1);
 }
 
+
 void Swap16(const FunctionCallbackInfo<Value>& args) {
   Environment* env = Environment::GetCurrent(args);
-  THROW_AND_RETURN_UNLESS_BUFFER(env, args.This());
-  SPREAD_ARG(args.This(), ts_obj);
+  THROW_AND_RETURN_UNLESS_BUFFER(env, args[0]);
+  SPREAD_ARG(args[0], ts_obj);
+
+  CHECK_EQ(ts_obj_length % 2, 0);
 
-  for (size_t i = 0; i < ts_obj_length; i += 2) {
-    SWAP_BYTES(ts_obj_data, i, i + 1);
+  int align = reinterpret_cast<uintptr_t>(ts_obj_data) % sizeof(uint16_t);
+
+  if (align == 0) {
+    uint16_t* data16 = reinterpret_cast<uint16_t*>(ts_obj_data);
+    size_t len16 = ts_obj_length / 2;
+    for (size_t i = 0; i < len16; i++) {
+      data16[i] = BSWAP_INTRINSIC_2(data16[i]);
+    }
+  } else {
+    for (size_t i = 0; i < ts_obj_length; i += 2) {
+      std::swap(ts_obj_data[i], ts_obj_data[i + 1]);
+    }
   }
-  args.GetReturnValue().Set(args.This());
+
+  args.GetReturnValue().Set(args[0]);
 }
 
+
 void Swap32(const FunctionCallbackInfo<Value>& args) {
   Environment* env = Environment::GetCurrent(args);
-  THROW_AND_RETURN_UNLESS_BUFFER(env, args.This());
-  SPREAD_ARG(args.This(), ts_obj);
+  THROW_AND_RETURN_UNLESS_BUFFER(env, args[0]);
+  SPREAD_ARG(args[0], ts_obj);
+
+  CHECK_EQ(ts_obj_length % 4, 0);
+
+  int align = reinterpret_cast<uintptr_t>(ts_obj_data) % sizeof(uint32_t);
 
-  for (size_t i = 0; i < ts_obj_length; i += 4) {
-    SWAP_BYTES(ts_obj_data, i, i + 3);
-    SWAP_BYTES(ts_obj_data, i + 1, i + 2);
+  if (align == 0) {
+    uint32_t* data32 = reinterpret_cast<uint32_t*>(ts_obj_data);
+    size_t len32 = ts_obj_length / 4;
+    for (size_t i = 0; i < len32; i++) {
+      data32[i] = BSWAP_INTRINSIC_4(data32[i]);
+    }
+  } else {
+    for (size_t i = 0; i < ts_obj_length; i += 4) {
+      std::swap(ts_obj_data[i], ts_obj_data[i + 3]);
+      std::swap(ts_obj_data[i + 1], ts_obj_data[i + 2]);
+    }
   }
-  args.GetReturnValue().Set(args.This());
+
+  args.GetReturnValue().Set(args[0]);
 }
 
+
+void Swap64(const FunctionCallbackInfo<Value>& args) {
+  Environment* env = Environment::GetCurrent(args);
+  THROW_AND_RETURN_UNLESS_BUFFER(env, args[0]);
+  SPREAD_ARG(args[0], ts_obj);
+
+  CHECK_EQ(ts_obj_length % 8, 0);
+
+  int align = reinterpret_cast<uintptr_t>(ts_obj_data) % sizeof(uint64_t);
+
+  if (align == 0) {
+    uint64_t* data64 = reinterpret_cast<uint64_t*>(ts_obj_data);
+    size_t len32 = ts_obj_length / 8;
+    for (size_t i = 0; i < len32; i++) {
+      data64[i] = BSWAP_INTRINSIC_8(data64[i]);
+    }
+  } else {
+    for (size_t i = 0; i < ts_obj_length; i += 8) {
+      std::swap(ts_obj_data[i], ts_obj_data[i + 7]);
+      std::swap(ts_obj_data[i + 1], ts_obj_data[i + 6]);
+      std::swap(ts_obj_data[i + 2], ts_obj_data[i + 5]);
+      // NOLINT added because current cpplint.py is old and doesn't know that
+      // std::swap() now lives in <utility> instead of <algorithm>.
+      std::swap(ts_obj_data[i + 3], ts_obj_data[i + 4]);  // NOLINT
+    }
+  }
+
+  args.GetReturnValue().Set(args[0]);
+}
+
+
 // pass Buffer object to load prototype methods
 void SetupBufferJS(const FunctionCallbackInfo<Value>& args) {
   Environment* env = Environment::GetCurrent(args);
@@ -1238,6 +1323,7 @@ void Initialize(Local<Object> target,
 
   env->SetMethod(target, "swap16", Swap16);
   env->SetMethod(target, "swap32", Swap32);
+  env->SetMethod(target, "swap64", Swap64);
 
   target->Set(env->context(),
               FIXED_ONE_BYTE_STRING(env->isolate(), "kMaxLength"),
diff --git a/test/parallel/test-buffer-swap.js b/test/parallel/test-buffer-swap.js
index c2a6901955..f72b638f75 100644
--- a/test/parallel/test-buffer-swap.js
+++ b/test/parallel/test-buffer-swap.js
@@ -3,31 +3,28 @@
 require('../common');
 const assert = require('assert');
 
-const buf = Buffer.from([0x1, 0x2, 0x3, 0x4]);
+// Test buffers small enough to use the JS implementation
+const buf = Buffer.from([0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09,
+                         0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10]);
 
 assert.strictEqual(buf, buf.swap16());
-assert.deepStrictEqual(buf, Buffer.from([0x2, 0x1, 0x4, 0x3]));
+assert.deepStrictEqual(buf, Buffer.from([0x02, 0x01, 0x04, 0x03, 0x06, 0x05,
+                                         0x08, 0x07, 0x0a, 0x09, 0x0c, 0x0b,
+                                         0x0e, 0x0d, 0x10, 0x0f]));
+buf.swap16(); // restore
 
 assert.strictEqual(buf, buf.swap32());
-assert.deepStrictEqual(buf, Buffer.from([0x3, 0x4, 0x1, 0x2]));
-
-const buf_array = [];
-for (var i = 1; i < 33; i++)
-  buf_array.push(i);
-const buf2 = Buffer.from(buf_array);
-buf2.swap32();
-assert.deepStrictEqual(buf2,
-    Buffer.from([0x04, 0x03, 0x02, 0x01, 0x08, 0x07, 0x06, 0x05, 0x0c,
-                 0x0b, 0x0a, 0x09, 0x10, 0x0f, 0x0e, 0x0d, 0x14, 0x13,
-                 0x12, 0x11, 0x18, 0x17, 0x16, 0x15, 0x1c, 0x1b, 0x1a,
-                 0x19, 0x20, 0x1f, 0x1e, 0x1d]));
-buf2.swap16();
-assert.deepStrictEqual(buf2,
-    Buffer.from([0x03, 0x04, 0x01, 0x02, 0x07, 0x08, 0x05, 0x06, 0x0b,
-                 0x0c, 0x09, 0x0a, 0x0f, 0x10, 0x0d, 0x0e, 0x13, 0x14,
-                 0x11, 0x12, 0x17, 0x18, 0x15, 0x16, 0x1b, 0x1c, 0x19,
-                 0x1a, 0x1f, 0x20, 0x1d, 0x1e]));
+assert.deepStrictEqual(buf, Buffer.from([0x04, 0x03, 0x02, 0x01, 0x08, 0x07,
+                                         0x06, 0x05, 0x0c, 0x0b, 0x0a, 0x09,
+                                         0x10, 0x0f, 0x0e, 0x0d]));
+buf.swap32(); // restore
 
+assert.strictEqual(buf, buf.swap64());
+assert.deepStrictEqual(buf, Buffer.from([0x08, 0x07, 0x06, 0x05, 0x04, 0x03,
+                                         0x02, 0x01, 0x10, 0x0f, 0x0e, 0x0d,
+                                         0x0c, 0x0b, 0x0a, 0x09]));
+
+// Operates in-place
 const buf3 = Buffer.from([0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7]);
 buf3.slice(1, 5).swap32();
 assert.deepStrictEqual(buf3, Buffer.from([0x1, 0x5, 0x4, 0x3, 0x2, 0x6, 0x7]));
@@ -35,26 +32,107 @@ assert.deepStrictEqual(buf3, Buffer.from([0x1, 0x5, 0x4, 0x3, 0x2, 0x6, 0x7]));
 buf3.slice(1, 5).swap16();
 assert.deepStrictEqual(buf3, Buffer.from([0x1, 0x4, 0x5, 0x2, 0x3, 0x6, 0x7]));
 
+const buf3_64 = Buffer.from([0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
+                             0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10,
+                             0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
+                             0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10]);
+buf3_64.slice(2, 18).swap64();
+assert.deepStrictEqual(buf3_64, Buffer.from([0x01, 0x02, 0x0a, 0x09, 0x08, 0x07,
+                                             0x06, 0x05, 0x04, 0x03, 0x02, 0x01,
+                                             0x10, 0x0f, 0x0e, 0x0d, 0x0c, 0x0b,
+                                             0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
+                                             0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e,
+                                             0x0f, 0x10]));
+
 // Force use of native code (Buffer size above threshold limit for js impl)
-const buf4 = Buffer.allocUnsafe(1024).fill([0x1, 0x2, 0x3, 0x4]);
-const buf5 = Buffer.allocUnsafe(1024).fill([0x2, 0x1, 0x4, 0x3]);
-const buf6 = Buffer.allocUnsafe(1024)
-                   .fill([0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8]);
-const buf7 = Buffer.allocUnsafe(1024)
-                   .fill([0x4, 0x3, 0x2, 0x1, 0x8, 0x7, 0x6, 0x5]);
+var buf4A = new Uint32Array(256).fill(0x04030201);
+var buf4 = Buffer.from(buf4A.buffer, buf4A.byteOffset);
+var buf5A = new Uint32Array(256).fill(0x03040102);
+var buf5 = Buffer.from(buf5A.buffer, buf5A.byteOffset);
 
 buf4.swap16();
 assert.deepStrictEqual(buf4, buf5);
 
+var buf6A = new Uint32Array(256).fill(0x04030201);
+var buf6 = Buffer.from(buf6A.buffer);
+var bu7A = new Uint32Array(256).fill(0x01020304);
+var buf7 = Buffer.from(bu7A.buffer, bu7A.byteOffset);
+
 buf6.swap32();
 assert.deepStrictEqual(buf6, buf7);
 
+var buf8A = new Uint8Array(256 * 8);
+var buf9A = new Uint8Array(256 * 8);
+for (let i = 0; i < buf8A.length; i++) {
+  buf8A[i] = i % 8;
+  buf9A[buf9A.length - i - 1] = i % 8;
+}
+var buf8 = Buffer.from(buf8A.buffer, buf8A.byteOffset);
+var buf9 = Buffer.from(buf9A.buffer, buf9A.byteOffset);
+
+buf8.swap64();
+assert.deepStrictEqual(buf8, buf9);
+
+// Test native code with buffers that are not memory-aligned
+var buf10A = new Uint8Array(256 * 8);
+var buf11A = new Uint8Array(256 * 8 - 2);
+for (let i = 0; i < buf10A.length; i++) {
+  buf10A[i] = i % 2;
+}
+for (let i = 1; i < buf11A.length; i++) {
+  buf11A[buf11A.length - i] = (i + 1) % 2;
+}
+var buf10 = Buffer.from(buf10A.buffer, buf10A.byteOffset);
+// 0|1 0|1 0|1...
+var buf11 = Buffer.from(buf11A.buffer, buf11A.byteOffset);
+// 0|0 1|0 1|0...
+
+buf10.slice(1, buf10.length - 1).swap16();
+assert.deepStrictEqual(buf10.slice(0, buf11.length), buf11);
+
+
+var buf12A = new Uint8Array(256 * 8);
+var buf13A = new Uint8Array(256 * 8 - 4);
+for (let i = 0; i < buf12A.length; i++) {
+  buf12A[i] = i % 4;
+}
+for (let i = 1; i < buf13A.length; i++) {
+  buf13A[buf13A.length - i] = (i + 1) % 4;
+}
+var buf12 = Buffer.from(buf12A.buffer, buf12A.byteOffset);
+// 0|1 2 3 0|1 2 3...
+var buf13 = Buffer.from(buf13A.buffer, buf13A.byteOffset);
+// 0|0 3 2 1|0 3 2...
+
+buf12.slice(1, buf12.length - 3).swap32();
+assert.deepStrictEqual(buf12.slice(0, buf13.length), buf13);
+
+
+var buf14A = new Uint8Array(256 * 8);
+var buf15A = new Uint8Array(256 * 8 - 8);
+for (let i = 0; i < buf14A.length; i++) {
+  buf14A[i] = i % 8;
+}
+for (let i = 1; i < buf15A.length; i++) {
+  buf15A[buf15A.length - i] = (i + 1) % 8;
+}
+var buf14 = Buffer.from(buf14A.buffer, buf14A.byteOffset);
+// 0|1 2 3 4 5 6 7 0|1 2 3 4...
+var buf15 = Buffer.from(buf15A.buffer, buf15A.byteOffset);
+// 0|0 7 6 5 4 3 2 1|0 7 6 5...
+
+buf14.slice(1, buf14.length - 7).swap64();
+assert.deepStrictEqual(buf14.slice(0, buf15.length), buf15);
 
+// Length assertions
 const re16 = /Buffer size must be a multiple of 16-bits/;
 const re32 = /Buffer size must be a multiple of 32-bits/;
+const re64 = /Buffer size must be a multiple of 64-bits/;
 
 assert.throws(() => Buffer.from(buf3).swap16(), re16);
 assert.throws(() => Buffer.alloc(1025).swap16(), re16);
 assert.throws(() => Buffer.from(buf3).swap32(), re32);
 assert.throws(() => buf3.slice(1, 3).swap32(), re32);
 assert.throws(() => Buffer.alloc(1025).swap32(), re32);
+assert.throws(() => buf3.slice(1, 3).swap64(), re64);
+assert.throws(() => Buffer.alloc(1025).swap64(), re64);