Improve predictability of writeUtf8/writeAscii performance (#10368)

Motivation:

writeUtf8 can suffer from inlining issues and/or megamorphic call-sites on the hot path due to ByteBuf hierarchy

Modifications:

Duplicate and specialize the code paths to reduce the need of polymorphic calls

Result:

Performance are more stable in user code

buffer/src/main/java/io/netty/buffer/AbstractByteBuf.java

@@ -705,7 +705,7 @@ public abstract class AbstractByteBuf extends ByteBuf {
             } else {
                 checkIndex(index, length);
             }
-            return ByteBufUtil.writeUtf8(this, index, sequence, sequence.length());
+            return ByteBufUtil.writeUtf8(this, index, length, sequence, sequence.length());
         }
         if (charset.equals(CharsetUtil.US_ASCII) || charset.equals(CharsetUtil.ISO_8859_1)) {
             int length = sequence.length();

buffer/src/main/java/io/netty/buffer/ByteBufUtil.java

@@ -570,7 +570,7 @@ public final class ByteBufUtil {
             } else if (buf instanceof AbstractByteBuf) {
                 AbstractByteBuf byteBuf = (AbstractByteBuf) buf;
                 byteBuf.ensureWritable0(reserveBytes);
-                int written = writeUtf8(byteBuf, byteBuf.writerIndex, seq, start, end);
+                int written = writeUtf8(byteBuf, byteBuf.writerIndex, reserveBytes, seq, start, end);
                 byteBuf.writerIndex += written;
                 return written;
             } else if (buf instanceof WrappedByteBuf) {
@@ -584,12 +584,111 @@ public final class ByteBufUtil {
         }
     }
 
-    static int writeUtf8(AbstractByteBuf buffer, int writerIndex, CharSequence seq, int len) {
+    static int writeUtf8(AbstractByteBuf buffer, int writerIndex, int reservedBytes, CharSequence seq, int len) {
-        return writeUtf8(buffer, writerIndex, seq, 0, len);
+        return writeUtf8(buffer, writerIndex, reservedBytes, seq, 0, len);
     }
 
     // Fast-Path implementation
-    static int writeUtf8(AbstractByteBuf buffer, int writerIndex, CharSequence seq, int start, int end) {
+    static int writeUtf8(AbstractByteBuf buffer, int writerIndex, int reservedBytes,
+                         CharSequence seq, int start, int end) {
+        if (seq instanceof AsciiString) {
+            writeAsciiString(buffer, writerIndex, (AsciiString) seq, start, end);
+            return end - start;
+        }
+        if (PlatformDependent.hasUnsafe()) {
+            if (buffer.hasArray()) {
+                return unsafeWriteUtf8(buffer.array(), PlatformDependent.byteArrayBaseOffset(),
+                                       buffer.arrayOffset() + writerIndex, seq, start, end);
+            }
+            if (buffer.hasMemoryAddress()) {
+                return unsafeWriteUtf8(null, buffer.memoryAddress(), writerIndex, seq, start, end);
+            }
+        } else {
+            if (buffer.hasArray()) {
+                return safeArrayWriteUtf8(buffer.array(), buffer.arrayOffset() + writerIndex, seq, start, end);
+            }
+            if (buffer.isDirect()) {
+                assert buffer.nioBufferCount() == 1;
+                final ByteBuffer internalDirectBuffer = buffer.internalNioBuffer(writerIndex, reservedBytes);
+                final int bufferPosition = internalDirectBuffer.position();
+                return safeDirectWriteUtf8(internalDirectBuffer, bufferPosition, seq, start, end);
+            }
+        }
+        return safeWriteUtf8(buffer, writerIndex, seq, start, end);
+    }
+
+    // AsciiString Fast-Path implementation - no explicit bound-checks
+    static void writeAsciiString(AbstractByteBuf buffer, int writerIndex, AsciiString seq, int start, int end) {
+        final int begin = seq.arrayOffset() + start;
+        final int length = end - start;
+        if (PlatformDependent.hasUnsafe()) {
+            if (buffer.hasArray()) {
+                PlatformDependent.copyMemory(seq.array(), begin,
+                                             buffer.array(), buffer.arrayOffset() + writerIndex, length);
+                return;
+            }
+            if (buffer.hasMemoryAddress()) {
+                PlatformDependent.copyMemory(seq.array(), begin, buffer.memoryAddress() + writerIndex, length);
+                return;
+            }
+        }
+        if (buffer.hasArray()) {
+            System.arraycopy(seq.array(), begin, buffer.array(), buffer.arrayOffset() + writerIndex, length);
+            return;
+        }
+        buffer.setBytes(writerIndex, seq.array(), begin, length);
+    }
+
+    // Safe off-heap Fast-Path implementation
+    private static int safeDirectWriteUtf8(ByteBuffer buffer, int writerIndex, CharSequence seq, int start, int end) {
+        assert !(seq instanceof AsciiString);
+        int oldWriterIndex = writerIndex;
+
+        // We can use the _set methods as these not need to do any index checks and reference checks.
+        // This is possible as we called ensureWritable(...) before.
+        for (int i = start; i < end; i++) {
+            char c = seq.charAt(i);
+            if (c < 0x80) {
+                buffer.put(writerIndex++, (byte) c);
+            } else if (c < 0x800) {
+                buffer.put(writerIndex++, (byte) (0xc0 | (c >> 6)));
+                buffer.put(writerIndex++, (byte) (0x80 | (c & 0x3f)));
+            } else if (isSurrogate(c)) {
+                if (!Character.isHighSurrogate(c)) {
+                    buffer.put(writerIndex++, WRITE_UTF_UNKNOWN);
+                    continue;
+                }
+                // Surrogate Pair consumes 2 characters.
+                if (++i == end) {
+                    buffer.put(writerIndex++, WRITE_UTF_UNKNOWN);
+                    break;
+                }
+                // Extra method is copied here to NOT allow inlining of writeUtf8
+                // and increase the chance to inline CharSequence::charAt instead
+                char c2 = seq.charAt(i);
+                if (!Character.isLowSurrogate(c2)) {
+                    buffer.put(writerIndex++, WRITE_UTF_UNKNOWN);
+                    buffer.put(writerIndex++, Character.isHighSurrogate(c2)? WRITE_UTF_UNKNOWN : (byte) c2);
+                } else {
+                    int codePoint = Character.toCodePoint(c, c2);
+                    // See http://www.unicode.org/versions/Unicode7.0.0/ch03.pdf#G2630.
+                    buffer.put(writerIndex++, (byte) (0xf0 | (codePoint >> 18)));
+                    buffer.put(writerIndex++, (byte) (0x80 | ((codePoint >> 12) & 0x3f)));
+                    buffer.put(writerIndex++, (byte) (0x80 | ((codePoint >> 6) & 0x3f)));
+                    buffer.put(writerIndex++, (byte) (0x80 | (codePoint & 0x3f)));
+                }
+            } else {
+                buffer.put(writerIndex++, (byte) (0xe0 | (c >> 12)));
+                buffer.put(writerIndex++, (byte) (0x80 | ((c >> 6) & 0x3f)));
+                buffer.put(writerIndex++, (byte) (0x80 | (c & 0x3f)));
+            }
+        }
+        return writerIndex - oldWriterIndex;
+    }
+
+    // Safe off-heap Fast-Path implementation
+    private static int safeWriteUtf8(AbstractByteBuf buffer, int writerIndex, CharSequence seq, int start, int end) {
+        assert !(seq instanceof AsciiString);
         int oldWriterIndex = writerIndex;
 
         // We can use the _set methods as these not need to do any index checks and reference checks.
@@ -611,8 +710,20 @@ public final class ByteBufUtil {
                     buffer._setByte(writerIndex++, WRITE_UTF_UNKNOWN);
                     break;
                 }
-                // Extra method to allow inlining the rest of writeUtf8 which is the most likely code path.
+                // Extra method is copied here to NOT allow inlining of writeUtf8
-                writerIndex = writeUtf8Surrogate(buffer, writerIndex, c, seq.charAt(i));
+                // and increase the chance to inline CharSequence::charAt instead
+                char c2 = seq.charAt(i);
+                if (!Character.isLowSurrogate(c2)) {
+                    buffer._setByte(writerIndex++, WRITE_UTF_UNKNOWN);
+                    buffer._setByte(writerIndex++, Character.isHighSurrogate(c2)? WRITE_UTF_UNKNOWN : c2);
+                } else {
+                    int codePoint = Character.toCodePoint(c, c2);
+                    // See http://www.unicode.org/versions/Unicode7.0.0/ch03.pdf#G2630.
+                    buffer._setByte(writerIndex++, (byte) (0xf0 | (codePoint >> 18)));
+                    buffer._setByte(writerIndex++, (byte) (0x80 | ((codePoint >> 12) & 0x3f)));
+                    buffer._setByte(writerIndex++, (byte) (0x80 | ((codePoint >> 6) & 0x3f)));
+                    buffer._setByte(writerIndex++, (byte) (0x80 | (codePoint & 0x3f)));
+                }
             } else {
                 buffer._setByte(writerIndex++, (byte) (0xe0 | (c >> 12)));
                 buffer._setByte(writerIndex++, (byte) (0x80 | ((c >> 6) & 0x3f)));
@@ -622,19 +733,94 @@ public final class ByteBufUtil {
         return writerIndex - oldWriterIndex;
     }
 
-    private static int writeUtf8Surrogate(AbstractByteBuf buffer, int writerIndex, char c, char c2) {
+    // safe byte[] Fast-Path implementation
-        if (!Character.isLowSurrogate(c2)) {
+    private static int safeArrayWriteUtf8(byte[] buffer, int writerIndex, CharSequence seq, int start, int end) {
-            buffer._setByte(writerIndex++, WRITE_UTF_UNKNOWN);
+        int oldWriterIndex = writerIndex;
-            buffer._setByte(writerIndex++, Character.isHighSurrogate(c2) ? WRITE_UTF_UNKNOWN : c2);
+        for (int i = start; i < end; i++) {
-            return writerIndex;
+            char c = seq.charAt(i);
+            if (c < 0x80) {
+                buffer[writerIndex++] = (byte) c;
+            } else if (c < 0x800) {
+                buffer[writerIndex++] = (byte) (0xc0 | (c >> 6));
+                buffer[writerIndex++] = (byte) (0x80 | (c & 0x3f));
+            } else if (isSurrogate(c)) {
+                if (!Character.isHighSurrogate(c)) {
+                    buffer[writerIndex++] = WRITE_UTF_UNKNOWN;
+                    continue;
+                }
+                // Surrogate Pair consumes 2 characters.
+                if (++i == end) {
+                    buffer[writerIndex++] = WRITE_UTF_UNKNOWN;
+                    break;
+                }
+                char c2 = seq.charAt(i);
+                // Extra method is copied here to NOT allow inlining of writeUtf8
+                // and increase the chance to inline CharSequence::charAt instead
+                if (!Character.isLowSurrogate(c2)) {
+                    buffer[writerIndex++] = WRITE_UTF_UNKNOWN;
+                    buffer[writerIndex++] = (byte) (Character.isHighSurrogate(c2)? WRITE_UTF_UNKNOWN : c2);
+                } else {
+                    int codePoint = Character.toCodePoint(c, c2);
+                    // See http://www.unicode.org/versions/Unicode7.0.0/ch03.pdf#G2630.
+                    buffer[writerIndex++] = (byte) (0xf0 | (codePoint >> 18));
+                    buffer[writerIndex++] = (byte) (0x80 | ((codePoint >> 12) & 0x3f));
+                    buffer[writerIndex++] = (byte) (0x80 | ((codePoint >> 6) & 0x3f));
+                    buffer[writerIndex++] = (byte) (0x80 | (codePoint & 0x3f));
+                }
+            } else {
+                buffer[writerIndex++] = (byte) (0xe0 | (c >> 12));
+                buffer[writerIndex++] = (byte) (0x80 | ((c >> 6) & 0x3f));
+                buffer[writerIndex++] = (byte) (0x80 | (c & 0x3f));
+            }
         }
-        int codePoint = Character.toCodePoint(c, c2);
+        return writerIndex - oldWriterIndex;
-        // See http://www.unicode.org/versions/Unicode7.0.0/ch03.pdf#G2630.
+    }
-        buffer._setByte(writerIndex++, (byte) (0xf0 | (codePoint >> 18)));
+
-        buffer._setByte(writerIndex++, (byte) (0x80 | ((codePoint >> 12) & 0x3f)));
+    // unsafe Fast-Path implementation
-        buffer._setByte(writerIndex++, (byte) (0x80 | ((codePoint >> 6) & 0x3f)));
+    private static int unsafeWriteUtf8(byte[] buffer, long memoryOffset, int writerIndex,
-        buffer._setByte(writerIndex++, (byte) (0x80 | (codePoint & 0x3f)));
+                                       CharSequence seq, int start, int end) {
-        return writerIndex;
+        assert !(seq instanceof AsciiString);
+        long writerOffset = memoryOffset + writerIndex;
+        final long oldWriterOffset = writerOffset;
+        for (int i = start; i < end; i++) {
+            char c = seq.charAt(i);
+            if (c < 0x80) {
+                PlatformDependent.putByte(buffer, writerOffset++, (byte) c);
+            } else if (c < 0x800) {
+                PlatformDependent.putByte(buffer, writerOffset++, (byte) (0xc0 | (c >> 6)));
+                PlatformDependent.putByte(buffer, writerOffset++, (byte) (0x80 | (c & 0x3f)));
+            } else if (isSurrogate(c)) {
+                if (!Character.isHighSurrogate(c)) {
+                    PlatformDependent.putByte(buffer, writerOffset++, WRITE_UTF_UNKNOWN);
+                    continue;
+                }
+                // Surrogate Pair consumes 2 characters.
+                if (++i == end) {
+                    PlatformDependent.putByte(buffer, writerOffset++, WRITE_UTF_UNKNOWN);
+                    break;
+                }
+                char c2 = seq.charAt(i);
+                // Extra method is copied here to NOT allow inlining of writeUtf8
+                // and increase the chance to inline CharSequence::charAt instead
+                if (!Character.isLowSurrogate(c2)) {
+                    PlatformDependent.putByte(buffer, writerOffset++, WRITE_UTF_UNKNOWN);
+                    PlatformDependent.putByte(buffer, writerOffset++,
+                                              (byte) (Character.isHighSurrogate(c2)? WRITE_UTF_UNKNOWN : c2));
+                } else {
+                    int codePoint = Character.toCodePoint(c, c2);
+                    // See http://www.unicode.org/versions/Unicode7.0.0/ch03.pdf#G2630.
+                    PlatformDependent.putByte(buffer, writerOffset++, (byte) (0xf0 | (codePoint >> 18)));
+                    PlatformDependent.putByte(buffer, writerOffset++, (byte) (0x80 | ((codePoint >> 12) & 0x3f)));
+                    PlatformDependent.putByte(buffer, writerOffset++, (byte) (0x80 | ((codePoint >> 6) & 0x3f)));
+                    PlatformDependent.putByte(buffer, writerOffset++, (byte) (0x80 | (codePoint & 0x3f)));
+                }
+            } else {
+                PlatformDependent.putByte(buffer, writerOffset++, (byte) (0xe0 | (c >> 12)));
+                PlatformDependent.putByte(buffer, writerOffset++, (byte) (0x80 | ((c >> 6) & 0x3f)));
+                PlatformDependent.putByte(buffer, writerOffset++, (byte) (0x80 | (c & 0x3f)));
+            }
+        }
+        return (int) (writerOffset - oldWriterOffset);
     }
 
     /**
@@ -742,32 +928,31 @@ public final class ByteBufUtil {
      */
     public static int writeAscii(ByteBuf buf, CharSequence seq) {
         // ASCII uses 1 byte per char
-        final int len = seq.length();
+        for (;;) {
-        if (seq instanceof AsciiString) {
+            if (buf instanceof WrappedCompositeByteBuf) {
-            AsciiString asciiString = (AsciiString) seq;
+                // WrappedCompositeByteBuf is a sub-class of AbstractByteBuf so it needs special handling.
-            buf.writeBytes(asciiString.array(), asciiString.arrayOffset(), len);
+                buf = buf.unwrap();
-        } else {
+            } else if (buf instanceof AbstractByteBuf) {
-            for (;;) {
+                final int len = seq.length();
-                if (buf instanceof WrappedCompositeByteBuf) {
+                AbstractByteBuf byteBuf = (AbstractByteBuf) buf;
-                    // WrappedCompositeByteBuf is a sub-class of AbstractByteBuf so it needs special handling.
+                byteBuf.ensureWritable0(len);
-                    buf = buf.unwrap();
+                if (seq instanceof AsciiString) {
-                } else if (buf instanceof AbstractByteBuf) {
+                    writeAsciiString(byteBuf, byteBuf.writerIndex, (AsciiString) seq, 0, len);
-                    AbstractByteBuf byteBuf = (AbstractByteBuf) buf;
-                    byteBuf.ensureWritable0(len);
-                    int written = writeAscii(byteBuf, byteBuf.writerIndex, seq, len);
-                    byteBuf.writerIndex += written;
-                    return written;
-                } else if (buf instanceof WrappedByteBuf) {
-                    // Unwrap as the wrapped buffer may be an AbstractByteBuf and so we can use fast-path.
-                    buf = buf.unwrap();
                 } else {
-                    byte[] bytes = seq.toString().getBytes(CharsetUtil.US_ASCII);
+                    final int written = writeAscii(byteBuf, byteBuf.writerIndex, seq, len);
-                    buf.writeBytes(bytes);
+                    assert written == len;
-                    return bytes.length;
                 }
+                byteBuf.writerIndex += len;
+                return len;
+            } else if (buf instanceof WrappedByteBuf) {
+                // Unwrap as the wrapped buffer may be an AbstractByteBuf and so we can use fast-path.
+                buf = buf.unwrap();
+            } else {
+                byte[] bytes = seq.toString().getBytes(CharsetUtil.US_ASCII);
+                buf.writeBytes(bytes);
+                return bytes.length;
             }
         }
-        return len;
     }
 
     // Fast-Path implementation

buffer/src/test/java/io/netty/buffer/ByteBufUtilTest.java

@@ -18,11 +18,14 @@ package io.netty.buffer;
 import io.netty.util.AsciiString;
 import io.netty.util.CharsetUtil;
 import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.Parameterized;
 
 import java.nio.ByteOrder;
 import java.nio.charset.Charset;
 import java.util.ArrayList;
 import java.util.Arrays;
+import java.util.Collection;
 import java.util.List;
 import java.util.Random;
 import java.util.concurrent.atomic.AtomicInteger;
@@ -31,16 +34,48 @@ import java.util.concurrent.atomic.AtomicReference;
 import static io.netty.buffer.Unpooled.unreleasableBuffer;
 import static org.hamcrest.Matchers.greaterThan;
 import static org.hamcrest.Matchers.is;
-import static org.junit.Assert.assertArrayEquals;
+import static org.junit.Assert.*;
-import static org.junit.Assert.assertEquals;
-import static org.junit.Assert.assertFalse;
-import static org.junit.Assert.assertTrue;
-import static org.junit.Assert.fail;
-import static org.junit.Assume.assumeFalse;
 import static org.junit.Assume.assumeThat;
-import static org.junit.Assume.assumeTrue;
 
+@RunWith(Parameterized.class)
 public class ByteBufUtilTest {
+
+    private enum BufferType {
+        DIRECT_UNPOOLED, DIRECT_POOLED, HEAP_POOLED, HEAP_UNPOOLED
+    }
+
+    private final BufferType bufferType;
+
+    public ByteBufUtilTest(BufferType bufferType) {
+        this.bufferType = bufferType;
+    }
+
+    private ByteBuf buffer(int capacity) {
+        switch (bufferType) {
+
+        case DIRECT_UNPOOLED:
+            return Unpooled.directBuffer(capacity);
+        case HEAP_UNPOOLED:
+            return Unpooled.buffer(capacity);
+        case DIRECT_POOLED:
+            return PooledByteBufAllocator.DEFAULT.directBuffer(capacity);
+        case HEAP_POOLED:
+            return PooledByteBufAllocator.DEFAULT.buffer(capacity);
+        default:
+            throw new AssertionError("unexpected buffer type: " + bufferType);
+        }
+    }
+
+    @Parameterized.Parameters(name = "bufferType = {0}")
+    public static Collection<Object[]> noUnsafe() {
+        return Arrays.asList(new Object[][] {
+                { BufferType.DIRECT_POOLED },
+                { BufferType.DIRECT_UNPOOLED },
+                { BufferType.HEAP_POOLED },
+                { BufferType.HEAP_UNPOOLED }
+        });
+    }
+
     @Test
     public void decodeRandomHexBytesWithEvenLength() {
         decodeRandomHexBytes(256);
@@ -144,14 +179,14 @@ public class ByteBufUtilTest {
     public void writeShortBE() {
         int expected = 0x1234;
 
-        ByteBuf buf = Unpooled.buffer(2).order(ByteOrder.BIG_ENDIAN);
+        ByteBuf buf = buffer(2).order(ByteOrder.BIG_ENDIAN);
         ByteBufUtil.writeShortBE(buf, expected);
         assertEquals(expected, buf.readShort());
         buf.resetReaderIndex();
         assertEquals(ByteBufUtil.swapShort((short) expected), buf.readShortLE());
         buf.release();
 
-        buf = Unpooled.buffer(2).order(ByteOrder.LITTLE_ENDIAN);
+        buf = buffer(2).order(ByteOrder.LITTLE_ENDIAN);
         ByteBufUtil.writeShortBE(buf, expected);
         assertEquals((short) expected, buf.readShortLE());
         buf.resetReaderIndex();
@@ -184,14 +219,14 @@ public class ByteBufUtilTest {
     public void writeMediumBE() {
         int mediumValue = 0x123456;
 
-        ByteBuf buf = Unpooled.buffer(4).order(ByteOrder.BIG_ENDIAN);
+        ByteBuf buf = buffer(4).order(ByteOrder.BIG_ENDIAN);
         ByteBufUtil.writeMediumBE(buf, mediumValue);
         assertEquals(mediumValue, buf.readMedium());
         buf.resetReaderIndex();
         assertEquals(ByteBufUtil.swapMedium(mediumValue), buf.readMediumLE());
         buf.release();
 
-        buf = Unpooled.buffer(4).order(ByteOrder.LITTLE_ENDIAN);
+        buf = buffer(4).order(ByteOrder.LITTLE_ENDIAN);
         ByteBufUtil.writeMediumBE(buf, mediumValue);
         assertEquals(mediumValue, buf.readMediumLE());
         buf.resetReaderIndex();
@@ -202,9 +237,9 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUsAscii() {
         String usAscii = "NettyRocks";
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(usAscii.getBytes(CharsetUtil.US_ASCII));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeAscii(buf2, usAscii);
 
         assertEquals(buf, buf2);
@@ -216,9 +251,9 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUsAsciiSwapped() {
         String usAscii = "NettyRocks";
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(usAscii.getBytes(CharsetUtil.US_ASCII));
-        SwappedByteBuf buf2 = new SwappedByteBuf(Unpooled.buffer(16));
+        SwappedByteBuf buf2 = new SwappedByteBuf(buffer(16));
         ByteBufUtil.writeAscii(buf2, usAscii);
 
         assertEquals(buf, buf2);
@@ -230,10 +265,10 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUsAsciiWrapped() {
         String usAscii = "NettyRocks";
-        ByteBuf buf = unreleasableBuffer(Unpooled.buffer(16));
+        ByteBuf buf = unreleasableBuffer(buffer(16));
         assertWrapped(buf);
         buf.writeBytes(usAscii.getBytes(CharsetUtil.US_ASCII));
-        ByteBuf buf2 = unreleasableBuffer(Unpooled.buffer(16));
+        ByteBuf buf2 = unreleasableBuffer(buffer(16));
         assertWrapped(buf2);
         ByteBufUtil.writeAscii(buf2, usAscii);
 
@@ -246,10 +281,10 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUsAsciiComposite() {
         String usAscii = "NettyRocks";
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(usAscii.getBytes(CharsetUtil.US_ASCII));
         ByteBuf buf2 = Unpooled.compositeBuffer().addComponent(
-                Unpooled.buffer(8)).addComponent(Unpooled.buffer(24));
+                buffer(8)).addComponent(buffer(24));
         // write some byte so we start writing with an offset.
         buf2.writeByte(1);
         ByteBufUtil.writeAscii(buf2, usAscii);
@@ -264,10 +299,10 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUsAsciiCompositeWrapped() {
         String usAscii = "NettyRocks";
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(usAscii.getBytes(CharsetUtil.US_ASCII));
         ByteBuf buf2 = new WrappedCompositeByteBuf(Unpooled.compositeBuffer().addComponent(
-                Unpooled.buffer(8)).addComponent(Unpooled.buffer(24)));
+                buffer(8)).addComponent(buffer(24)));
         // write some byte so we start writing with an offset.
         buf2.writeByte(1);
         ByteBufUtil.writeAscii(buf2, usAscii);
@@ -282,9 +317,9 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUtf8() {
         String usAscii = "Some UTF-8 like äÄ∏ŒŒ";
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(usAscii.getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeUtf8(buf2, usAscii);
 
         assertEquals(buf, buf2);
@@ -296,10 +331,10 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUtf8Composite() {
         String utf8 = "Some UTF-8 like äÄ∏ŒŒ";
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(utf8.getBytes(CharsetUtil.UTF_8));
         ByteBuf buf2 = Unpooled.compositeBuffer().addComponent(
-                Unpooled.buffer(8)).addComponent(Unpooled.buffer(24));
+                buffer(8)).addComponent(buffer(24));
         // write some byte so we start writing with an offset.
         buf2.writeByte(1);
         ByteBufUtil.writeUtf8(buf2, utf8);
@@ -314,10 +349,10 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUtf8CompositeWrapped() {
         String utf8 = "Some UTF-8 like äÄ∏ŒŒ";
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(utf8.getBytes(CharsetUtil.UTF_8));
         ByteBuf buf2 = new WrappedCompositeByteBuf(Unpooled.compositeBuffer().addComponent(
-                Unpooled.buffer(8)).addComponent(Unpooled.buffer(24)));
+                buffer(8)).addComponent(buffer(24)));
         // write some byte so we start writing with an offset.
         buf2.writeByte(1);
         ByteBufUtil.writeUtf8(buf2, utf8);
@@ -338,9 +373,9 @@ public class ByteBufUtilTest {
                                 .append('\uDC00')
                                 .append('b')
                                 .toString();
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(surrogateString.getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeUtf8(buf2, surrogateString);
 
         assertEquals(buf, buf2);
@@ -357,9 +392,9 @@ public class ByteBufUtilTest {
                                 .append('\uDC00')
                                 .append('b')
                                 .toString();
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(surrogateString.getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeUtf8(buf2, surrogateString);
 
         assertEquals(buf, buf2);
@@ -376,9 +411,9 @@ public class ByteBufUtilTest {
                                 .append('\uD800')
                                 .append('b')
                                 .toString();
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(surrogateString.getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeUtf8(buf2, surrogateString);
 
         assertEquals(buf, buf2);
@@ -396,9 +431,9 @@ public class ByteBufUtilTest {
                                 .append('\uD800')
                                 .append('b')
                                 .toString();
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(surrogateString.getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeUtf8(buf2, surrogateString);
 
         assertEquals(buf, buf2);
@@ -416,9 +451,9 @@ public class ByteBufUtilTest {
                                 .append('\uD800')
                                 .append('b')
                                 .toString();
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(surrogateString.getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeUtf8(buf2, surrogateString);
 
         assertEquals(buf, buf2);
@@ -435,9 +470,9 @@ public class ByteBufUtilTest {
                                 .append('\uDC00')
                                 .append('b')
                                 .toString();
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(surrogateString.getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeUtf8(buf2, surrogateString);
 
         assertEquals(buf, buf2);
@@ -452,9 +487,9 @@ public class ByteBufUtilTest {
         String surrogateString = new StringBuilder(2)
                                 .append('\uD800')
                                 .toString();
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(surrogateString.getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeUtf8(buf2, surrogateString);
 
         assertEquals(buf, buf2);
@@ -469,9 +504,9 @@ public class ByteBufUtilTest {
         String surrogateString = new StringBuilder(2)
                                 .append('\uDC00')
                                 .toString();
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(surrogateString.getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeUtf8(buf2, surrogateString);
 
         assertEquals(buf, buf2);
@@ -484,9 +519,10 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUsAsciiString() {
         AsciiString usAscii = new AsciiString("NettyRocks");
-        ByteBuf buf = Unpooled.buffer(16);
+        int expectedCapacity = usAscii.length();
+        ByteBuf buf = buffer(expectedCapacity);
         buf.writeBytes(usAscii.toString().getBytes(CharsetUtil.US_ASCII));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(expectedCapacity);
         ByteBufUtil.writeAscii(buf2, usAscii);
 
         assertEquals(buf, buf2);
@@ -498,17 +534,17 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUtf8Wrapped() {
         String usAscii = "Some UTF-8 like äÄ∏ŒŒ";
-        ByteBuf buf = unreleasableBuffer(Unpooled.buffer(16));
+        ByteBuf buf = unreleasableBuffer(buffer(16));
         assertWrapped(buf);
         buf.writeBytes(usAscii.getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = unreleasableBuffer(Unpooled.buffer(16));
+        ByteBuf buf2 = unreleasableBuffer(buffer(16));
         assertWrapped(buf2);
         ByteBufUtil.writeUtf8(buf2, usAscii);
 
         assertEquals(buf, buf2);
 
-        buf.release();
+        buf.unwrap().release();
-        buf2.release();
+        buf2.unwrap().release();
     }
 
     private static void assertWrapped(ByteBuf buf) {
@@ -518,9 +554,9 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUtf8Subsequence() {
         String usAscii = "Some UTF-8 like äÄ∏ŒŒ";
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(usAscii.substring(5, 18).getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeUtf8(buf2, usAscii, 5, 18);
 
         assertEquals(buf, buf2);
@@ -532,9 +568,9 @@ public class ByteBufUtilTest {
     @Test
     public void testWriteUtf8SubsequenceSplitSurrogate() {
         String usAscii = "\uD800\uDC00"; // surrogate pair: one code point, two chars
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(usAscii.substring(0, 1).getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         ByteBufUtil.writeUtf8(buf2, usAscii, 0, 1);
 
         assertEquals(buf, buf2);
@@ -546,9 +582,9 @@ public class ByteBufUtilTest {
     @Test
     public void testReserveAndWriteUtf8Subsequence() {
         String usAscii = "Some UTF-8 like äÄ∏ŒŒ";
-        ByteBuf buf = Unpooled.buffer(16);
+        ByteBuf buf = buffer(16);
         buf.writeBytes(usAscii.substring(5, 18).getBytes(CharsetUtil.UTF_8));
-        ByteBuf buf2 = Unpooled.buffer(16);
+        ByteBuf buf2 = buffer(16);
         int count = ByteBufUtil.reserveAndWriteUtf8(buf2, usAscii, 5, 18, 16);
 
         assertEquals(buf, buf2);
@@ -575,7 +611,7 @@ public class ByteBufUtilTest {
 
     private void testInvalidSubsequences(TestMethod method) {
         for (int [] range : INVALID_RANGES) {
-            ByteBuf buf = Unpooled.buffer(16);
+            ByteBuf buf = buffer(16);
             try {
                 method.invoke(buf, "Some UTF-8 like äÄ∏ŒŒ", range[0], range[1]);
                 fail("Did not throw IndexOutOfBoundsException for range (" + range[0] + ", " + range[1] + ")");
@@ -641,8 +677,8 @@ public class ByteBufUtilTest {
         CompositeByteBuf buffer = Unpooled.compositeBuffer();
         try {
             byte[] bytes = "1234".getBytes(CharsetUtil.UTF_8);
-            buffer.addComponent(Unpooled.buffer(bytes.length).writeBytes(bytes));
+            buffer.addComponent(buffer(bytes.length).writeBytes(bytes));
-            buffer.addComponent(Unpooled.buffer(bytes.length).writeBytes(bytes));
+            buffer.addComponent(buffer(bytes.length).writeBytes(bytes));
             assertEquals("1234", buffer.toString(bytes.length, bytes.length, CharsetUtil.UTF_8));
         } finally {
             buffer.release();
@@ -703,7 +739,7 @@ public class ByteBufUtilTest {
 
     @Test
     public void testIsTextWithInvalidIndexAndLength() {
-        ByteBuf buffer = Unpooled.buffer();
+        ByteBuf buffer = buffer(4);
         try {
             buffer.writeBytes(new byte[4]);
             int[][] validIndexLengthPairs = {
@@ -760,8 +796,8 @@ public class ByteBufUtilTest {
         checkUtf8Bytes(s);
     }
 
-    private static void checkUtf8Bytes(final CharSequence charSequence) {
+    private void checkUtf8Bytes(final CharSequence charSequence) {
-        final ByteBuf buf = Unpooled.buffer(ByteBufUtil.utf8MaxBytes(charSequence));
+        final ByteBuf buf = buffer(ByteBufUtil.utf8MaxBytes(charSequence));
         try {
             final int writtenBytes = ByteBufUtil.writeUtf8(buf, charSequence);
             final int utf8Bytes = ByteBufUtil.utf8Bytes(charSequence);
@@ -771,8 +807,8 @@ public class ByteBufUtilTest {
         }
     }
 
-    private static void assertIsText(byte[] bytes, boolean expected, Charset charset) {
+    private void assertIsText(byte[] bytes, boolean expected, Charset charset) {
-        ByteBuf buffer = Unpooled.buffer();
+        ByteBuf buffer = buffer(bytes.length);
         try {
             buffer.writeBytes(bytes);
             assertEquals(expected, ByteBufUtil.isText(buffer, charset));
@@ -783,6 +819,7 @@ public class ByteBufUtilTest {
 
     @Test
     public void testIsTextMultiThreaded() throws Throwable {
+        assumeThat(bufferType, is(BufferType.HEAP_UNPOOLED));
         final ByteBuf buffer = Unpooled.copiedBuffer("Hello, World!", CharsetUtil.ISO_8859_1);
 
         try {
@@ -822,21 +859,9 @@ public class ByteBufUtilTest {
     }
 
     @Test
-    public void testGetBytesHeap() {
+    public void testGetBytes() {
-        final ByteBuf buf = Unpooled.buffer(4);
+        final ByteBuf buf = buffer(4);
         try {
-            assumeTrue(buf.hasArray());
-            checkGetBytes(buf);
-        } finally {
-            buf.release();
-        }
-    }
-
-    @Test
-    public void testGetBytesDirect() {
-        final ByteBuf buf = Unpooled.directBuffer(4);
-        try {
-            assumeFalse(buf.hasArray());
             checkGetBytes(buf);
         } finally {
             buf.release();
@@ -845,16 +870,17 @@ public class ByteBufUtilTest {
 
     @Test
     public void testGetBytesHeapWithNonZeroArrayOffset() {
-        final ByteBuf buf = Unpooled.buffer(5);
+        assumeThat(bufferType, is(BufferType.HEAP_UNPOOLED));
+        final ByteBuf buf = buffer(5);
         try {
             buf.setByte(0, 0x05);
 
             final ByteBuf slice = buf.slice(1, 4);
             slice.writerIndex(0);
 
-            assumeTrue(slice.hasArray());
+            assertTrue(slice.hasArray());
-            assumeThat(slice.arrayOffset(), is(1));
+            assertThat(slice.arrayOffset(), is(1));
-            assumeThat(slice.array().length, is(buf.capacity()));
+            assertThat(slice.array().length, is(buf.capacity()));
 
             checkGetBytes(slice);
         } finally {
@@ -864,16 +890,17 @@ public class ByteBufUtilTest {
 
     @Test
     public void testGetBytesHeapWithArrayLengthGreaterThanCapacity() {
-        final ByteBuf buf = Unpooled.buffer(5);
+        assumeThat(bufferType, is(BufferType.HEAP_UNPOOLED));
+        final ByteBuf buf = buffer(5);
         try {
             buf.setByte(4, 0x05);
 
             final ByteBuf slice = buf.slice(0, 4);
             slice.writerIndex(0);
 
-            assumeTrue(slice.hasArray());
+            assertTrue(slice.hasArray());
-            assumeThat(slice.arrayOffset(), is(0));
+            assertThat(slice.arrayOffset(), is(0));
-            assumeThat(slice.array().length, greaterThan(slice.capacity()));
+            assertThat(slice.array().length, greaterThan(slice.capacity()));
 
             checkGetBytes(slice);
         } finally {

common/src/main/java/io/netty/util/internal/PlatformDependent.java

@@ -273,6 +273,10 @@ public final class PlatformDependent {
         LINUX_OS_CLASSIFIERS = Collections.unmodifiableSet(availableClassifiers);
     }
 
+    public static long byteArrayBaseOffset() {
+        return BYTE_ARRAY_BASE_OFFSET;
+    }
+
     public static boolean hasDirectBufferNoCleanerConstructor() {
         return PlatformDependent0.hasDirectBufferNoCleanerConstructor();
     }
@@ -659,6 +663,10 @@ public final class PlatformDependent {
         PlatformDependent0.putByte(data, index, value);
     }
 
+    public static void putByte(Object data, long offset, byte value) {
+        PlatformDependent0.putByte(data, offset, value);
+    }
+
     public static void putShort(byte[] data, int index, short value) {
         PlatformDependent0.putShort(data, index, value);
     }
@@ -687,6 +695,11 @@ public final class PlatformDependent {
         PlatformDependent0.copyMemory(src, BYTE_ARRAY_BASE_OFFSET + srcIndex, null, dstAddr, length);
     }
 
+    public static void copyMemory(byte[] src, int srcIndex, byte[] dst, int dstIndex, long length) {
+        PlatformDependent0.copyMemory(src, BYTE_ARRAY_BASE_OFFSET + srcIndex,
+                                      dst, BYTE_ARRAY_BASE_OFFSET + dstIndex, length);
+    }
+
     public static void copyMemory(long srcAddr, byte[] dst, int dstIndex, long length) {
         PlatformDependent0.copyMemory(null, srcAddr, dst, BYTE_ARRAY_BASE_OFFSET + dstIndex, length);
     }

common/src/main/java/io/netty/util/internal/PlatformDependent0.java

@@ -590,6 +590,10 @@ final class PlatformDependent0 {
         UNSAFE.putByte(data, BYTE_ARRAY_BASE_OFFSET + index, value);
     }
 
+    static void putByte(Object data, long offset, byte value) {
+        UNSAFE.putByte(data, offset, value);
+    }
+
     static void putShort(byte[] data, int index, short value) {
         UNSAFE.putShort(data, BYTE_ARRAY_BASE_OFFSET + index, value);
     }

microbench/src/main/java/io/netty/microbench/buffer/ByteBufUtilBenchmark.java

@@ -22,6 +22,7 @@ import io.netty.microbench.util.AbstractMicrobenchmark;
 import io.netty.util.CharsetUtil;
 import org.openjdk.jmh.annotations.Benchmark;
 import org.openjdk.jmh.annotations.Measurement;
+import org.openjdk.jmh.annotations.Param;
 import org.openjdk.jmh.annotations.Scope;
 import org.openjdk.jmh.annotations.Setup;
 import org.openjdk.jmh.annotations.State;
@@ -34,6 +35,11 @@ import org.openjdk.jmh.annotations.Warmup;
 @Measurement(iterations = 10)
 public class
     ByteBufUtilBenchmark extends AbstractMicrobenchmark {
+
+    @Param({ "true", "false" })
+    private boolean direct;
+    @Param({ "8", "16", "64", "128" })
+    private int length;
     private ByteBuf buffer;
     private ByteBuf wrapped;
     private ByteBuf asciiBuffer;
@@ -48,18 +54,19 @@ public class
     @Setup
     public void setup() {
         // Use buffer sizes that will also allow to write UTF-8 without grow the buffer
-        buffer = Unpooled.directBuffer(512);
+        final int maxBytes = ByteBufUtil.utf8MaxBytes(length);
-        wrapped = Unpooled.unreleasableBuffer(Unpooled.directBuffer(512));
+        buffer = direct? Unpooled.directBuffer(maxBytes) : Unpooled.buffer(maxBytes);
-        asciiSequence = new StringBuilder(128);
+        wrapped = Unpooled.unreleasableBuffer(direct? Unpooled.directBuffer(maxBytes) : Unpooled.buffer(maxBytes));
-        for (int i = 0; i < 128; i++) {
+        asciiSequence = new StringBuilder(length);
+        for (int i = 0; i < length; i++) {
             asciiSequence.append('a');
         }
         ascii = asciiSequence.toString();
 
         // Generate some mixed UTF-8 String for benchmark
-        utf8Sequence = new StringBuilder(128);
+        utf8Sequence = new StringBuilder(length);
         char[] chars = "Some UTF-8 like äÄ∏ŒŒ".toCharArray();
-        for (int i = 0; i < 128; i++) {
+        for (int i = 0; i < length; i++) {
             utf8Sequence.append(chars[i % chars.length]);
         }
         utf8 = utf8Sequence.toString();

microbench/src/main/java/io/netty/microbench/buffer/Utf8EncodingBenchmark.java

@@ -0,0 +1,277 @@
+/*
+ * Copyright 2020 The Netty Project
+ *
+ * The Netty Project licenses this file to you under the Apache License,
+ * version 2.0 (the "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at:
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ * License for the specific language governing permissions and limitations
+ * under the License.
+ */
+package io.netty.microbench.buffer;
+
+import io.netty.buffer.ByteBuf;
+import io.netty.buffer.ByteBufUtil;
+import io.netty.buffer.Unpooled;
+import io.netty.microbench.util.AbstractMicrobenchmark;
+import io.netty.util.AsciiString;
+import org.openjdk.jmh.annotations.Benchmark;
+import org.openjdk.jmh.annotations.CompilerControl;
+import org.openjdk.jmh.annotations.CompilerControl.Mode;
+import org.openjdk.jmh.annotations.Fork;
+import org.openjdk.jmh.annotations.Param;
+import org.openjdk.jmh.annotations.Setup;
+
+import java.io.BufferedReader;
+import java.io.IOException;
+import java.io.InputStream;
+import java.io.InputStreamReader;
+import java.io.Reader;
+import java.io.UnsupportedEncodingException;
+import java.util.ArrayList;
+import java.util.List;
+
+@Fork(value = 2, jvmArgsAppend = "-XX:MaxInlineLevel=9")
+public class Utf8EncodingBenchmark extends AbstractMicrobenchmark {
+    private static class AnotherCharSequence implements CharSequence {
+        private final char[] chars;
+
+        AnotherCharSequence(String chars) {
+            this.chars = new char[chars.length()];
+            chars.getChars(0, chars.length(), this.chars, 0);
+        }
+
+        @Override
+        public int length() {
+            return chars.length;
+        }
+
+        @Override
+        public char charAt(int i) {
+            return chars[i];
+        }
+
+        @Override
+        public CharSequence subSequence(int start, int end) {
+            throw new UnsupportedOperationException();
+        }
+
+        @Override
+        public String toString() {
+            throw new UnsupportedOperationException();
+        }
+    }
+
+    // experiment test input
+    private String[] strings;
+    private StringBuilder[] stringBuilders;
+    private AnotherCharSequence[] anotherCharSequences;
+    private AsciiString[] asciiStrings;
+    @Param({ "false", "true" })
+    private boolean direct;
+    private ByteBuf buffer;
+    @Param({ "false", "true" })
+    private boolean noUnsafe;
+    private int dataSetLength;
+
+    @Setup
+    public void init() {
+        System.setProperty("io.netty.noUnsafe", Boolean.valueOf(noUnsafe).toString());
+        InputStream testTextStream = null;
+        InputStreamReader inStreamReader = null;
+        BufferedReader buffReader = null;
+        int maxExpectedSize = 0;
+        List<String> strings = new ArrayList<String>();
+        List<StringBuilder> stringBuilders = new ArrayList<StringBuilder>();
+        List<AnotherCharSequence> anotherCharSequenceList = new ArrayList<AnotherCharSequence>();
+        List<AsciiString> asciiStrings = new ArrayList<AsciiString>();
+        try {
+            testTextStream = getClass().getResourceAsStream("/Utf8Samples.txt");
+            inStreamReader = new InputStreamReader(testTextStream, "UTF-8");
+            buffReader = new BufferedReader(inStreamReader);
+            String line;
+            while ((line = buffReader.readLine()) != null) {
+                strings.add(line);
+                stringBuilders.add(new StringBuilder(line));
+                anotherCharSequenceList.add(new AnotherCharSequence(line));
+                asciiStrings.add(new AsciiString(line));
+                maxExpectedSize = Math.max(maxExpectedSize, ByteBufUtil.utf8MaxBytes(line.length()));
+            }
+        } catch (Exception e) {
+            throw new RuntimeException(e);
+        } finally {
+            closeStream(testTextStream);
+            closeReader(inStreamReader);
+            closeReader(buffReader);
+        }
+        buffer = direct? Unpooled.directBuffer(maxExpectedSize, maxExpectedSize) :
+                Unpooled.buffer(maxExpectedSize, maxExpectedSize);
+        buffer.setByte(maxExpectedSize - 1, 0);
+        this.strings = strings.toArray(new String[strings.size()]);
+        this.stringBuilders = stringBuilders.toArray(new StringBuilder[stringBuilders.size()]);
+        this.anotherCharSequences =
+                anotherCharSequenceList.toArray(new AnotherCharSequence[anotherCharSequenceList.size()]);
+        this.asciiStrings = asciiStrings.toArray(new AsciiString[asciiStrings.size()]);
+        this.dataSetLength = this.strings.length;
+    }
+
+    private static void closeStream(InputStream inStream) {
+        if (inStream != null) {
+            try {
+                inStream.close();
+            } catch (IOException e) {
+                throw new RuntimeException(e);
+            }
+        }
+    }
+
+    private static void closeReader(Reader buffReader) {
+        if (buffReader != null) {
+            try {
+                buffReader.close();
+            } catch (IOException e) {
+                throw new RuntimeException(e);
+            }
+        }
+    }
+
+    @Benchmark
+    @CompilerControl(Mode.DONT_INLINE)
+    public int nestedByteBufUtilWriteUtf8String() {
+        int countBytes = 0;
+        for (String string : strings) {
+            countBytes += nestedByteBufUtilWriteUtf8String1(string);
+        }
+        return countBytes;
+    }
+
+    private int nestedByteBufUtilWriteUtf8String1(String string) {
+        return nestedByteBufUtilWriteUtf8String2(string);
+    }
+
+    private int nestedByteBufUtilWriteUtf8String2(String string) {
+        return nestedByteBufUtilWriteUtf8String3(string);
+    }
+
+    private int nestedByteBufUtilWriteUtf8String3(String string) {
+        return nestedByteBufUtilWriteUtf8String4(string);
+    }
+
+    private int nestedByteBufUtilWriteUtf8String4(String string) {
+        return nestedByteBufUtilWriteUtf8String5(string);
+    }
+
+    private int nestedByteBufUtilWriteUtf8String5(String string) {
+        return nestedByteBufUtilWriteUtf8String6(string);
+    }
+
+    private int nestedByteBufUtilWriteUtf8String6(String string) {
+        // this calls should be inlined but...what happen to the subsequent calls > MaxInlineLevel?
+        buffer.resetWriterIndex();
+        ByteBufUtil.writeUtf8(buffer, string, 0, string.length());
+        return buffer.writerIndex();
+    }
+
+    @Benchmark
+    @CompilerControl(Mode.DONT_INLINE)
+    public int byteBufUtilWriteUtf8String() {
+        int countBytes = 0;
+        for (String string : strings) {
+            buffer.resetWriterIndex();
+            ByteBufUtil.writeUtf8(buffer, string, 0, string.length());
+            countBytes += buffer.writerIndex();
+        }
+        return countBytes;
+    }
+
+    @Benchmark
+    @CompilerControl(Mode.DONT_INLINE)
+    public int byteBufUtilWriteUtf8Bimorphic() {
+        int countBytes = 0;
+        for (int i = 0, size = dataSetLength; i < size; i++) {
+            final StringBuilder stringBuilder = stringBuilders[i];
+            final String string = strings[i];
+            buffer.resetWriterIndex();
+            ByteBufUtil.writeUtf8(buffer, stringBuilder, 0, stringBuilder.length());
+            countBytes += buffer.writerIndex();
+            buffer.resetWriterIndex();
+            ByteBufUtil.writeUtf8(buffer, string, 0, string.length());
+            countBytes += buffer.writerIndex();
+        }
+        return countBytes;
+    }
+
+    @Benchmark
+    @CompilerControl(Mode.DONT_INLINE)
+    public int byteBufUtilWriteUtf8Megamorphic() {
+        int countBytes = 0;
+        for (int i = 0, size = dataSetLength; i < size; i++) {
+            final StringBuilder stringBuilder = stringBuilders[i];
+            final String string = strings[i];
+            final AnotherCharSequence anotherCharSequence = anotherCharSequences[i];
+            buffer.resetWriterIndex();
+            ByteBufUtil.writeUtf8(buffer, stringBuilder, 0, stringBuilder.length());
+            countBytes += buffer.writerIndex();
+            buffer.resetWriterIndex();
+            ByteBufUtil.writeUtf8(buffer, string, 0, string.length());
+            countBytes += buffer.writerIndex();
+            buffer.resetWriterIndex();
+            ByteBufUtil.writeUtf8(buffer, anotherCharSequence, 0, anotherCharSequence.length());
+            countBytes += buffer.writerIndex();
+        }
+        return countBytes;
+    }
+
+    @Benchmark
+    @CompilerControl(Mode.DONT_INLINE)
+    public int byteBufUtilWriteUtf8CommonCharSequences() {
+        int countBytes = 0;
+        for (int i = 0, size = dataSetLength; i < size; i++) {
+            final StringBuilder stringBuilder = stringBuilders[i];
+            final String string = strings[i];
+            final AsciiString asciiString = asciiStrings[i];
+            buffer.resetWriterIndex();
+            ByteBufUtil.writeUtf8(buffer, stringBuilder, 0, stringBuilder.length());
+            countBytes += buffer.writerIndex();
+            buffer.resetWriterIndex();
+            ByteBufUtil.writeUtf8(buffer, string, 0, string.length());
+            countBytes += buffer.writerIndex();
+            buffer.resetWriterIndex();
+            ByteBufUtil.writeUtf8(buffer, asciiString, 0, asciiString.length());
+            countBytes += buffer.writerIndex();
+        }
+        return countBytes;
+    }
+
+    @Benchmark
+    @CompilerControl(Mode.DONT_INLINE)
+    public int byteBufUtilWriteUtf8AsciiString() {
+        int countBytes = 0;
+        for (int i = 0, size = dataSetLength; i < size; i++) {
+            final AsciiString asciiString = asciiStrings[i];
+            buffer.resetWriterIndex();
+            ByteBufUtil.writeUtf8(buffer, asciiString, 0, asciiString.length());
+            countBytes += buffer.writerIndex();
+        }
+        return countBytes;
+    }
+
+    @Benchmark
+    @CompilerControl(Mode.DONT_INLINE)
+    public int writeGetBytes() throws UnsupportedEncodingException {
+        int countBytes = 0;
+        for (String string : strings) {
+            buffer.resetWriterIndex();
+            final byte[] bytes = string.getBytes("UTF-8");
+            buffer.writeBytes(bytes);
+            countBytes += buffer.writerIndex();
+        }
+        return countBytes;
+    }
+
+}