@@ -123,89 +123,85 @@
 	var ctx context
 	ctx.input = make([]byte, 0, 8)
 
 	return decompressor(func(output []byte) (int, error) {
 		var (
-			err                  error
-			flags                byte
-			readCount, available int
+			err                      error
+			flags                    byte
+			rc, available, predicted int
 		)
 
 		// Sanity check for space to read into
 		if len(output) == 0 {
 			return 0, nil
 		}
 
 		// Check whether we have leftover data in the buffer
 		if len(ctx.input) > 0 {
-			readCount = copy(output, ctx.input)
+			rc = copy(output, ctx.input)
 
 			// Check whether we still have leftover data in the buffer :)
-			if readCount < len(ctx.input) {
-				ctx.input = ctx.input[:copy(ctx.input, ctx.input[readCount:])]
+			if rc < len(ctx.input) {
+				ctx.input = ctx.input[:copy(ctx.input, ctx.input[rc:])]
 			}
-			return readCount, nil
+			return rc, nil
 		}
 
 		// Read the next prediction header
-		readCount, err = reader.Read(ctx.input[:1])
+		rc, err = reader.Read(ctx.input[:1])
 		// Fail on error unless it is EOF
 		if err != nil && err != io.EOF {
 			return 0, err
-		} else if readCount == 0 {
+		} else if rc == 0 {
 			return 0, err
 		}
 
 		// Extend the buffer, copy the prediction header
 		//  and calculate the number of subsequent bytes to read
 		ctx.input = ctx.input[:8]
 		flags = ctx.input[0]
-		available = 8 - int(bits.Hamming(flags))
+		predicted = int(bits.Hamming(flags))
+		available = 8 - predicted
 
-		// Read the non-predicted bytes according to header.
-		readCount, err = reader.Read(ctx.input[:available])
+		// Read the non-predicted bytes and place them in the end of the buffer
+		rc, err = reader.Read(ctx.input[predicted:])
 	retryData:
-		if readCount < int(available) && err == nil {
+		if rc < int(available) && err == nil {
 			// Retry the read if we have fewer bytes than what the prediction header indicates
 			var rc int
-			rc, err = reader.Read(ctx.input[readCount:available])
-			readCount += rc
+			rc, err = reader.Read(ctx.input[predicted+rc:])
+			rc += rc
 			goto retryData
 		} // Continue on any error, try to decompress and return it along the result
 
-		// Spread the read bytes right to left to avoid overlapping
-		for i, a := 7, available-1; i >= 0; i-- {
-			if ((flags >> uint(i)) & 1) == 0 {
-				ctx.input[i] = ctx.input[a]
-				a--
-			}
-		}
-
-		// Walk the buffer, fill in the predicted blanks and update the guess table
-		for i := uint(0); i < 8; i++ {
+		// Walk the buffer, filling in the predicted blanks,
+		// relocating read bytes and and updating the guess table
+		for i, a := uint(0), predicted; i < 8; i++ {
 			if (flags & (1 << i)) > 0 {
 				// Guess succeeded, fill in from the table
 				ctx.input[i] = ctx.table[ctx.hash]
-				readCount++
+				rc++
 			} else {
+				// Relocate a read byte
+				ctx.input[i], a = ctx.input[a], a+1
 				// Guess failed, update the table
 				ctx.table[ctx.hash] = ctx.input[i]
 			}
 			// Update the hash
 			ctx.hash = (ctx.hash << 4) ^ uint16(ctx.input[i])
 		}
 
-		// readCount now contains the precise amount of populated data
-		ctx.input = ctx.input[:readCount]
+		// rc now contains the precise amount of populated data
+		ctx.input = ctx.input[:rc]
 		available = copy(output, ctx.input)
 
 		// Check for remaining bytes that dont fit in the output buffer
-		if available < readCount {
+		if available < rc {
 			ctx.input = ctx.input[:copy(ctx.input, ctx.input[available:])]
 		} else {
 			// Clear the buffer
 			ctx.input = ctx.input[:0]
 		}
 
 		return available, err
 	})
 }