diff --git a/vendor/github.com/klauspost/compress/huff0/bitreader.go b/vendor/github.com/klauspost/compress/huff0/bitreader.go
index a4979e8..bfc7a52 100644
--- a/vendor/github.com/klauspost/compress/huff0/bitreader.go
+++ b/vendor/github.com/klauspost/compress/huff0/bitreader.go
@@ -6,120 +6,16 @@
 package huff0
 
 import (
-	"encoding/binary"
 	"errors"
+	"fmt"
 	"io"
+
+	"github.com/klauspost/compress/internal/le"
 )
 
 // bitReader reads a bitstream in reverse.
 // The last set bit indicates the start of the stream and is used
 // for aligning the input.
-type bitReader struct {
-	in       []byte
-	off      uint // next byte to read is at in[off - 1]
-	value    uint64
-	bitsRead uint8
-}
-
-// init initializes and resets the bit reader.
-func (b *bitReader) init(in []byte) error {
-	if len(in) < 1 {
-		return errors.New("corrupt stream: too short")
-	}
-	b.in = in
-	b.off = uint(len(in))
-	// The highest bit of the last byte indicates where to start
-	v := in[len(in)-1]
-	if v == 0 {
-		return errors.New("corrupt stream, did not find end of stream")
-	}
-	b.bitsRead = 64
-	b.value = 0
-	if len(in) >= 8 {
-		b.fillFastStart()
-	} else {
-		b.fill()
-		b.fill()
-	}
-	b.bitsRead += 8 - uint8(highBit32(uint32(v)))
-	return nil
-}
-
-// peekBitsFast requires that at least one bit is requested every time.
-// There are no checks if the buffer is filled.
-func (b *bitReader) peekBitsFast(n uint8) uint16 {
-	const regMask = 64 - 1
-	v := uint16((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask))
-	return v
-}
-
-// fillFast() will make sure at least 32 bits are available.
-// There must be at least 4 bytes available.
-func (b *bitReader) fillFast() {
-	if b.bitsRead < 32 {
-		return
-	}
-
-	// 2 bounds checks.
-	v := b.in[b.off-4 : b.off]
-	v = v[:4]
-	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-	b.value = (b.value << 32) | uint64(low)
-	b.bitsRead -= 32
-	b.off -= 4
-}
-
-func (b *bitReader) advance(n uint8) {
-	b.bitsRead += n
-}
-
-// fillFastStart() assumes the bitreader is empty and there is at least 8 bytes to read.
-func (b *bitReader) fillFastStart() {
-	// Do single re-slice to avoid bounds checks.
-	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
-	b.bitsRead = 0
-	b.off -= 8
-}
-
-// fill() will make sure at least 32 bits are available.
-func (b *bitReader) fill() {
-	if b.bitsRead < 32 {
-		return
-	}
-	if b.off > 4 {
-		v := b.in[b.off-4:]
-		v = v[:4]
-		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-		b.value = (b.value << 32) | uint64(low)
-		b.bitsRead -= 32
-		b.off -= 4
-		return
-	}
-	for b.off > 0 {
-		b.value = (b.value << 8) | uint64(b.in[b.off-1])
-		b.bitsRead -= 8
-		b.off--
-	}
-}
-
-// finished returns true if all bits have been read from the bit stream.
-func (b *bitReader) finished() bool {
-	return b.off == 0 && b.bitsRead >= 64
-}
-
-// close the bitstream and returns an error if out-of-buffer reads occurred.
-func (b *bitReader) close() error {
-	// Release reference.
-	b.in = nil
-	if b.bitsRead > 64 {
-		return io.ErrUnexpectedEOF
-	}
-	return nil
-}
-
-// bitReader reads a bitstream in reverse.
-// The last set bit indicates the start of the stream and is used
-// for aligning the input.
 type bitReaderBytes struct {
 	in       []byte
 	off      uint // next byte to read is at in[off - 1]
@@ -151,7 +47,7 @@
 	return nil
 }
 
-// peekBitsFast requires that at least one bit is requested every time.
+// peekByteFast requires that at least one byte is requested every time.
 // There are no checks if the buffer is filled.
 func (b *bitReaderBytes) peekByteFast() uint8 {
 	got := uint8(b.value >> 56)
@@ -171,9 +67,7 @@
 	}
 
 	// 2 bounds checks.
-	v := b.in[b.off-4 : b.off]
-	v = v[:4]
-	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+	low := le.Load32(b.in, b.off-4)
 	b.value |= uint64(low) << (b.bitsRead - 32)
 	b.bitsRead -= 32
 	b.off -= 4
@@ -182,7 +76,7 @@
 // fillFastStart() assumes the bitReaderBytes is empty and there is at least 8 bytes to read.
 func (b *bitReaderBytes) fillFastStart() {
 	// Do single re-slice to avoid bounds checks.
-	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
+	b.value = le.Load64(b.in, b.off-8)
 	b.bitsRead = 0
 	b.off -= 8
 }
@@ -192,10 +86,8 @@
 	if b.bitsRead < 32 {
 		return
 	}
-	if b.off > 4 {
-		v := b.in[b.off-4:]
-		v = v[:4]
-		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+	if b.off >= 4 {
+		low := le.Load32(b.in, b.off-4)
 		b.value |= uint64(low) << (b.bitsRead - 32)
 		b.bitsRead -= 32
 		b.off -= 4
@@ -213,10 +105,17 @@
 	return b.off == 0 && b.bitsRead >= 64
 }
 
+func (b *bitReaderBytes) remaining() uint {
+	return b.off*8 + uint(64-b.bitsRead)
+}
+
 // close the bitstream and returns an error if out-of-buffer reads occurred.
 func (b *bitReaderBytes) close() error {
 	// Release reference.
 	b.in = nil
+	if b.remaining() > 0 {
+		return fmt.Errorf("corrupt input: %d bits remain on stream", b.remaining())
+	}
 	if b.bitsRead > 64 {
 		return io.ErrUnexpectedEOF
 	}
@@ -275,10 +174,7 @@
 		return
 	}
 
-	// 2 bounds checks.
-	v := b.in[b.off-4 : b.off]
-	v = v[:4]
-	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+	low := le.Load32(b.in, b.off-4)
 	b.value |= uint64(low) << ((b.bitsRead - 32) & 63)
 	b.bitsRead -= 32
 	b.off -= 4
@@ -286,8 +182,7 @@
 
 // fillFastStart() assumes the bitReaderShifted is empty and there is at least 8 bytes to read.
 func (b *bitReaderShifted) fillFastStart() {
-	// Do single re-slice to avoid bounds checks.
-	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
+	b.value = le.Load64(b.in, b.off-8)
 	b.bitsRead = 0
 	b.off -= 8
 }
@@ -298,9 +193,7 @@
 		return
 	}
 	if b.off > 4 {
-		v := b.in[b.off-4:]
-		v = v[:4]
-		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+		low := le.Load32(b.in, b.off-4)
 		b.value |= uint64(low) << ((b.bitsRead - 32) & 63)
 		b.bitsRead -= 32
 		b.off -= 4
@@ -313,15 +206,17 @@
 	}
 }
 
-// finished returns true if all bits have been read from the bit stream.
-func (b *bitReaderShifted) finished() bool {
-	return b.off == 0 && b.bitsRead >= 64
+func (b *bitReaderShifted) remaining() uint {
+	return b.off*8 + uint(64-b.bitsRead)
 }
 
 // close the bitstream and returns an error if out-of-buffer reads occurred.
 func (b *bitReaderShifted) close() error {
 	// Release reference.
 	b.in = nil
+	if b.remaining() > 0 {
+		return fmt.Errorf("corrupt input: %d bits remain on stream", b.remaining())
+	}
 	if b.bitsRead > 64 {
 		return io.ErrUnexpectedEOF
 	}
diff --git a/vendor/github.com/klauspost/compress/huff0/bitwriter.go b/vendor/github.com/klauspost/compress/huff0/bitwriter.go
index 6bce4e8..0ebc9aa 100644
--- a/vendor/github.com/klauspost/compress/huff0/bitwriter.go
+++ b/vendor/github.com/klauspost/compress/huff0/bitwriter.go
@@ -5,8 +5,6 @@
 
 package huff0
 
-import "fmt"
-
 // bitWriter will write bits.
 // First bit will be LSB of the first byte of output.
 type bitWriter struct {
@@ -15,22 +13,6 @@
 	out          []byte
 }
 
-// bitMask16 is bitmasks. Has extra to avoid bounds check.
-var bitMask16 = [32]uint16{
-	0, 1, 3, 7, 0xF, 0x1F,
-	0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF,
-	0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0xFFFF,
-	0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF,
-	0xFFFF, 0xFFFF} /* up to 16 bits */
-
-// addBits16NC will add up to 16 bits.
-// It will not check if there is space for them,
-// so the caller must ensure that it has flushed recently.
-func (b *bitWriter) addBits16NC(value uint16, bits uint8) {
-	b.bitContainer |= uint64(value&bitMask16[bits&31]) << (b.nBits & 63)
-	b.nBits += bits
-}
-
 // addBits16Clean will add up to 16 bits. value may not contain more set bits than indicated.
 // It will not check if there is space for them, so the caller must ensure that it has flushed recently.
 func (b *bitWriter) addBits16Clean(value uint16, bits uint8) {
@@ -70,102 +52,20 @@
 	b.nBits += encA.nBits + encB.nBits
 }
 
-// addBits16ZeroNC will add up to 16 bits.
+// encFourSymbols adds up to 32 bits from four symbols.
 // It will not check if there is space for them,
-// so the caller must ensure that it has flushed recently.
-// This is fastest if bits can be zero.
-func (b *bitWriter) addBits16ZeroNC(value uint16, bits uint8) {
-	if bits == 0 {
-		return
-	}
-	value <<= (16 - bits) & 15
-	value >>= (16 - bits) & 15
-	b.bitContainer |= uint64(value) << (b.nBits & 63)
-	b.nBits += bits
-}
-
-// flush will flush all pending full bytes.
-// There will be at least 56 bits available for writing when this has been called.
-// Using flush32 is faster, but leaves less space for writing.
-func (b *bitWriter) flush() {
-	v := b.nBits >> 3
-	switch v {
-	case 0:
-		return
-	case 1:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-		)
-		b.bitContainer >>= 1 << 3
-	case 2:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-		)
-		b.bitContainer >>= 2 << 3
-	case 3:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-		)
-		b.bitContainer >>= 3 << 3
-	case 4:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-		)
-		b.bitContainer >>= 4 << 3
-	case 5:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-		)
-		b.bitContainer >>= 5 << 3
-	case 6:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-		)
-		b.bitContainer >>= 6 << 3
-	case 7:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-		)
-		b.bitContainer >>= 7 << 3
-	case 8:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-			byte(b.bitContainer>>56),
-		)
-		b.bitContainer = 0
-		b.nBits = 0
-		return
-	default:
-		panic(fmt.Errorf("bits (%d) > 64", b.nBits))
-	}
-	b.nBits &= 7
+// so the caller must ensure that b has been flushed recently.
+func (b *bitWriter) encFourSymbols(encA, encB, encC, encD cTableEntry) {
+	bitsA := encA.nBits
+	bitsB := bitsA + encB.nBits
+	bitsC := bitsB + encC.nBits
+	bitsD := bitsC + encD.nBits
+	combined := uint64(encA.val) |
+		(uint64(encB.val) << (bitsA & 63)) |
+		(uint64(encC.val) << (bitsB & 63)) |
+		(uint64(encD.val) << (bitsC & 63))
+	b.bitContainer |= combined << (b.nBits & 63)
+	b.nBits += bitsD
 }
 
 // flush32 will flush out, so there are at least 32 bits available for writing.
@@ -194,17 +94,9 @@
 
 // close will write the alignment bit and write the final byte(s)
 // to the output.
-func (b *bitWriter) close() error {
+func (b *bitWriter) close() {
 	// End mark
 	b.addBits16Clean(1, 1)
 	// flush until next byte.
 	b.flushAlign()
-	return nil
-}
-
-// reset and continue writing by appending to out.
-func (b *bitWriter) reset(out []byte) {
-	b.bitContainer = 0
-	b.nBits = 0
-	b.out = out
 }
diff --git a/vendor/github.com/klauspost/compress/huff0/bytereader.go b/vendor/github.com/klauspost/compress/huff0/bytereader.go
deleted file mode 100644
index 50bcdf6..0000000
--- a/vendor/github.com/klauspost/compress/huff0/bytereader.go
+++ /dev/null
@@ -1,54 +0,0 @@
-// Copyright 2018 Klaus Post. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-// Based on work Copyright (c) 2013, Yann Collet, released under BSD License.
-
-package huff0
-
-// byteReader provides a byte reader that reads
-// little endian values from a byte stream.
-// The input stream is manually advanced.
-// The reader performs no bounds checks.
-type byteReader struct {
-	b   []byte
-	off int
-}
-
-// init will initialize the reader and set the input.
-func (b *byteReader) init(in []byte) {
-	b.b = in
-	b.off = 0
-}
-
-// advance the stream b n bytes.
-func (b *byteReader) advance(n uint) {
-	b.off += int(n)
-}
-
-// Int32 returns a little endian int32 starting at current offset.
-func (b byteReader) Int32() int32 {
-	v3 := int32(b.b[b.off+3])
-	v2 := int32(b.b[b.off+2])
-	v1 := int32(b.b[b.off+1])
-	v0 := int32(b.b[b.off])
-	return (v3 << 24) | (v2 << 16) | (v1 << 8) | v0
-}
-
-// Uint32 returns a little endian uint32 starting at current offset.
-func (b byteReader) Uint32() uint32 {
-	v3 := uint32(b.b[b.off+3])
-	v2 := uint32(b.b[b.off+2])
-	v1 := uint32(b.b[b.off+1])
-	v0 := uint32(b.b[b.off])
-	return (v3 << 24) | (v2 << 16) | (v1 << 8) | v0
-}
-
-// unread returns the unread portion of the input.
-func (b byteReader) unread() []byte {
-	return b.b[b.off:]
-}
-
-// remain will return the number of bytes remaining.
-func (b byteReader) remain() int {
-	return len(b.b) - b.off
-}
diff --git a/vendor/github.com/klauspost/compress/huff0/compress.go b/vendor/github.com/klauspost/compress/huff0/compress.go
index 0823c92..84aa3d1 100644
--- a/vendor/github.com/klauspost/compress/huff0/compress.go
+++ b/vendor/github.com/klauspost/compress/huff0/compress.go
@@ -2,6 +2,7 @@
 
 import (
 	"fmt"
+	"math"
 	"runtime"
 	"sync"
 )
@@ -161,11 +162,75 @@
 	return s.Out, false, nil
 }
 
-func (s *Scratch) compress1X(src []byte) ([]byte, error) {
-	return s.compress1xDo(s.Out, src)
+// EstimateSizes will estimate the data sizes
+func EstimateSizes(in []byte, s *Scratch) (tableSz, dataSz, reuseSz int, err error) {
+	s, err = s.prepare(in)
+	if err != nil {
+		return 0, 0, 0, err
+	}
+
+	// Create histogram, if none was provided.
+	tableSz, dataSz, reuseSz = -1, -1, -1
+	maxCount := s.maxCount
+	var canReuse = false
+	if maxCount == 0 {
+		maxCount, canReuse = s.countSimple(in)
+	} else {
+		canReuse = s.canUseTable(s.prevTable)
+	}
+
+	// We want the output size to be less than this:
+	wantSize := len(in)
+	if s.WantLogLess > 0 {
+		wantSize -= wantSize >> s.WantLogLess
+	}
+
+	// Reset for next run.
+	s.clearCount = true
+	s.maxCount = 0
+	if maxCount >= len(in) {
+		if maxCount > len(in) {
+			return 0, 0, 0, fmt.Errorf("maxCount (%d) > length (%d)", maxCount, len(in))
+		}
+		if len(in) == 1 {
+			return 0, 0, 0, ErrIncompressible
+		}
+		// One symbol, use RLE
+		return 0, 0, 0, ErrUseRLE
+	}
+	if maxCount == 1 || maxCount < (len(in)>>7) {
+		// Each symbol present maximum once or too well distributed.
+		return 0, 0, 0, ErrIncompressible
+	}
+
+	// Calculate new table.
+	err = s.buildCTable()
+	if err != nil {
+		return 0, 0, 0, err
+	}
+
+	if false && !s.canUseTable(s.cTable) {
+		panic("invalid table generated")
+	}
+
+	tableSz, err = s.cTable.estTableSize(s)
+	if err != nil {
+		return 0, 0, 0, err
+	}
+	if canReuse {
+		reuseSz = s.prevTable.estimateSize(s.count[:s.symbolLen])
+	}
+	dataSz = s.cTable.estimateSize(s.count[:s.symbolLen])
+
+	// Restore
+	return tableSz, dataSz, reuseSz, nil
 }
 
-func (s *Scratch) compress1xDo(dst, src []byte) ([]byte, error) {
+func (s *Scratch) compress1X(src []byte) ([]byte, error) {
+	return s.compress1xDo(s.Out, src), nil
+}
+
+func (s *Scratch) compress1xDo(dst, src []byte) []byte {
 	var bw = bitWriter{out: dst}
 
 	// N is length divisible by 4.
@@ -183,8 +248,7 @@
 			tmp := src[n : n+4]
 			// tmp should be len 4
 			bw.flush32()
-			bw.encTwoSymbols(cTable, tmp[3], tmp[2])
-			bw.encTwoSymbols(cTable, tmp[1], tmp[0])
+			bw.encFourSymbols(cTable[tmp[3]], cTable[tmp[2]], cTable[tmp[1]], cTable[tmp[0]])
 		}
 	} else {
 		for ; n >= 0; n -= 4 {
@@ -196,8 +260,8 @@
 			bw.encTwoSymbols(cTable, tmp[1], tmp[0])
 		}
 	}
-	err := bw.close()
-	return bw.out, err
+	bw.close()
+	return bw.out
 }
 
 var sixZeros [6]byte
@@ -219,11 +283,11 @@
 		}
 		src = src[len(toDo):]
 
-		var err error
 		idx := len(s.Out)
-		s.Out, err = s.compress1xDo(s.Out, toDo)
-		if err != nil {
-			return nil, err
+		s.Out = s.compress1xDo(s.Out, toDo)
+		if len(s.Out)-idx > math.MaxUint16 {
+			// We cannot store the size in the jump table
+			return nil, ErrIncompressible
 		}
 		// Write compressed length as little endian before block.
 		if i < 3 {
@@ -247,7 +311,6 @@
 
 	segmentSize := (len(src) + 3) / 4
 	var wg sync.WaitGroup
-	var errs [4]error
 	wg.Add(4)
 	for i := 0; i < 4; i++ {
 		toDo := src
@@ -258,16 +321,17 @@
 
 		// Separate goroutine for each block.
 		go func(i int) {
-			s.tmpOut[i], errs[i] = s.compress1xDo(s.tmpOut[i][:0], toDo)
+			s.tmpOut[i] = s.compress1xDo(s.tmpOut[i][:0], toDo)
 			wg.Done()
 		}(i)
 	}
 	wg.Wait()
 	for i := 0; i < 4; i++ {
-		if errs[i] != nil {
-			return nil, errs[i]
-		}
 		o := s.tmpOut[i]
+		if len(o) > math.MaxUint16 {
+			// We cannot store the size in the jump table
+			return nil, ErrIncompressible
+		}
 		// Write compressed length as little endian before block.
 		if i < 3 {
 			// Last length is not written.
@@ -286,35 +350,36 @@
 // Does not update s.clearCount.
 func (s *Scratch) countSimple(in []byte) (max int, reuse bool) {
 	reuse = true
+	_ = s.count // Assert that s != nil to speed up the following loop.
 	for _, v := range in {
 		s.count[v]++
 	}
 	m := uint32(0)
 	if len(s.prevTable) > 0 {
 		for i, v := range s.count[:] {
+			if v == 0 {
+				continue
+			}
 			if v > m {
 				m = v
 			}
-			if v > 0 {
-				s.symbolLen = uint16(i) + 1
-				if i >= len(s.prevTable) {
-					reuse = false
-				} else {
-					if s.prevTable[i].nBits == 0 {
-						reuse = false
-					}
-				}
+			s.symbolLen = uint16(i) + 1
+			if i >= len(s.prevTable) {
+				reuse = false
+			} else if s.prevTable[i].nBits == 0 {
+				reuse = false
 			}
 		}
 		return int(m), reuse
 	}
 	for i, v := range s.count[:] {
+		if v == 0 {
+			continue
+		}
 		if v > m {
 			m = v
 		}
-		if v > 0 {
-			s.symbolLen = uint16(i) + 1
-		}
+		s.symbolLen = uint16(i) + 1
 	}
 	return int(m), false
 }
@@ -331,6 +396,7 @@
 	return true
 }
 
+//lint:ignore U1000 used for debugging
 func (s *Scratch) validateTable(c cTable) bool {
 	if len(c) < int(s.symbolLen) {
 		return false
@@ -350,7 +416,7 @@
 
 // minTableLog provides the minimum logSize to safely represent a distribution.
 func (s *Scratch) minTableLog() uint8 {
-	minBitsSrc := highBit32(uint32(s.br.remain())) + 1
+	minBitsSrc := highBit32(uint32(s.srcLen)) + 1
 	minBitsSymbols := highBit32(uint32(s.symbolLen-1)) + 2
 	if minBitsSrc < minBitsSymbols {
 		return uint8(minBitsSrc)
@@ -362,7 +428,7 @@
 func (s *Scratch) optimalTableLog() {
 	tableLog := s.TableLog
 	minBits := s.minTableLog()
-	maxBitsSrc := uint8(highBit32(uint32(s.br.remain()-1))) - 1
+	maxBitsSrc := uint8(highBit32(uint32(s.srcLen-1))) - 1
 	if maxBitsSrc < tableLog {
 		// Accuracy can be reduced
 		tableLog = maxBitsSrc
@@ -410,34 +476,35 @@
 	// Different from reference implementation.
 	huffNode0 := s.nodes[0 : huffNodesLen+1]
 
-	for huffNode[nonNullRank].count == 0 {
+	for huffNode[nonNullRank].count() == 0 {
 		nonNullRank--
 	}
 
 	lowS := int16(nonNullRank)
 	nodeRoot := nodeNb + lowS - 1
 	lowN := nodeNb
-	huffNode[nodeNb].count = huffNode[lowS].count + huffNode[lowS-1].count
-	huffNode[lowS].parent, huffNode[lowS-1].parent = uint16(nodeNb), uint16(nodeNb)
+	huffNode[nodeNb].setCount(huffNode[lowS].count() + huffNode[lowS-1].count())
+	huffNode[lowS].setParent(nodeNb)
+	huffNode[lowS-1].setParent(nodeNb)
 	nodeNb++
 	lowS -= 2
 	for n := nodeNb; n <= nodeRoot; n++ {
-		huffNode[n].count = 1 << 30
+		huffNode[n].setCount(1 << 30)
 	}
 	// fake entry, strong barrier
-	huffNode0[0].count = 1 << 31
+	huffNode0[0].setCount(1 << 31)
 
 	// create parents
 	for nodeNb <= nodeRoot {
 		var n1, n2 int16
-		if huffNode0[lowS+1].count < huffNode0[lowN+1].count {
+		if huffNode0[lowS+1].count() < huffNode0[lowN+1].count() {
 			n1 = lowS
 			lowS--
 		} else {
 			n1 = lowN
 			lowN++
 		}
-		if huffNode0[lowS+1].count < huffNode0[lowN+1].count {
+		if huffNode0[lowS+1].count() < huffNode0[lowN+1].count() {
 			n2 = lowS
 			lowS--
 		} else {
@@ -445,18 +512,19 @@
 			lowN++
 		}
 
-		huffNode[nodeNb].count = huffNode0[n1+1].count + huffNode0[n2+1].count
-		huffNode0[n1+1].parent, huffNode0[n2+1].parent = uint16(nodeNb), uint16(nodeNb)
+		huffNode[nodeNb].setCount(huffNode0[n1+1].count() + huffNode0[n2+1].count())
+		huffNode0[n1+1].setParent(nodeNb)
+		huffNode0[n2+1].setParent(nodeNb)
 		nodeNb++
 	}
 
 	// distribute weights (unlimited tree height)
-	huffNode[nodeRoot].nbBits = 0
+	huffNode[nodeRoot].setNbBits(0)
 	for n := nodeRoot - 1; n >= startNode; n-- {
-		huffNode[n].nbBits = huffNode[huffNode[n].parent].nbBits + 1
+		huffNode[n].setNbBits(huffNode[huffNode[n].parent()].nbBits() + 1)
 	}
 	for n := uint16(0); n <= nonNullRank; n++ {
-		huffNode[n].nbBits = huffNode[huffNode[n].parent].nbBits + 1
+		huffNode[n].setNbBits(huffNode[huffNode[n].parent()].nbBits() + 1)
 	}
 	s.actualTableLog = s.setMaxHeight(int(nonNullRank))
 	maxNbBits := s.actualTableLog
@@ -468,7 +536,7 @@
 	var nbPerRank [tableLogMax + 1]uint16
 	var valPerRank [16]uint16
 	for _, v := range huffNode[:nonNullRank+1] {
-		nbPerRank[v.nbBits]++
+		nbPerRank[v.nbBits()]++
 	}
 	// determine stating value per rank
 	{
@@ -483,7 +551,7 @@
 
 	// push nbBits per symbol, symbol order
 	for _, v := range huffNode[:nonNullRank+1] {
-		s.cTable[v.symbol].nBits = v.nbBits
+		s.cTable[v.symbol()].nBits = v.nbBits()
 	}
 
 	// assign value within rank, symbol order
@@ -529,12 +597,12 @@
 		pos := rank[r].current
 		rank[r].current++
 		prev := nodes[(pos-1)&huffNodesMask]
-		for pos > rank[r].base && c > prev.count {
+		for pos > rank[r].base && c > prev.count() {
 			nodes[pos&huffNodesMask] = prev
 			pos--
 			prev = nodes[(pos-1)&huffNodesMask]
 		}
-		nodes[pos&huffNodesMask] = nodeElt{count: c, symbol: byte(n)}
+		nodes[pos&huffNodesMask] = makeNodeElt(c, byte(n))
 	}
 }
 
@@ -543,7 +611,7 @@
 	huffNode := s.nodes[1 : huffNodesLen+1]
 	//huffNode = huffNode[: huffNodesLen]
 
-	largestBits := huffNode[lastNonNull].nbBits
+	largestBits := huffNode[lastNonNull].nbBits()
 
 	// early exit : no elt > maxNbBits
 	if largestBits <= maxNbBits {
@@ -553,14 +621,14 @@
 	baseCost := int(1) << (largestBits - maxNbBits)
 	n := uint32(lastNonNull)
 
-	for huffNode[n].nbBits > maxNbBits {
-		totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits))
-		huffNode[n].nbBits = maxNbBits
+	for huffNode[n].nbBits() > maxNbBits {
+		totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits()))
+		huffNode[n].setNbBits(maxNbBits)
 		n--
 	}
 	// n stops at huffNode[n].nbBits <= maxNbBits
 
-	for huffNode[n].nbBits == maxNbBits {
+	for huffNode[n].nbBits() == maxNbBits {
 		n--
 	}
 	// n end at index of smallest symbol using < maxNbBits
@@ -581,10 +649,10 @@
 		{
 			currentNbBits := maxNbBits
 			for pos := int(n); pos >= 0; pos-- {
-				if huffNode[pos].nbBits >= currentNbBits {
+				if huffNode[pos].nbBits() >= currentNbBits {
 					continue
 				}
-				currentNbBits = huffNode[pos].nbBits // < maxNbBits
+				currentNbBits = huffNode[pos].nbBits() // < maxNbBits
 				rankLast[maxNbBits-currentNbBits] = uint32(pos)
 			}
 		}
@@ -601,8 +669,8 @@
 				if lowPos == noSymbol {
 					break
 				}
-				highTotal := huffNode[highPos].count
-				lowTotal := 2 * huffNode[lowPos].count
+				highTotal := huffNode[highPos].count()
+				lowTotal := 2 * huffNode[lowPos].count()
 				if highTotal <= lowTotal {
 					break
 				}
@@ -618,13 +686,14 @@
 				// this rank is no longer empty
 				rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease]
 			}
-			huffNode[rankLast[nBitsToDecrease]].nbBits++
+			huffNode[rankLast[nBitsToDecrease]].setNbBits(1 +
+				huffNode[rankLast[nBitsToDecrease]].nbBits())
 			if rankLast[nBitsToDecrease] == 0 {
 				/* special case, reached largest symbol */
 				rankLast[nBitsToDecrease] = noSymbol
 			} else {
 				rankLast[nBitsToDecrease]--
-				if huffNode[rankLast[nBitsToDecrease]].nbBits != maxNbBits-nBitsToDecrease {
+				if huffNode[rankLast[nBitsToDecrease]].nbBits() != maxNbBits-nBitsToDecrease {
 					rankLast[nBitsToDecrease] = noSymbol /* this rank is now empty */
 				}
 			}
@@ -632,15 +701,15 @@
 
 		for totalCost < 0 { /* Sometimes, cost correction overshoot */
 			if rankLast[1] == noSymbol { /* special case : no rank 1 symbol (using maxNbBits-1); let's create one from largest rank 0 (using maxNbBits) */
-				for huffNode[n].nbBits == maxNbBits {
+				for huffNode[n].nbBits() == maxNbBits {
 					n--
 				}
-				huffNode[n+1].nbBits--
+				huffNode[n+1].setNbBits(huffNode[n+1].nbBits() - 1)
 				rankLast[1] = n + 1
 				totalCost++
 				continue
 			}
-			huffNode[rankLast[1]+1].nbBits--
+			huffNode[rankLast[1]+1].setNbBits(huffNode[rankLast[1]+1].nbBits() - 1)
 			rankLast[1]++
 			totalCost++
 		}
@@ -648,9 +717,26 @@
 	return maxNbBits
 }
 
-type nodeElt struct {
-	count  uint32
-	parent uint16
-	symbol byte
-	nbBits uint8
+// A nodeElt is the fields
+//
+//	count  uint32
+//	parent uint16
+//	symbol byte
+//	nbBits uint8
+//
+// in some order, all squashed into an integer so that the compiler
+// always loads and stores entire nodeElts instead of separate fields.
+type nodeElt uint64
+
+func makeNodeElt(count uint32, symbol byte) nodeElt {
+	return nodeElt(count) | nodeElt(symbol)<<48
 }
+
+func (e *nodeElt) count() uint32  { return uint32(*e) }
+func (e *nodeElt) parent() uint16 { return uint16(*e >> 32) }
+func (e *nodeElt) symbol() byte   { return byte(*e >> 48) }
+func (e *nodeElt) nbBits() uint8  { return uint8(*e >> 56) }
+
+func (e *nodeElt) setCount(c uint32) { *e = (*e)&0xffffffff00000000 | nodeElt(c) }
+func (e *nodeElt) setParent(p int16) { *e = (*e)&0xffff0000ffffffff | nodeElt(uint16(p))<<32 }
+func (e *nodeElt) setNbBits(n uint8) { *e = (*e)&0x00ffffffffffffff | nodeElt(n)<<56 }
diff --git a/vendor/github.com/klauspost/compress/huff0/decompress.go b/vendor/github.com/klauspost/compress/huff0/decompress.go
index 41703bb..0f56b02 100644
--- a/vendor/github.com/klauspost/compress/huff0/decompress.go
+++ b/vendor/github.com/klauspost/compress/huff0/decompress.go
@@ -4,13 +4,13 @@
 	"errors"
 	"fmt"
 	"io"
+	"sync"
 
 	"github.com/klauspost/compress/fse"
 )
 
 type dTable struct {
 	single []dEntrySingle
-	double []dEntryDouble
 }
 
 // single-symbols decoding
@@ -18,13 +18,6 @@
 	entry uint16
 }
 
-// double-symbols decoding
-type dEntryDouble struct {
-	seq   uint16
-	nBits uint8
-	len   uint8
-}
-
 // Uses special code for all tables that are < 8 bits.
 const use8BitTables = true
 
@@ -34,7 +27,7 @@
 // If no Scratch is provided a new one is allocated.
 // The returned Scratch can be used for encoding or decoding input using this table.
 func ReadTable(in []byte, s *Scratch) (s2 *Scratch, remain []byte, err error) {
-	s, err = s.prepare(in)
+	s, err = s.prepare(nil)
 	if err != nil {
 		return s, nil, err
 	}
@@ -68,7 +61,7 @@
 		b, err := fse.Decompress(in[:iSize], s.fse)
 		s.fse.Out = nil
 		if err != nil {
-			return s, nil, err
+			return s, nil, fmt.Errorf("fse decompress returned: %w", err)
 		}
 		if len(b) > 255 {
 			return s, nil, errors.New("corrupt input: output table too large")
@@ -216,6 +209,7 @@
 	return &Decoder{
 		dt:             s.dt,
 		actualTableLog: s.actualTableLog,
+		bufs:           &s.decPool,
 	}
 }
 
@@ -223,103 +217,15 @@
 type Decoder struct {
 	dt             dTable
 	actualTableLog uint8
+	bufs           *sync.Pool
 }
 
-// Decompress1X will decompress a 1X encoded stream.
-// The cap of the output buffer will be the maximum decompressed size.
-// The length of the supplied input must match the end of a block exactly.
-func (d *Decoder) Decompress1X(dst, src []byte) ([]byte, error) {
-	if len(d.dt.single) == 0 {
-		return nil, errors.New("no table loaded")
+func (d *Decoder) buffer() *[4][256]byte {
+	buf, ok := d.bufs.Get().(*[4][256]byte)
+	if ok {
+		return buf
 	}
-	if use8BitTables && d.actualTableLog <= 8 {
-		return d.decompress1X8Bit(dst, src)
-	}
-	var br bitReaderShifted
-	err := br.init(src)
-	if err != nil {
-		return dst, err
-	}
-	maxDecodedSize := cap(dst)
-	dst = dst[:0]
-
-	// Avoid bounds check by always having full sized table.
-	const tlSize = 1 << tableLogMax
-	const tlMask = tlSize - 1
-	dt := d.dt.single[:tlSize]
-
-	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
-	var off uint8
-
-	for br.off >= 8 {
-		br.fillFast()
-		v := dt[br.peekBitsFast(d.actualTableLog)&tlMask]
-		br.advance(uint8(v.entry))
-		buf[off+0] = uint8(v.entry >> 8)
-
-		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
-		br.advance(uint8(v.entry))
-		buf[off+1] = uint8(v.entry >> 8)
-
-		// Refill
-		br.fillFast()
-
-		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
-		br.advance(uint8(v.entry))
-		buf[off+2] = uint8(v.entry >> 8)
-
-		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
-		br.advance(uint8(v.entry))
-		buf[off+3] = uint8(v.entry >> 8)
-
-		off += 4
-		if off == 0 {
-			if len(dst)+256 > maxDecodedSize {
-				br.close()
-				return nil, ErrMaxDecodedSizeExceeded
-			}
-			dst = append(dst, buf[:]...)
-		}
-	}
-
-	if len(dst)+int(off) > maxDecodedSize {
-		br.close()
-		return nil, ErrMaxDecodedSizeExceeded
-	}
-	dst = append(dst, buf[:off]...)
-
-	// br < 8, so uint8 is fine
-	bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead
-	for bitsLeft > 0 {
-		br.fill()
-		if false && br.bitsRead >= 32 {
-			if br.off >= 4 {
-				v := br.in[br.off-4:]
-				v = v[:4]
-				low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-				br.value = (br.value << 32) | uint64(low)
-				br.bitsRead -= 32
-				br.off -= 4
-			} else {
-				for br.off > 0 {
-					br.value = (br.value << 8) | uint64(br.in[br.off-1])
-					br.bitsRead -= 8
-					br.off--
-				}
-			}
-		}
-		if len(dst) >= maxDecodedSize {
-			br.close()
-			return nil, ErrMaxDecodedSizeExceeded
-		}
-		v := d.dt.single[br.peekBitsFast(d.actualTableLog)&tlMask]
-		nBits := uint8(v.entry)
-		br.advance(nBits)
-		bitsLeft -= nBits
-		dst = append(dst, uint8(v.entry>>8))
-	}
-	return dst, br.close()
+	return &[4][256]byte{}
 }
 
 // decompress1X8Bit will decompress a 1X encoded stream with tablelog <= 8.
@@ -341,41 +247,258 @@
 	dt := d.dt.single[:256]
 
 	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
+	bufs := d.buffer()
+	buf := &bufs[0]
 	var off uint8
 
-	shift := (8 - d.actualTableLog) & 7
+	switch d.actualTableLog {
+	case 8:
+		const shift = 0
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
 
-	//fmt.Printf("mask: %b, tl:%d\n", mask, d.actualTableLog)
-	for br.off >= 4 {
-		br.fillFast()
-		v := dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+0] = uint8(v.entry >> 8)
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
 
-		v = dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+1] = uint8(v.entry >> 8)
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
 
-		v = dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+2] = uint8(v.entry >> 8)
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
 
-		v = dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+3] = uint8(v.entry >> 8)
-
-		off += 4
-		if off == 0 {
-			if len(dst)+256 > maxDecodedSize {
-				br.close()
-				return nil, ErrMaxDecodedSizeExceeded
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					br.close()
+					d.bufs.Put(bufs)
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
 			}
-			dst = append(dst, buf[:]...)
 		}
+	case 7:
+		const shift = 8 - 7
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					br.close()
+					d.bufs.Put(bufs)
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 6:
+		const shift = 8 - 6
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 5:
+		const shift = 8 - 5
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 4:
+		const shift = 8 - 4
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 3:
+		const shift = 8 - 3
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 2:
+		const shift = 8 - 2
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 1:
+		const shift = 8 - 1
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	default:
+		d.bufs.Put(bufs)
+		return nil, fmt.Errorf("invalid tablelog: %d", d.actualTableLog)
 	}
 
 	if len(dst)+int(off) > maxDecodedSize {
+		d.bufs.Put(bufs)
 		br.close()
 		return nil, ErrMaxDecodedSizeExceeded
 	}
@@ -383,6 +506,8 @@
 
 	// br < 4, so uint8 is fine
 	bitsLeft := int8(uint8(br.off)*8 + (64 - br.bitsRead))
+	shift := (8 - d.actualTableLog) & 7
+
 	for bitsLeft > 0 {
 		if br.bitsRead >= 64-8 {
 			for br.off > 0 {
@@ -393,6 +518,7 @@
 		}
 		if len(dst) >= maxDecodedSize {
 			br.close()
+			d.bufs.Put(bufs)
 			return nil, ErrMaxDecodedSizeExceeded
 		}
 		v := dt[br.peekByteFast()>>shift]
@@ -401,6 +527,7 @@
 		bitsLeft -= int8(nBits)
 		dst = append(dst, uint8(v.entry>>8))
 	}
+	d.bufs.Put(bufs)
 	return dst, br.close()
 }
 
@@ -420,33 +547,35 @@
 	dt := d.dt.single[:256]
 
 	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
+	bufs := d.buffer()
+	buf := &bufs[0]
 	var off uint8
 
-	const shift = 0
+	const shift = 56
 
 	//fmt.Printf("mask: %b, tl:%d\n", mask, d.actualTableLog)
 	for br.off >= 4 {
 		br.fillFast()
-		v := dt[br.peekByteFast()>>shift]
+		v := dt[uint8(br.value>>shift)]
 		br.advance(uint8(v.entry))
 		buf[off+0] = uint8(v.entry >> 8)
 
-		v = dt[br.peekByteFast()>>shift]
+		v = dt[uint8(br.value>>shift)]
 		br.advance(uint8(v.entry))
 		buf[off+1] = uint8(v.entry >> 8)
 
-		v = dt[br.peekByteFast()>>shift]
+		v = dt[uint8(br.value>>shift)]
 		br.advance(uint8(v.entry))
 		buf[off+2] = uint8(v.entry >> 8)
 
-		v = dt[br.peekByteFast()>>shift]
+		v = dt[uint8(br.value>>shift)]
 		br.advance(uint8(v.entry))
 		buf[off+3] = uint8(v.entry >> 8)
 
 		off += 4
 		if off == 0 {
 			if len(dst)+256 > maxDecodedSize {
+				d.bufs.Put(bufs)
 				br.close()
 				return nil, ErrMaxDecodedSizeExceeded
 			}
@@ -455,6 +584,7 @@
 	}
 
 	if len(dst)+int(off) > maxDecodedSize {
+		d.bufs.Put(bufs)
 		br.close()
 		return nil, ErrMaxDecodedSizeExceeded
 	}
@@ -471,15 +601,17 @@
 			}
 		}
 		if len(dst) >= maxDecodedSize {
+			d.bufs.Put(bufs)
 			br.close()
 			return nil, ErrMaxDecodedSizeExceeded
 		}
-		v := dt[br.peekByteFast()>>shift]
+		v := dt[br.peekByteFast()]
 		nBits := uint8(v.entry)
 		br.advance(nBits)
 		bitsLeft -= int8(nBits)
 		dst = append(dst, uint8(v.entry>>8))
 	}
+	d.bufs.Put(bufs)
 	return dst, br.close()
 }
 
@@ -487,196 +619,6 @@
 // The length of the supplied input must match the end of a block exactly.
 // The *capacity* of the dst slice must match the destination size of
 // the uncompressed data exactly.
-func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) {
-	if len(d.dt.single) == 0 {
-		return nil, errors.New("no table loaded")
-	}
-	if len(src) < 6+(4*1) {
-		return nil, errors.New("input too small")
-	}
-	if use8BitTables && d.actualTableLog <= 8 {
-		return d.decompress4X8bit(dst, src)
-	}
-
-	var br [4]bitReaderShifted
-	start := 6
-	for i := 0; i < 3; i++ {
-		length := int(src[i*2]) | (int(src[i*2+1]) << 8)
-		if start+length >= len(src) {
-			return nil, errors.New("truncated input (or invalid offset)")
-		}
-		err := br[i].init(src[start : start+length])
-		if err != nil {
-			return nil, err
-		}
-		start += length
-	}
-	err := br[3].init(src[start:])
-	if err != nil {
-		return nil, err
-	}
-
-	// destination, offset to match first output
-	dstSize := cap(dst)
-	dst = dst[:dstSize]
-	out := dst
-	dstEvery := (dstSize + 3) / 4
-
-	const tlSize = 1 << tableLogMax
-	const tlMask = tlSize - 1
-	single := d.dt.single[:tlSize]
-
-	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
-	var off uint8
-	var decoded int
-
-	// Decode 2 values from each decoder/loop.
-	const bufoff = 256 / 4
-	for {
-		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
-			break
-		}
-
-		{
-			const stream = 0
-			const stream2 = 1
-			br[stream].fillFast()
-			br[stream2].fillFast()
-
-			val := br[stream].peekBitsFast(d.actualTableLog)
-			v := single[val&tlMask]
-			br[stream].advance(uint8(v.entry))
-			buf[off+bufoff*stream] = uint8(v.entry >> 8)
-
-			val2 := br[stream2].peekBitsFast(d.actualTableLog)
-			v2 := single[val2&tlMask]
-			br[stream2].advance(uint8(v2.entry))
-			buf[off+bufoff*stream2] = uint8(v2.entry >> 8)
-
-			val = br[stream].peekBitsFast(d.actualTableLog)
-			v = single[val&tlMask]
-			br[stream].advance(uint8(v.entry))
-			buf[off+bufoff*stream+1] = uint8(v.entry >> 8)
-
-			val2 = br[stream2].peekBitsFast(d.actualTableLog)
-			v2 = single[val2&tlMask]
-			br[stream2].advance(uint8(v2.entry))
-			buf[off+bufoff*stream2+1] = uint8(v2.entry >> 8)
-		}
-
-		{
-			const stream = 2
-			const stream2 = 3
-			br[stream].fillFast()
-			br[stream2].fillFast()
-
-			val := br[stream].peekBitsFast(d.actualTableLog)
-			v := single[val&tlMask]
-			br[stream].advance(uint8(v.entry))
-			buf[off+bufoff*stream] = uint8(v.entry >> 8)
-
-			val2 := br[stream2].peekBitsFast(d.actualTableLog)
-			v2 := single[val2&tlMask]
-			br[stream2].advance(uint8(v2.entry))
-			buf[off+bufoff*stream2] = uint8(v2.entry >> 8)
-
-			val = br[stream].peekBitsFast(d.actualTableLog)
-			v = single[val&tlMask]
-			br[stream].advance(uint8(v.entry))
-			buf[off+bufoff*stream+1] = uint8(v.entry >> 8)
-
-			val2 = br[stream2].peekBitsFast(d.actualTableLog)
-			v2 = single[val2&tlMask]
-			br[stream2].advance(uint8(v2.entry))
-			buf[off+bufoff*stream2+1] = uint8(v2.entry >> 8)
-		}
-
-		off += 2
-
-		if off == bufoff {
-			if bufoff > dstEvery {
-				return nil, errors.New("corruption detected: stream overrun 1")
-			}
-			copy(out, buf[:bufoff])
-			copy(out[dstEvery:], buf[bufoff:bufoff*2])
-			copy(out[dstEvery*2:], buf[bufoff*2:bufoff*3])
-			copy(out[dstEvery*3:], buf[bufoff*3:bufoff*4])
-			off = 0
-			out = out[bufoff:]
-			decoded += 256
-			// There must at least be 3 buffers left.
-			if len(out) < dstEvery*3 {
-				return nil, errors.New("corruption detected: stream overrun 2")
-			}
-		}
-	}
-	if off > 0 {
-		ioff := int(off)
-		if len(out) < dstEvery*3+ioff {
-			return nil, errors.New("corruption detected: stream overrun 3")
-		}
-		copy(out, buf[:off])
-		copy(out[dstEvery:dstEvery+ioff], buf[bufoff:bufoff*2])
-		copy(out[dstEvery*2:dstEvery*2+ioff], buf[bufoff*2:bufoff*3])
-		copy(out[dstEvery*3:dstEvery*3+ioff], buf[bufoff*3:bufoff*4])
-		decoded += int(off) * 4
-		out = out[off:]
-	}
-
-	// Decode remaining.
-	for i := range br {
-		offset := dstEvery * i
-		br := &br[i]
-		bitsLeft := br.off*8 + uint(64-br.bitsRead)
-		for bitsLeft > 0 {
-			br.fill()
-			if false && br.bitsRead >= 32 {
-				if br.off >= 4 {
-					v := br.in[br.off-4:]
-					v = v[:4]
-					low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-					br.value = (br.value << 32) | uint64(low)
-					br.bitsRead -= 32
-					br.off -= 4
-				} else {
-					for br.off > 0 {
-						br.value = (br.value << 8) | uint64(br.in[br.off-1])
-						br.bitsRead -= 8
-						br.off--
-					}
-				}
-			}
-			// end inline...
-			if offset >= len(out) {
-				return nil, errors.New("corruption detected: stream overrun 4")
-			}
-
-			// Read value and increment offset.
-			val := br.peekBitsFast(d.actualTableLog)
-			v := single[val&tlMask].entry
-			nBits := uint8(v)
-			br.advance(nBits)
-			bitsLeft -= uint(nBits)
-			out[offset] = uint8(v >> 8)
-			offset++
-		}
-		decoded += offset - dstEvery*i
-		err = br.close()
-		if err != nil {
-			return nil, err
-		}
-	}
-	if dstSize != decoded {
-		return nil, errors.New("corruption detected: short output block")
-	}
-	return dst, nil
-}
-
-// Decompress4X will decompress a 4X encoded stream.
-// The length of the supplied input must match the end of a block exactly.
-// The *capacity* of the dst slice must match the destination size of
-// the uncompressed data exactly.
 func (d *Decoder) decompress4X8bit(dst, src []byte) ([]byte, error) {
 	if d.actualTableLog == 8 {
 		return d.decompress4X8bitExactly(dst, src)
@@ -706,19 +648,18 @@
 	out := dst
 	dstEvery := (dstSize + 3) / 4
 
-	shift := (8 - d.actualTableLog) & 7
+	shift := (56 + (8 - d.actualTableLog)) & 63
 
 	const tlSize = 1 << 8
-	const tlMask = tlSize - 1
 	single := d.dt.single[:tlSize]
 
 	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
+	buf := d.buffer()
 	var off uint8
 	var decoded int
 
 	// Decode 4 values from each decoder/loop.
-	const bufoff = 256 / 4
+	const bufoff = 256
 	for {
 		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
 			break
@@ -728,120 +669,144 @@
 			// Interleave 2 decodes.
 			const stream = 0
 			const stream2 = 1
-			br[stream].fillFast()
-			br[stream2].fillFast()
+			br1 := &br[stream]
+			br2 := &br[stream2]
+			br1.fillFast()
+			br2.fillFast()
 
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v := single[uint8(br1.value>>shift)].entry
+			v2 := single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off] = uint8(v >> 8)
+			buf[stream2][off] = uint8(v2 >> 8)
 
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+1] = uint8(v >> 8)
+			buf[stream2][off+1] = uint8(v2 >> 8)
 
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+2] = uint8(v >> 8)
+			buf[stream2][off+2] = uint8(v2 >> 8)
 
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+2] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+3] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+3] = uint8(v >> 8)
+			buf[stream2][off+3] = uint8(v2 >> 8)
 		}
 
 		{
 			const stream = 2
 			const stream2 = 3
-			br[stream].fillFast()
-			br[stream2].fillFast()
+			br1 := &br[stream]
+			br2 := &br[stream2]
+			br1.fillFast()
+			br2.fillFast()
 
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v := single[uint8(br1.value>>shift)].entry
+			v2 := single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off] = uint8(v >> 8)
+			buf[stream2][off] = uint8(v2 >> 8)
 
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+1] = uint8(v >> 8)
+			buf[stream2][off+1] = uint8(v2 >> 8)
 
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+2] = uint8(v >> 8)
+			buf[stream2][off+2] = uint8(v2 >> 8)
 
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+2] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+3] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+3] = uint8(v >> 8)
+			buf[stream2][off+3] = uint8(v2 >> 8)
 		}
 
 		off += 4
 
-		if off == bufoff {
+		if off == 0 {
 			if bufoff > dstEvery {
+				d.bufs.Put(buf)
 				return nil, errors.New("corruption detected: stream overrun 1")
 			}
-			copy(out, buf[:bufoff])
-			copy(out[dstEvery:], buf[bufoff:bufoff*2])
-			copy(out[dstEvery*2:], buf[bufoff*2:bufoff*3])
-			copy(out[dstEvery*3:], buf[bufoff*3:bufoff*4])
-			off = 0
-			out = out[bufoff:]
-			decoded += 256
 			// There must at least be 3 buffers left.
-			if len(out) < dstEvery*3 {
+			if len(out)-bufoff < dstEvery*3 {
+				d.bufs.Put(buf)
 				return nil, errors.New("corruption detected: stream overrun 2")
 			}
+			//copy(out, buf[0][:])
+			//copy(out[dstEvery:], buf[1][:])
+			//copy(out[dstEvery*2:], buf[2][:])
+			*(*[bufoff]byte)(out) = buf[0]
+			*(*[bufoff]byte)(out[dstEvery:]) = buf[1]
+			*(*[bufoff]byte)(out[dstEvery*2:]) = buf[2]
+			*(*[bufoff]byte)(out[dstEvery*3:]) = buf[3]
+			out = out[bufoff:]
+			decoded += bufoff * 4
 		}
 	}
 	if off > 0 {
 		ioff := int(off)
 		if len(out) < dstEvery*3+ioff {
+			d.bufs.Put(buf)
 			return nil, errors.New("corruption detected: stream overrun 3")
 		}
-		copy(out, buf[:off])
-		copy(out[dstEvery:dstEvery+ioff], buf[bufoff:bufoff*2])
-		copy(out[dstEvery*2:dstEvery*2+ioff], buf[bufoff*2:bufoff*3])
-		copy(out[dstEvery*3:dstEvery*3+ioff], buf[bufoff*3:bufoff*4])
+		copy(out, buf[0][:off])
+		copy(out[dstEvery:], buf[1][:off])
+		copy(out[dstEvery*2:], buf[2][:off])
+		copy(out[dstEvery*3:], buf[3][:off])
 		decoded += int(off) * 4
 		out = out[off:]
 	}
 
 	// Decode remaining.
+	// Decode remaining.
+	remainBytes := dstEvery - (decoded / 4)
 	for i := range br {
 		offset := dstEvery * i
+		endsAt := offset + remainBytes
+		if endsAt > len(out) {
+			endsAt = len(out)
+		}
 		br := &br[i]
-		bitsLeft := int(br.off*8) + int(64-br.bitsRead)
+		bitsLeft := br.remaining()
 		for bitsLeft > 0 {
 			if br.finished() {
+				d.bufs.Put(buf)
 				return nil, io.ErrUnexpectedEOF
 			}
 			if br.bitsRead >= 56 {
@@ -861,24 +826,31 @@
 				}
 			}
 			// end inline...
-			if offset >= len(out) {
+			if offset >= endsAt {
+				d.bufs.Put(buf)
 				return nil, errors.New("corruption detected: stream overrun 4")
 			}
 
 			// Read value and increment offset.
-			v := single[br.peekByteFast()>>shift].entry
+			v := single[uint8(br.value>>shift)].entry
 			nBits := uint8(v)
 			br.advance(nBits)
-			bitsLeft -= int(nBits)
+			bitsLeft -= uint(nBits)
 			out[offset] = uint8(v >> 8)
 			offset++
 		}
+		if offset != endsAt {
+			d.bufs.Put(buf)
+			return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt)
+		}
 		decoded += offset - dstEvery*i
 		err = br.close()
 		if err != nil {
+			d.bufs.Put(buf)
 			return nil, err
 		}
 	}
+	d.bufs.Put(buf)
 	if dstSize != decoded {
 		return nil, errors.New("corruption detected: short output block")
 	}
@@ -914,18 +886,17 @@
 	out := dst
 	dstEvery := (dstSize + 3) / 4
 
-	const shift = 0
+	const shift = 56
 	const tlSize = 1 << 8
-	const tlMask = tlSize - 1
 	single := d.dt.single[:tlSize]
 
 	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
+	buf := d.buffer()
 	var off uint8
 	var decoded int
 
 	// Decode 4 values from each decoder/loop.
-	const bufoff = 256 / 4
+	const bufoff = 256
 	for {
 		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
 			break
@@ -935,98 +906,116 @@
 			// Interleave 2 decodes.
 			const stream = 0
 			const stream2 = 1
-			br[stream].fillFast()
-			br[stream2].fillFast()
+			br1 := &br[stream]
+			br2 := &br[stream2]
+			br1.fillFast()
+			br2.fillFast()
 
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v := single[uint8(br1.value>>shift)].entry
+			v2 := single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off] = uint8(v >> 8)
+			buf[stream2][off] = uint8(v2 >> 8)
 
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+1] = uint8(v >> 8)
+			buf[stream2][off+1] = uint8(v2 >> 8)
 
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+2] = uint8(v >> 8)
+			buf[stream2][off+2] = uint8(v2 >> 8)
 
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+2] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+3] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+3] = uint8(v >> 8)
+			buf[stream2][off+3] = uint8(v2 >> 8)
 		}
 
 		{
 			const stream = 2
 			const stream2 = 3
-			br[stream].fillFast()
-			br[stream2].fillFast()
+			br1 := &br[stream]
+			br2 := &br[stream2]
+			br1.fillFast()
+			br2.fillFast()
 
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v := single[uint8(br1.value>>shift)].entry
+			v2 := single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off] = uint8(v >> 8)
+			buf[stream2][off] = uint8(v2 >> 8)
 
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+1] = uint8(v >> 8)
+			buf[stream2][off+1] = uint8(v2 >> 8)
 
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+2] = uint8(v >> 8)
+			buf[stream2][off+2] = uint8(v2 >> 8)
 
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+2] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+3] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+3] = uint8(v >> 8)
+			buf[stream2][off+3] = uint8(v2 >> 8)
 		}
 
 		off += 4
 
-		if off == bufoff {
+		if off == 0 {
 			if bufoff > dstEvery {
+				d.bufs.Put(buf)
 				return nil, errors.New("corruption detected: stream overrun 1")
 			}
-			copy(out, buf[:bufoff])
-			copy(out[dstEvery:], buf[bufoff:bufoff*2])
-			copy(out[dstEvery*2:], buf[bufoff*2:bufoff*3])
-			copy(out[dstEvery*3:], buf[bufoff*3:bufoff*4])
-			off = 0
-			out = out[bufoff:]
-			decoded += 256
 			// There must at least be 3 buffers left.
-			if len(out) < dstEvery*3 {
+			if len(out)-bufoff < dstEvery*3 {
+				d.bufs.Put(buf)
 				return nil, errors.New("corruption detected: stream overrun 2")
 			}
+
+			//copy(out, buf[0][:])
+			//copy(out[dstEvery:], buf[1][:])
+			//copy(out[dstEvery*2:], buf[2][:])
+			// copy(out[dstEvery*3:], buf[3][:])
+			*(*[bufoff]byte)(out) = buf[0]
+			*(*[bufoff]byte)(out[dstEvery:]) = buf[1]
+			*(*[bufoff]byte)(out[dstEvery*2:]) = buf[2]
+			*(*[bufoff]byte)(out[dstEvery*3:]) = buf[3]
+			out = out[bufoff:]
+			decoded += bufoff * 4
 		}
 	}
 	if off > 0 {
@@ -1034,21 +1023,27 @@
 		if len(out) < dstEvery*3+ioff {
 			return nil, errors.New("corruption detected: stream overrun 3")
 		}
-		copy(out, buf[:off])
-		copy(out[dstEvery:dstEvery+ioff], buf[bufoff:bufoff*2])
-		copy(out[dstEvery*2:dstEvery*2+ioff], buf[bufoff*2:bufoff*3])
-		copy(out[dstEvery*3:dstEvery*3+ioff], buf[bufoff*3:bufoff*4])
+		copy(out, buf[0][:off])
+		copy(out[dstEvery:], buf[1][:off])
+		copy(out[dstEvery*2:], buf[2][:off])
+		copy(out[dstEvery*3:], buf[3][:off])
 		decoded += int(off) * 4
 		out = out[off:]
 	}
 
 	// Decode remaining.
+	remainBytes := dstEvery - (decoded / 4)
 	for i := range br {
 		offset := dstEvery * i
+		endsAt := offset + remainBytes
+		if endsAt > len(out) {
+			endsAt = len(out)
+		}
 		br := &br[i]
-		bitsLeft := int(br.off*8) + int(64-br.bitsRead)
+		bitsLeft := br.remaining()
 		for bitsLeft > 0 {
 			if br.finished() {
+				d.bufs.Put(buf)
 				return nil, io.ErrUnexpectedEOF
 			}
 			if br.bitsRead >= 56 {
@@ -1068,24 +1063,32 @@
 				}
 			}
 			// end inline...
-			if offset >= len(out) {
+			if offset >= endsAt {
+				d.bufs.Put(buf)
 				return nil, errors.New("corruption detected: stream overrun 4")
 			}
 
 			// Read value and increment offset.
-			v := single[br.peekByteFast()>>shift].entry
+			v := single[br.peekByteFast()].entry
 			nBits := uint8(v)
 			br.advance(nBits)
-			bitsLeft -= int(nBits)
+			bitsLeft -= uint(nBits)
 			out[offset] = uint8(v >> 8)
 			offset++
 		}
+		if offset != endsAt {
+			d.bufs.Put(buf)
+			return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt)
+		}
+
 		decoded += offset - dstEvery*i
 		err = br.close()
 		if err != nil {
+			d.bufs.Put(buf)
 			return nil, err
 		}
 	}
+	d.bufs.Put(buf)
 	if dstSize != decoded {
 		return nil, errors.New("corruption detected: short output block")
 	}
@@ -1133,7 +1136,7 @@
 			errs++
 		}
 		if errs > 0 {
-			fmt.Fprintf(w, "%d errros in base, stopping\n", errs)
+			fmt.Fprintf(w, "%d errors in base, stopping\n", errs)
 			continue
 		}
 		// Ensure that all combinations are covered.
@@ -1149,7 +1152,7 @@
 				errs++
 			}
 			if errs > 20 {
-				fmt.Fprintf(w, "%d errros, stopping\n", errs)
+				fmt.Fprintf(w, "%d errors, stopping\n", errs)
 				break
 			}
 		}
diff --git a/vendor/github.com/klauspost/compress/huff0/decompress_amd64.go b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.go
new file mode 100644
index 0000000..ba7e8e6
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.go
@@ -0,0 +1,226 @@
+//go:build amd64 && !appengine && !noasm && gc
+// +build amd64,!appengine,!noasm,gc
+
+// This file contains the specialisation of Decoder.Decompress4X
+// and Decoder.Decompress1X that use an asm implementation of thir main loops.
+package huff0
+
+import (
+	"errors"
+	"fmt"
+
+	"github.com/klauspost/compress/internal/cpuinfo"
+)
+
+// decompress4x_main_loop_x86 is an x86 assembler implementation
+// of Decompress4X when tablelog > 8.
+//
+//go:noescape
+func decompress4x_main_loop_amd64(ctx *decompress4xContext)
+
+// decompress4x_8b_loop_x86 is an x86 assembler implementation
+// of Decompress4X when tablelog <= 8 which decodes 4 entries
+// per loop.
+//
+//go:noescape
+func decompress4x_8b_main_loop_amd64(ctx *decompress4xContext)
+
+// fallback8BitSize is the size where using Go version is faster.
+const fallback8BitSize = 800
+
+type decompress4xContext struct {
+	pbr      *[4]bitReaderShifted
+	peekBits uint8
+	out      *byte
+	dstEvery int
+	tbl      *dEntrySingle
+	decoded  int
+	limit    *byte
+}
+
+// Decompress4X will decompress a 4X encoded stream.
+// The length of the supplied input must match the end of a block exactly.
+// The *capacity* of the dst slice must match the destination size of
+// the uncompressed data exactly.
+func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) {
+	if len(d.dt.single) == 0 {
+		return nil, errors.New("no table loaded")
+	}
+	if len(src) < 6+(4*1) {
+		return nil, errors.New("input too small")
+	}
+
+	use8BitTables := d.actualTableLog <= 8
+	if cap(dst) < fallback8BitSize && use8BitTables {
+		return d.decompress4X8bit(dst, src)
+	}
+
+	var br [4]bitReaderShifted
+	// Decode "jump table"
+	start := 6
+	for i := 0; i < 3; i++ {
+		length := int(src[i*2]) | (int(src[i*2+1]) << 8)
+		if start+length >= len(src) {
+			return nil, errors.New("truncated input (or invalid offset)")
+		}
+		err := br[i].init(src[start : start+length])
+		if err != nil {
+			return nil, err
+		}
+		start += length
+	}
+	err := br[3].init(src[start:])
+	if err != nil {
+		return nil, err
+	}
+
+	// destination, offset to match first output
+	dstSize := cap(dst)
+	dst = dst[:dstSize]
+	out := dst
+	dstEvery := (dstSize + 3) / 4
+
+	const tlSize = 1 << tableLogMax
+	const tlMask = tlSize - 1
+	single := d.dt.single[:tlSize]
+
+	var decoded int
+
+	if len(out) > 4*4 && !(br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4) {
+		ctx := decompress4xContext{
+			pbr:      &br,
+			peekBits: uint8((64 - d.actualTableLog) & 63), // see: bitReaderShifted.peekBitsFast()
+			out:      &out[0],
+			dstEvery: dstEvery,
+			tbl:      &single[0],
+			limit:    &out[dstEvery-4], // Always stop decoding when first buffer gets here to avoid writing OOB on last.
+		}
+		if use8BitTables {
+			decompress4x_8b_main_loop_amd64(&ctx)
+		} else {
+			decompress4x_main_loop_amd64(&ctx)
+		}
+
+		decoded = ctx.decoded
+		out = out[decoded/4:]
+	}
+
+	// Decode remaining.
+	remainBytes := dstEvery - (decoded / 4)
+	for i := range br {
+		offset := dstEvery * i
+		endsAt := offset + remainBytes
+		if endsAt > len(out) {
+			endsAt = len(out)
+		}
+		br := &br[i]
+		bitsLeft := br.remaining()
+		for bitsLeft > 0 {
+			br.fill()
+			if offset >= endsAt {
+				return nil, errors.New("corruption detected: stream overrun 4")
+			}
+
+			// Read value and increment offset.
+			val := br.peekBitsFast(d.actualTableLog)
+			v := single[val&tlMask].entry
+			nBits := uint8(v)
+			br.advance(nBits)
+			bitsLeft -= uint(nBits)
+			out[offset] = uint8(v >> 8)
+			offset++
+		}
+		if offset != endsAt {
+			return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt)
+		}
+		decoded += offset - dstEvery*i
+		err = br.close()
+		if err != nil {
+			return nil, err
+		}
+	}
+	if dstSize != decoded {
+		return nil, errors.New("corruption detected: short output block")
+	}
+	return dst, nil
+}
+
+// decompress4x_main_loop_x86 is an x86 assembler implementation
+// of Decompress1X when tablelog > 8.
+//
+//go:noescape
+func decompress1x_main_loop_amd64(ctx *decompress1xContext)
+
+// decompress4x_main_loop_x86 is an x86 with BMI2 assembler implementation
+// of Decompress1X when tablelog > 8.
+//
+//go:noescape
+func decompress1x_main_loop_bmi2(ctx *decompress1xContext)
+
+type decompress1xContext struct {
+	pbr      *bitReaderShifted
+	peekBits uint8
+	out      *byte
+	outCap   int
+	tbl      *dEntrySingle
+	decoded  int
+}
+
+// Error reported by asm implementations
+const error_max_decoded_size_exeeded = -1
+
+// Decompress1X will decompress a 1X encoded stream.
+// The cap of the output buffer will be the maximum decompressed size.
+// The length of the supplied input must match the end of a block exactly.
+func (d *Decoder) Decompress1X(dst, src []byte) ([]byte, error) {
+	if len(d.dt.single) == 0 {
+		return nil, errors.New("no table loaded")
+	}
+	var br bitReaderShifted
+	err := br.init(src)
+	if err != nil {
+		return dst, err
+	}
+	maxDecodedSize := cap(dst)
+	dst = dst[:maxDecodedSize]
+
+	const tlSize = 1 << tableLogMax
+	const tlMask = tlSize - 1
+
+	if maxDecodedSize >= 4 {
+		ctx := decompress1xContext{
+			pbr:      &br,
+			out:      &dst[0],
+			outCap:   maxDecodedSize,
+			peekBits: uint8((64 - d.actualTableLog) & 63), // see: bitReaderShifted.peekBitsFast()
+			tbl:      &d.dt.single[0],
+		}
+
+		if cpuinfo.HasBMI2() {
+			decompress1x_main_loop_bmi2(&ctx)
+		} else {
+			decompress1x_main_loop_amd64(&ctx)
+		}
+		if ctx.decoded == error_max_decoded_size_exeeded {
+			return nil, ErrMaxDecodedSizeExceeded
+		}
+
+		dst = dst[:ctx.decoded]
+	}
+
+	// br < 8, so uint8 is fine
+	bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead
+	for bitsLeft > 0 {
+		br.fill()
+		if len(dst) >= maxDecodedSize {
+			br.close()
+			return nil, ErrMaxDecodedSizeExceeded
+		}
+		v := d.dt.single[br.peekBitsFast(d.actualTableLog)&tlMask]
+		nBits := uint8(v.entry)
+		br.advance(nBits)
+		bitsLeft -= nBits
+		dst = append(dst, uint8(v.entry>>8))
+	}
+	return dst, br.close()
+}
diff --git a/vendor/github.com/klauspost/compress/huff0/decompress_amd64.s b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.s
new file mode 100644
index 0000000..c4c7ab2
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.s
@@ -0,0 +1,830 @@
+// Code generated by command: go run gen.go -out ../decompress_amd64.s -pkg=huff0. DO NOT EDIT.
+
+//go:build amd64 && !appengine && !noasm && gc
+
+// func decompress4x_main_loop_amd64(ctx *decompress4xContext)
+TEXT ·decompress4x_main_loop_amd64(SB), $0-8
+	// Preload values
+	MOVQ    ctx+0(FP), AX
+	MOVBQZX 8(AX), DI
+	MOVQ    16(AX), BX
+	MOVQ    48(AX), SI
+	MOVQ    24(AX), R8
+	MOVQ    32(AX), R9
+	MOVQ    (AX), R10
+
+	// Main loop
+main_loop:
+	XORL  DX, DX
+	CMPQ  BX, SI
+	SETGE DL
+
+	// br0.fillFast32()
+	MOVQ    32(R10), R11
+	MOVBQZX 40(R10), R12
+	CMPQ    R12, $0x20
+	JBE     skip_fill0
+	MOVQ    24(R10), AX
+	SUBQ    $0x20, R12
+	SUBQ    $0x04, AX
+	MOVQ    (R10), R13
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (AX)(R13*1), R13
+	MOVQ R12, CX
+	SHLQ CL, R13
+	MOVQ AX, 24(R10)
+	ORQ  R13, R11
+
+	// exhausted += (br0.off < 4)
+	CMPQ AX, $0x04
+	ADCB $+0, DL
+
+skip_fill0:
+	// val0 := br0.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v0 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br0.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val1 := br0.peekTopBits(peekBits)
+	MOVQ DI, CX
+	MOVQ R11, R13
+	SHRQ CL, R13
+
+	// v1 := table[val1&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br0.advance(uint8(v1.entry))
+	MOVB CH, AH
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// these two writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	MOVW AX, (BX)
+
+	// update the bitreader structure
+	MOVQ R11, 32(R10)
+	MOVB R12, 40(R10)
+
+	// br1.fillFast32()
+	MOVQ    80(R10), R11
+	MOVBQZX 88(R10), R12
+	CMPQ    R12, $0x20
+	JBE     skip_fill1
+	MOVQ    72(R10), AX
+	SUBQ    $0x20, R12
+	SUBQ    $0x04, AX
+	MOVQ    48(R10), R13
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (AX)(R13*1), R13
+	MOVQ R12, CX
+	SHLQ CL, R13
+	MOVQ AX, 72(R10)
+	ORQ  R13, R11
+
+	// exhausted += (br1.off < 4)
+	CMPQ AX, $0x04
+	ADCB $+0, DL
+
+skip_fill1:
+	// val0 := br1.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v0 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br1.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val1 := br1.peekTopBits(peekBits)
+	MOVQ DI, CX
+	MOVQ R11, R13
+	SHRQ CL, R13
+
+	// v1 := table[val1&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br1.advance(uint8(v1.entry))
+	MOVB CH, AH
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// these two writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	MOVW AX, (BX)(R8*1)
+
+	// update the bitreader structure
+	MOVQ R11, 80(R10)
+	MOVB R12, 88(R10)
+
+	// br2.fillFast32()
+	MOVQ    128(R10), R11
+	MOVBQZX 136(R10), R12
+	CMPQ    R12, $0x20
+	JBE     skip_fill2
+	MOVQ    120(R10), AX
+	SUBQ    $0x20, R12
+	SUBQ    $0x04, AX
+	MOVQ    96(R10), R13
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (AX)(R13*1), R13
+	MOVQ R12, CX
+	SHLQ CL, R13
+	MOVQ AX, 120(R10)
+	ORQ  R13, R11
+
+	// exhausted += (br2.off < 4)
+	CMPQ AX, $0x04
+	ADCB $+0, DL
+
+skip_fill2:
+	// val0 := br2.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v0 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br2.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val1 := br2.peekTopBits(peekBits)
+	MOVQ DI, CX
+	MOVQ R11, R13
+	SHRQ CL, R13
+
+	// v1 := table[val1&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br2.advance(uint8(v1.entry))
+	MOVB CH, AH
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// these two writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	MOVW AX, (BX)(R8*2)
+
+	// update the bitreader structure
+	MOVQ R11, 128(R10)
+	MOVB R12, 136(R10)
+
+	// br3.fillFast32()
+	MOVQ    176(R10), R11
+	MOVBQZX 184(R10), R12
+	CMPQ    R12, $0x20
+	JBE     skip_fill3
+	MOVQ    168(R10), AX
+	SUBQ    $0x20, R12
+	SUBQ    $0x04, AX
+	MOVQ    144(R10), R13
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (AX)(R13*1), R13
+	MOVQ R12, CX
+	SHLQ CL, R13
+	MOVQ AX, 168(R10)
+	ORQ  R13, R11
+
+	// exhausted += (br3.off < 4)
+	CMPQ AX, $0x04
+	ADCB $+0, DL
+
+skip_fill3:
+	// val0 := br3.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v0 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br3.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val1 := br3.peekTopBits(peekBits)
+	MOVQ DI, CX
+	MOVQ R11, R13
+	SHRQ CL, R13
+
+	// v1 := table[val1&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br3.advance(uint8(v1.entry))
+	MOVB CH, AH
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// these two writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	LEAQ (R8)(R8*2), CX
+	MOVW AX, (BX)(CX*1)
+
+	// update the bitreader structure
+	MOVQ  R11, 176(R10)
+	MOVB  R12, 184(R10)
+	ADDQ  $0x02, BX
+	TESTB DL, DL
+	JZ    main_loop
+	MOVQ  ctx+0(FP), AX
+	SUBQ  16(AX), BX
+	SHLQ  $0x02, BX
+	MOVQ  BX, 40(AX)
+	RET
+
+// func decompress4x_8b_main_loop_amd64(ctx *decompress4xContext)
+TEXT ·decompress4x_8b_main_loop_amd64(SB), $0-8
+	// Preload values
+	MOVQ    ctx+0(FP), CX
+	MOVBQZX 8(CX), DI
+	MOVQ    16(CX), BX
+	MOVQ    48(CX), SI
+	MOVQ    24(CX), R8
+	MOVQ    32(CX), R9
+	MOVQ    (CX), R10
+
+	// Main loop
+main_loop:
+	XORL  DX, DX
+	CMPQ  BX, SI
+	SETGE DL
+
+	// br0.fillFast32()
+	MOVQ    32(R10), R11
+	MOVBQZX 40(R10), R12
+	CMPQ    R12, $0x20
+	JBE     skip_fill0
+	MOVQ    24(R10), R13
+	SUBQ    $0x20, R12
+	SUBQ    $0x04, R13
+	MOVQ    (R10), R14
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (R13)(R14*1), R14
+	MOVQ R12, CX
+	SHLQ CL, R14
+	MOVQ R13, 24(R10)
+	ORQ  R14, R11
+
+	// exhausted += (br0.off < 4)
+	CMPQ R13, $0x04
+	ADCB $+0, DL
+
+skip_fill0:
+	// val0 := br0.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v0 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br0.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val1 := br0.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v1 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br0.advance(uint8(v1.entry)
+	MOVB   CH, AH
+	SHLQ   CL, R11
+	ADDB   CL, R12
+	BSWAPL AX
+
+	// val2 := br0.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v2 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br0.advance(uint8(v2.entry)
+	MOVB CH, AH
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val3 := br0.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v3 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br0.advance(uint8(v3.entry)
+	MOVB   CH, AL
+	SHLQ   CL, R11
+	ADDB   CL, R12
+	BSWAPL AX
+
+	// these four writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	// out[id * dstEvery + 3] = uint8(v2.entry >> 8)
+	// out[id * dstEvery + 4] = uint8(v3.entry >> 8)
+	MOVL AX, (BX)
+
+	// update the bitreader structure
+	MOVQ R11, 32(R10)
+	MOVB R12, 40(R10)
+
+	// br1.fillFast32()
+	MOVQ    80(R10), R11
+	MOVBQZX 88(R10), R12
+	CMPQ    R12, $0x20
+	JBE     skip_fill1
+	MOVQ    72(R10), R13
+	SUBQ    $0x20, R12
+	SUBQ    $0x04, R13
+	MOVQ    48(R10), R14
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (R13)(R14*1), R14
+	MOVQ R12, CX
+	SHLQ CL, R14
+	MOVQ R13, 72(R10)
+	ORQ  R14, R11
+
+	// exhausted += (br1.off < 4)
+	CMPQ R13, $0x04
+	ADCB $+0, DL
+
+skip_fill1:
+	// val0 := br1.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v0 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br1.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val1 := br1.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v1 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br1.advance(uint8(v1.entry)
+	MOVB   CH, AH
+	SHLQ   CL, R11
+	ADDB   CL, R12
+	BSWAPL AX
+
+	// val2 := br1.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v2 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br1.advance(uint8(v2.entry)
+	MOVB CH, AH
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val3 := br1.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v3 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br1.advance(uint8(v3.entry)
+	MOVB   CH, AL
+	SHLQ   CL, R11
+	ADDB   CL, R12
+	BSWAPL AX
+
+	// these four writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	// out[id * dstEvery + 3] = uint8(v2.entry >> 8)
+	// out[id * dstEvery + 4] = uint8(v3.entry >> 8)
+	MOVL AX, (BX)(R8*1)
+
+	// update the bitreader structure
+	MOVQ R11, 80(R10)
+	MOVB R12, 88(R10)
+
+	// br2.fillFast32()
+	MOVQ    128(R10), R11
+	MOVBQZX 136(R10), R12
+	CMPQ    R12, $0x20
+	JBE     skip_fill2
+	MOVQ    120(R10), R13
+	SUBQ    $0x20, R12
+	SUBQ    $0x04, R13
+	MOVQ    96(R10), R14
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (R13)(R14*1), R14
+	MOVQ R12, CX
+	SHLQ CL, R14
+	MOVQ R13, 120(R10)
+	ORQ  R14, R11
+
+	// exhausted += (br2.off < 4)
+	CMPQ R13, $0x04
+	ADCB $+0, DL
+
+skip_fill2:
+	// val0 := br2.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v0 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br2.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val1 := br2.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v1 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br2.advance(uint8(v1.entry)
+	MOVB   CH, AH
+	SHLQ   CL, R11
+	ADDB   CL, R12
+	BSWAPL AX
+
+	// val2 := br2.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v2 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br2.advance(uint8(v2.entry)
+	MOVB CH, AH
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val3 := br2.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v3 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br2.advance(uint8(v3.entry)
+	MOVB   CH, AL
+	SHLQ   CL, R11
+	ADDB   CL, R12
+	BSWAPL AX
+
+	// these four writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	// out[id * dstEvery + 3] = uint8(v2.entry >> 8)
+	// out[id * dstEvery + 4] = uint8(v3.entry >> 8)
+	MOVL AX, (BX)(R8*2)
+
+	// update the bitreader structure
+	MOVQ R11, 128(R10)
+	MOVB R12, 136(R10)
+
+	// br3.fillFast32()
+	MOVQ    176(R10), R11
+	MOVBQZX 184(R10), R12
+	CMPQ    R12, $0x20
+	JBE     skip_fill3
+	MOVQ    168(R10), R13
+	SUBQ    $0x20, R12
+	SUBQ    $0x04, R13
+	MOVQ    144(R10), R14
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (R13)(R14*1), R14
+	MOVQ R12, CX
+	SHLQ CL, R14
+	MOVQ R13, 168(R10)
+	ORQ  R14, R11
+
+	// exhausted += (br3.off < 4)
+	CMPQ R13, $0x04
+	ADCB $+0, DL
+
+skip_fill3:
+	// val0 := br3.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v0 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br3.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val1 := br3.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v1 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br3.advance(uint8(v1.entry)
+	MOVB   CH, AH
+	SHLQ   CL, R11
+	ADDB   CL, R12
+	BSWAPL AX
+
+	// val2 := br3.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v2 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br3.advance(uint8(v2.entry)
+	MOVB CH, AH
+	SHLQ CL, R11
+	ADDB CL, R12
+
+	// val3 := br3.peekTopBits(peekBits)
+	MOVQ R11, R13
+	MOVQ DI, CX
+	SHRQ CL, R13
+
+	// v3 := table[val0&mask]
+	MOVW (R9)(R13*2), CX
+
+	// br3.advance(uint8(v3.entry)
+	MOVB   CH, AL
+	SHLQ   CL, R11
+	ADDB   CL, R12
+	BSWAPL AX
+
+	// these four writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	// out[id * dstEvery + 3] = uint8(v2.entry >> 8)
+	// out[id * dstEvery + 4] = uint8(v3.entry >> 8)
+	LEAQ (R8)(R8*2), CX
+	MOVL AX, (BX)(CX*1)
+
+	// update the bitreader structure
+	MOVQ  R11, 176(R10)
+	MOVB  R12, 184(R10)
+	ADDQ  $0x04, BX
+	TESTB DL, DL
+	JZ    main_loop
+	MOVQ  ctx+0(FP), AX
+	SUBQ  16(AX), BX
+	SHLQ  $0x02, BX
+	MOVQ  BX, 40(AX)
+	RET
+
+// func decompress1x_main_loop_amd64(ctx *decompress1xContext)
+TEXT ·decompress1x_main_loop_amd64(SB), $0-8
+	MOVQ    ctx+0(FP), CX
+	MOVQ    16(CX), DX
+	MOVQ    24(CX), BX
+	CMPQ    BX, $0x04
+	JB      error_max_decoded_size_exceeded
+	LEAQ    (DX)(BX*1), BX
+	MOVQ    (CX), SI
+	MOVQ    (SI), R8
+	MOVQ    24(SI), R9
+	MOVQ    32(SI), R10
+	MOVBQZX 40(SI), R11
+	MOVQ    32(CX), SI
+	MOVBQZX 8(CX), DI
+	JMP     loop_condition
+
+main_loop:
+	// Check if we have room for 4 bytes in the output buffer
+	LEAQ 4(DX), CX
+	CMPQ CX, BX
+	JGE  error_max_decoded_size_exceeded
+
+	// Decode 4 values
+	CMPQ R11, $0x20
+	JL   bitReader_fillFast_1_end
+	SUBQ $0x20, R11
+	SUBQ $0x04, R9
+	MOVL (R8)(R9*1), R12
+	MOVQ R11, CX
+	SHLQ CL, R12
+	ORQ  R12, R10
+
+bitReader_fillFast_1_end:
+	MOVQ    DI, CX
+	MOVQ    R10, R12
+	SHRQ    CL, R12
+	MOVW    (SI)(R12*2), CX
+	MOVB    CH, AL
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLQ    CL, R10
+	MOVQ    DI, CX
+	MOVQ    R10, R12
+	SHRQ    CL, R12
+	MOVW    (SI)(R12*2), CX
+	MOVB    CH, AH
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLQ    CL, R10
+	BSWAPL  AX
+	CMPQ    R11, $0x20
+	JL      bitReader_fillFast_2_end
+	SUBQ    $0x20, R11
+	SUBQ    $0x04, R9
+	MOVL    (R8)(R9*1), R12
+	MOVQ    R11, CX
+	SHLQ    CL, R12
+	ORQ     R12, R10
+
+bitReader_fillFast_2_end:
+	MOVQ    DI, CX
+	MOVQ    R10, R12
+	SHRQ    CL, R12
+	MOVW    (SI)(R12*2), CX
+	MOVB    CH, AH
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLQ    CL, R10
+	MOVQ    DI, CX
+	MOVQ    R10, R12
+	SHRQ    CL, R12
+	MOVW    (SI)(R12*2), CX
+	MOVB    CH, AL
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLQ    CL, R10
+	BSWAPL  AX
+
+	// Store the decoded values
+	MOVL AX, (DX)
+	ADDQ $0x04, DX
+
+loop_condition:
+	CMPQ R9, $0x08
+	JGE  main_loop
+
+	// Update ctx structure
+	MOVQ ctx+0(FP), AX
+	SUBQ 16(AX), DX
+	MOVQ DX, 40(AX)
+	MOVQ (AX), AX
+	MOVQ R9, 24(AX)
+	MOVQ R10, 32(AX)
+	MOVB R11, 40(AX)
+	RET
+
+	// Report error
+error_max_decoded_size_exceeded:
+	MOVQ ctx+0(FP), AX
+	MOVQ $-1, CX
+	MOVQ CX, 40(AX)
+	RET
+
+// func decompress1x_main_loop_bmi2(ctx *decompress1xContext)
+// Requires: BMI2
+TEXT ·decompress1x_main_loop_bmi2(SB), $0-8
+	MOVQ    ctx+0(FP), CX
+	MOVQ    16(CX), DX
+	MOVQ    24(CX), BX
+	CMPQ    BX, $0x04
+	JB      error_max_decoded_size_exceeded
+	LEAQ    (DX)(BX*1), BX
+	MOVQ    (CX), SI
+	MOVQ    (SI), R8
+	MOVQ    24(SI), R9
+	MOVQ    32(SI), R10
+	MOVBQZX 40(SI), R11
+	MOVQ    32(CX), SI
+	MOVBQZX 8(CX), DI
+	JMP     loop_condition
+
+main_loop:
+	// Check if we have room for 4 bytes in the output buffer
+	LEAQ 4(DX), CX
+	CMPQ CX, BX
+	JGE  error_max_decoded_size_exceeded
+
+	// Decode 4 values
+	CMPQ  R11, $0x20
+	JL    bitReader_fillFast_1_end
+	SUBQ  $0x20, R11
+	SUBQ  $0x04, R9
+	MOVL  (R8)(R9*1), CX
+	SHLXQ R11, CX, CX
+	ORQ   CX, R10
+
+bitReader_fillFast_1_end:
+	SHRXQ   DI, R10, CX
+	MOVW    (SI)(CX*2), CX
+	MOVB    CH, AL
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLXQ   CX, R10, R10
+	SHRXQ   DI, R10, CX
+	MOVW    (SI)(CX*2), CX
+	MOVB    CH, AH
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLXQ   CX, R10, R10
+	BSWAPL  AX
+	CMPQ    R11, $0x20
+	JL      bitReader_fillFast_2_end
+	SUBQ    $0x20, R11
+	SUBQ    $0x04, R9
+	MOVL    (R8)(R9*1), CX
+	SHLXQ   R11, CX, CX
+	ORQ     CX, R10
+
+bitReader_fillFast_2_end:
+	SHRXQ   DI, R10, CX
+	MOVW    (SI)(CX*2), CX
+	MOVB    CH, AH
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLXQ   CX, R10, R10
+	SHRXQ   DI, R10, CX
+	MOVW    (SI)(CX*2), CX
+	MOVB    CH, AL
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLXQ   CX, R10, R10
+	BSWAPL  AX
+
+	// Store the decoded values
+	MOVL AX, (DX)
+	ADDQ $0x04, DX
+
+loop_condition:
+	CMPQ R9, $0x08
+	JGE  main_loop
+
+	// Update ctx structure
+	MOVQ ctx+0(FP), AX
+	SUBQ 16(AX), DX
+	MOVQ DX, 40(AX)
+	MOVQ (AX), AX
+	MOVQ R9, 24(AX)
+	MOVQ R10, 32(AX)
+	MOVB R11, 40(AX)
+	RET
+
+	// Report error
+error_max_decoded_size_exceeded:
+	MOVQ ctx+0(FP), AX
+	MOVQ $-1, CX
+	MOVQ CX, 40(AX)
+	RET
diff --git a/vendor/github.com/klauspost/compress/huff0/decompress_generic.go b/vendor/github.com/klauspost/compress/huff0/decompress_generic.go
new file mode 100644
index 0000000..908c17d
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/huff0/decompress_generic.go
@@ -0,0 +1,299 @@
+//go:build !amd64 || appengine || !gc || noasm
+// +build !amd64 appengine !gc noasm
+
+// This file contains a generic implementation of Decoder.Decompress4X.
+package huff0
+
+import (
+	"errors"
+	"fmt"
+)
+
+// Decompress4X will decompress a 4X encoded stream.
+// The length of the supplied input must match the end of a block exactly.
+// The *capacity* of the dst slice must match the destination size of
+// the uncompressed data exactly.
+func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) {
+	if len(d.dt.single) == 0 {
+		return nil, errors.New("no table loaded")
+	}
+	if len(src) < 6+(4*1) {
+		return nil, errors.New("input too small")
+	}
+	if use8BitTables && d.actualTableLog <= 8 {
+		return d.decompress4X8bit(dst, src)
+	}
+
+	var br [4]bitReaderShifted
+	// Decode "jump table"
+	start := 6
+	for i := 0; i < 3; i++ {
+		length := int(src[i*2]) | (int(src[i*2+1]) << 8)
+		if start+length >= len(src) {
+			return nil, errors.New("truncated input (or invalid offset)")
+		}
+		err := br[i].init(src[start : start+length])
+		if err != nil {
+			return nil, err
+		}
+		start += length
+	}
+	err := br[3].init(src[start:])
+	if err != nil {
+		return nil, err
+	}
+
+	// destination, offset to match first output
+	dstSize := cap(dst)
+	dst = dst[:dstSize]
+	out := dst
+	dstEvery := (dstSize + 3) / 4
+
+	const tlSize = 1 << tableLogMax
+	const tlMask = tlSize - 1
+	single := d.dt.single[:tlSize]
+
+	// Use temp table to avoid bound checks/append penalty.
+	buf := d.buffer()
+	var off uint8
+	var decoded int
+
+	// Decode 2 values from each decoder/loop.
+	const bufoff = 256
+	for {
+		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
+			break
+		}
+
+		{
+			const stream = 0
+			const stream2 = 1
+			br[stream].fillFast()
+			br[stream2].fillFast()
+
+			val := br[stream].peekBitsFast(d.actualTableLog)
+			val2 := br[stream2].peekBitsFast(d.actualTableLog)
+			v := single[val&tlMask]
+			v2 := single[val2&tlMask]
+			br[stream].advance(uint8(v.entry))
+			br[stream2].advance(uint8(v2.entry))
+			buf[stream][off] = uint8(v.entry >> 8)
+			buf[stream2][off] = uint8(v2.entry >> 8)
+
+			val = br[stream].peekBitsFast(d.actualTableLog)
+			val2 = br[stream2].peekBitsFast(d.actualTableLog)
+			v = single[val&tlMask]
+			v2 = single[val2&tlMask]
+			br[stream].advance(uint8(v.entry))
+			br[stream2].advance(uint8(v2.entry))
+			buf[stream][off+1] = uint8(v.entry >> 8)
+			buf[stream2][off+1] = uint8(v2.entry >> 8)
+		}
+
+		{
+			const stream = 2
+			const stream2 = 3
+			br[stream].fillFast()
+			br[stream2].fillFast()
+
+			val := br[stream].peekBitsFast(d.actualTableLog)
+			val2 := br[stream2].peekBitsFast(d.actualTableLog)
+			v := single[val&tlMask]
+			v2 := single[val2&tlMask]
+			br[stream].advance(uint8(v.entry))
+			br[stream2].advance(uint8(v2.entry))
+			buf[stream][off] = uint8(v.entry >> 8)
+			buf[stream2][off] = uint8(v2.entry >> 8)
+
+			val = br[stream].peekBitsFast(d.actualTableLog)
+			val2 = br[stream2].peekBitsFast(d.actualTableLog)
+			v = single[val&tlMask]
+			v2 = single[val2&tlMask]
+			br[stream].advance(uint8(v.entry))
+			br[stream2].advance(uint8(v2.entry))
+			buf[stream][off+1] = uint8(v.entry >> 8)
+			buf[stream2][off+1] = uint8(v2.entry >> 8)
+		}
+
+		off += 2
+
+		if off == 0 {
+			if bufoff > dstEvery {
+				d.bufs.Put(buf)
+				return nil, errors.New("corruption detected: stream overrun 1")
+			}
+			// There must at least be 3 buffers left.
+			if len(out)-bufoff < dstEvery*3 {
+				d.bufs.Put(buf)
+				return nil, errors.New("corruption detected: stream overrun 2")
+			}
+			//copy(out, buf[0][:])
+			//copy(out[dstEvery:], buf[1][:])
+			//copy(out[dstEvery*2:], buf[2][:])
+			//copy(out[dstEvery*3:], buf[3][:])
+			*(*[bufoff]byte)(out) = buf[0]
+			*(*[bufoff]byte)(out[dstEvery:]) = buf[1]
+			*(*[bufoff]byte)(out[dstEvery*2:]) = buf[2]
+			*(*[bufoff]byte)(out[dstEvery*3:]) = buf[3]
+			out = out[bufoff:]
+			decoded += bufoff * 4
+		}
+	}
+	if off > 0 {
+		ioff := int(off)
+		if len(out) < dstEvery*3+ioff {
+			d.bufs.Put(buf)
+			return nil, errors.New("corruption detected: stream overrun 3")
+		}
+		copy(out, buf[0][:off])
+		copy(out[dstEvery:], buf[1][:off])
+		copy(out[dstEvery*2:], buf[2][:off])
+		copy(out[dstEvery*3:], buf[3][:off])
+		decoded += int(off) * 4
+		out = out[off:]
+	}
+
+	// Decode remaining.
+	remainBytes := dstEvery - (decoded / 4)
+	for i := range br {
+		offset := dstEvery * i
+		endsAt := offset + remainBytes
+		if endsAt > len(out) {
+			endsAt = len(out)
+		}
+		br := &br[i]
+		bitsLeft := br.remaining()
+		for bitsLeft > 0 {
+			br.fill()
+			if offset >= endsAt {
+				d.bufs.Put(buf)
+				return nil, errors.New("corruption detected: stream overrun 4")
+			}
+
+			// Read value and increment offset.
+			val := br.peekBitsFast(d.actualTableLog)
+			v := single[val&tlMask].entry
+			nBits := uint8(v)
+			br.advance(nBits)
+			bitsLeft -= uint(nBits)
+			out[offset] = uint8(v >> 8)
+			offset++
+		}
+		if offset != endsAt {
+			d.bufs.Put(buf)
+			return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt)
+		}
+		decoded += offset - dstEvery*i
+		err = br.close()
+		if err != nil {
+			return nil, err
+		}
+	}
+	d.bufs.Put(buf)
+	if dstSize != decoded {
+		return nil, errors.New("corruption detected: short output block")
+	}
+	return dst, nil
+}
+
+// Decompress1X will decompress a 1X encoded stream.
+// The cap of the output buffer will be the maximum decompressed size.
+// The length of the supplied input must match the end of a block exactly.
+func (d *Decoder) Decompress1X(dst, src []byte) ([]byte, error) {
+	if len(d.dt.single) == 0 {
+		return nil, errors.New("no table loaded")
+	}
+	if use8BitTables && d.actualTableLog <= 8 {
+		return d.decompress1X8Bit(dst, src)
+	}
+	var br bitReaderShifted
+	err := br.init(src)
+	if err != nil {
+		return dst, err
+	}
+	maxDecodedSize := cap(dst)
+	dst = dst[:0]
+
+	// Avoid bounds check by always having full sized table.
+	const tlSize = 1 << tableLogMax
+	const tlMask = tlSize - 1
+	dt := d.dt.single[:tlSize]
+
+	// Use temp table to avoid bound checks/append penalty.
+	bufs := d.buffer()
+	buf := &bufs[0]
+	var off uint8
+
+	for br.off >= 8 {
+		br.fillFast()
+		v := dt[br.peekBitsFast(d.actualTableLog)&tlMask]
+		br.advance(uint8(v.entry))
+		buf[off+0] = uint8(v.entry >> 8)
+
+		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
+		br.advance(uint8(v.entry))
+		buf[off+1] = uint8(v.entry >> 8)
+
+		// Refill
+		br.fillFast()
+
+		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
+		br.advance(uint8(v.entry))
+		buf[off+2] = uint8(v.entry >> 8)
+
+		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
+		br.advance(uint8(v.entry))
+		buf[off+3] = uint8(v.entry >> 8)
+
+		off += 4
+		if off == 0 {
+			if len(dst)+256 > maxDecodedSize {
+				br.close()
+				d.bufs.Put(bufs)
+				return nil, ErrMaxDecodedSizeExceeded
+			}
+			dst = append(dst, buf[:]...)
+		}
+	}
+
+	if len(dst)+int(off) > maxDecodedSize {
+		d.bufs.Put(bufs)
+		br.close()
+		return nil, ErrMaxDecodedSizeExceeded
+	}
+	dst = append(dst, buf[:off]...)
+
+	// br < 8, so uint8 is fine
+	bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead
+	for bitsLeft > 0 {
+		br.fill()
+		if false && br.bitsRead >= 32 {
+			if br.off >= 4 {
+				v := br.in[br.off-4:]
+				v = v[:4]
+				low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+				br.value = (br.value << 32) | uint64(low)
+				br.bitsRead -= 32
+				br.off -= 4
+			} else {
+				for br.off > 0 {
+					br.value = (br.value << 8) | uint64(br.in[br.off-1])
+					br.bitsRead -= 8
+					br.off--
+				}
+			}
+		}
+		if len(dst) >= maxDecodedSize {
+			d.bufs.Put(bufs)
+			br.close()
+			return nil, ErrMaxDecodedSizeExceeded
+		}
+		v := d.dt.single[br.peekBitsFast(d.actualTableLog)&tlMask]
+		nBits := uint8(v.entry)
+		br.advance(nBits)
+		bitsLeft -= nBits
+		dst = append(dst, uint8(v.entry>>8))
+	}
+	d.bufs.Put(bufs)
+	return dst, br.close()
+}
diff --git a/vendor/github.com/klauspost/compress/huff0/huff0.go b/vendor/github.com/klauspost/compress/huff0/huff0.go
index 7ec2022..77ecd68 100644
--- a/vendor/github.com/klauspost/compress/huff0/huff0.go
+++ b/vendor/github.com/klauspost/compress/huff0/huff0.go
@@ -8,6 +8,7 @@
 	"fmt"
 	"math"
 	"math/bits"
+	"sync"
 
 	"github.com/klauspost/compress/fse"
 )
@@ -87,7 +88,7 @@
 	// Decoders will return ErrMaxDecodedSizeExceeded is this limit is exceeded.
 	MaxDecodedSize int
 
-	br byteReader
+	srcLen int
 
 	// MaxSymbolValue will override the maximum symbol value of the next block.
 	MaxSymbolValue uint8
@@ -116,6 +117,7 @@
 	nodes          []nodeElt
 	tmpOut         [4][]byte
 	fse            *fse.Scratch
+	decPool        sync.Pool // *[4][256]byte buffers.
 	huffWeight     [maxSymbolValue + 1]byte
 }
 
@@ -168,7 +170,7 @@
 	if s.fse == nil {
 		s.fse = &fse.Scratch{}
 	}
-	s.br.init(in)
+	s.srcLen = len(in)
 
 	return s, nil
 }
@@ -245,6 +247,68 @@
 	return nil
 }
 
+func (c cTable) estTableSize(s *Scratch) (sz int, err error) {
+	var (
+		// precomputed conversion table
+		bitsToWeight [tableLogMax + 1]byte
+		huffLog      = s.actualTableLog
+		// last weight is not saved.
+		maxSymbolValue = uint8(s.symbolLen - 1)
+		huffWeight     = s.huffWeight[:256]
+	)
+	const (
+		maxFSETableLog = 6
+	)
+	// convert to weight
+	bitsToWeight[0] = 0
+	for n := uint8(1); n < huffLog+1; n++ {
+		bitsToWeight[n] = huffLog + 1 - n
+	}
+
+	// Acquire histogram for FSE.
+	hist := s.fse.Histogram()
+	hist = hist[:256]
+	for i := range hist[:16] {
+		hist[i] = 0
+	}
+	for n := uint8(0); n < maxSymbolValue; n++ {
+		v := bitsToWeight[c[n].nBits] & 15
+		huffWeight[n] = v
+		hist[v]++
+	}
+
+	// FSE compress if feasible.
+	if maxSymbolValue >= 2 {
+		huffMaxCnt := uint32(0)
+		huffMax := uint8(0)
+		for i, v := range hist[:16] {
+			if v == 0 {
+				continue
+			}
+			huffMax = byte(i)
+			if v > huffMaxCnt {
+				huffMaxCnt = v
+			}
+		}
+		s.fse.HistogramFinished(huffMax, int(huffMaxCnt))
+		s.fse.TableLog = maxFSETableLog
+		b, err := fse.Compress(huffWeight[:maxSymbolValue], s.fse)
+		if err == nil && len(b) < int(s.symbolLen>>1) {
+			sz += 1 + len(b)
+			return sz, nil
+		}
+		// Unable to compress (RLE/uncompressible)
+	}
+	// write raw values as 4-bits (max : 15)
+	if maxSymbolValue > (256 - 128) {
+		// should not happen : likely means source cannot be compressed
+		return 0, ErrIncompressible
+	}
+	// special case, pack weights 4 bits/weight.
+	sz += 1 + int(maxSymbolValue/2)
+	return sz, nil
+}
+
 // estimateSize returns the estimated size in bytes of the input represented in the
 // histogram supplied.
 func (c cTable) estimateSize(hist []uint32) int {