SuperSeriousBusiness/GoToSocial
1
1
Fork 0
mirror of https://github.com/superseriousbusiness/gotosocial synced 2025-06-05 21:59:39 +02:00
Code Issues Projects Releases Wiki Activity

bump go-store version (includes minio) (#1657)

Browse Source 
Signed-off-by: kim <grufwub@gmail.com>
This commit is contained in:
kim
2023-03-30 22:39:55 +01:00
committed by GitHub
parent 0746ef741a
commit a5c920a50b
33 changed files with 14876 additions and 8512 deletions
Download Patch File Download Diff File Expand all files Collapse all files

186
vendor/github.com/klauspost/compress/s2/README.md generated vendored
View File

@ -20,11 +20,12 @@ This is important, so you don't have to worry about spending CPU cycles on alrea
* Concurrent stream compression
* Faster decompression, even for Snappy compatible content
* Concurrent Snappy/S2 stream decompression
* Ability to quickly skip forward in compressed stream
* Skip forward in compressed stream
* Random seeking with indexes
* Compatible with reading Snappy compressed content
* Smaller block size overhead on incompressible blocks
* Block concatenation
* Block Dictionary support
* Uncompressed stream mode
* Automatic stream size padding
* Snappy compatible block compression
@ -594,6 +595,123 @@ Best... 10737418240 -> 4210602774 [39.21%]; 42.96s, 254.4MB/s
Decompression speed should be around the same as using the 'better' compression mode.
## Dictionaries
*Note: S2 dictionary compression is currently at an early implementation stage, with no assembly for
neither encoding nor decoding. Performance improvements can be expected in the future.*
Adding dictionaries allow providing a custom dictionary that will serve as lookup in the beginning of blocks.
The same dictionary *must* be used for both encoding and decoding.
S2 does not keep track of whether the same dictionary is used,
and using the wrong dictionary will most often not result in an error when decompressing.
Blocks encoded *without* dictionaries can be decompressed seamlessly *with* a dictionary.
This means it is possible to switch from an encoding without dictionaries to an encoding with dictionaries
and treat the blocks similarly.
Similar to [zStandard dictionaries](https://github.com/facebook/zstd#the-case-for-small-data-compression),
the same usage scenario applies to S2 dictionaries.
> Training works if there is some correlation in a family of small data samples. The more data-specific a dictionary is, the more efficient it is (there is no universal dictionary). Hence, deploying one dictionary per type of data will provide the greatest benefits. Dictionary gains are mostly effective in the first few KB. Then, the compression algorithm will gradually use previously decoded content to better compress the rest of the file.
S2 further limits the dictionary to only be enabled on the first 64KB of a block.
This will remove any negative (speed) impacts of the dictionaries on bigger blocks.
### Compression
Using the [github_users_sample_set](https://github.com/facebook/zstd/releases/download/v1.1.3/github_users_sample_set.tar.zst)
and a 64KB dictionary trained with zStandard the following sizes can be achieved.
| | Default | Better | Best |
|--------------------|------------------|------------------|-----------------------|
| Without Dictionary | 3362023 (44.92%) | 3083163 (41.19%) | 3057944 (40.86%) |
| With Dictionary | 921524 (12.31%) | 873154 (11.67%) | 785503 bytes (10.49%) |
So for highly repetitive content, this case provides an almost 3x reduction in size.
For less uniform data we will use the Go source code tree.
Compressing First 64KB of all `.go` files in `go/src`, Go 1.19.5, 8912 files, 51253563 bytes input:
| | Default | Better | Best |
|--------------------|-------------------|-------------------|-------------------|
| Without Dictionary | 22955767 (44.79%) | 20189613 (39.39% | 19482828 (38.01%) |
| With Dictionary | 19654568 (38.35%) | 16289357 (31.78%) | 15184589 (29.63%) |
| Saving/file | 362 bytes | 428 bytes | 472 bytes |
### Creating Dictionaries
There are no tools to create dictionaries in S2.
However, there are multiple ways to create a useful dictionary:
#### Using a Sample File
If your input is very uniform, you can just use a sample file as the dictionary.
For example in the `github_users_sample_set` above, the average compression only goes up from
10.49% to 11.48% by using the first file as dictionary compared to using a dedicated dictionary.
```Go
// Read a sample
sample, err := os.ReadFile("sample.json")
// Create a dictionary.
dict := s2.MakeDict(sample, nil)
// b := dict.Bytes() will provide a dictionary that can be saved
// and reloaded with s2.NewDict(b).
// To encode:
encoded := dict.Encode(nil, file)
// To decode:
decoded, err := dict.Decode(nil, file)
```
#### Using Zstandard
Zstandard dictionaries can easily be converted to S2 dictionaries.
This can be helpful to generate dictionaries for files that don't have a fixed structure.
Example, with training set files placed in `./training-set`:
`λ zstd -r --train-fastcover training-set/* --maxdict=65536 -o name.dict`
This will create a dictionary of 64KB, that can be converted to a dictionary like this:
```Go
// Decode the Zstandard dictionary.
insp, err := zstd.InspectDictionary(zdict)
if err != nil {
panic(err)
}
// We are only interested in the contents.
// Assume that files start with "// Copyright (c) 2023".
// Search for the longest match for that.
// This may save a few bytes.
dict := s2.MakeDict(insp.Content(), []byte("// Copyright (c) 2023"))
// b := dict.Bytes() will provide a dictionary that can be saved
// and reloaded with s2.NewDict(b).
// We can now encode using this dictionary
encodedWithDict := dict.Encode(nil, payload)
// To decode content:
decoded, err := dict.Decode(nil, encodedWithDict)
```
It is recommended to save the dictionary returned by ` b:= dict.Bytes()`, since that will contain only the S2 dictionary.
This dictionary can later be loaded using `s2.NewDict(b)`. The dictionary then no longer requires `zstd` to be initialized.
Also note how `s2.MakeDict` allows you to search for a common starting sequence of your files.
This can be omitted, at the expense of a few bytes.
# Snappy Compatibility
S2 now offers full compatibility with Snappy.
@ -929,6 +1047,72 @@ The first copy of a block cannot be a repeat offset and the offset is reset on e
Default streaming block size is 1MB.
# Dictionary Encoding
Adding dictionaries allow providing a custom dictionary that will serve as lookup in the beginning of blocks.
A dictionary provides an initial repeat value that can be used to point to a common header.
Other than that the dictionary contains values that can be used as back-references.
Often used data should be placed at the *end* of the dictionary since offsets < 2048 bytes will be smaller.
## Format
Dictionary *content* must at least 16 bytes and less or equal to 64KiB (65536 bytes).
Encoding: `[repeat value (uvarint)][dictionary content...]`
Before the dictionary content, an unsigned base-128 (uvarint) encoded value specifying the initial repeat offset.
This value is an offset into the dictionary content and not a back-reference offset,
so setting this to 0 will make the repeat value point to the first value of the dictionary.
The value must be less than the dictionary length-8
## Encoding
From the decoder point of view the dictionary content is seen as preceding the encoded content.
`[dictionary content][decoded output]`
Backreferences to the dictionary are encoded as ordinary backreferences that have an offset before the start of the decoded block.
Matches copying from the dictionary are **not** allowed to cross from the dictionary into the decoded data.
However, if a copy ends at the end of the dictionary the next repeat will point to the start of the decoded buffer, which is allowed.
The first match can be a repeat value, which will use the repeat offset stored in the dictionary.
When 64KB (65536 bytes) has been en/decoded it is no longer allowed to reference the dictionary,
neither by a copy nor repeat operations.
If the boundary is crossed while copying from the dictionary, the operation should complete,
but the next instruction is not allowed to reference the dictionary.
Valid blocks encoded *without* a dictionary can be decoded with any dictionary.
There are no checks whether the supplied dictionary is the correct for a block.
Because of this there is no overhead by using a dictionary.
## Example
This is the dictionary content. Elements are separated by `[]`.
Dictionary: `[0x0a][Yesterday 25 bananas were added to Benjamins brown bag]`.
Initial repeat offset is set at 10, which is the letter `2`.
Encoded `[LIT "10"][REPEAT len=10][LIT "hich"][MATCH off=50 len=6][MATCH off=31 len=6][MATCH off=61 len=10]`
Decoded: `[10][ bananas w][hich][ were ][brown ][were added]`
Output: `10 bananas which were brown were added`
## Streams
For streams each block can use the dictionary.
The dictionary cannot not currently be provided on the stream.
# LICENSE
This code is based on the [Snappy-Go](https://github.com/golang/snappy) implementation.

491
vendor/github.com/klauspost/compress/s2/decode.go generated vendored
View File

@ -13,6 +13,7 @@ import (
"io/ioutil"
"math"
"runtime"
"strconv"
"sync"
)
@ -880,15 +881,20 @@ func (r *Reader) Skip(n int64) error {
// See Reader.ReadSeeker
type ReadSeeker struct {
*Reader
readAtMu sync.Mutex
}
// ReadSeeker will return an io.ReadSeeker compatible version of the reader.
// ReadSeeker will return an io.ReadSeeker and io.ReaderAt
// compatible version of the reader.
// If 'random' is specified the returned io.Seeker can be used for
// random seeking, otherwise only forward seeking is supported.
// Enabling random seeking requires the original input to support
// the io.Seeker interface.
// A custom index can be specified which will be used if supplied.
// When using a custom index, it will not be read from the input stream.
// The ReadAt position will affect regular reads and the current position of Seek.
// So using Read after ReadAt will continue from where the ReadAt stopped.
// No functions should be used concurrently.
// The returned ReadSeeker contains a shallow reference to the existing Reader,
// meaning changes performed to one is reflected in the other.
func (r *Reader) ReadSeeker(random bool, index []byte) (*ReadSeeker, error) {
@ -958,42 +964,55 @@ func (r *ReadSeeker) Seek(offset int64, whence int) (int64, error) {
// Reset on EOF
r.err = nil
}
if offset == 0 && whence == io.SeekCurrent {
return r.blockStart + int64(r.i), nil
}
if !r.readHeader {
// Make sure we read the header.
_, r.err = r.Read([]byte{})
}
rs, ok := r.r.(io.ReadSeeker)
if r.index == nil || !ok {
if whence == io.SeekCurrent && offset >= 0 {
err := r.Skip(offset)
return r.blockStart + int64(r.i), err
}
if whence == io.SeekStart && offset >= r.blockStart+int64(r.i) {
err := r.Skip(offset - r.blockStart - int64(r.i))
return r.blockStart + int64(r.i), err
}
return 0, ErrUnsupported
}
// Calculate absolute offset.
absOffset := offset
switch whence {
case io.SeekStart:
case io.SeekCurrent:
offset += r.blockStart + int64(r.i)
absOffset = r.blockStart + int64(r.i) + offset
case io.SeekEnd:
if offset > 0 {
return 0, errors.New("seek after end of file")
if r.index == nil {
return 0, ErrUnsupported
}
offset = r.index.TotalUncompressed + offset
absOffset = r.index.TotalUncompressed + offset
default:
r.err = ErrUnsupported
return 0, r.err
}
if offset < 0 {
if absOffset < 0 {
return 0, errors.New("seek before start of file")
}
c, u, err := r.index.Find(offset)
if !r.readHeader {
// Make sure we read the header.
_, r.err = r.Read([]byte{})
if r.err != nil {
return 0, r.err
}
}
// If we are inside current block no need to seek.
// This includes no offset changes.
if absOffset >= r.blockStart && absOffset < r.blockStart+int64(r.j) {
r.i = int(absOffset - r.blockStart)
return r.blockStart + int64(r.i), nil
}
rs, ok := r.r.(io.ReadSeeker)
if r.index == nil || !ok {
currOffset := r.blockStart + int64(r.i)
if absOffset >= currOffset {
err := r.Skip(absOffset - currOffset)
return r.blockStart + int64(r.i), err
}
return 0, ErrUnsupported
}
// We can seek and we have an index.
c, u, err := r.index.Find(absOffset)
if err != nil {
return r.blockStart + int64(r.i), err
}
@ -1004,12 +1023,57 @@ func (r *ReadSeeker) Seek(offset int64, whence int) (int64, error) {
return 0, err
}
r.i = r.j // Remove rest of current block.
if u < offset {
r.i = r.j // Remove rest of current block.
r.blockStart = u - int64(r.j) // Adjust current block start for accounting.
if u < absOffset {
// Forward inside block
return offset, r.Skip(offset - u)
return absOffset, r.Skip(absOffset - u)
}
return offset, nil
if u > absOffset {
return 0, fmt.Errorf("s2 seek: (internal error) u (%d) > absOffset (%d)", u, absOffset)
}
return absOffset, nil
}
// ReadAt reads len(p) bytes into p starting at offset off in the
// underlying input source. It returns the number of bytes
// read (0 <= n <= len(p)) and any error encountered.
//
// When ReadAt returns n < len(p), it returns a non-nil error
// explaining why more bytes were not returned. In this respect,
// ReadAt is stricter than Read.
//
// Even if ReadAt returns n < len(p), it may use all of p as scratch
// space during the call. If some data is available but not len(p) bytes,
// ReadAt blocks until either all the data is available or an error occurs.
// In this respect ReadAt is different from Read.
//
// If the n = len(p) bytes returned by ReadAt are at the end of the
// input source, ReadAt may return either err == EOF or err == nil.
//
// If ReadAt is reading from an input source with a seek offset,
// ReadAt should not affect nor be affected by the underlying
// seek offset.
//
// Clients of ReadAt can execute parallel ReadAt calls on the
// same input source. This is however not recommended.
func (r *ReadSeeker) ReadAt(p []byte, offset int64) (int, error) {
r.readAtMu.Lock()
defer r.readAtMu.Unlock()
_, err := r.Seek(offset, io.SeekStart)
if err != nil {
return 0, err
}
n := 0
for n < len(p) {
n2, err := r.Read(p[n:])
if err != nil {
// This will include io.EOF
return n + n2, err
}
n += n2
}
return n, nil
}
// ReadByte satisfies the io.ByteReader interface.
@ -1048,3 +1112,370 @@ func (r *Reader) SkippableCB(id uint8, fn func(r io.Reader) error) error {
r.skippableCB[id] = fn
return nil
}
// s2DecodeDict writes the decoding of src to dst. It assumes that the varint-encoded
// length of the decompressed bytes has already been read, and that len(dst)
// equals that length.
//
// It returns 0 on success or a decodeErrCodeXxx error code on failure.
func s2DecodeDict(dst, src []byte, dict *Dict) int {
if dict == nil {
return s2Decode(dst, src)
}
const debug = false
const debugErrs = debug
if debug {
fmt.Println("Starting decode, dst len:", len(dst))
}
var d, s, length int
offset := len(dict.dict) - dict.repeat
// As long as we can read at least 5 bytes...
for s < len(src)-5 {
// Removing bounds checks is SLOWER, when if doing
// in := src[s:s+5]
// Checked on Go 1.18
switch src[s] & 0x03 {
case tagLiteral:
x := uint32(src[s] >> 2)
switch {
case x < 60:
s++
case x == 60:
s += 2
x = uint32(src[s-1])
case x == 61:
in := src[s : s+3]
x = uint32(in[1]) | uint32(in[2])<<8
s += 3
case x == 62:
in := src[s : s+4]
// Load as 32 bit and shift down.
x = uint32(in[0]) | uint32(in[1])<<8 | uint32(in[2])<<16 | uint32(in[3])<<24
x >>= 8
s += 4
case x == 63:
in := src[s : s+5]
x = uint32(in[1]) | uint32(in[2])<<8 | uint32(in[3])<<16 | uint32(in[4])<<24
s += 5
}
length = int(x) + 1
if debug {
fmt.Println("literals, length:", length, "d-after:", d+length)
}
if length > len(dst)-d || length > len(src)-s || (strconv.IntSize == 32 && length <= 0) {
if debugErrs {
fmt.Println("corrupt literal: length:", length, "d-left:", len(dst)-d, "src-left:", len(src)-s)
}
return decodeErrCodeCorrupt
}
copy(dst[d:], src[s:s+length])
d += length
s += length
continue
case tagCopy1:
s += 2
toffset := int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1]))
length = int(src[s-2]) >> 2 & 0x7
if toffset == 0 {
if debug {
fmt.Print("(repeat) ")
}
// keep last offset
switch length {
case 5:
length = int(src[s]) + 4
s += 1
case 6:
in := src[s : s+2]
length = int(uint32(in[0])|(uint32(in[1])<<8)) + (1 << 8)
s += 2
case 7:
in := src[s : s+3]
length = int((uint32(in[2])<<16)|(uint32(in[1])<<8)|uint32(in[0])) + (1 << 16)
s += 3
default: // 0-> 4
}
} else {
offset = toffset
}
length += 4
case tagCopy2:
in := src[s : s+3]
offset = int(uint32(in[1]) | uint32(in[2])<<8)
length = 1 + int(in[0])>>2
s += 3
case tagCopy4:
in := src[s : s+5]
offset = int(uint32(in[1]) | uint32(in[2])<<8 | uint32(in[3])<<16 | uint32(in[4])<<24)
length = 1 + int(in[0])>>2
s += 5
}
if offset <= 0 || length > len(dst)-d {
if debugErrs {
fmt.Println("match error; offset:", offset, "length:", length, "dst-left:", len(dst)-d)
}
return decodeErrCodeCorrupt
}
// copy from dict
if d < offset {
if d > MaxDictSrcOffset {
if debugErrs {
fmt.Println("dict after", MaxDictSrcOffset, "d:", d, "offset:", offset, "length:", length)
}
return decodeErrCodeCorrupt
}
startOff := len(dict.dict) - offset + d
if startOff < 0 || startOff+length > len(dict.dict) {
if debugErrs {
fmt.Printf("offset (%d) + length (%d) bigger than dict (%d)\n", offset, length, len(dict.dict))
}
return decodeErrCodeCorrupt
}
if debug {
fmt.Println("dict copy, length:", length, "offset:", offset, "d-after:", d+length, "dict start offset:", startOff)
}
copy(dst[d:d+length], dict.dict[startOff:])
d += length
continue
}
if debug {
fmt.Println("copy, length:", length, "offset:", offset, "d-after:", d+length)
}
// Copy from an earlier sub-slice of dst to a later sub-slice.
// If no overlap, use the built-in copy:
if offset > length {
copy(dst[d:d+length], dst[d-offset:])
d += length
continue
}
// Unlike the built-in copy function, this byte-by-byte copy always runs
// forwards, even if the slices overlap. Conceptually, this is:
//
// d += forwardCopy(dst[d:d+length], dst[d-offset:])
//
// We align the slices into a and b and show the compiler they are the same size.
// This allows the loop to run without bounds checks.
a := dst[d : d+length]
b := dst[d-offset:]
b = b[:len(a)]
for i := range a {
a[i] = b[i]
}
d += length
}
// Remaining with extra checks...
for s < len(src) {
switch src[s] & 0x03 {
case tagLiteral:
x := uint32(src[s] >> 2)
switch {
case x < 60:
s++
case x == 60:
s += 2
if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
if debugErrs {
fmt.Println("src went oob")
}
return decodeErrCodeCorrupt
}
x = uint32(src[s-1])
case x == 61:
s += 3
if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
if debugErrs {
fmt.Println("src went oob")
}
return decodeErrCodeCorrupt
}
x = uint32(src[s-2]) | uint32(src[s-1])<<8
case x == 62:
s += 4
if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
if debugErrs {
fmt.Println("src went oob")
}
return decodeErrCodeCorrupt
}
x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16
case x == 63:
s += 5
if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
if debugErrs {
fmt.Println("src went oob")
}
return decodeErrCodeCorrupt
}
x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24
}
length = int(x) + 1
if length > len(dst)-d || length > len(src)-s || (strconv.IntSize == 32 && length <= 0) {
if debugErrs {
fmt.Println("corrupt literal: length:", length, "d-left:", len(dst)-d, "src-left:", len(src)-s)
}
return decodeErrCodeCorrupt
}
if debug {
fmt.Println("literals, length:", length, "d-after:", d+length)
}
copy(dst[d:], src[s:s+length])
d += length
s += length
continue
case tagCopy1:
s += 2
if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
if debugErrs {
fmt.Println("src went oob")
}
return decodeErrCodeCorrupt
}
length = int(src[s-2]) >> 2 & 0x7
toffset := int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1]))
if toffset == 0 {
if debug {
fmt.Print("(repeat) ")
}
// keep last offset
switch length {
case 5:
s += 1
if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
if debugErrs {
fmt.Println("src went oob")
}
return decodeErrCodeCorrupt
}
length = int(uint32(src[s-1])) + 4
case 6:
s += 2
if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
if debugErrs {
fmt.Println("src went oob")
}
return decodeErrCodeCorrupt
}
length = int(uint32(src[s-2])|(uint32(src[s-1])<<8)) + (1 << 8)
case 7:
s += 3
if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
if debugErrs {
fmt.Println("src went oob")
}
return decodeErrCodeCorrupt
}
length = int(uint32(src[s-3])|(uint32(src[s-2])<<8)|(uint32(src[s-1])<<16)) + (1 << 16)
default: // 0-> 4
}
} else {
offset = toffset
}
length += 4
case tagCopy2:
s += 3
if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
if debugErrs {
fmt.Println("src went oob")
}
return decodeErrCodeCorrupt
}
length = 1 + int(src[s-3])>>2
offset = int(uint32(src[s-2]) | uint32(src[s-1])<<8)
case tagCopy4:
s += 5
if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
if debugErrs {
fmt.Println("src went oob")
}
return decodeErrCodeCorrupt
}
length = 1 + int(src[s-5])>>2
offset = int(uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24)
}
if offset <= 0 || length > len(dst)-d {
if debugErrs {
fmt.Println("match error; offset:", offset, "length:", length, "dst-left:", len(dst)-d)
}
return decodeErrCodeCorrupt
}
// copy from dict
if d < offset {
if d > MaxDictSrcOffset {
if debugErrs {
fmt.Println("dict after", MaxDictSrcOffset, "d:", d, "offset:", offset, "length:", length)
}
return decodeErrCodeCorrupt
}
rOff := len(dict.dict) - (offset - d)
if debug {
fmt.Println("starting dict entry from dict offset", len(dict.dict)-rOff)
}
if rOff+length > len(dict.dict) {
if debugErrs {
fmt.Println("err: END offset", rOff+length, "bigger than dict", len(dict.dict), "dict offset:", rOff, "length:", length)
}
return decodeErrCodeCorrupt
}
if rOff < 0 {
if debugErrs {
fmt.Println("err: START offset", rOff, "less than 0", len(dict.dict), "dict offset:", rOff, "length:", length)
}
return decodeErrCodeCorrupt
}
copy(dst[d:d+length], dict.dict[rOff:])
d += length
continue
}
if debug {
fmt.Println("copy, length:", length, "offset:", offset, "d-after:", d+length)
}
// Copy from an earlier sub-slice of dst to a later sub-slice.
// If no overlap, use the built-in copy:
if offset > length {
copy(dst[d:d+length], dst[d-offset:])
d += length
continue
}
// Unlike the built-in copy function, this byte-by-byte copy always runs
// forwards, even if the slices overlap. Conceptually, this is:
//
// d += forwardCopy(dst[d:d+length], dst[d-offset:])
//
// We align the slices into a and b and show the compiler they are the same size.
// This allows the loop to run without bounds checks.
a := dst[d : d+length]
b := dst[d-offset:]
b = b[:len(a)]
for i := range a {
a[i] = b[i]
}
d += length
}
if d != len(dst) {
if debugErrs {
fmt.Println("wanted length", len(dst), "got", d)
}
return decodeErrCodeCorrupt
}
return 0
}

13
vendor/github.com/klauspost/compress/s2/decode_other.go generated vendored
View File

@ -57,6 +57,9 @@ func s2Decode(dst, src []byte) int {
}
length = int(x) + 1
if length > len(dst)-d || length > len(src)-s || (strconv.IntSize == 32 && length <= 0) {
if debug {
fmt.Println("corrupt: lit size", length)
}
return decodeErrCodeCorrupt
}
if debug {
@ -109,6 +112,10 @@ func s2Decode(dst, src []byte) int {
}
if offset <= 0 || d < offset || length > len(dst)-d {
if debug {
fmt.Println("corrupt: match, length", length, "offset:", offset, "dst avail:", len(dst)-d, "dst pos:", d)
}
return decodeErrCodeCorrupt
}
@ -175,6 +182,9 @@ func s2Decode(dst, src []byte) int {
}
length = int(x) + 1
if length > len(dst)-d || length > len(src)-s || (strconv.IntSize == 32 && length <= 0) {
if debug {
fmt.Println("corrupt: lit size", length)
}
return decodeErrCodeCorrupt
}
if debug {
@ -241,6 +251,9 @@ func s2Decode(dst, src []byte) int {
}
if offset <= 0 || d < offset || length > len(dst)-d {
if debug {
fmt.Println("corrupt: match, length", length, "offset:", offset, "dst avail:", len(dst)-d, "dst pos:", d)
}
return decodeErrCodeCorrupt
}

331
vendor/github.com/klauspost/compress/s2/dict.go generated vendored Normal file
View File

@ -0,0 +1,331 @@
// Copyright (c) 2022+ 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.
package s2
import (
"bytes"
"encoding/binary"
"sync"
)
const (
// MinDictSize is the minimum dictionary size when repeat has been read.
MinDictSize = 16
// MaxDictSize is the maximum dictionary size when repeat has been read.
MaxDictSize = 65536
// MaxDictSrcOffset is the maximum offset where a dictionary entry can start.
MaxDictSrcOffset = 65535
)
// Dict contains a dictionary that can be used for encoding and decoding s2
type Dict struct {
dict []byte
repeat int // Repeat as index of dict
fast, better, best sync.Once
fastTable *[1 << 14]uint16
betterTableShort *[1 << 14]uint16
betterTableLong *[1 << 17]uint16
bestTableShort *[1 << 16]uint32
bestTableLong *[1 << 19]uint32
}
// NewDict will read a dictionary.
// It will return nil if the dictionary is invalid.
func NewDict(dict []byte) *Dict {
if len(dict) == 0 {
return nil
}
var d Dict
// Repeat is the first value of the dict
r, n := binary.Uvarint(dict)
if n <= 0 {
return nil
}
dict = dict[n:]
d.dict = dict
if cap(d.dict) < len(d.dict)+16 {
d.dict = append(make([]byte, 0, len(d.dict)+16), d.dict...)
}
if len(dict) < MinDictSize || len(dict) > MaxDictSize {
return nil
}
d.repeat = int(r)
if d.repeat > len(dict) {
return nil
}
return &d
}
// Bytes will return a serialized version of the dictionary.
// The output can be sent to NewDict.
func (d *Dict) Bytes() []byte {
dst := make([]byte, binary.MaxVarintLen16+len(d.dict))
return append(dst[:binary.PutUvarint(dst, uint64(d.repeat))], d.dict...)
}
// MakeDict will create a dictionary.
// 'data' must be at least MinDictSize.
// If data is longer than MaxDictSize only the last MaxDictSize bytes will be used.
// If searchStart is set the start repeat value will be set to the last
// match of this content.
// If no matches are found, it will attempt to find shorter matches.
// This content should match the typical start of a block.
// If at least 4 bytes cannot be matched, repeat is set to start of block.
func MakeDict(data []byte, searchStart []byte) *Dict {
if len(data) == 0 {
return nil
}
if len(data) > MaxDictSize {
data = data[len(data)-MaxDictSize:]
}
var d Dict
dict := data
d.dict = dict
if cap(d.dict) < len(d.dict)+16 {
d.dict = append(make([]byte, 0, len(d.dict)+16), d.dict...)
}
if len(dict) < MinDictSize {
return nil
}
// Find the longest match possible, last entry if multiple.
for s := len(searchStart); s > 4; s-- {
if idx := bytes.LastIndex(data, searchStart[:s]); idx >= 0 && idx <= len(data)-8 {
d.repeat = idx
break
}
}
return &d
}
// Encode returns the encoded form of src. The returned slice may be a sub-
// slice of dst if dst was large enough to hold the entire encoded block.
// Otherwise, a newly allocated slice will be returned.
//
// The dst and src must not overlap. It is valid to pass a nil dst.
//
// The blocks will require the same amount of memory to decode as encoding,
// and does not make for concurrent decoding.
// Also note that blocks do not contain CRC information, so corruption may be undetected.
//
// If you need to encode larger amounts of data, consider using
// the streaming interface which gives all of these features.
func (d *Dict) Encode(dst, src []byte) []byte {
if n := MaxEncodedLen(len(src)); n < 0 {
panic(ErrTooLarge)
} else if cap(dst) < n {
dst = make([]byte, n)
} else {
dst = dst[:n]
}
// The block starts with the varint-encoded length of the decompressed bytes.
dstP := binary.PutUvarint(dst, uint64(len(src)))
if len(src) == 0 {
return dst[:dstP]
}
if len(src) < minNonLiteralBlockSize {
dstP += emitLiteral(dst[dstP:], src)
return dst[:dstP]
}
n := encodeBlockDictGo(dst[dstP:], src, d)
if n > 0 {
dstP += n
return dst[:dstP]
}
// Not compressible
dstP += emitLiteral(dst[dstP:], src)
return dst[:dstP]
}
// EncodeBetter returns the encoded form of src. The returned slice may be a sub-
// slice of dst if dst was large enough to hold the entire encoded block.
// Otherwise, a newly allocated slice will be returned.
//
// EncodeBetter compresses better than Encode but typically with a
// 10-40% speed decrease on both compression and decompression.
//
// The dst and src must not overlap. It is valid to pass a nil dst.
//
// The blocks will require the same amount of memory to decode as encoding,
// and does not make for concurrent decoding.
// Also note that blocks do not contain CRC information, so corruption may be undetected.
//
// If you need to encode larger amounts of data, consider using
// the streaming interface which gives all of these features.
func (d *Dict) EncodeBetter(dst, src []byte) []byte {
if n := MaxEncodedLen(len(src)); n < 0 {
panic(ErrTooLarge)
} else if len(dst) < n {
dst = make([]byte, n)
}
// The block starts with the varint-encoded length of the decompressed bytes.
dstP := binary.PutUvarint(dst, uint64(len(src)))
if len(src) == 0 {
return dst[:dstP]
}
if len(src) < minNonLiteralBlockSize {
dstP += emitLiteral(dst[dstP:], src)
return dst[:dstP]
}
n := encodeBlockBetterDict(dst[dstP:], src, d)
if n > 0 {
dstP += n
return dst[:dstP]
}
// Not compressible
dstP += emitLiteral(dst[dstP:], src)
return dst[:dstP]
}
// EncodeBest returns the encoded form of src. The returned slice may be a sub-
// slice of dst if dst was large enough to hold the entire encoded block.
// Otherwise, a newly allocated slice will be returned.
//
// EncodeBest compresses as good as reasonably possible but with a
// big speed decrease.
//
// The dst and src must not overlap. It is valid to pass a nil dst.
//
// The blocks will require the same amount of memory to decode as encoding,
// and does not make for concurrent decoding.
// Also note that blocks do not contain CRC information, so corruption may be undetected.
//
// If you need to encode larger amounts of data, consider using
// the streaming interface which gives all of these features.
func (d *Dict) EncodeBest(dst, src []byte) []byte {
if n := MaxEncodedLen(len(src)); n < 0 {
panic(ErrTooLarge)
} else if len(dst) < n {
dst = make([]byte, n)
}
// The block starts with the varint-encoded length of the decompressed bytes.
dstP := binary.PutUvarint(dst, uint64(len(src)))
if len(src) == 0 {
return dst[:dstP]
}
if len(src) < minNonLiteralBlockSize {
dstP += emitLiteral(dst[dstP:], src)
return dst[:dstP]
}
n := encodeBlockBest(dst[dstP:], src, d)
if n > 0 {
dstP += n
return dst[:dstP]
}
// Not compressible
dstP += emitLiteral(dst[dstP:], src)
return dst[:dstP]
}
// Decode returns the decoded form of src. The returned slice may be a sub-
// slice of dst if dst was large enough to hold the entire decoded block.
// Otherwise, a newly allocated slice will be returned.
//
// The dst and src must not overlap. It is valid to pass a nil dst.
func (d *Dict) Decode(dst, src []byte) ([]byte, error) {
dLen, s, err := decodedLen(src)
if err != nil {
return nil, err
}
if dLen <= cap(dst) {
dst = dst[:dLen]
} else {
dst = make([]byte, dLen)
}
if s2DecodeDict(dst, src[s:], d) != 0 {
return nil, ErrCorrupt
}
return dst, nil
}
func (d *Dict) initFast() {
d.fast.Do(func() {
const (
tableBits = 14
maxTableSize = 1 << tableBits
)
var table [maxTableSize]uint16
// We stop so any entry of length 8 can always be read.
for i := 0; i < len(d.dict)-8-2; i += 3 {
x0 := load64(d.dict, i)
h0 := hash6(x0, tableBits)
h1 := hash6(x0>>8, tableBits)
h2 := hash6(x0>>16, tableBits)
table[h0] = uint16(i)
table[h1] = uint16(i + 1)
table[h2] = uint16(i + 2)
}
d.fastTable = &table
})
}
func (d *Dict) initBetter() {
d.better.Do(func() {
const (
// Long hash matches.
lTableBits = 17
maxLTableSize = 1 << lTableBits
// Short hash matches.
sTableBits = 14
maxSTableSize = 1 << sTableBits
)
var lTable [maxLTableSize]uint16
var sTable [maxSTableSize]uint16
// We stop so any entry of length 8 can always be read.
for i := 0; i < len(d.dict)-8; i++ {
cv := load64(d.dict, i)
lTable[hash7(cv, lTableBits)] = uint16(i)
sTable[hash4(cv, sTableBits)] = uint16(i)
}
d.betterTableShort = &sTable
d.betterTableLong = &lTable
})
}
func (d *Dict) initBest() {
d.best.Do(func() {
const (
// Long hash matches.
lTableBits = 19
maxLTableSize = 1 << lTableBits
// Short hash matches.
sTableBits = 16
maxSTableSize = 1 << sTableBits
)
var lTable [maxLTableSize]uint32
var sTable [maxSTableSize]uint32
// We stop so any entry of length 8 can always be read.
for i := 0; i < len(d.dict)-8; i++ {
cv := load64(d.dict, i)
hashL := hash8(cv, lTableBits)
hashS := hash4(cv, sTableBits)
candidateL := lTable[hashL]
candidateS := sTable[hashS]
lTable[hashL] = uint32(i) | candidateL<<16
sTable[hashS] = uint32(i) | candidateS<<16
}
d.bestTableShort = &sTable
d.bestTableLong = &lTable
})
}

54
vendor/github.com/klauspost/compress/s2/encode.go generated vendored
View File

@ -58,6 +58,32 @@ func Encode(dst, src []byte) []byte {
return dst[:d]
}
// EstimateBlockSize will perform a very fast compression
// without outputting the result and return the compressed output size.
// The function returns -1 if no improvement could be achieved.
// Using actual compression will most often produce better compression than the estimate.
func EstimateBlockSize(src []byte) (d int) {
if len(src) < 6 || int64(len(src)) > 0xffffffff {
return -1
}
if len(src) <= 1024 {
d = calcBlockSizeSmall(src)
} else {
d = calcBlockSize(src)
}
if d == 0 {
return -1
}
// Size of the varint encoded block size.
d += (bits.Len64(uint64(len(src))) + 7) / 7
if d >= len(src) {
return -1
}
return d
}
// EncodeBetter returns the encoded form of src. The returned slice may be a sub-
// slice of dst if dst was large enough to hold the entire encoded block.
// Otherwise, a newly allocated slice will be returned.
@ -132,7 +158,7 @@ func EncodeBest(dst, src []byte) []byte {
d += emitLiteral(dst[d:], src)
return dst[:d]
}
n := encodeBlockBest(dst[d:], src)
n := encodeBlockBest(dst[d:], src, nil)
if n > 0 {
d += n
return dst[:d]
@ -404,10 +430,11 @@ type Writer struct {
buffers sync.Pool
pad int
writer io.Writer
randSrc io.Reader
writerWg sync.WaitGroup
index Index
writer io.Writer
randSrc io.Reader
writerWg sync.WaitGroup
index Index
customEnc func(dst, src []byte) int
// wroteStreamHeader is whether we have written the stream header.
wroteStreamHeader bool
@ -773,6 +800,9 @@ func (w *Writer) EncodeBuffer(buf []byte) (err error) {
}
func (w *Writer) encodeBlock(obuf, uncompressed []byte) int {
if w.customEnc != nil {
return w.customEnc(obuf, uncompressed)
}
if w.snappy {
switch w.level {
case levelFast:
@ -790,7 +820,7 @@ func (w *Writer) encodeBlock(obuf, uncompressed []byte) int {
case levelBetter:
return encodeBlockBetter(obuf, uncompressed)
case levelBest:
return encodeBlockBest(obuf, uncompressed)
return encodeBlockBest(obuf, uncompressed, nil)
}
return 0
}
@ -1339,3 +1369,15 @@ func WriterFlushOnWrite() WriterOption {
return nil
}
}
// WriterCustomEncoder allows to override the encoder for blocks on the stream.
// The function must compress 'src' into 'dst' and return the bytes used in dst as an integer.
// Block size (initial varint) should not be added by the encoder.
// Returning value 0 indicates the block could not be compressed.
// The function should expect to be called concurrently.
func WriterCustomEncoder(fn func(dst, src []byte) int) WriterOption {
return func(w *Writer) error {
w.customEnc = fn
return nil
}
}

592
vendor/github.com/klauspost/compress/s2/encode_all.go generated vendored
View File

@ -8,6 +8,7 @@ package s2
import (
"bytes"
"encoding/binary"
"fmt"
"math/bits"
)
@ -455,3 +456,594 @@ emitRemainder:
}
return d
}
// encodeBlockGo encodes a non-empty src to a guaranteed-large-enough dst. It
// assumes that the varint-encoded length of the decompressed bytes has already
// been written.
//
// It also assumes that:
//
// len(dst) >= MaxEncodedLen(len(src)) &&
// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize
func encodeBlockDictGo(dst, src []byte, dict *Dict) (d int) {
// Initialize the hash table.
const (
tableBits = 14
maxTableSize = 1 << tableBits
maxAhead = 8 // maximum bytes ahead without checking sLimit
debug = false
)
dict.initFast()
var table [maxTableSize]uint32
// sLimit is when to stop looking for offset/length copies. The inputMargin
// lets us use a fast path for emitLiteral in the main loop, while we are
// looking for copies.
sLimit := len(src) - inputMargin
if sLimit > MaxDictSrcOffset-maxAhead {
sLimit = MaxDictSrcOffset - maxAhead
}
// Bail if we can't compress to at least this.
dstLimit := len(src) - len(src)>>5 - 5
// nextEmit is where in src the next emitLiteral should start from.
nextEmit := 0
// The encoded form can start with a dict entry (copy or repeat).
s := 0
// Convert dict repeat to offset
repeat := len(dict.dict) - dict.repeat
cv := load64(src, 0)
// While in dict
searchDict:
for {
// Next src position to check
nextS := s + (s-nextEmit)>>6 + 4
hash0 := hash6(cv, tableBits)
hash1 := hash6(cv>>8, tableBits)
if nextS > sLimit {
if debug {
fmt.Println("slimit reached", s, nextS)
}
break searchDict
}
candidateDict := int(dict.fastTable[hash0])
candidateDict2 := int(dict.fastTable[hash1])
candidate2 := int(table[hash1])
candidate := int(table[hash0])
table[hash0] = uint32(s)
table[hash1] = uint32(s + 1)
hash2 := hash6(cv>>16, tableBits)
// Check repeat at offset checkRep.
const checkRep = 1
if repeat > s {
candidate := len(dict.dict) - repeat + s
if repeat-s >= 4 && uint32(cv) == load32(dict.dict, candidate) {
// Extend back
base := s
for i := candidate; base > nextEmit && i > 0 && dict.dict[i-1] == src[base-1]; {
i--
base--
}
d += emitLiteral(dst[d:], src[nextEmit:base])
if debug && nextEmit != base {
fmt.Println("emitted ", base-nextEmit, "literals")
}
s += 4
candidate += 4
for candidate < len(dict.dict)-8 && s <= len(src)-8 {
if diff := load64(src, s) ^ load64(dict.dict, candidate); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidate += 8
}
d += emitRepeat(dst[d:], repeat, s-base)
if debug {
fmt.Println("emitted dict repeat length", s-base, "offset:", repeat, "s:", s)
}
nextEmit = s
if s >= sLimit {
break searchDict
}
cv = load64(src, s)
continue
}
} else if uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) {
base := s + checkRep
// Extend back
for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; {
i--
base--
}
d += emitLiteral(dst[d:], src[nextEmit:base])
if debug && nextEmit != base {
fmt.Println("emitted ", base-nextEmit, "literals")
}
// Extend forward
candidate := s - repeat + 4 + checkRep
s += 4 + checkRep
for s <= sLimit {
if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidate += 8
}
if debug {
// Validate match.
if s <= candidate {
panic("s <= candidate")
}
a := src[base:s]
b := src[base-repeat : base-repeat+(s-base)]
if !bytes.Equal(a, b) {
panic("mismatch")
}
}
if nextEmit > 0 {
// same as `add := emitCopy(dst[d:], repeat, s-base)` but skips storing offset.
d += emitRepeat(dst[d:], repeat, s-base)
} else {
// First match, cannot be repeat.
d += emitCopy(dst[d:], repeat, s-base)
}
nextEmit = s
if s >= sLimit {
break searchDict
}
if debug {
fmt.Println("emitted reg repeat", s-base, "s:", s)
}
cv = load64(src, s)
continue searchDict
}
if s == 0 {
cv = load64(src, nextS)
s = nextS
continue searchDict
}
// Start with table. These matches will always be closer.
if uint32(cv) == load32(src, candidate) {
goto emitMatch
}
candidate = int(table[hash2])
if uint32(cv>>8) == load32(src, candidate2) {
table[hash2] = uint32(s + 2)
candidate = candidate2
s++
goto emitMatch
}
// Check dict. Dicts have longer offsets, so we want longer matches.
if cv == load64(dict.dict, candidateDict) {
table[hash2] = uint32(s + 2)
goto emitDict
}
candidateDict = int(dict.fastTable[hash2])
// Check if upper 7 bytes match
if candidateDict2 >= 1 {
if cv^load64(dict.dict, candidateDict2-1) < (1 << 8) {
table[hash2] = uint32(s + 2)
candidateDict = candidateDict2
s++
goto emitDict
}
}
table[hash2] = uint32(s + 2)
if uint32(cv>>16) == load32(src, candidate) {
s += 2
goto emitMatch
}
if candidateDict >= 2 {
// Check if upper 6 bytes match
if cv^load64(dict.dict, candidateDict-2) < (1 << 16) {
s += 2
goto emitDict
}
}
cv = load64(src, nextS)
s = nextS
continue searchDict
emitDict:
{
if debug {
if load32(dict.dict, candidateDict) != load32(src, s) {
panic("dict emit mismatch")
}
}
// Extend backwards.
// The top bytes will be rechecked to get the full match.
for candidateDict > 0 && s > nextEmit && dict.dict[candidateDict-1] == src[s-1] {
candidateDict--
s--
}
// Bail if we exceed the maximum size.
if d+(s-nextEmit) > dstLimit {
return 0
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
d += emitLiteral(dst[d:], src[nextEmit:s])
if debug && nextEmit != s {
fmt.Println("emitted ", s-nextEmit, "literals")
}
{
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
base := s
repeat = s + (len(dict.dict)) - candidateDict
// Extend the 4-byte match as long as possible.
s += 4
candidateDict += 4
for s <= len(src)-8 && len(dict.dict)-candidateDict >= 8 {
if diff := load64(src, s) ^ load64(dict.dict, candidateDict); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidateDict += 8
}
// Matches longer than 64 are split.
if s <= sLimit || s-base < 8 {
d += emitCopy(dst[d:], repeat, s-base)
} else {
// Split to ensure we don't start a copy within next block
d += emitCopy(dst[d:], repeat, 4)
d += emitRepeat(dst[d:], repeat, s-base-4)
}
if false {
// Validate match.
if s <= candidate {
panic("s <= candidate")
}
a := src[base:s]
b := dict.dict[base-repeat : base-repeat+(s-base)]
if !bytes.Equal(a, b) {
panic("mismatch")
}
}
if debug {
fmt.Println("emitted dict copy, length", s-base, "offset:", repeat, "s:", s)
}
nextEmit = s
if s >= sLimit {
break searchDict
}
if d > dstLimit {
// Do we have space for more, if not bail.
return 0
}
// Index and continue loop to try new candidate.
x := load64(src, s-2)
m2Hash := hash6(x, tableBits)
currHash := hash6(x>>8, tableBits)
candidate = int(table[currHash])
table[m2Hash] = uint32(s - 2)
table[currHash] = uint32(s - 1)
cv = load64(src, s)
}
continue
}
emitMatch:
// Extend backwards.
// The top bytes will be rechecked to get the full match.
for candidate > 0 && s > nextEmit && src[candidate-1] == src[s-1] {
candidate--
s--
}
// Bail if we exceed the maximum size.
if d+(s-nextEmit) > dstLimit {
return 0
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
d += emitLiteral(dst[d:], src[nextEmit:s])
if debug && nextEmit != s {
fmt.Println("emitted ", s-nextEmit, "literals")
}
// Call emitCopy, and then see if another emitCopy could be our next
// move. Repeat until we find no match for the input immediately after
// what was consumed by the last emitCopy call.
//
// If we exit this loop normally then we need to call emitLiteral next,
// though we don't yet know how big the literal will be. We handle that
// by proceeding to the next iteration of the main loop. We also can
// exit this loop via goto if we get close to exhausting the input.
for {
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
base := s
repeat = base - candidate
// Extend the 4-byte match as long as possible.
s += 4
candidate += 4
for s <= len(src)-8 {
if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidate += 8
}
d += emitCopy(dst[d:], repeat, s-base)
if debug {
// Validate match.
if s <= candidate {
panic("s <= candidate")
}
a := src[base:s]
b := src[base-repeat : base-repeat+(s-base)]
if !bytes.Equal(a, b) {
panic("mismatch")
}
}
if debug {
fmt.Println("emitted src copy, length", s-base, "offset:", repeat, "s:", s)
}
nextEmit = s
if s >= sLimit {
break searchDict
}
if d > dstLimit {
// Do we have space for more, if not bail.
return 0
}
// Check for an immediate match, otherwise start search at s+1
x := load64(src, s-2)
m2Hash := hash6(x, tableBits)
currHash := hash6(x>>16, tableBits)
candidate = int(table[currHash])
table[m2Hash] = uint32(s - 2)
table[currHash] = uint32(s)
if debug && s == candidate {
panic("s == candidate")
}
if uint32(x>>16) != load32(src, candidate) {
cv = load64(src, s+1)
s++
break
}
}
}
// Search without dict:
if repeat > s {
repeat = 0
}
// No more dict
sLimit = len(src) - inputMargin
if s >= sLimit {
goto emitRemainder
}
if debug {
fmt.Println("non-dict matching at", s, "repeat:", repeat)
}
cv = load64(src, s)
if debug {
fmt.Println("now", s, "->", sLimit, "out:", d, "left:", len(src)-s, "nextemit:", nextEmit, "dstLimit:", dstLimit, "s:", s)
}
for {
candidate := 0
for {
// Next src position to check
nextS := s + (s-nextEmit)>>6 + 4
if nextS > sLimit {
goto emitRemainder
}
hash0 := hash6(cv, tableBits)
hash1 := hash6(cv>>8, tableBits)
candidate = int(table[hash0])
candidate2 := int(table[hash1])
table[hash0] = uint32(s)
table[hash1] = uint32(s + 1)
hash2 := hash6(cv>>16, tableBits)
// Check repeat at offset checkRep.
const checkRep = 1
if repeat > 0 && uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) {
base := s + checkRep
// Extend back
for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; {
i--
base--
}
d += emitLiteral(dst[d:], src[nextEmit:base])
if debug && nextEmit != base {
fmt.Println("emitted ", base-nextEmit, "literals")
}
// Extend forward
candidate := s - repeat + 4 + checkRep
s += 4 + checkRep
for s <= sLimit {
if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidate += 8
}
if debug {
// Validate match.
if s <= candidate {
panic("s <= candidate")
}
a := src[base:s]
b := src[base-repeat : base-repeat+(s-base)]
if !bytes.Equal(a, b) {
panic("mismatch")
}
}
if nextEmit > 0 {
// same as `add := emitCopy(dst[d:], repeat, s-base)` but skips storing offset.
d += emitRepeat(dst[d:], repeat, s-base)
} else {
// First match, cannot be repeat.
d += emitCopy(dst[d:], repeat, s-base)
}
if debug {
fmt.Println("emitted src repeat length", s-base, "offset:", repeat, "s:", s)
}
nextEmit = s
if s >= sLimit {
goto emitRemainder
}
cv = load64(src, s)
continue
}
if uint32(cv) == load32(src, candidate) {
break
}
candidate = int(table[hash2])
if uint32(cv>>8) == load32(src, candidate2) {
table[hash2] = uint32(s + 2)
candidate = candidate2
s++
break
}
table[hash2] = uint32(s + 2)
if uint32(cv>>16) == load32(src, candidate) {
s += 2
break
}
cv = load64(src, nextS)
s = nextS
}
// Extend backwards.
// The top bytes will be rechecked to get the full match.
for candidate > 0 && s > nextEmit && src[candidate-1] == src[s-1] {
candidate--
s--
}
// Bail if we exceed the maximum size.
if d+(s-nextEmit) > dstLimit {
return 0
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
d += emitLiteral(dst[d:], src[nextEmit:s])
if debug && nextEmit != s {
fmt.Println("emitted ", s-nextEmit, "literals")
}
// Call emitCopy, and then see if another emitCopy could be our next
// move. Repeat until we find no match for the input immediately after
// what was consumed by the last emitCopy call.
//
// If we exit this loop normally then we need to call emitLiteral next,
// though we don't yet know how big the literal will be. We handle that
// by proceeding to the next iteration of the main loop. We also can
// exit this loop via goto if we get close to exhausting the input.
for {
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
base := s
repeat = base - candidate
// Extend the 4-byte match as long as possible.
s += 4
candidate += 4
for s <= len(src)-8 {
if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidate += 8
}
d += emitCopy(dst[d:], repeat, s-base)
if debug {
// Validate match.
if s <= candidate {
panic("s <= candidate")
}
a := src[base:s]
b := src[base-repeat : base-repeat+(s-base)]
if !bytes.Equal(a, b) {
panic("mismatch")
}
}
if debug {
fmt.Println("emitted src copy, length", s-base, "offset:", repeat, "s:", s)
}
nextEmit = s
if s >= sLimit {
goto emitRemainder
}
if d > dstLimit {
// Do we have space for more, if not bail.
return 0
}
// Check for an immediate match, otherwise start search at s+1
x := load64(src, s-2)
m2Hash := hash6(x, tableBits)
currHash := hash6(x>>16, tableBits)
candidate = int(table[currHash])
table[m2Hash] = uint32(s - 2)
table[currHash] = uint32(s)
if debug && s == candidate {
panic("s == candidate")
}
if uint32(x>>16) != load32(src, candidate) {
cv = load64(src, s+1)
s++
break
}
}
}
emitRemainder:
if nextEmit < len(src) {
// Bail if we exceed the maximum size.
if d+len(src)-nextEmit > dstLimit {
return 0
}
d += emitLiteral(dst[d:], src[nextEmit:])
if debug && nextEmit != s {
fmt.Println("emitted ", len(src)-nextEmit, "literals")
}
}
return d
}

2
vendor/github.com/klauspost/compress/s2/encode_amd64.go generated vendored
View File

@ -3,6 +3,8 @@
package s2
const hasAmd64Asm = true
// encodeBlock encodes a non-empty src to a guaranteed-large-enough dst. It
// assumes that the varint-encoded length of the decompressed bytes has already
// been written.

190
vendor/github.com/klauspost/compress/s2/encode_best.go generated vendored
View File

@ -7,6 +7,7 @@ package s2
import (
"fmt"
"math"
"math/bits"
)
@ -18,7 +19,7 @@ import (
//
// len(dst) >= MaxEncodedLen(len(src)) &&
// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize
func encodeBlockBest(dst, src []byte) (d int) {
func encodeBlockBest(dst, src []byte, dict *Dict) (d int) {
// Initialize the hash tables.
const (
// Long hash matches.
@ -30,6 +31,8 @@ func encodeBlockBest(dst, src []byte) (d int) {
maxSTableSize = 1 << sTableBits
inputMargin = 8 + 2
debug = false
)
// sLimit is when to stop looking for offset/length copies. The inputMargin
@ -39,6 +42,10 @@ func encodeBlockBest(dst, src []byte) (d int) {
if len(src) < minNonLiteralBlockSize {
return 0
}
sLimitDict := len(src) - inputMargin
if sLimitDict > MaxDictSrcOffset-inputMargin {
sLimitDict = MaxDictSrcOffset - inputMargin
}
var lTable [maxLTableSize]uint64
var sTable [maxSTableSize]uint64
@ -52,10 +59,15 @@ func encodeBlockBest(dst, src []byte) (d int) {
// The encoded form must start with a literal, as there are no previous
// bytes to copy, so we start looking for hash matches at s == 1.
s := 1
repeat := 1
if dict != nil {
dict.initBest()
s = 0
repeat = len(dict.dict) - dict.repeat
}
cv := load64(src, s)
// We search for a repeat at -1, but don't output repeats when nextEmit == 0
repeat := 1
const lowbitMask = 0xffffffff
getCur := func(x uint64) int {
return int(x & lowbitMask)
@ -67,11 +79,11 @@ func encodeBlockBest(dst, src []byte) (d int) {
for {
type match struct {
offset int
s int
length int
score int
rep bool
offset int
s int
length int
score int
rep, dict bool
}
var best match
for {
@ -85,6 +97,12 @@ func encodeBlockBest(dst, src []byte) (d int) {
if nextS > sLimit {
goto emitRemainder
}
if dict != nil && s >= MaxDictSrcOffset {
dict = nil
if repeat > s {
repeat = math.MinInt32
}
}
hashL := hash8(cv, lTableBits)
hashS := hash4(cv, sTableBits)
candidateL := lTable[hashL]
@ -114,7 +132,15 @@ func encodeBlockBest(dst, src []byte) (d int) {
}
m := match{offset: offset, s: s, length: 4 + offset, rep: rep}
s += 4
for s <= sLimit {
for s < len(src) {
if len(src)-s < 8 {
if src[s] == src[m.length] {
m.length++
s++
continue
}
break
}
if diff := load64(src, s) ^ load64(src, m.length); diff != 0 {
m.length += bits.TrailingZeros64(diff) >> 3
break
@ -130,6 +156,62 @@ func encodeBlockBest(dst, src []byte) (d int) {
}
return m
}
matchDict := func(candidate, s int, first uint32, rep bool) match {
// Calculate offset as if in continuous array with s
offset := -len(dict.dict) + candidate
if best.length != 0 && best.s-best.offset == s-offset && !rep {
// Don't retest if we have the same offset.
return match{offset: offset, s: s}
}
if load32(dict.dict, candidate) != first {
return match{offset: offset, s: s}
}
m := match{offset: offset, s: s, length: 4 + candidate, rep: rep, dict: true}
s += 4
if !rep {
for s < sLimitDict && m.length < len(dict.dict) {
if len(src)-s < 8 || len(dict.dict)-m.length < 8 {
if src[s] == dict.dict[m.length] {
m.length++
s++
continue
}
break
}
if diff := load64(src, s) ^ load64(dict.dict, m.length); diff != 0 {
m.length += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
m.length += 8
}
} else {
for s < len(src) && m.length < len(dict.dict) {
if len(src)-s < 8 || len(dict.dict)-m.length < 8 {
if src[s] == dict.dict[m.length] {
m.length++
s++
continue
}
break
}
if diff := load64(src, s) ^ load64(dict.dict, m.length); diff != 0 {
m.length += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
m.length += 8
}
}
m.length -= candidate
m.score = score(m)
if m.score <= -m.s {
// Eliminate if no savings, we might find a better one.
m.length = 0
}
return m
}
bestOf := func(a, b match) match {
if b.length == 0 {
@ -146,35 +228,82 @@ func encodeBlockBest(dst, src []byte) (d int) {
return b
}
best = bestOf(matchAt(getCur(candidateL), s, uint32(cv), false), matchAt(getPrev(candidateL), s, uint32(cv), false))
best = bestOf(best, matchAt(getCur(candidateS), s, uint32(cv), false))
best = bestOf(best, matchAt(getPrev(candidateS), s, uint32(cv), false))
if s > 0 {
best = bestOf(matchAt(getCur(candidateL), s, uint32(cv), false), matchAt(getPrev(candidateL), s, uint32(cv), false))
best = bestOf(best, matchAt(getCur(candidateS), s, uint32(cv), false))
best = bestOf(best, matchAt(getPrev(candidateS), s, uint32(cv), false))
}
if dict != nil {
candidateL := dict.bestTableLong[hashL]
candidateS := dict.bestTableShort[hashS]
best = bestOf(best, matchDict(int(candidateL&0xffff), s, uint32(cv), false))
best = bestOf(best, matchDict(int(candidateL>>16), s, uint32(cv), false))
best = bestOf(best, matchDict(int(candidateS&0xffff), s, uint32(cv), false))
best = bestOf(best, matchDict(int(candidateS>>16), s, uint32(cv), false))
}
{
best = bestOf(best, matchAt(s-repeat+1, s+1, uint32(cv>>8), true))
if (dict == nil || repeat <= s) && repeat > 0 {
best = bestOf(best, matchAt(s-repeat+1, s+1, uint32(cv>>8), true))
} else if s-repeat < -4 && dict != nil {
candidate := len(dict.dict) - (repeat - s)
best = bestOf(best, matchDict(candidate, s, uint32(cv), true))
candidate++
best = bestOf(best, matchDict(candidate, s+1, uint32(cv>>8), true))
}
if best.length > 0 {
hashS := hash4(cv>>8, sTableBits)
// s+1
nextShort := sTable[hash4(cv>>8, sTableBits)]
nextShort := sTable[hashS]
s := s + 1
cv := load64(src, s)
nextLong := lTable[hash8(cv, lTableBits)]
hashL := hash8(cv, lTableBits)
nextLong := lTable[hashL]
best = bestOf(best, matchAt(getCur(nextShort), s, uint32(cv), false))
best = bestOf(best, matchAt(getPrev(nextShort), s, uint32(cv), false))
best = bestOf(best, matchAt(getCur(nextLong), s, uint32(cv), false))
best = bestOf(best, matchAt(getPrev(nextLong), s, uint32(cv), false))
// Repeat at + 2
best = bestOf(best, matchAt(s-repeat+1, s+1, uint32(cv>>8), true))
// Dict at + 1
if dict != nil {
candidateL := dict.bestTableLong[hashL]
candidateS := dict.bestTableShort[hashS]
best = bestOf(best, matchDict(int(candidateL&0xffff), s, uint32(cv), false))
best = bestOf(best, matchDict(int(candidateS&0xffff), s, uint32(cv), false))
}
// s+2
if true {
nextShort = sTable[hash4(cv>>8, sTableBits)]
hashS := hash4(cv>>8, sTableBits)
nextShort = sTable[hashS]
s++
cv = load64(src, s)
nextLong = lTable[hash8(cv, lTableBits)]
hashL := hash8(cv, lTableBits)
nextLong = lTable[hashL]
if (dict == nil || repeat <= s) && repeat > 0 {
// Repeat at + 2
best = bestOf(best, matchAt(s-repeat, s, uint32(cv), true))
} else if repeat-s > 4 && dict != nil {
candidate := len(dict.dict) - (repeat - s)
best = bestOf(best, matchDict(candidate, s, uint32(cv), true))
}
best = bestOf(best, matchAt(getCur(nextShort), s, uint32(cv), false))
best = bestOf(best, matchAt(getPrev(nextShort), s, uint32(cv), false))
best = bestOf(best, matchAt(getCur(nextLong), s, uint32(cv), false))
best = bestOf(best, matchAt(getPrev(nextLong), s, uint32(cv), false))
// Dict at +2
// Very small gain
if dict != nil {
candidateL := dict.bestTableLong[hashL]
candidateS := dict.bestTableShort[hashS]
best = bestOf(best, matchDict(int(candidateL&0xffff), s, uint32(cv), false))
best = bestOf(best, matchDict(int(candidateS&0xffff), s, uint32(cv), false))
}
}
// Search for a match at best match end, see if that is better.
// Allow some bytes at the beginning to mismatch.
@ -227,7 +356,7 @@ func encodeBlockBest(dst, src []byte) (d int) {
// Extend backwards, not needed for repeats...
s = best.s
if !best.rep {
if !best.rep && !best.dict {
for best.offset > 0 && s > nextEmit && src[best.offset-1] == src[s-1] {
best.offset--
best.length++
@ -244,7 +373,6 @@ func encodeBlockBest(dst, src []byte) (d int) {
base := s
offset := s - best.offset
s += best.length
if offset > 65535 && s-base <= 5 && !best.rep {
@ -256,16 +384,28 @@ func encodeBlockBest(dst, src []byte) (d int) {
cv = load64(src, s)
continue
}
if debug && nextEmit != base {
fmt.Println("EMIT", base-nextEmit, "literals. base-after:", base)
}
d += emitLiteral(dst[d:], src[nextEmit:base])
if best.rep {
if nextEmit > 0 {
if nextEmit > 0 || best.dict {
if debug {
fmt.Println("REPEAT, length", best.length, "offset:", offset, "s-after:", s, "dict:", best.dict, "best:", best)
}
// same as `add := emitCopy(dst[d:], repeat, s-base)` but skips storing offset.
d += emitRepeat(dst[d:], offset, best.length)
} else {
// First match, cannot be repeat.
// First match without dict cannot be a repeat.
if debug {
fmt.Println("COPY, length", best.length, "offset:", offset, "s-after:", s, "dict:", best.dict, "best:", best)
}
d += emitCopy(dst[d:], offset, best.length)
}
} else {
if debug {
fmt.Println("COPY, length", best.length, "offset:", offset, "s-after:", s, "dict:", best.dict, "best:", best)
}
d += emitCopy(dst[d:], offset, best.length)
}
repeat = offset
@ -296,6 +436,9 @@ emitRemainder:
if d+len(src)-nextEmit > dstLimit {
return 0
}
if debug && nextEmit != s {
fmt.Println("emitted ", len(src)-nextEmit, "literals")
}
d += emitLiteral(dst[d:], src[nextEmit:])
}
return d
@ -642,7 +785,6 @@ func emitRepeatSize(offset, length int) int {
left := 0
if length > maxRepeat {
left = length - maxRepeat + 4
length = maxRepeat - 4
}
if left > 0 {
return 5 + emitRepeatSize(offset, left)

622
vendor/github.com/klauspost/compress/s2/encode_better.go generated vendored
View File

@ -6,6 +6,8 @@
package s2
import (
"bytes"
"fmt"
"math/bits"
)
@ -476,3 +478,623 @@ emitRemainder:
}
return d
}
// encodeBlockBetterDict encodes a non-empty src to a guaranteed-large-enough dst. It
// assumes that the varint-encoded length of the decompressed bytes has already
// been written.
//
// It also assumes that:
//
// len(dst) >= MaxEncodedLen(len(src)) &&
// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize
func encodeBlockBetterDict(dst, src []byte, dict *Dict) (d int) {
// sLimit is when to stop looking for offset/length copies. The inputMargin
// lets us use a fast path for emitLiteral in the main loop, while we are
// looking for copies.
// Initialize the hash tables.
const (
// Long hash matches.
lTableBits = 17
maxLTableSize = 1 << lTableBits
// Short hash matches.
sTableBits = 14
maxSTableSize = 1 << sTableBits
maxAhead = 8 // maximum bytes ahead without checking sLimit
debug = false
)
sLimit := len(src) - inputMargin
if sLimit > MaxDictSrcOffset-maxAhead {
sLimit = MaxDictSrcOffset - maxAhead
}
if len(src) < minNonLiteralBlockSize {
return 0
}
dict.initBetter()
var lTable [maxLTableSize]uint32
var sTable [maxSTableSize]uint32
// Bail if we can't compress to at least this.
dstLimit := len(src) - len(src)>>5 - 6
// nextEmit is where in src the next emitLiteral should start from.
nextEmit := 0
// The encoded form must start with a literal, as there are no previous
// bytes to copy, so we start looking for hash matches at s == 1.
s := 0
cv := load64(src, s)
// We initialize repeat to 0, so we never match on first attempt
repeat := len(dict.dict) - dict.repeat
// While in dict
searchDict:
for {
candidateL := 0
nextS := 0
for {
// Next src position to check
nextS = s + (s-nextEmit)>>7 + 1
if nextS > sLimit {
break searchDict
}
hashL := hash7(cv, lTableBits)
hashS := hash4(cv, sTableBits)
candidateL = int(lTable[hashL])
candidateS := int(sTable[hashS])
dictL := int(dict.betterTableLong[hashL])
dictS := int(dict.betterTableShort[hashS])
lTable[hashL] = uint32(s)
sTable[hashS] = uint32(s)
valLong := load64(src, candidateL)
valShort := load64(src, candidateS)
// If long matches at least 8 bytes, use that.
if s != 0 {
if cv == valLong {
goto emitMatch
}
if cv == valShort {
candidateL = candidateS
goto emitMatch
}
}
// Check dict repeat.
if repeat >= s+4 {
candidate := len(dict.dict) - repeat + s
if candidate > 0 && uint32(cv) == load32(dict.dict, candidate) {
// Extend back
base := s
for i := candidate; base > nextEmit && i > 0 && dict.dict[i-1] == src[base-1]; {
i--
base--
}
d += emitLiteral(dst[d:], src[nextEmit:base])
if debug && nextEmit != base {
fmt.Println("emitted ", base-nextEmit, "literals")
}
s += 4
candidate += 4
for candidate < len(dict.dict)-8 && s <= len(src)-8 {
if diff := load64(src, s) ^ load64(dict.dict, candidate); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidate += 8
}
d += emitRepeat(dst[d:], repeat, s-base)
if debug {
fmt.Println("emitted dict repeat length", s-base, "offset:", repeat, "s:", s)
}
nextEmit = s
if s >= sLimit {
break searchDict
}
cv = load64(src, s)
// Index in-between
index0 := base + 1
index1 := s - 2
cv = load64(src, s)
for index0 < index1 {
cv0 := load64(src, index0)
cv1 := load64(src, index1)
lTable[hash7(cv0, lTableBits)] = uint32(index0)
sTable[hash4(cv0>>8, sTableBits)] = uint32(index0 + 1)
lTable[hash7(cv1, lTableBits)] = uint32(index1)
sTable[hash4(cv1>>8, sTableBits)] = uint32(index1 + 1)
index0 += 2
index1 -= 2
}
continue
}
}
// Don't try to find match at s==0
if s == 0 {
cv = load64(src, nextS)
s = nextS
continue
}
// Long likely matches 7, so take that.
if uint32(cv) == uint32(valLong) {
goto emitMatch
}
// Long dict...
if uint32(cv) == load32(dict.dict, dictL) {
candidateL = dictL
goto emitDict
}
// Check our short candidate
if uint32(cv) == uint32(valShort) {
// Try a long candidate at s+1
hashL = hash7(cv>>8, lTableBits)
candidateL = int(lTable[hashL])
lTable[hashL] = uint32(s + 1)
if uint32(cv>>8) == load32(src, candidateL) {
s++
goto emitMatch
}
// Use our short candidate.
candidateL = candidateS
goto emitMatch
}
if uint32(cv) == load32(dict.dict, dictS) {
// Try a long candidate at s+1
hashL = hash7(cv>>8, lTableBits)
candidateL = int(lTable[hashL])
lTable[hashL] = uint32(s + 1)
if uint32(cv>>8) == load32(src, candidateL) {
s++
goto emitMatch
}
candidateL = dictS
goto emitDict
}
cv = load64(src, nextS)
s = nextS
}
emitDict:
{
if debug {
if load32(dict.dict, candidateL) != load32(src, s) {
panic("dict emit mismatch")
}
}
// Extend backwards.
// The top bytes will be rechecked to get the full match.
for candidateL > 0 && s > nextEmit && dict.dict[candidateL-1] == src[s-1] {
candidateL--
s--
}
// Bail if we exceed the maximum size.
if d+(s-nextEmit) > dstLimit {
return 0
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
d += emitLiteral(dst[d:], src[nextEmit:s])
if debug && nextEmit != s {
fmt.Println("emitted ", s-nextEmit, "literals")
}
{
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
base := s
offset := s + (len(dict.dict)) - candidateL
// Extend the 4-byte match as long as possible.
s += 4
candidateL += 4
for s <= len(src)-8 && len(dict.dict)-candidateL >= 8 {
if diff := load64(src, s) ^ load64(dict.dict, candidateL); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidateL += 8
}
if repeat == offset {
if debug {
fmt.Println("emitted dict repeat, length", s-base, "offset:", offset, "s:", s, "dict offset:", candidateL)
}
d += emitRepeat(dst[d:], offset, s-base)
} else {
if debug {
fmt.Println("emitted dict copy, length", s-base, "offset:", offset, "s:", s, "dict offset:", candidateL)
}
// Matches longer than 64 are split.
if s <= sLimit || s-base < 8 {
d += emitCopy(dst[d:], offset, s-base)
} else {
// Split to ensure we don't start a copy within next block.
d += emitCopy(dst[d:], offset, 4)
d += emitRepeat(dst[d:], offset, s-base-4)
}
repeat = offset
}
if false {
// Validate match.
if s <= candidateL {
panic("s <= candidate")
}
a := src[base:s]
b := dict.dict[base-repeat : base-repeat+(s-base)]
if !bytes.Equal(a, b) {
panic("mismatch")
}
}
nextEmit = s
if s >= sLimit {
break searchDict
}
if d > dstLimit {
// Do we have space for more, if not bail.
return 0
}
// Index short & long
index0 := base + 1
index1 := s - 2
cv0 := load64(src, index0)
cv1 := load64(src, index1)
lTable[hash7(cv0, lTableBits)] = uint32(index0)
sTable[hash4(cv0>>8, sTableBits)] = uint32(index0 + 1)
lTable[hash7(cv1, lTableBits)] = uint32(index1)
sTable[hash4(cv1>>8, sTableBits)] = uint32(index1 + 1)
index0 += 1
index1 -= 1
cv = load64(src, s)
// index every second long in between.
for index0 < index1 {
lTable[hash7(load64(src, index0), lTableBits)] = uint32(index0)
lTable[hash7(load64(src, index1), lTableBits)] = uint32(index1)
index0 += 2
index1 -= 2
}
}
continue
}
emitMatch:
// Extend backwards
for candidateL > 0 && s > nextEmit && src[candidateL-1] == src[s-1] {
candidateL--
s--
}
// Bail if we exceed the maximum size.
if d+(s-nextEmit) > dstLimit {
return 0
}
base := s
offset := base - candidateL
// Extend the 4-byte match as long as possible.
s += 4
candidateL += 4
for s < len(src) {
if len(src)-s < 8 {
if src[s] == src[candidateL] {
s++
candidateL++
continue
}
break
}
if diff := load64(src, s) ^ load64(src, candidateL); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidateL += 8
}
if offset > 65535 && s-base <= 5 && repeat != offset {
// Bail if the match is equal or worse to the encoding.
s = nextS + 1
if s >= sLimit {
goto emitRemainder
}
cv = load64(src, s)
continue
}
d += emitLiteral(dst[d:], src[nextEmit:base])
if debug && nextEmit != s {
fmt.Println("emitted ", s-nextEmit, "literals")
}
if repeat == offset {
if debug {
fmt.Println("emitted match repeat, length", s-base, "offset:", offset, "s:", s)
}
d += emitRepeat(dst[d:], offset, s-base)
} else {
if debug {
fmt.Println("emitted match copy, length", s-base, "offset:", offset, "s:", s)
}
d += emitCopy(dst[d:], offset, s-base)
repeat = offset
}
nextEmit = s
if s >= sLimit {
goto emitRemainder
}
if d > dstLimit {
// Do we have space for more, if not bail.
return 0
}
// Index short & long
index0 := base + 1
index1 := s - 2
cv0 := load64(src, index0)
cv1 := load64(src, index1)
lTable[hash7(cv0, lTableBits)] = uint32(index0)
sTable[hash4(cv0>>8, sTableBits)] = uint32(index0 + 1)
lTable[hash7(cv1, lTableBits)] = uint32(index1)
sTable[hash4(cv1>>8, sTableBits)] = uint32(index1 + 1)
index0 += 1
index1 -= 1
cv = load64(src, s)
// index every second long in between.
for index0 < index1 {
lTable[hash7(load64(src, index0), lTableBits)] = uint32(index0)
lTable[hash7(load64(src, index1), lTableBits)] = uint32(index1)
index0 += 2
index1 -= 2
}
}
// Search without dict:
if repeat > s {
repeat = 0
}
// No more dict
sLimit = len(src) - inputMargin
if s >= sLimit {
goto emitRemainder
}
cv = load64(src, s)
if debug {
fmt.Println("now", s, "->", sLimit, "out:", d, "left:", len(src)-s, "nextemit:", nextEmit, "dstLimit:", dstLimit, "s:", s)
}
for {
candidateL := 0
nextS := 0
for {
// Next src position to check
nextS = s + (s-nextEmit)>>7 + 1
if nextS > sLimit {
goto emitRemainder
}
hashL := hash7(cv, lTableBits)
hashS := hash4(cv, sTableBits)
candidateL = int(lTable[hashL])
candidateS := int(sTable[hashS])
lTable[hashL] = uint32(s)
sTable[hashS] = uint32(s)
valLong := load64(src, candidateL)
valShort := load64(src, candidateS)
// If long matches at least 8 bytes, use that.
if cv == valLong {
break
}
if cv == valShort {
candidateL = candidateS
break
}
// Check repeat at offset checkRep.
const checkRep = 1
// Minimum length of a repeat. Tested with various values.
// While 4-5 offers improvements in some, 6 reduces
// regressions significantly.
const wantRepeatBytes = 6
const repeatMask = ((1 << (wantRepeatBytes * 8)) - 1) << (8 * checkRep)
if false && repeat > 0 && cv&repeatMask == load64(src, s-repeat)&repeatMask {
base := s + checkRep
// Extend back
for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; {
i--
base--
}
d += emitLiteral(dst[d:], src[nextEmit:base])
// Extend forward
candidate := s - repeat + wantRepeatBytes + checkRep
s += wantRepeatBytes + checkRep
for s < len(src) {
if len(src)-s < 8 {
if src[s] == src[candidate] {
s++
candidate++
continue
}
break
}
if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidate += 8
}
// same as `add := emitCopy(dst[d:], repeat, s-base)` but skips storing offset.
d += emitRepeat(dst[d:], repeat, s-base)
nextEmit = s
if s >= sLimit {
goto emitRemainder
}
// Index in-between
index0 := base + 1
index1 := s - 2
cv = load64(src, s)
for index0 < index1 {
cv0 := load64(src, index0)
cv1 := load64(src, index1)
lTable[hash7(cv0, lTableBits)] = uint32(index0)
sTable[hash4(cv0>>8, sTableBits)] = uint32(index0 + 1)
lTable[hash7(cv1, lTableBits)] = uint32(index1)
sTable[hash4(cv1>>8, sTableBits)] = uint32(index1 + 1)
index0 += 2
index1 -= 2
}
cv = load64(src, s)
continue
}
// Long likely matches 7, so take that.
if uint32(cv) == uint32(valLong) {
break
}
// Check our short candidate
if uint32(cv) == uint32(valShort) {
// Try a long candidate at s+1
hashL = hash7(cv>>8, lTableBits)
candidateL = int(lTable[hashL])
lTable[hashL] = uint32(s + 1)
if uint32(cv>>8) == load32(src, candidateL) {
s++
break
}
// Use our short candidate.
candidateL = candidateS
break
}
cv = load64(src, nextS)
s = nextS
}
// Extend backwards
for candidateL > 0 && s > nextEmit && src[candidateL-1] == src[s-1] {
candidateL--
s--
}
// Bail if we exceed the maximum size.
if d+(s-nextEmit) > dstLimit {
return 0
}
base := s
offset := base - candidateL
// Extend the 4-byte match as long as possible.
s += 4
candidateL += 4
for s < len(src) {
if len(src)-s < 8 {
if src[s] == src[candidateL] {
s++
candidateL++
continue
}
break
}
if diff := load64(src, s) ^ load64(src, candidateL); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidateL += 8
}
if offset > 65535 && s-base <= 5 && repeat != offset {
// Bail if the match is equal or worse to the encoding.
s = nextS + 1
if s >= sLimit {
goto emitRemainder
}
cv = load64(src, s)
continue
}
d += emitLiteral(dst[d:], src[nextEmit:base])
if repeat == offset {
d += emitRepeat(dst[d:], offset, s-base)
} else {
d += emitCopy(dst[d:], offset, s-base)
repeat = offset
}
nextEmit = s
if s >= sLimit {
goto emitRemainder
}
if d > dstLimit {
// Do we have space for more, if not bail.
return 0
}
// Index short & long
index0 := base + 1
index1 := s - 2
cv0 := load64(src, index0)
cv1 := load64(src, index1)
lTable[hash7(cv0, lTableBits)] = uint32(index0)
sTable[hash4(cv0>>8, sTableBits)] = uint32(index0 + 1)
lTable[hash7(cv1, lTableBits)] = uint32(index1)
sTable[hash4(cv1>>8, sTableBits)] = uint32(index1 + 1)
index0 += 1
index1 -= 1
cv = load64(src, s)
// index every second long in between.
for index0 < index1 {
lTable[hash7(load64(src, index0), lTableBits)] = uint32(index0)
lTable[hash7(load64(src, index1), lTableBits)] = uint32(index1)
index0 += 2
index1 -= 2
}
}
emitRemainder:
if nextEmit < len(src) {
// Bail if we exceed the maximum size.
if d+len(src)-nextEmit > dstLimit {
return 0
}
d += emitLiteral(dst[d:], src[nextEmit:])
}
return d
}

413
vendor/github.com/klauspost/compress/s2/encode_go.go generated vendored
View File

@ -4,9 +4,12 @@
package s2
import (
"bytes"
"math/bits"
)
const hasAmd64Asm = false
// encodeBlock encodes a non-empty src to a guaranteed-large-enough dst. It
// assumes that the varint-encoded length of the decompressed bytes has already
// been written.
@ -312,3 +315,413 @@ func matchLen(a []byte, b []byte) int {
}
return len(a) + checked
}
func calcBlockSize(src []byte) (d int) {
// Initialize the hash table.
const (
tableBits = 13
maxTableSize = 1 << tableBits
)
var table [maxTableSize]uint32
// sLimit is when to stop looking for offset/length copies. The inputMargin
// lets us use a fast path for emitLiteral in the main loop, while we are
// looking for copies.
sLimit := len(src) - inputMargin
// Bail if we can't compress to at least this.
dstLimit := len(src) - len(src)>>5 - 5
// nextEmit is where in src the next emitLiteral should start from.
nextEmit := 0
// The encoded form must start with a literal, as there are no previous
// bytes to copy, so we start looking for hash matches at s == 1.
s := 1
cv := load64(src, s)
// We search for a repeat at -1, but don't output repeats when nextEmit == 0
repeat := 1
for {
candidate := 0
for {
// Next src position to check
nextS := s + (s-nextEmit)>>6 + 4
if nextS > sLimit {
goto emitRemainder
}
hash0 := hash6(cv, tableBits)
hash1 := hash6(cv>>8, tableBits)
candidate = int(table[hash0])
candidate2 := int(table[hash1])
table[hash0] = uint32(s)
table[hash1] = uint32(s + 1)
hash2 := hash6(cv>>16, tableBits)
// Check repeat at offset checkRep.
const checkRep = 1
if uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) {
base := s + checkRep
// Extend back
for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; {
i--
base--
}
d += emitLiteralSize(src[nextEmit:base])
// Extend forward
candidate := s - repeat + 4 + checkRep
s += 4 + checkRep
for s <= sLimit {
if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidate += 8
}
d += emitCopyNoRepeatSize(repeat, s-base)
nextEmit = s
if s >= sLimit {
goto emitRemainder
}
cv = load64(src, s)
continue
}
if uint32(cv) == load32(src, candidate) {
break
}
candidate = int(table[hash2])
if uint32(cv>>8) == load32(src, candidate2) {
table[hash2] = uint32(s + 2)
candidate = candidate2
s++
break
}
table[hash2] = uint32(s + 2)
if uint32(cv>>16) == load32(src, candidate) {
s += 2
break
}
cv = load64(src, nextS)
s = nextS
}
// Extend backwards
for candidate > 0 && s > nextEmit && src[candidate-1] == src[s-1] {
candidate--
s--
}
// Bail if we exceed the maximum size.
if d+(s-nextEmit) > dstLimit {
return 0
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
d += emitLiteralSize(src[nextEmit:s])
// Call emitCopy, and then see if another emitCopy could be our next
// move. Repeat until we find no match for the input immediately after
// what was consumed by the last emitCopy call.
//
// If we exit this loop normally then we need to call emitLiteral next,
// though we don't yet know how big the literal will be. We handle that
// by proceeding to the next iteration of the main loop. We also can
// exit this loop via goto if we get close to exhausting the input.
for {
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
base := s
repeat = base - candidate
// Extend the 4-byte match as long as possible.
s += 4
candidate += 4
for s <= len(src)-8 {
if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidate += 8
}
d += emitCopyNoRepeatSize(repeat, s-base)
if false {
// Validate match.
a := src[base:s]
b := src[base-repeat : base-repeat+(s-base)]
if !bytes.Equal(a, b) {
panic("mismatch")
}
}
nextEmit = s
if s >= sLimit {
goto emitRemainder
}
if d > dstLimit {
// Do we have space for more, if not bail.
return 0
}
// Check for an immediate match, otherwise start search at s+1
x := load64(src, s-2)
m2Hash := hash6(x, tableBits)
currHash := hash6(x>>16, tableBits)
candidate = int(table[currHash])
table[m2Hash] = uint32(s - 2)
table[currHash] = uint32(s)
if uint32(x>>16) != load32(src, candidate) {
cv = load64(src, s+1)
s++
break
}
}
}
emitRemainder:
if nextEmit < len(src) {
// Bail if we exceed the maximum size.
if d+len(src)-nextEmit > dstLimit {
return 0
}
d += emitLiteralSize(src[nextEmit:])
}
return d
}
func calcBlockSizeSmall(src []byte) (d int) {
// Initialize the hash table.
const (
tableBits = 9
maxTableSize = 1 << tableBits
)
var table [maxTableSize]uint32
// sLimit is when to stop looking for offset/length copies. The inputMargin
// lets us use a fast path for emitLiteral in the main loop, while we are
// looking for copies.
sLimit := len(src) - inputMargin
// Bail if we can't compress to at least this.
dstLimit := len(src) - len(src)>>5 - 5
// nextEmit is where in src the next emitLiteral should start from.
nextEmit := 0
// The encoded form must start with a literal, as there are no previous
// bytes to copy, so we start looking for hash matches at s == 1.
s := 1
cv := load64(src, s)
// We search for a repeat at -1, but don't output repeats when nextEmit == 0
repeat := 1
for {
candidate := 0
for {
// Next src position to check
nextS := s + (s-nextEmit)>>6 + 4
if nextS > sLimit {
goto emitRemainder
}
hash0 := hash6(cv, tableBits)
hash1 := hash6(cv>>8, tableBits)
candidate = int(table[hash0])
candidate2 := int(table[hash1])
table[hash0] = uint32(s)
table[hash1] = uint32(s + 1)
hash2 := hash6(cv>>16, tableBits)
// Check repeat at offset checkRep.
const checkRep = 1
if uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) {
base := s + checkRep
// Extend back
for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; {
i--
base--
}
d += emitLiteralSize(src[nextEmit:base])
// Extend forward
candidate := s - repeat + 4 + checkRep
s += 4 + checkRep
for s <= sLimit {
if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidate += 8
}
d += emitCopyNoRepeatSize(repeat, s-base)
nextEmit = s
if s >= sLimit {
goto emitRemainder
}
cv = load64(src, s)
continue
}
if uint32(cv) == load32(src, candidate) {
break
}
candidate = int(table[hash2])
if uint32(cv>>8) == load32(src, candidate2) {
table[hash2] = uint32(s + 2)
candidate = candidate2
s++
break
}
table[hash2] = uint32(s + 2)
if uint32(cv>>16) == load32(src, candidate) {
s += 2
break
}
cv = load64(src, nextS)
s = nextS
}
// Extend backwards
for candidate > 0 && s > nextEmit && src[candidate-1] == src[s-1] {
candidate--
s--
}
// Bail if we exceed the maximum size.
if d+(s-nextEmit) > dstLimit {
return 0
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
d += emitLiteralSize(src[nextEmit:s])
// Call emitCopy, and then see if another emitCopy could be our next
// move. Repeat until we find no match for the input immediately after
// what was consumed by the last emitCopy call.
//
// If we exit this loop normally then we need to call emitLiteral next,
// though we don't yet know how big the literal will be. We handle that
// by proceeding to the next iteration of the main loop. We also can
// exit this loop via goto if we get close to exhausting the input.
for {
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
base := s
repeat = base - candidate
// Extend the 4-byte match as long as possible.
s += 4
candidate += 4
for s <= len(src)-8 {
if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
s += bits.TrailingZeros64(diff) >> 3
break
}
s += 8
candidate += 8
}
d += emitCopyNoRepeatSize(repeat, s-base)
if false {
// Validate match.
a := src[base:s]
b := src[base-repeat : base-repeat+(s-base)]
if !bytes.Equal(a, b) {
panic("mismatch")
}
}
nextEmit = s
if s >= sLimit {
goto emitRemainder
}
if d > dstLimit {
// Do we have space for more, if not bail.
return 0
}
// Check for an immediate match, otherwise start search at s+1
x := load64(src, s-2)
m2Hash := hash6(x, tableBits)
currHash := hash6(x>>16, tableBits)
candidate = int(table[currHash])
table[m2Hash] = uint32(s - 2)
table[currHash] = uint32(s)
if uint32(x>>16) != load32(src, candidate) {
cv = load64(src, s+1)
s++
break
}
}
}
emitRemainder:
if nextEmit < len(src) {
// Bail if we exceed the maximum size.
if d+len(src)-nextEmit > dstLimit {
return 0
}
d += emitLiteralSize(src[nextEmit:])
}
return d
}
// emitLiteral writes a literal chunk and returns the number of bytes written.
//
// It assumes that:
//
// dst is long enough to hold the encoded bytes
// 0 <= len(lit) && len(lit) <= math.MaxUint32
func emitLiteralSize(lit []byte) int {
if len(lit) == 0 {
return 0
}
switch {
case len(lit) <= 60:
return len(lit) + 1
case len(lit) <= 1<<8:
return len(lit) + 2
case len(lit) <= 1<<16:
return len(lit) + 3
case len(lit) <= 1<<24:
return len(lit) + 4
default:
return len(lit) + 5
}
}
func cvtLZ4BlockAsm(dst []byte, src []byte) (uncompressed int, dstUsed int) {
panic("cvtLZ4BlockAsm should be unreachable")
}
func cvtLZ4BlockSnappyAsm(dst []byte, src []byte) (uncompressed int, dstUsed int) {
panic("cvtLZ4BlockSnappyAsm should be unreachable")
}
func cvtLZ4sBlockAsm(dst []byte, src []byte) (uncompressed int, dstUsed int) {
panic("cvtLZ4sBlockAsm should be unreachable")
}
func cvtLZ4sBlockSnappyAsm(dst []byte, src []byte) (uncompressed int, dstUsed int) {
panic("cvtLZ4sBlockSnappyAsm should be unreachable")
}

34
vendor/github.com/klauspost/compress/s2/encodeblock_amd64.go generated vendored
View File

@ -146,6 +146,20 @@ func encodeSnappyBetterBlockAsm10B(dst []byte, src []byte) int
//go:noescape
func encodeSnappyBetterBlockAsm8B(dst []byte, src []byte) int
// calcBlockSize encodes a non-empty src to a guaranteed-large-enough dst.
// Maximum input 4294967295 bytes.
// It assumes that the varint-encoded length of the decompressed bytes has already been written.
//
//go:noescape
func calcBlockSize(src []byte) int
// calcBlockSizeSmall encodes a non-empty src to a guaranteed-large-enough dst.
// Maximum input 1024 bytes.
// It assumes that the varint-encoded length of the decompressed bytes has already been written.
//
//go:noescape
func calcBlockSizeSmall(src []byte) int
// emitLiteral writes a literal chunk and returns the number of bytes written.
//
// It assumes that:
@ -192,3 +206,23 @@ func emitCopyNoRepeat(dst []byte, offset int, length int) int
//
//go:noescape
func matchLen(a []byte, b []byte) int
// cvtLZ4Block converts an LZ4 block to S2
//
//go:noescape
func cvtLZ4BlockAsm(dst []byte, src []byte) (uncompressed int, dstUsed int)
// cvtLZ4sBlock converts an LZ4s block to S2
//
//go:noescape
func cvtLZ4sBlockAsm(dst []byte, src []byte) (uncompressed int, dstUsed int)
// cvtLZ4Block converts an LZ4 block to Snappy
//
//go:noescape
func cvtLZ4BlockSnappyAsm(dst []byte, src []byte) (uncompressed int, dstUsed int)
// cvtLZ4sBlock converts an LZ4s block to Snappy
//
//go:noescape
func cvtLZ4sBlockSnappyAsm(dst []byte, src []byte) (uncompressed int, dstUsed int)

18843
vendor/github.com/klauspost/compress/s2/encodeblock_amd64.s generated vendored
View File

File diff suppressed because it is too large Load Diff

585
vendor/github.com/klauspost/compress/s2/lz4convert.go generated vendored Normal file
View File

@ -0,0 +1,585 @@
// Copyright (c) 2022 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.
package s2
import (
"encoding/binary"
"errors"
"fmt"
)
// LZ4Converter provides conversion from LZ4 blocks as defined here:
// https://github.com/lz4/lz4/blob/dev/doc/lz4_Block_format.md
type LZ4Converter struct {
}
// ErrDstTooSmall is returned when provided destination is too small.
var ErrDstTooSmall = errors.New("s2: destination too small")
// ConvertBlock will convert an LZ4 block and append it as an S2
// block without block length to dst.
// The uncompressed size is returned as well.
// dst must have capacity to contain the entire compressed block.
func (l *LZ4Converter) ConvertBlock(dst, src []byte) ([]byte, int, error) {
if len(src) == 0 {
return dst, 0, nil
}
const debug = false
const inline = true
const lz4MinMatch = 4
s, d := 0, len(dst)
dst = dst[:cap(dst)]
if !debug && hasAmd64Asm {
res, sz := cvtLZ4BlockAsm(dst[d:], src)
if res < 0 {
const (
errCorrupt = -1
errDstTooSmall = -2
)
switch res {
case errCorrupt:
return nil, 0, ErrCorrupt
case errDstTooSmall:
return nil, 0, ErrDstTooSmall
default:
return nil, 0, fmt.Errorf("unexpected result: %d", res)
}
}
if d+sz > len(dst) {
return nil, 0, ErrDstTooSmall
}
return dst[:d+sz], res, nil
}
dLimit := len(dst) - 10
var lastOffset uint16
var uncompressed int
if debug {
fmt.Printf("convert block start: len(src): %d, len(dst):%d \n", len(src), len(dst))
}
for {
if s >= len(src) {
return dst[:d], 0, ErrCorrupt
}
// Read literal info
token := src[s]
ll := int(token >> 4)
ml := int(lz4MinMatch + (token & 0xf))
// If upper nibble is 15, literal length is extended
if token >= 0xf0 {
for {
s++
if s >= len(src) {
if debug {
fmt.Printf("error reading ll: s (%d) >= len(src) (%d)\n", s, len(src))
}
return dst[:d], 0, ErrCorrupt
}
val := src[s]
ll += int(val)
if val != 255 {
break
}
}
}
// Skip past token
if s+ll >= len(src) {
if debug {
fmt.Printf("error literals: s+ll (%d+%d) >= len(src) (%d)\n", s, ll, len(src))
}
return nil, 0, ErrCorrupt
}
s++
if ll > 0 {
if d+ll > dLimit {
return nil, 0, ErrDstTooSmall
}
if debug {
fmt.Printf("emit %d literals\n", ll)
}
d += emitLiteralGo(dst[d:], src[s:s+ll])
s += ll
uncompressed += ll
}
// Check if we are done...
if s == len(src) && ml == lz4MinMatch {
break
}
// 2 byte offset
if s >= len(src)-2 {
if debug {
fmt.Printf("s (%d) >= len(src)-2 (%d)", s, len(src)-2)
}
return nil, 0, ErrCorrupt
}
offset := binary.LittleEndian.Uint16(src[s:])
s += 2
if offset == 0 {
if debug {
fmt.Printf("error: offset 0, ml: %d, len(src)-s: %d\n", ml, len(src)-s)
}
return nil, 0, ErrCorrupt
}
if int(offset) > uncompressed {
if debug {
fmt.Printf("error: offset (%d)> uncompressed (%d)\n", offset, uncompressed)
}
return nil, 0, ErrCorrupt
}
if ml == lz4MinMatch+15 {
for {
if s >= len(src) {
if debug {
fmt.Printf("error reading ml: s (%d) >= len(src) (%d)\n", s, len(src))
}
return nil, 0, ErrCorrupt
}
val := src[s]
s++
ml += int(val)
if val != 255 {
if s >= len(src) {
if debug {
fmt.Printf("error reading ml: s (%d) >= len(src) (%d)\n", s, len(src))
}
return nil, 0, ErrCorrupt
}
break
}
}
}
if offset == lastOffset {
if debug {
fmt.Printf("emit repeat, length: %d, offset: %d\n", ml, offset)
}
if !inline {
d += emitRepeat16(dst[d:], offset, ml)
} else {
length := ml
dst := dst[d:]
for len(dst) > 5 {
// Repeat offset, make length cheaper
length -= 4
if length <= 4 {
dst[0] = uint8(length)<<2 | tagCopy1
dst[1] = 0
d += 2
break
}
if length < 8 && offset < 2048 {
// Encode WITH offset
dst[1] = uint8(offset)
dst[0] = uint8(offset>>8)<<5 | uint8(length)<<2 | tagCopy1
d += 2
break
}
if length < (1<<8)+4 {
length -= 4
dst[2] = uint8(length)
dst[1] = 0
dst[0] = 5<<2 | tagCopy1
d += 3
break
}
if length < (1<<16)+(1<<8) {
length -= 1 << 8
dst[3] = uint8(length >> 8)
dst[2] = uint8(length >> 0)
dst[1] = 0
dst[0] = 6<<2 | tagCopy1
d += 4
break
}
const maxRepeat = (1 << 24) - 1
length -= 1 << 16
left := 0
if length > maxRepeat {
left = length - maxRepeat + 4
length = maxRepeat - 4
}
dst[4] = uint8(length >> 16)
dst[3] = uint8(length >> 8)
dst[2] = uint8(length >> 0)
dst[1] = 0
dst[0] = 7<<2 | tagCopy1
if left > 0 {
d += 5 + emitRepeat16(dst[5:], offset, left)
break
}
d += 5
break
}
}
} else {
if debug {
fmt.Printf("emit copy, length: %d, offset: %d\n", ml, offset)
}
if !inline {
d += emitCopy16(dst[d:], offset, ml)
} else {
length := ml
dst := dst[d:]
for len(dst) > 5 {
// Offset no more than 2 bytes.
if length > 64 {
off := 3
if offset < 2048 {
// emit 8 bytes as tagCopy1, rest as repeats.
dst[1] = uint8(offset)
dst[0] = uint8(offset>>8)<<5 | uint8(8-4)<<2 | tagCopy1
length -= 8
off = 2
} else {
// Emit a length 60 copy, encoded as 3 bytes.
// Emit remaining as repeat value (minimum 4 bytes).
dst[2] = uint8(offset >> 8)
dst[1] = uint8(offset)
dst[0] = 59<<2 | tagCopy2
length -= 60
}
// Emit remaining as repeats, at least 4 bytes remain.
d += off + emitRepeat16(dst[off:], offset, length)
break
}
if length >= 12 || offset >= 2048 {
// Emit the remaining copy, encoded as 3 bytes.
dst[2] = uint8(offset >> 8)
dst[1] = uint8(offset)
dst[0] = uint8(length-1)<<2 | tagCopy2
d += 3
break
}
// Emit the remaining copy, encoded as 2 bytes.
dst[1] = uint8(offset)
dst[0] = uint8(offset>>8)<<5 | uint8(length-4)<<2 | tagCopy1
d += 2
break
}
}
lastOffset = offset
}
uncompressed += ml
if d > dLimit {
return nil, 0, ErrDstTooSmall
}
}
return dst[:d], uncompressed, nil
}
// ConvertBlockSnappy will convert an LZ4 block and append it
// as a Snappy block without block length to dst.
// The uncompressed size is returned as well.
// dst must have capacity to contain the entire compressed block.
func (l *LZ4Converter) ConvertBlockSnappy(dst, src []byte) ([]byte, int, error) {
if len(src) == 0 {
return dst, 0, nil
}
const debug = false
const lz4MinMatch = 4
s, d := 0, len(dst)
dst = dst[:cap(dst)]
// Use assembly when possible
if !debug && hasAmd64Asm {
res, sz := cvtLZ4BlockSnappyAsm(dst[d:], src)
if res < 0 {
const (
errCorrupt = -1
errDstTooSmall = -2
)
switch res {
case errCorrupt:
return nil, 0, ErrCorrupt
case errDstTooSmall:
return nil, 0, ErrDstTooSmall
default:
return nil, 0, fmt.Errorf("unexpected result: %d", res)
}
}
if d+sz > len(dst) {
return nil, 0, ErrDstTooSmall
}
return dst[:d+sz], res, nil
}
dLimit := len(dst) - 10
var uncompressed int
if debug {
fmt.Printf("convert block start: len(src): %d, len(dst):%d \n", len(src), len(dst))
}
for {
if s >= len(src) {
return nil, 0, ErrCorrupt
}
// Read literal info
token := src[s]
ll := int(token >> 4)
ml := int(lz4MinMatch + (token & 0xf))
// If upper nibble is 15, literal length is extended
if token >= 0xf0 {
for {
s++
if s >= len(src) {
if debug {
fmt.Printf("error reading ll: s (%d) >= len(src) (%d)\n", s, len(src))
}
return nil, 0, ErrCorrupt
}
val := src[s]
ll += int(val)
if val != 255 {
break
}
}
}
// Skip past token
if s+ll >= len(src) {
if debug {
fmt.Printf("error literals: s+ll (%d+%d) >= len(src) (%d)\n", s, ll, len(src))
}
return nil, 0, ErrCorrupt
}
s++
if ll > 0 {
if d+ll > dLimit {
return nil, 0, ErrDstTooSmall
}
if debug {
fmt.Printf("emit %d literals\n", ll)
}
d += emitLiteralGo(dst[d:], src[s:s+ll])
s += ll
uncompressed += ll
}
// Check if we are done...
if s == len(src) && ml == lz4MinMatch {
break
}
// 2 byte offset
if s >= len(src)-2 {
if debug {
fmt.Printf("s (%d) >= len(src)-2 (%d)", s, len(src)-2)
}
return nil, 0, ErrCorrupt
}
offset := binary.LittleEndian.Uint16(src[s:])
s += 2
if offset == 0 {
if debug {
fmt.Printf("error: offset 0, ml: %d, len(src)-s: %d\n", ml, len(src)-s)
}
return nil, 0, ErrCorrupt
}
if int(offset) > uncompressed {
if debug {
fmt.Printf("error: offset (%d)> uncompressed (%d)\n", offset, uncompressed)
}
return nil, 0, ErrCorrupt
}
if ml == lz4MinMatch+15 {
for {
if s >= len(src) {
if debug {
fmt.Printf("error reading ml: s (%d) >= len(src) (%d)\n", s, len(src))
}
return nil, 0, ErrCorrupt
}
val := src[s]
s++
ml += int(val)
if val != 255 {
if s >= len(src) {
if debug {
fmt.Printf("error reading ml: s (%d) >= len(src) (%d)\n", s, len(src))
}
return nil, 0, ErrCorrupt
}
break
}
}
}
if debug {
fmt.Printf("emit copy, length: %d, offset: %d\n", ml, offset)
}
length := ml
// d += emitCopyNoRepeat(dst[d:], int(offset), ml)
for length > 0 {
if d >= dLimit {
return nil, 0, ErrDstTooSmall
}
// Offset no more than 2 bytes.
if length > 64 {
// Emit a length 64 copy, encoded as 3 bytes.
dst[d+2] = uint8(offset >> 8)
dst[d+1] = uint8(offset)
dst[d+0] = 63<<2 | tagCopy2
length -= 64
d += 3
continue
}
if length >= 12 || offset >= 2048 || length < 4 {
// Emit the remaining copy, encoded as 3 bytes.
dst[d+2] = uint8(offset >> 8)
dst[d+1] = uint8(offset)
dst[d+0] = uint8(length-1)<<2 | tagCopy2
d += 3
break
}
// Emit the remaining copy, encoded as 2 bytes.
dst[d+1] = uint8(offset)
dst[d+0] = uint8(offset>>8)<<5 | uint8(length-4)<<2 | tagCopy1
d += 2
break
}
uncompressed += ml
if d > dLimit {
return nil, 0, ErrDstTooSmall
}
}
return dst[:d], uncompressed, nil
}
// emitRepeat writes a repeat chunk and returns the number of bytes written.
// Length must be at least 4 and < 1<<24
func emitRepeat16(dst []byte, offset uint16, length int) int {
// Repeat offset, make length cheaper
length -= 4
if length <= 4 {
dst[0] = uint8(length)<<2 | tagCopy1
dst[1] = 0
return 2
}
if length < 8 && offset < 2048 {
// Encode WITH offset
dst[1] = uint8(offset)
dst[0] = uint8(offset>>8)<<5 | uint8(length)<<2 | tagCopy1
return 2
}
if length < (1<<8)+4 {
length -= 4
dst[2] = uint8(length)
dst[1] = 0
dst[0] = 5<<2 | tagCopy1
return 3
}
if length < (1<<16)+(1<<8) {
length -= 1 << 8
dst[3] = uint8(length >> 8)
dst[2] = uint8(length >> 0)
dst[1] = 0
dst[0] = 6<<2 | tagCopy1
return 4
}
const maxRepeat = (1 << 24) - 1
length -= 1 << 16
left := 0
if length > maxRepeat {
left = length - maxRepeat + 4
length = maxRepeat - 4
}
dst[4] = uint8(length >> 16)
dst[3] = uint8(length >> 8)
dst[2] = uint8(length >> 0)
dst[1] = 0
dst[0] = 7<<2 | tagCopy1
if left > 0 {
return 5 + emitRepeat16(dst[5:], offset, left)
}
return 5
}
// emitCopy writes a copy chunk and returns the number of bytes written.
//
// It assumes that:
//
// dst is long enough to hold the encoded bytes
// 1 <= offset && offset <= math.MaxUint16
// 4 <= length && length <= math.MaxUint32
func emitCopy16(dst []byte, offset uint16, length int) int {
// Offset no more than 2 bytes.
if length > 64 {
off := 3
if offset < 2048 {
// emit 8 bytes as tagCopy1, rest as repeats.
dst[1] = uint8(offset)
dst[0] = uint8(offset>>8)<<5 | uint8(8-4)<<2 | tagCopy1
length -= 8
off = 2
} else {
// Emit a length 60 copy, encoded as 3 bytes.
// Emit remaining as repeat value (minimum 4 bytes).
dst[2] = uint8(offset >> 8)
dst[1] = uint8(offset)
dst[0] = 59<<2 | tagCopy2
length -= 60
}
// Emit remaining as repeats, at least 4 bytes remain.
return off + emitRepeat16(dst[off:], offset, length)
}
if length >= 12 || offset >= 2048 {
// Emit the remaining copy, encoded as 3 bytes.
dst[2] = uint8(offset >> 8)
dst[1] = uint8(offset)
dst[0] = uint8(length-1)<<2 | tagCopy2
return 3
}
// Emit the remaining copy, encoded as 2 bytes.
dst[1] = uint8(offset)
dst[0] = uint8(offset>>8)<<5 | uint8(length-4)<<2 | tagCopy1
return 2
}
// emitLiteral writes a literal chunk and returns the number of bytes written.
//
// It assumes that:
//
// dst is long enough to hold the encoded bytes
// 0 <= len(lit) && len(lit) <= math.MaxUint32
func emitLiteralGo(dst, lit []byte) int {
if len(lit) == 0 {
return 0
}
i, n := 0, uint(len(lit)-1)
switch {
case n < 60:
dst[0] = uint8(n)<<2 | tagLiteral
i = 1
case n < 1<<8:
dst[1] = uint8(n)
dst[0] = 60<<2 | tagLiteral
i = 2
case n < 1<<16:
dst[2] = uint8(n >> 8)
dst[1] = uint8(n)
dst[0] = 61<<2 | tagLiteral
i = 3
case n < 1<<24:
dst[3] = uint8(n >> 16)
dst[2] = uint8(n >> 8)
dst[1] = uint8(n)
dst[0] = 62<<2 | tagLiteral
i = 4
default:
dst[4] = uint8(n >> 24)
dst[3] = uint8(n >> 16)
dst[2] = uint8(n >> 8)
dst[1] = uint8(n)
dst[0] = 63<<2 | tagLiteral
i = 5
}
return i + copy(dst[i:], lit)
}

467
vendor/github.com/klauspost/compress/s2/lz4sconvert.go generated vendored Normal file
View File

@ -0,0 +1,467 @@
// Copyright (c) 2022 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.
package s2
import (
"encoding/binary"
"fmt"
)
// LZ4sConverter provides conversion from LZ4s.
// (Intel modified LZ4 Blocks)
// https://cdrdv2-public.intel.com/743912/743912-qat-programmers-guide-v2.0.pdf
// LZ4s is a variant of LZ4 block format. LZ4s should be considered as an intermediate compressed block format.
// The LZ4s format is selected when the application sets the compType to CPA_DC_LZ4S in CpaDcSessionSetupData.
// The LZ4s block returned by the Intel® QAT hardware can be used by an external
// software post-processing to generate other compressed data formats.
// The following table lists the differences between LZ4 and LZ4s block format. LZ4s block format uses
// the same high-level formatting as LZ4 block format with the following encoding changes:
// For Min Match of 4 bytes, Copy length value 1-15 means length 4-18 with 18 bytes adding an extra byte.
// ONLY "Min match of 4 bytes" is supported.
type LZ4sConverter struct {
}
// ConvertBlock will convert an LZ4s block and append it as an S2
// block without block length to dst.
// The uncompressed size is returned as well.
// dst must have capacity to contain the entire compressed block.
func (l *LZ4sConverter) ConvertBlock(dst, src []byte) ([]byte, int, error) {
if len(src) == 0 {
return dst, 0, nil
}
const debug = false
const inline = true
const lz4MinMatch = 3
s, d := 0, len(dst)
dst = dst[:cap(dst)]
if !debug && hasAmd64Asm {
res, sz := cvtLZ4sBlockAsm(dst[d:], src)
if res < 0 {
const (
errCorrupt = -1
errDstTooSmall = -2
)
switch res {
case errCorrupt:
return nil, 0, ErrCorrupt
case errDstTooSmall:
return nil, 0, ErrDstTooSmall
default:
return nil, 0, fmt.Errorf("unexpected result: %d", res)
}
}
if d+sz > len(dst) {
return nil, 0, ErrDstTooSmall
}
return dst[:d+sz], res, nil
}
dLimit := len(dst) - 10
var lastOffset uint16
var uncompressed int
if debug {
fmt.Printf("convert block start: len(src): %d, len(dst):%d \n", len(src), len(dst))
}
for {
if s >= len(src) {
return dst[:d], 0, ErrCorrupt
}
// Read literal info
token := src[s]
ll := int(token >> 4)
ml := int(lz4MinMatch + (token & 0xf))
// If upper nibble is 15, literal length is extended
if token >= 0xf0 {
for {
s++
if s >= len(src) {
if debug {
fmt.Printf("error reading ll: s (%d) >= len(src) (%d)\n", s, len(src))
}
return dst[:d], 0, ErrCorrupt
}
val := src[s]
ll += int(val)
if val != 255 {
break
}
}
}
// Skip past token
if s+ll >= len(src) {
if debug {
fmt.Printf("error literals: s+ll (%d+%d) >= len(src) (%d)\n", s, ll, len(src))
}
return nil, 0, ErrCorrupt
}
s++
if ll > 0 {
if d+ll > dLimit {
return nil, 0, ErrDstTooSmall
}
if debug {
fmt.Printf("emit %d literals\n", ll)
}
d += emitLiteralGo(dst[d:], src[s:s+ll])
s += ll
uncompressed += ll
}
// Check if we are done...
if ml == lz4MinMatch {
if s == len(src) {
break
}
// 0 bytes.
continue
}
// 2 byte offset
if s >= len(src)-2 {
if debug {
fmt.Printf("s (%d) >= len(src)-2 (%d)", s, len(src)-2)
}
return nil, 0, ErrCorrupt
}
offset := binary.LittleEndian.Uint16(src[s:])
s += 2
if offset == 0 {
if debug {
fmt.Printf("error: offset 0, ml: %d, len(src)-s: %d\n", ml, len(src)-s)
}
return nil, 0, ErrCorrupt
}
if int(offset) > uncompressed {
if debug {
fmt.Printf("error: offset (%d)> uncompressed (%d)\n", offset, uncompressed)
}
return nil, 0, ErrCorrupt
}
if ml == lz4MinMatch+15 {
for {
if s >= len(src) {
if debug {
fmt.Printf("error reading ml: s (%d) >= len(src) (%d)\n", s, len(src))
}
return nil, 0, ErrCorrupt
}
val := src[s]
s++
ml += int(val)
if val != 255 {
if s >= len(src) {
if debug {
fmt.Printf("error reading ml: s (%d) >= len(src) (%d)\n", s, len(src))
}
return nil, 0, ErrCorrupt
}
break
}
}
}
if offset == lastOffset {
if debug {
fmt.Printf("emit repeat, length: %d, offset: %d\n", ml, offset)
}
if !inline {
d += emitRepeat16(dst[d:], offset, ml)
} else {
length := ml
dst := dst[d:]
for len(dst) > 5 {
// Repeat offset, make length cheaper
length -= 4
if length <= 4 {
dst[0] = uint8(length)<<2 | tagCopy1
dst[1] = 0
d += 2
break
}
if length < 8 && offset < 2048 {
// Encode WITH offset
dst[1] = uint8(offset)
dst[0] = uint8(offset>>8)<<5 | uint8(length)<<2 | tagCopy1
d += 2
break
}
if length < (1<<8)+4 {
length -= 4
dst[2] = uint8(length)
dst[1] = 0
dst[0] = 5<<2 | tagCopy1
d += 3
break
}
if length < (1<<16)+(1<<8) {
length -= 1 << 8
dst[3] = uint8(length >> 8)
dst[2] = uint8(length >> 0)
dst[1] = 0
dst[0] = 6<<2 | tagCopy1
d += 4
break
}
const maxRepeat = (1 << 24) - 1
length -= 1 << 16
left := 0
if length > maxRepeat {
left = length - maxRepeat + 4
length = maxRepeat - 4
}
dst[4] = uint8(length >> 16)
dst[3] = uint8(length >> 8)
dst[2] = uint8(length >> 0)
dst[1] = 0
dst[0] = 7<<2 | tagCopy1
if left > 0 {
d += 5 + emitRepeat16(dst[5:], offset, left)
break
}
d += 5
break
}
}
} else {
if debug {
fmt.Printf("emit copy, length: %d, offset: %d\n", ml, offset)
}
if !inline {
d += emitCopy16(dst[d:], offset, ml)
} else {
length := ml
dst := dst[d:]
for len(dst) > 5 {
// Offset no more than 2 bytes.
if length > 64 {
off := 3
if offset < 2048 {
// emit 8 bytes as tagCopy1, rest as repeats.
dst[1] = uint8(offset)
dst[0] = uint8(offset>>8)<<5 | uint8(8-4)<<2 | tagCopy1
length -= 8
off = 2
} else {
// Emit a length 60 copy, encoded as 3 bytes.
// Emit remaining as repeat value (minimum 4 bytes).
dst[2] = uint8(offset >> 8)
dst[1] = uint8(offset)
dst[0] = 59<<2 | tagCopy2
length -= 60
}
// Emit remaining as repeats, at least 4 bytes remain.
d += off + emitRepeat16(dst[off:], offset, length)
break
}
if length >= 12 || offset >= 2048 {
// Emit the remaining copy, encoded as 3 bytes.
dst[2] = uint8(offset >> 8)
dst[1] = uint8(offset)
dst[0] = uint8(length-1)<<2 | tagCopy2
d += 3
break
}
// Emit the remaining copy, encoded as 2 bytes.
dst[1] = uint8(offset)
dst[0] = uint8(offset>>8)<<5 | uint8(length-4)<<2 | tagCopy1
d += 2
break
}
}
lastOffset = offset
}
uncompressed += ml
if d > dLimit {
return nil, 0, ErrDstTooSmall
}
}
return dst[:d], uncompressed, nil
}
// ConvertBlockSnappy will convert an LZ4s block and append it
// as a Snappy block without block length to dst.
// The uncompressed size is returned as well.
// dst must have capacity to contain the entire compressed block.
func (l *LZ4sConverter) ConvertBlockSnappy(dst, src []byte) ([]byte, int, error) {
if len(src) == 0 {
return dst, 0, nil
}
const debug = false
const lz4MinMatch = 3
s, d := 0, len(dst)
dst = dst[:cap(dst)]
// Use assembly when possible
if !debug && hasAmd64Asm {
res, sz := cvtLZ4sBlockSnappyAsm(dst[d:], src)
if res < 0 {
const (
errCorrupt = -1
errDstTooSmall = -2
)
switch res {
case errCorrupt:
return nil, 0, ErrCorrupt
case errDstTooSmall:
return nil, 0, ErrDstTooSmall
default:
return nil, 0, fmt.Errorf("unexpected result: %d", res)
}
}
if d+sz > len(dst) {
return nil, 0, ErrDstTooSmall
}
return dst[:d+sz], res, nil
}
dLimit := len(dst) - 10
var uncompressed int
if debug {
fmt.Printf("convert block start: len(src): %d, len(dst):%d \n", len(src), len(dst))
}
for {
if s >= len(src) {
return nil, 0, ErrCorrupt
}
// Read literal info
token := src[s]
ll := int(token >> 4)
ml := int(lz4MinMatch + (token & 0xf))
// If upper nibble is 15, literal length is extended
if token >= 0xf0 {
for {
s++
if s >= len(src) {
if debug {
fmt.Printf("error reading ll: s (%d) >= len(src) (%d)\n", s, len(src))
}
return nil, 0, ErrCorrupt
}
val := src[s]
ll += int(val)
if val != 255 {
break
}
}
}
// Skip past token
if s+ll >= len(src) {
if debug {
fmt.Printf("error literals: s+ll (%d+%d) >= len(src) (%d)\n", s, ll, len(src))
}
return nil, 0, ErrCorrupt
}
s++
if ll > 0 {
if d+ll > dLimit {
return nil, 0, ErrDstTooSmall
}
if debug {
fmt.Printf("emit %d literals\n", ll)
}
d += emitLiteralGo(dst[d:], src[s:s+ll])
s += ll
uncompressed += ll
}
// Check if we are done...
if ml == lz4MinMatch {
if s == len(src) {
break
}
// 0 bytes.
continue
}
// 2 byte offset
if s >= len(src)-2 {
if debug {
fmt.Printf("s (%d) >= len(src)-2 (%d)", s, len(src)-2)
}
return nil, 0, ErrCorrupt
}
offset := binary.LittleEndian.Uint16(src[s:])
s += 2
if offset == 0 {
if debug {
fmt.Printf("error: offset 0, ml: %d, len(src)-s: %d\n", ml, len(src)-s)
}
return nil, 0, ErrCorrupt
}
if int(offset) > uncompressed {
if debug {
fmt.Printf("error: offset (%d)> uncompressed (%d)\n", offset, uncompressed)
}
return nil, 0, ErrCorrupt
}
if ml == lz4MinMatch+15 {
for {
if s >= len(src) {
if debug {
fmt.Printf("error reading ml: s (%d) >= len(src) (%d)\n", s, len(src))
}
return nil, 0, ErrCorrupt
}
val := src[s]
s++
ml += int(val)
if val != 255 {
if s >= len(src) {
if debug {
fmt.Printf("error reading ml: s (%d) >= len(src) (%d)\n", s, len(src))
}
return nil, 0, ErrCorrupt
}
break
}
}
}
if debug {
fmt.Printf("emit copy, length: %d, offset: %d\n", ml, offset)
}
length := ml
// d += emitCopyNoRepeat(dst[d:], int(offset), ml)
for length > 0 {
if d >= dLimit {
return nil, 0, ErrDstTooSmall
}
// Offset no more than 2 bytes.
if length > 64 {
// Emit a length 64 copy, encoded as 3 bytes.
dst[d+2] = uint8(offset >> 8)
dst[d+1] = uint8(offset)
dst[d+0] = 63<<2 | tagCopy2
length -= 64
d += 3
continue
}
if length >= 12 || offset >= 2048 || length < 4 {
// Emit the remaining copy, encoded as 3 bytes.
dst[d+2] = uint8(offset >> 8)
dst[d+1] = uint8(offset)
dst[d+0] = uint8(length-1)<<2 | tagCopy2
d += 3
break
}
// Emit the remaining copy, encoded as 2 bytes.
dst[d+1] = uint8(offset)
dst[d+0] = uint8(offset>>8)<<5 | uint8(length-4)<<2 | tagCopy1
d += 2
break
}
uncompressed += ml
if d > dLimit {
return nil, 0, ErrDstTooSmall
}
}
return dst[:d], uncompressed, nil
}