/// SynLZ Compression routines
// - licensed under a MPL/GPL/LGPL tri-license; version 1.18
unit SynLZ;

{
    This file is part of Synopse SynLZ Compression.

    Synopse SynLZ Compression. Copyright (C) 2019 Arnaud Bouchez
      Synopse Informatique - https://synopse.info

  *** BEGIN LICENSE BLOCK *****
  Version: MPL 1.1/GPL 2.0/LGPL 2.1

  The contents of this file are subject to the Mozilla Public License Version
  1.1 (the "License"); you may not use this file except in compliance with
  the License. You may obtain a copy of the License at
  http://www.mozilla.org/MPL

  Software distributed under the License is distributed on an "AS IS" basis,
  WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License
  for the specific language governing rights and limitations under the License.

  The Original Code is Synopse SynLZ Compression.

  The Initial Developer of the Original Code is Arnaud Bouchez.

  Portions created by the Initial Developer are Copyright (C) 2019
  the Initial Developer. All Rights Reserved.

  Contributor(s):

  Alternatively, the contents of this file may be used under the terms of
  either the GNU General Public License Version 2 or later (the "GPL"), or
  the GNU Lesser General Public License Version 2.1 or later (the "LGPL"),
  in which case the provisions of the GPL or the LGPL are applicable instead
  of those above. If you wish to allow use of your version of this file only
  under the terms of either the GPL or the LGPL, and not to allow others to
  use your version of this file under the terms of the MPL, indicate your
  decision by deleting the provisions above and replace them with the notice
  and other provisions required by the GPL or the LGPL. If you do not delete
  the provisions above, a recipient may use your version of this file under
  the terms of any one of the MPL, the GPL or the LGPL.

  ***** END LICENSE BLOCK *****


     SynLZ Compression / Decompression library
     =========================================
      by Arnaud Bouchez http://bouchez.info

    * SynLZ is a very FAST lossless data compression library
      written in optimized pascal code for Delphi 3 up to Delphi 2009
      with a tuned asm version available
    * symetrical compression and decompression speed (which is
      very rare above all other compression algorithms in the wild)
    * good compression rate (usualy better than LZO)
    * fastest averrage compression speed (ideal for xml/text communication, e.g.)

    SynLZ implements a new compression algorithm with the following features:
    * hashing+dictionary compression in one pass, with no huffman table
    * optimized 32bits control word, embedded in the data stream
    * in-memory compression (the dictionary is the input stream itself)
    * compression and decompression have the same speed (both use hashing)
    * thread safe and lossless algorithm
    * supports overlapping compression and in-place decompression
    * code size for compression/decompression functions is smaller than LZO's

    The contents of this file are subject to the Mozilla Public License
    Version 1.1 (the "License"); you may not use this file except in
    compliance with the License. You may obtain a copy of the License at
    http://www.mozilla.org/MPL
    Software distributed under the License is distributed on an "AS IS"
    basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See the
    License for the specific language governing rights and limitations
    under the License.

    The Initial Developer of the Original Code is Arnaud Bouchez.
    This work is Copyright (C)2008 Arnaud Bouchez - http://bouchez.info

    Conversion notes:
    - this format is NOT stream compatible with any lz* official format
       => use it internally in your application, not as exchange format
    - very small code size (less than 1KB for both compressor/decompressor)
    - the uncompressed data length is stored in the beginning of the stream
       and can be retrieved easily for proper out_p memory allocation
    - please give correct data to the decompressor (i.e. first CRC in_p data)
       => we recommend our very fast Adler32 procedure, or a zip-like container
    - a 2nd more tuned algorithm is included, but is somewhat slower in practice
       => use SynLZ[de]compres1*() functions in your applications
    - tested and benchmarked with a lot of data types/sizes
       => use the asm code, which is very tuned: SynLZ[de]compress1asm()
    - tested under Delphi 7, Delphi 2007 and Delphi 2009
    - a hashing limitation makes SynLZ sometimes unable to pack continuous
       blocks of same byte -> SynLZ is perfect for xml/text, but SynLZO
       is prefered for database files
    - if you include it in your application, please give me some credits:
       "use SynLZ compression by http://bouchez.info"
    - use at your own risk!

    Some benchmark on a Sempron computer:
    - compression is 20 times faster than zip, decompression 3 times
    - same compression ratio as lzo algo, but faster (up to 2x) on compression
    - the R and W intermediate speed are at the compressed stream level, i.e.
      the speed which used for disk usage -> you see that SynLZ behaves
      very well for real-time data compression, for backup purpose e.g.
      (a typical SATA disk drive has a speed of 50-70 MB/s)

 KLOG.xml 6034 bytes
 lz1 asm     1287 21.3% R  256 MB/s W   54 MB/s R   71 MB/s W 334 MB/s
 lz1 pas     1287 21.3% R  184 MB/s W   39 MB/s R   58 MB/s W 274 MB/s
 lz2 pas     1274 21.1% R  173 MB/s W   36 MB/s R   57 MB/s W 274 MB/s
   lzo C     1347 22.3% R  185 MB/s W   41 MB/s R  111 MB/s W 501 MB/s
     zip      806 13.4% R   14 MB/s W    1 MB/s R   14 MB/s W 110 MB/s

 MiniLZO.cs 25252 bytes
 lz1 asm     5775 22.9% R  246 MB/s W   56 MB/s R   70 MB/s W 306 MB/s
 lz1 pas     5775 22.9% R  178 MB/s W   40 MB/s R   58 MB/s W 253 MB/s
 lz2 pas     5762 22.8% R  166 MB/s W   37 MB/s R   57 MB/s W 250 MB/s
   lzo C     5846 23.2% R  164 MB/s W   38 MB/s R  103 MB/s W 448 MB/s
     zip     3707 14.7% R   15 MB/s W    2 MB/s R   22 MB/s W 154 MB/s

 TestLZO.exe 158720 bytes
 lz1 asm   110686 69.7% R  127 MB/s W   88 MB/s R   80 MB/s W 115 MB/s
 lz1 pas   110686 69.7% R   98 MB/s W   68 MB/s R   63 MB/s W  90 MB/s
 lz2 pas   109004 68.7% R   88 MB/s W   60 MB/s R   60 MB/s W  88 MB/s
   lzo C   108202 68.2% R   40 MB/s W   27 MB/s R  164 MB/s W 241 MB/s
     zip    88786 55.9% R    5 MB/s W    3 MB/s R   33 MB/s W  60 MB/s

 Browsing.sq3db 46047232 bytes (46MB)
 lz1 asm 19766884 42.9% R  171 MB/s W   73 MB/s R   73 MB/s W 171 MB/s
 lz1 pas 19766884 42.9% R  130 MB/s W   56 MB/s R   59 MB/s W 139 MB/s
 lz2 pas 19707346 42.8% R  123 MB/s W   52 MB/s R   59 MB/s W 139 MB/s
 lzo asm 20629084 44.8% R   89 MB/s W   40 MB/s R  135 MB/s W 302 MB/s
   lzo C 20629083 44.8% R   66 MB/s W   29 MB/s R  145 MB/s W 325 MB/s
     zip 15564126 33.8% R    6 MB/s W    2 MB/s R   30 MB/s W  91 MB/s

 TRKCHG.DBF 4572297 bytes (4MB)
 lz1 asm   265782 5.8% R  430 MB/s W   25 MB/s R   29 MB/s W 510 MB/s
 lz1 pas   265782 5.8% R  296 MB/s W   17 MB/s R   28 MB/s W 483 MB/s
 lz2 pas   274773 6.0% R  258 MB/s W   15 MB/s R   27 MB/s W 450 MB/s
   lzo C   266897 5.8% R  318 MB/s W   18 MB/s R   41 MB/s W 702 MB/s
     zip   158408 3.5% R   25 MB/s W    0 MB/s R   11 MB/s W 318 MB/s

 CATENA5.TXT 6358752 bytes
 lz1 asm  3275269 51.5% R  132 MB/s W   68 MB/s R   66 MB/s W 129 MB/s
 lz1 pas  3275269 51.5% R  103 MB/s W   53 MB/s R   57 MB/s W 112 MB/s
 lz2 pas  3277397 51.5% R   95 MB/s W   49 MB/s R   57 MB/s W 112 MB/s
   lzo C  3289373 51.7% R   63 MB/s W   33 MB/s R   90 MB/s W 175 MB/s
     zip  2029096 31.9% R    4 MB/s W    1 MB/s R   29 MB/s W  91 MB/s


  Benchmark update - introducing LZ4 at http://code.google.com/p/lz4
  190 MB file containing pascal sources, on a Core 2 duo PC, using x86 asm:
   LZ4     compression = 1.25 sec, comp. size = 71 MB, decompression = 0.44 sec
   SynLZ   compression = 1.09 sec, comp. size = 63 MB, decompression = 0.51 sec
   zip (1) compression = 6.44 sec, comp. size = 52 MB, decompression = 1.49 sec
   zip (6) compression = 20.1 sec, comp. size = 42 MB, decompression = 1.35 sec
   Note: zip decompression here uses fast asm optimized version of SynZip.pas
  Decompression is slower in SynLZ, due to the algorithm used: it does recreate
   the hash table even at decompression, while it is not needed by LZ4.
  Having the hash table at hand allows more patterns to be available, so
   compression ratio is better, at the expand of a slower speed.

  Conclusion:
   SynLZ compresses better than LZ4, SynLZ is faster to compress than LZ4,
   but slower to decompress than LZ4. So SynLZ is still very competitive for
   our Client-Server mORMot purpose, since it is a simple pascal unit with
   no external .obj/.o/.dll dependency. ;)

  Updated benchmarks on a Core i7, with the 2017/08 x86 and x64 optimized asm:
    Win32 Processing devpcm.log = 98.7 MB
       Snappy compress in 125.07ms, ratio=84%, 789.3 MB/s
       Snappy uncompress in 70.35ms, 1.3 GB/s
       SynLZ compress in 103.61ms, ratio=93%, 952.8 MB/s
       SynLZ uncompress in 68.71ms, 1.4 GB/s
    Win64 Processing devpcm.log = 98.7 MB
       Snappy compress in 107.13ms, ratio=84%, 921.5 MB/s
       Snappy uncompress in 61.06ms, 1.5 GB/s
       SynLZ compress in 97.25ms, ratio=93%, 1015.1 MB/s
       SynLZ uncompress in 61.27ms, 1.5 GB/s


  Revision history

  Version 1.6
  - first release, associated with the main Synopse SQLite3 framework

  Version 1.13
  - code modifications to compile with Delphi 5 compiler
  - comment refactoring (mostly for inclusion in SynProject documentation)
  - new CompressSynLZ function, for THttpSocket.RegisterCompress - this
    function will return 'synlzo' as "ACCEPT-ENCODING:" HTTP header parameter

  Version 1.15
  - force ignore asm version of the code if PUREPASCAL conditional is defined

  Version 1.16
  - fixed potential GPF issue in Hash32() function

  Version 1.17
  - Use RawByteString type for CompressSynLZ() function prototype

  Version 1.18
  - unit fixed and tested with Delphi XE2 and up 64-bit compiler
  - introducing SynLZCompress1/SynLZDecompress1 low-level functions
  - added SynLZdecompress1partial() function for partial and secure (but slower)
    decompression - implements feature request [82ca067959]
  - removed several compilation hints when assertions are set to off
  - some performance optimization, especially when using a 64bit CPU, thanks
    to a new tuned x64 asm revision, and 8 bytes chunk copy for smallest blocks

}

interface

{$I Synopse.inc}

/// get maximum possible (worse) compressed size for out_p
function SynLZcompressdestlen(in_len: integer): integer;

/// get uncompressed size from lz-compressed buffer (to reserve memory, e.g.)
function SynLZdecompressdestlen(in_p: PAnsiChar): integer;

/// 1st compression algorithm uses hashing with a 32bits control word
function SynLZcompress1pas(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;

/// 1st compression algorithm uses hashing with a 32bits control word
// - this is the fastest pure pascal implementation
function SynLZdecompress1pas(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;

/// 1st compression algorithm uses hashing with a 32bits control word
// - this overload function is slower, but will allow to uncompress only the start
// of the content (e.g. to read some metadata header)
// - it will also check for dst buffer overflow, so will be more secure than
// other functions, which expect the content to be verified (e.g. via CRC)
function SynLZdecompress1partial(src: PAnsiChar; size: integer; dst: PAnsiChar;
  maxDst: integer): integer;

{$ifdef CPUINTEL}
/// optimized x86/x64 asm version of the 1st compression algorithm
function SynLZcompress1(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
/// optimized x86/x64 asm version of the 1st compression algorithm
function SynLZdecompress1(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
{$else}
var
  /// fastest available SynLZ compression (using 1st algorithm)
  SynLZCompress1: function(
    src: PAnsiChar; size: integer; dst: PAnsiChar): integer = SynLZcompress1pas;

  /// fastest available SynLZ decompression (using 1st algorithm)
  SynLZDecompress1: function(
    src: PAnsiChar; size: integer; dst: PAnsiChar): integer = SynLZDecompress1pas;
{$endif CPUINTEL}

/// 2nd compression algorithm optimizing pattern copy
// - this algorithm is a bit smaller, but slower, so the 1st method is preferred
function SynLZcompress2(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
/// 2nd compression algorithm optimizing pattern copy
// - this algorithm is a bit smaller, but slower, so the 1st method is preferred
function SynLZdecompress2(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;


implementation

function SynLZcompressdestlen(in_len: integer): integer;
// get maximum possible (worse) compressed size for out_p
begin
  result := in_len+in_len shr 3+16; // worse case
end;

type
  {$ifndef FPC}
  PtrUInt = {$ifdef CPU64}NativeUInt{$else}cardinal{$endif};
  {$endif}
  {$ifdef DELPHI5OROLDER} // Delphi 5 doesn't have those base types defined :(
  PByte = ^Byte;
  PWord = ^Word;
  PInteger = ^integer;
  PCardinal = ^Cardinal;
  IntegerArray  = array[0..$effffff] of integer;
  PIntegerArray = ^IntegerArray;
  {$endif}
  TOffsets = array[0..4095] of PAnsiChar; // 16KB/32KB hashing code

function SynLZdecompressdestlen(in_p: PAnsiChar): integer;
// get uncompressed size from lz-compressed buffer (to reserve memory, e.g.)
begin
  result := PWord(in_p)^;
  if result and $8000<>0 then
    result := (result and $7fff) or (integer(PWord(in_p+2)^) shl 15);
end;

{$ifdef CPUINTEL}
// using direct x86 jmp also circumvents Internal Error C11715 for Delphi 5
{$ifdef CPUX86}
function SynLZcompress1(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
asm
        push    ebp
        push    ebx
        push    esi
        push    edi
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -32
        mov     esi, eax // esi=src
        mov     edi, ecx // edi=dst
        mov     [esp+08H], ecx
        mov     eax, edx
        cmp     eax, 32768
        jl      @@0889
        or      ax, 8000H
        mov     [edi], eax
        mov     eax, edx
        shr     eax, 15
        mov     [edi+2], eax
        add     edi, 4
        jmp     @@0891
@@0890: mov     eax, 2
        jmp     @@0904
@@0889: mov     [edi], eax
        test    eax, eax
        jz      @@0890
        add     edi, 2
@@0891: lea     eax, [edx+esi]
        mov     [esp+18H], edi
        mov     [esp+0CH], eax
        sub     eax, 11
        mov     [esp+4], eax
        lea     ebx, [esp+24H]
        xor     eax, eax
        mov     ecx, 1024
@@089I: mov     [ebx], eax // faster than FillChar / stosb
        mov     [ebx+4], eax
        mov     [ebx+8], eax
        mov     [ebx+12], eax
        add     ebx, 16
        dec     ecx
        jnz     @@089I
        mov     [edi], eax
        add     edi, 4
        mov     ebx, 1 // ebx=1 shl CWbit
        // main loop:
        cmp     esi, [esp+4]
        ja      @@0900
@@0892: mov     edx, [esi]
        mov     eax, edx
        shr     edx, 12
        xor     edx, eax
        and     edx, 0FFFH
        mov     ebp, [esp+edx*4+24H]
        mov     ecx, [esp+edx*4+4024H]
        mov     [esp+edx*4+24H], esi
        xor     ecx, eax
        test    ecx, 0FFFFFFH
        mov     [esp+edx*4+4024H], eax
        jnz     @@0897
        mov     eax, esi
        or      ebp, ebp
        jz      @@0897
        sub     eax, ebp
        mov     ecx, [esp+18H]
        cmp     eax, 2
        jle     @@0897
        lea     esi, [esi+2]
        or      dword ptr[ecx], ebx
        mov     ecx, [esp+0CH]
        add     ebp, 2
        mov     eax, 1
        sub     ecx, esi
        dec     ecx
        mov     [esp], ecx
        cmp     ecx, 271
        jl      @@0894
        mov     dword ptr [esp], 271
        jmp     @@0894
@@0893: inc     eax
@@0894: mov     ecx, [ebp+eax]
        cmp     cl, [esi+eax]
        jnz     @@0895
        cmp     eax, [esp]
        jge     @@0895
        inc     eax
        cmp     ch, [esi+eax]
        jnz     @@0895
        shr     ecx, 16
        cmp     eax, [esp]
        jge     @@0895
        inc     eax
        cmp     cl, [esi+eax]
        jnz     @@0895
        cmp     eax, [esp]
        jge     @@0895
        inc     eax
        cmp     ch, [esi+eax]
        jnz     @@0895
        cmp     eax, [esp]
        jl      @@0893
@@0895: add     esi, eax
        shl     edx, 4
        cmp     eax, 15
        jg      @@0896
        or      eax, edx
        mov     word ptr [edi], ax
        add     edi, 2
        jmp     @@0898
@@0896: sub     eax, 16
        mov     [edi], dx
        mov     [edi+2H], al
        add     edi, 3
        jmp     @@0898
@@0897: mov     al, [esi] // movsb is actually slower!
        mov     [edi], al
        inc     esi
        inc     edi
@@0898: add     ebx, ebx
        jz      @@0899
        cmp     esi, [esp+4]
        jbe     @@0892
        jmp     @@0900
@@0899: mov     [esp+18H], edi
        mov     [edi], ebx
        inc     ebx
        add     edi, 4
        cmp     esi, [esp+4]
        jbe     @@0892
@@0900: cmp     esi, [esp+0CH]
        jnc     @@0903
@@0901: mov     al, [esi]
        mov     [edi], al
        inc     esi
        inc     edi
        add     ebx, ebx
        jz      @@0902
        cmp     esi, [esp+0CH]
        jc      @@0901
        jmp     @@0903
@@0902: mov     [edi], ebx
        inc     ebx
        add     edi, 4
        cmp     esi, [esp+0CH]
        jc      @@0901
@@0903: mov     eax, edi
        sub     eax, [esp+08H]
@@0904: add     esp, 32804
        pop     edi
        pop     esi
        pop     ebx
        pop     ebp
{$else CPUX86}
function SynLZcompress1(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
var off: TOffsets;
    cache: array[0..4095] of cardinal; // uses 32KB+16KB=48KB on stack
asm // rcx=src, edx=size, r8=dest
        {$ifndef win64} // Linux 64-bit ABI
        mov     r8, rdx
        mov     rdx, rsi
        mov     rcx, rdi
        {$endif win64}
        push    rbx
        push    rdi
        push    rsi
        push    r12
        push    r13
        push    r14
        push    r15
        mov     r15, r8   // r8=dest r15=dst_beg
        mov     rbx, rcx  // rbx=src
        cmp     edx, 32768
        jc      @03
        mov     eax, edx
        and     eax, 7FFFH
        or      eax, 8000H
        mov     word ptr [r8], ax
        mov     eax, edx
        shr     eax, 15
        mov     word ptr [r8+2H], ax
        add     r8, 4
        jmp     @05
@03:    mov     word ptr [r8], dx
        test    edx, edx
        jnz     @04
        mov     r15d, 2
        jmp     @19
        nop
@04:    add     r8, 2
@05:    lea     r9, [rdx+rbx] // r9=src_end
        lea     r10, [r9-0BH] // r10=src_endmatch
        mov     ecx, 1        // ecx=CWBits
        mov     r11, r8       // r11=CWpoint
        mov     dword ptr [r8], 0
        add     r8, 4
        pxor    xmm0, xmm0
        mov     eax, 32768-64
@06:    movdqa  dqword ptr [off+rax-48], xmm0 // stack is aligned to 16 bytes
        movdqa  dqword ptr [off+rax-32], xmm0
        movdqa  dqword ptr [off+rax-16], xmm0
        movdqa  dqword ptr [off+rax], xmm0
        sub     eax, 64
        jae     @06
        cmp     rbx, r10
        ja      @15
@07:    mov     edx, dword ptr [rbx]
        mov     rax, rdx
        mov     r12, rdx
        shr     rax, 12
        xor     rax, rdx
        and     rax, 0FFFH // rax=h
        mov     r14, qword ptr [off+rax*8] // r14=o
        mov     edx, dword ptr [cache+rax*4]
        mov     qword ptr [off+rax*8], rbx
        mov     dword ptr [cache+rax*4], r12d
        xor     rdx, r12
        test    r14, r14
        lea     rdi, [r9-1]
        je      @12
        and     rdx, 0FFFFFFH
        jne     @12
        mov     rdx, rbx
        sub     rdx, r14
        cmp     rdx, 2
        jbe     @12
        or      dword ptr[r11], ecx
        add     rbx, 2
        add     r14, 2
        mov     esi, 1
        sub     rdi, rbx
        cmp     rdi, 271
        jc      @09
        mov     edi, 271
        jmp     @09
@08:    inc     rsi
@09:    mov     edx, dword ptr [r14+rsi]
        cmp     dl, byte ptr [rbx+rsi]
        jnz     @10
        cmp     rsi, rdi
        jge     @10
        inc     rsi
        cmp     dh, byte ptr [rbx+rsi]
        jnz     @10
        shr     edx, 16
        cmp     rsi, rdi
        jge     @10
        inc     rsi
        cmp     dl, byte ptr [rbx+rsi]
        jnz     @10
        cmp     rsi, rdi
        jge     @10
        inc     rsi
        cmp     dh, byte ptr [rbx+rsi]
        jnz     @10
        cmp     rsi, rdi
        jc      @08
@10:    add     rbx, rsi
        shl     rax, 4
        cmp     rsi, 15
        ja      @11
        or      rax, rsi
        mov     word ptr [r8], ax
        add     r8, 2
        jmp     @13
@11:    sub     rsi, 16
        mov     word ptr [r8], ax
        mov     byte ptr [r8+2H], sil
        add     r8, 3
        jmp     @13
@12:    mov     al, byte ptr [rbx]
        mov     byte ptr [r8], al
        add     rbx, 1
        add     r8, 1
@13:    add     ecx, ecx
        jnz     @14
        mov     r11, r8
        mov     [r8], ecx
        add     r8, 4
        add     ecx, 1
@14:    cmp     rbx, r10
        jbe     @07
@15:    cmp     rbx, r9
        jnc     @18
@16:    mov     al, byte ptr [rbx]
        mov     byte ptr [r8], al
        add     rbx, 1
        add     r8, 1
        add     ecx, ecx
        jnz     @17
        mov     [r8], ecx
        add     r8, 4
        add     ecx, 1
@17:    cmp     rbx, r9
        jc      @16
@18:    sub     r8, r15
        mov     r15, r8
@19:    mov     rax, r15
        pop     r15
        pop     r14
        pop     r13
        pop     r12
        pop     rsi
        pop     rdi
        pop     rbx
{$endif CPUX86}
end;
{$endif CPUINTEL}

function SynLZcompress1pas(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
var dst_beg,          // initial dst value
    src_end,          // real last byte available in src
    src_endmatch,     // last byte to try for hashing
    o: PAnsiChar;
    CWbit: byte;
    CWpoint: PCardinal;
    v, h, cached, t, tmax: PtrUInt;
    offset: TOffsets;
    cache: array[0..4095] of cardinal; // 16KB+16KB=32KB on stack (48KB under Win64)
begin
  dst_beg := dst;
  // 1. store in_len
  if size>=$8000 then begin // size in 32KB..2GB -> stored as integer
    PWord(dst)^ := $8000 or (size and $7fff);
    PWord(dst+2)^ := size shr 15;
    inc(dst,4);
  end else begin
    PWord(dst)^ := size ; // size<32768 -> stored as word
    if size=0 then begin
      result := 2;
      exit;
    end;
    inc(dst,2);
  end;
  // 2. compress
  src_end := src+size;
  src_endmatch := src_end-(6+5);
  CWbit := 0;
  CWpoint := pointer(dst);
  PCardinal(dst)^ := 0;
  inc(dst,sizeof(CWpoint^));
  fillchar(offset,sizeof(offset),0); // fast 16KB reset to 0
  // 1. main loop to search using hash[]
  if src<=src_endmatch then
  repeat
    v := PCardinal(src)^;
    h := ((v shr 12) xor v) and 4095;
    o := offset[h];
    offset[h] := src;
    cached := v xor cache[h]; // o=nil if cache[h] is uninitialized
    cache[h] := v;
    if (cached and $00ffffff=0) and (o<>nil) and (src-o>2) then begin
      CWpoint^ := CWpoint^ or (cardinal(1) shl CWbit);
      inc(src,2);
      inc(o,2);
      t := 1;
      tmax := src_end-src-1;
      if tmax>=(255+16) then
        tmax := (255+16);
      while (o[t]=src[t]) and (t<tmax) do
        inc(t);
      inc(src,t);
      h := h shl 4;
      // here we have always t>0
      if t<=15 then begin // mark 2 to 17 bytes -> size=1..15
        PWord(dst)^ := integer(t or h);
        inc(dst,2);
      end else begin // mark 18 to (255+16) bytes -> size=0, next byte=t
        dec(t,16);
        PWord(dst)^ := h; // size=0
        dst[2] := ansichar(t);
        inc(dst,3);
      end;
    end else begin
      dst^ := src^;
      inc(src);
      inc(dst);
    end;
    if CWbit<31 then begin
      inc(CWbit);
      if src<=src_endmatch then continue else break;
    end else begin
      CWpoint := pointer(dst);
      PCardinal(dst)^ := 0;
      inc(dst,sizeof(CWpoint^));
      CWbit := 0;
      if src<=src_endmatch then continue else break;
    end;
  until false;
  // 2. store remaining bytes
  if src<src_end then
  repeat
    dst^ := src^;
    inc(src);
    inc(dst);
    if CWbit<31 then begin
      inc(CWbit);
      if src<src_end then continue else break;
    end else begin
      PCardinal(dst)^ := 0;
      inc(dst,4);
      CWbit := 0;
      if src<src_end then continue else break;
    end;
  until false;
  result := dst-dst_beg;
end;

procedure movechars(s,d: PAnsiChar; t: PtrUInt); {$ifdef HASINLINE}inline;{$endif}
// fast code for unaligned and overlapping (see {$define WT}) small blocks
// this code is sometimes used rather than system.move()
begin
  dec(PtrUInt(s), PtrUInt(d));
  inc(t, PtrUInt(d));
  repeat
    d^ := s[PtrUInt(d)];
    inc(d);
  until PtrUInt(d)=t;
end;

const
  bitlut: array[0..15] of integer =
    (4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0);

function SynLZdecompress1b(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
// this routine was trying to improve speed, but was slower
var last_hashed: PAnsiChar; // initial src and dst value
    src_end: PAnsiChar;
    CWbit: integer;
    CW, v, t, h: integer;
    offset: TOffsets;
label nextCW;
begin
//  src_beg := src;
//  dst_beg := dst;
  src_end := src+size;
  // 1. retrieve out_len
  result := PWord(src)^;
  if result=0 then exit;
  inc(src,2);
  if result and $8000<>0 then begin
    result := (result and $7fff) or (integer(PWord(src)^) shl 15);
    inc(src,2);
  end;
  // 2. decompress
  last_hashed := dst-1;
  CWbit := 32;
nextCW:
  CW := PCardinal(src)^;
  inc(src,4);
  CWbit := CWbit-32;
  if src<src_end then
  repeat
    if CW and 1=0 then begin
      if CWbit<(32-4) then begin
        PCardinal(dst)^ := PCardinal(src)^;
        v := bitlut[CW and 15];
        inc(src,v);
        inc(dst,v);
        inc(CWbit,v);
        CW := CW shr v;
        if src>=src_end then break;
        while last_hashed<dst-3 do begin
          inc(last_hashed);
          v := PCardinal(last_hashed)^;
          offset[((v shr 12) xor v) and 4095] := last_hashed;
        end;
      end else begin
        dst^ := src^;
        inc(src);
        inc(dst);
        if src>=src_end then break;
        if last_hashed<dst-3 then begin
          inc(last_hashed);
          v := PCardinal(last_hashed)^;
          offset[((v shr 12) xor v) and 4095] := last_hashed;
        end;
        inc(CWbit);
        CW := CW shr 1;
        if CWbit<32 then
          continue else
          goto nextCW;
      end;
    end else begin
      h := PWord(src)^;
      inc(src,2);
      t := (h and 15)+2;
      h := h shr 4;
      if t=2 then begin
        t := ord(src^)+(16+2);
        inc(src);
      end;
      if dst-offset[h]<t then
        movechars(offset[h],dst,t) else
        move(offset[h]^,dst^,t);
      if last_hashed<dst then
        repeat
          inc(last_hashed);
          v := PCardinal(last_hashed)^;
          offset[((v shr 12) xor v) and 4095] := last_hashed;
        until last_hashed>=dst;
      inc(dst,t);
      if src>=src_end then break;
      last_hashed := dst-1;
      inc(CWbit);
      CW := CW shr 1;
      if CWbit<32 then
        continue else
        goto nextCW;
    end;
  until false;
//  assert(result=dst-dst_beg);
end;

{$ifdef CPUINTEL}
{$ifdef CPUX86}
// using direct x86 jmp also circumvents Internal Error C11715 for Delphi 5
function SynLZdecompress1(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
asm
        push    ebp
        push    ebx
        push    esi
        push    edi
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -4092
        push    eax
        add     esp, -24
        mov     esi, ecx
        mov     ebx, eax
        add     edx, eax
        mov     [esp+8H], esi
        mov     [esp+10H], edx
        movzx   eax, word ptr [ebx]
        mov     [esp], eax
        or      eax,eax
        je      @@0917
        add     ebx, 2
        mov     eax, [esp]
        test    ah, 80H
        jz      @@0907
        and     eax, 7FFFH
        movzx   edx, word ptr [ebx]
        shl     edx, 15
        or      eax, edx
        mov     [esp], eax
        add     ebx, 2
@@0907: lea     ebp, [esi-1]
@@0908: mov     ecx, [ebx]
        add     ebx, 4
        mov     [esp+14H], ecx
        mov     edi, 1             // edi=CWbit
        cmp     ebx, [esp+10H]
        jnc     @@0917
@@0909: mov     ecx, [esp+14H]
@@090A: test    ecx, edi
        jnz     @@0911
        mov     al, [ebx]
        inc     ebx
        mov     [esi], al
        inc     esi
        cmp     ebx, [esp+10H]
        lea     eax, [esi-3]
        jnc     @@0917
        cmp     eax, ebp
        jbe     @@0910
        inc     ebp
        mov     eax, [ebp]
        mov     edx, eax
        shr     eax, 12
        xor     eax, edx
        and     eax, 0FFFH
        mov     [esp+eax*4+1CH], ebp
@@0910: add     edi, edi
        jnz     @@090A
        jmp     @@0908
@@0911: movzx   edx, word ptr [ebx]
        add     ebx, 2
        mov     eax, edx
        and     edx, 0FH
        add     edx, 2
        shr     eax, 4
        cmp     edx,2
        jnz     @@0912
        movzx   edx, byte ptr [ebx]
        inc     ebx
        add     edx, 18
@@0912: mov     eax, [esp+eax*4+1CH]
        mov     ecx, esi
        mov     [esp+18H], edx
        sub     ecx, eax
        cmp     ecx, edx
        jl      @@0913
        cmp     edx, 32            // inlined optimized move()
        ja      @large
        sub     edx, 8
        jg      @9_32
        mov     ecx, [eax]
        mov     eax, [eax+4]       // always copy 8 bytes for 0..8
        mov     [esi], ecx         // safe since src_endmatch := src_end-(6+5)
        mov     [esi+4], eax
        jmp     @movend
@9_32:  fild    qword ptr[eax+edx]
        fild    qword ptr[eax]
        cmp     edx, 8
        jle     @16
        fild    qword ptr[eax+8]
        cmp     edx, 16
        jle     @24
        fild    qword ptr[eax+16]
        fistp   qword ptr[esi+16]
@24:    fistp   qword ptr[esi+8]
@16:    fistp   qword ptr[esi]
        fistp   qword ptr[esi+edx]
        jmp     @movend
        nop
@large: push    esi
        fild    qword ptr[eax]
        lea     eax, [eax+edx-8]
        lea     edx, [esi+edx-8]
        fild    qword ptr[eax]
        push    edx
        neg     edx
        and     esi,  -8
        lea     edx, [edx+esi+8]
        pop     esi
@lrgnxt:fild    qword ptr[eax+edx]
        fistp   qword ptr[esi+edx]
        add     edx, 8
        jl      @lrgnxt
        fistp   qword ptr[esi]
        pop     esi
        fistp   qword ptr[esi]
@movend:cmp     esi, ebp
        jbe     @@0916
@@0915: inc     ebp
        mov     edx, [ebp]
        mov     eax, edx
        shr     edx, 12
        xor     eax, edx
        and     eax, 0FFFH
        mov     [esp+eax*4+1CH], ebp
        cmp     esi, ebp
        ja      @@0915
@@0916: add     esi, [esp+18H]
        cmp     ebx, [esp+10H]
        jnc     @@0917
        add     edi, edi
        lea     ebp, [esi-1]
        jz      @@0908
        jmp     @@0909
@@0913: push    ebx
        xor     ecx, ecx
@s:     dec     edx
        mov     bl, [eax+ecx]
        mov     [esi+ecx], bl
        lea     ecx,[ecx+1]
        jnz     @s
        pop     ebx
        jmp     @movend
@@0917: mov     eax, [esp]
        add     esp, 16412
        pop     edi
        pop     esi
        pop     ebx
        pop     ebp
{$else CPUX86}
function SynLZdecompress1(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
var off: TOffsets;
asm // rcx=src, edx=size, r8=dest
        {$ifndef win64} // Linux 64-bit ABI
        mov     r8, rdx
        mov     rdx, rsi
        mov     rcx, rdi
        {$endif win64}
        push    rbx
        push    rsi
        push    rdi
        push    r12
        push    r13
        push    r14
        push    r15
        movzx   eax, word ptr [rcx] // rcx=src   eax=result
        lea     r9, [rcx+rdx] // r9=src_end
        test    eax, eax
        je      @35
        add     rcx, 2
        mov     r10d, eax
        and     r10d, 8000H
        jz      @21
        movzx   ebx, word ptr [rcx]
        shl     ebx, 15
        mov     r10d, eax
        and     r10d, 7FFFH
        or      r10d, ebx
        mov     eax, r10d
        add     rcx, 2
@21:    lea     r10, [r8-1H]  // r10=last_hashed  r8=dest
@22:    mov     edi, dword ptr [rcx]  // edi=CW
        add     rcx, 4
        mov     r13d, 1 // r13d=CWBit
        cmp     rcx, r9
        jnc     @35
@23:    test    r13d, edi
        jnz     @25
        mov     bl, byte ptr [rcx]
        mov     byte ptr [r8], bl
        add     rcx, 1
        lea     rbx, [r8-2H]
        add     r8, 1
        cmp     rcx, r9
        jnc     @35
        cmp     rbx, r10
        jbe     @24
        add     r10, 1
        mov     esi, dword ptr [r10]
        mov     rbx, rsi
        shr     esi, 12
        xor     ebx, esi
        and     ebx, 0FFFH
        mov     qword ptr [off+rbx*8], r10
@24:    shl     r13d, 1
        jnz     @23
        jmp     @22
@25:    movzx   r11, word ptr [rcx] // r11=t
        add     rcx, 2
        mov     ebx, r11d // ebx=h
        shr     ebx, 4
        and     r11, 0FH
        lea     r11, [r11+2H]
        jnz     @26
        movzx   r11, byte ptr [rcx]
        add     rcx, 1
        lea     r11, [r11+12H]
@26:    mov     r14, qword ptr [off+rbx*8] // r14=o
        mov     rbx, r8
        xor     rsi, rsi
        sub     rbx, r14
        mov     r12, r11
        mov     r15, r11
        cmp     rbx, r11
        jc      @29
        shr     r12, 3
        jz      @30
@27:    mov     rbx, qword ptr [r14+rsi]
        mov     qword ptr [r8+rsi], rbx
        add     rsi, 8
        dec     r12
        jnz     @27
        mov     rbx, qword ptr [r14+rsi]
        and     r15, 7
        jz      @31
@28:    mov     byte ptr [r8+rsi], bl
        shr     rbx, 8
        inc     rsi
        dec     r15
        jnz     @28
        jmp     @31
@29:    mov     bl, byte ptr [r14+rsi]
        mov     byte ptr [r8+rsi], bl
        inc     rsi
        dec     r12
        jnz     @29
        cmp     rcx, r9
        jnz     @33
        jmp     @35
@30:    mov     rbx, qword ptr [r14]
        mov     qword ptr [r8], rbx
@31:    cmp     rcx, r9
        jz      @35
        cmp     r10, r8
        jnc     @34
@32:    add     r10, 1
        mov     ebx, dword ptr [r10]
        mov     rsi, rbx
        shr     ebx, 12
        xor     esi, ebx
        and     esi, 0FFFH
        mov     qword ptr [off+rsi*8], r10
@33:    cmp     r10, r8
        jc      @32
@34:    add     r8, r11
        lea     r10, [r8-1H]
        shl     r13d, 1
        jnz     @23
        jmp     @22
@35:    pop     r15
        pop     r14
        pop     r13
        pop     r12
        pop     rdi
        pop     rsi
        pop     rbx
{$endif CPUX86}
end;
{$endif CPUINTEL}

// better code generation with sub-functions for raw decoding
procedure SynLZdecompress1passub(src, src_end, dst: PAnsiChar; var offset: TOffsets);
var last_hashed: PAnsiChar; // initial src and dst value
    {$ifdef CPU64}
    o: PAnsiChar;
    {$endif}
    CW, CWbit: cardinal;
    v, t, h: PtrUInt;
label nextCW;
begin
  last_hashed := dst-1;
nextCW:
  CW := PCardinal(src)^;
  inc(src,4);
  CWbit := 1;
  if src<src_end then
  repeat
    if CW and CWbit=0 then begin
      dst^ := src^;
      inc(src);
      inc(dst);
      if src>=src_end then break;
      if last_hashed<dst-3 then begin
        inc(last_hashed);
        v := PCardinal(last_hashed)^;
        offset[((v shr 12) xor v) and 4095] := last_hashed;
      end;
      CWbit := CWbit shl 1;
      if CWbit<>0 then
        continue else
        goto nextCW;
    end else begin
      h := PWord(src)^;
      inc(src,2);
      t := (h and 15)+2;
      h := h shr 4;
      if t=2 then begin
        t := ord(src^)+(16+2);
        inc(src);
      end;
      {$ifdef CPU64}
      o := offset[h];
      if PtrUInt(dst-o)<t then
        movechars(o,dst,t) else
        if t<=8 then
          PInt64(dst)^ := PInt64(o)^ else
          move(o^,dst^,t);
      {$else}
      if PtrUInt(dst-offset[h])<t then
        movechars(offset[h],dst,t) else
        if t>8 then // safe since src_endmatch := src_end-(6+5)
          move(offset[h]^,dst^,t) else
          PInt64(dst)^ := PInt64(offset[h])^; // much faster in practice
      {$endif}
      if src>=src_end then break;
      if last_hashed<dst then
        repeat
          inc(last_hashed);
          v := PCardinal(last_hashed)^;
          offset[((v shr 12) xor v) and 4095] := last_hashed;
        until last_hashed>=dst;
      inc(dst,t);
      last_hashed := dst-1;
      CWbit := CWbit shl 1;
      if CWbit<>0 then
        continue else
        goto nextCW;
    end;
  until false;
end;

function SynLZdecompress1pas(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
var offset: TOffsets;
    src_end: PAnsiChar;
begin
  src_end := src+size;
  result := PWord(src)^;
  if result=0 then exit;
  inc(src,2);
  if result and $8000<>0 then begin
    result := (result and $7fff) or (integer(PWord(src)^) shl 15);
    inc(src,2);
  end;
  SynLZdecompress1passub(src, src_end, dst, offset);
end;

procedure SynLZdecompress1partialsub(src, dst, src_end, dst_end: PAnsiChar; var offset: TOffsets);
var last_hashed: PAnsiChar; // initial src and dst value
    CWbit, CW: integer;
    v, t, h: PtrUInt;
    {$ifdef CPU64}
    o: PAnsiChar;
    {$endif}
label nextCW;
begin
  last_hashed := dst-1;
nextCW:
  CW := PCardinal(src)^;
  inc(src,4);
  CWbit := 1;
  if src<src_end then
  repeat
    if CW and CWbit=0 then begin
      dst^ := src^;
      inc(src);
      inc(dst);
      if (src>=src_end) or (dst>=dst_end) then
        break;
      if last_hashed<dst-3 then begin
        inc(last_hashed);
        v := PCardinal(last_hashed)^;
        offset[((v shr 12) xor v) and 4095] := last_hashed;
      end;
      CWbit := CWbit shl 1;
      if CWbit<>0 then
        continue else
        goto nextCW;
    end else begin
      h := PWord(src)^;
      inc(src,2);
      t := (h and 15)+2;
      h := h shr 4;
      if t=2 then begin
        t := ord(src^)+(16+2);
        inc(src);
      end;
      if dst+t>=dst_end then begin // avoid buffer overflow by all means
        movechars(offset[h],dst,dst_end-dst);
        break;
      end;
      {$ifdef CPU64}
      o := offset[h];
      if (t<=8) or (PtrUInt(dst-o)<t) then
        movechars(o,dst,t) else
        move(o^,dst^,t);
      {$else}
      if (t<=8) or (PtrUInt(dst-offset[h])<t) then
        movechars(offset[h],dst,t) else
        move(offset[h]^,dst^,t);
      {$endif}
      if src>=src_end then
        break;
      if last_hashed<dst then
        repeat
          inc(last_hashed);
          v := PCardinal(last_hashed)^;
          offset[((v shr 12) xor v) and 4095] := last_hashed;
        until last_hashed>=dst;
      inc(dst,t);
      last_hashed := dst-1;
      CWbit := CWbit shl 1;
      if CWbit<>0 then
        continue else
        goto nextCW;
    end;
  until false;
end;

function SynLZdecompress1partial(src: PAnsiChar; size: integer; dst: PAnsiChar; maxDst: integer): integer;
var offset: TOffsets;
    src_end: PAnsiChar;
begin
  src_end := src+size;
  result := PWord(src)^;
  if result=0 then exit;
  inc(src,2);
  if result and $8000<>0 then begin
    result := (result and $7fff) or (integer(PWord(src)^) shl 15);
    inc(src,2);
  end;
  if maxDst<result then
    result := maxDst;
  if result>0 then
    SynLZdecompress1partialsub(src, dst, src_end, dst+result, offset);
end;


function SynLZcompress2(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
var dst_beg,      // initial dst value
    src_end,      // real last byte available in src
    src_endmatch, // last byte to try for hashing
    o: PAnsiChar;
    CWbit: byte;
    CWpoint: PCardinal;
    h, v, cached: integer;
    t, tmax, tdiff, i: integer;
    offset: TOffsets; // 16KB+16KB=32KB hashing code
    cache: array[0..4095] of integer;
    label dotdiff;
begin
  dst_beg := dst;
  // 1. store in_len
  if size>=$8000 then begin
    PWord(dst)^ := $8000 or (size and $7fff);
    PWord(dst+2)^ := size shr 15;
    inc(dst,4);
  end else begin
    PWord(dst)^ := size ; // src<32768 -> stored as word, otherwise as integer
    if size=0 then begin
      result := 2;
      exit;
    end;
    inc(dst,2);
  end;
  // 2. compress
  src_end := src+size;
  src_endmatch := src_end-(6+5);
  CWbit := 0;
  CWpoint := pointer(dst);
  PCardinal(dst)^ := 0;
  inc(dst,sizeof(CWpoint^));
  tdiff := 0;
  fillchar(offset,sizeof(offset),0); // fast 16KB reset to 0
  // 1. main loop to search using hash[]
  if src<=src_endmatch then
  repeat
    v := PCardinal(src)^;
    h := ((v shr 12) xor v) and 4095;
    o := offset[h];
    offset[h] := src;
    cached := v xor cache[h];
    cache[h] := v;
    if (cached and $00ffffff=0) and (o<>nil) and (src-o>2) then begin
//      SetBit(CWpoint,CWbit);
//      asm movzx eax,byte ptr CWbit; bts [CWpoint],eax; end
      if tdiff<>0 then begin
        dec(src,tdiff);
dotdiff:v := tdiff;
        if v<=8 then begin
          if CWBit+v>31 then begin
            for i := CWBit to 31 do begin
              dst^ := src^;
              inc(dst);
              inc(src);
            end;
            CWpoint := pointer(dst);
            PCardinal(dst)^ := 0;
            inc(dst,4);
            CWBit := (CWBit+v) and 31;
            for i := 1 to CWBit do begin
              dst^ := src^;
              inc(dst);
              inc(src);
            end;
          end else begin
            inc(CWBit,v);
            for i := 1 to v do begin
              dst^ := src^;
              inc(dst);
              inc(src);
            end;
          end;
        end else begin
          CWpoint^ := CWpoint^ or (cardinal(1) shl CWbit);
          dec(v,9);
          if v>15 then begin
            v := 15; // v=9..24 -> h=0..15
            dst^ := #$ff; // size=15 -> tdiff
          end else
            dst^ := ansichar((v shl 4) or 15); // size=15 -> tdiff
          inc(dst);
          pInt64(dst)^ := pInt64(src)^;
          inc(dst,8);
          inc(src,8);
          for i := 1 to v+1 do begin
            dst^ := src^;
            inc(dst);
            inc(src);
          end;
          if CWBit<31 then
            inc(CWBit) else begin
            CWpoint := pointer(dst);
            PCardinal(dst)^ := 0;
            inc(dst,4);
            CWbit := 0;
          end;
          dec(tdiff,24);
          if tdiff>0 then
            goto dotdiff;
        end;
      end;
//      assert(PWord(o)^=PWord(src)^);
      tdiff := 0;
      CWpoint^ := CWpoint^ or (cardinal(1) shl CWbit);
      inc(src,2);
      inc(o,2);
      t := 0; // t=matchlen-2
      tmax := src_end-src;
      if tmax>=(255+15) then
        tmax := (255+15);
      while (o[t]=src[t]) and (t<tmax) do
        inc(t);
      inc(src,t);
      h := h shl 4;
//      assert(t>0);
      // here we have always t>0
      if t<15 then begin // store t=1..14 -> size=t=1..14
        PWord(dst)^ := integer(t or h);
        inc(dst,2);
      end else begin // store t=15..255+15 -> size=0, next byte=matchlen-15-2
        dst[2] := ansichar(t-15);
        PWord(dst)^ := h; // size=0
        inc(dst,3);
      end;
      if CWbit<31 then begin
        inc(CWbit);
        if src<=src_endmatch then continue else break;
      end else begin
        CWpoint := pointer(dst);
        PCardinal(dst)^ := 0;
        inc(dst,4);
        CWbit := 0;
        if src<=src_endmatch then continue else break;
      end;
    end else begin
      inc(src);
      inc(tdiff);
      if src<=src_endmatch then continue else break;
    end;
  until false;
  // 2. store remaining bytes
  dec(src,tdiff); // force store trailing bytes
  if src<src_end then
  repeat
    dst^ := src^;
    inc(src);
    inc(dst);
    if CWbit<31 then begin
      inc(CWbit);
      if src<src_end then continue else break;
    end else begin
      PCardinal(dst)^ := 0;
      inc(dst,4);
      CWbit := 0;
      if src<src_end then continue else break;
    end;
  until false;
  result := dst-dst_beg;
end;

function SynLZdecompress2(src: PAnsiChar; size: integer; dst: PAnsiChar): integer;
var {$ifopt C+}dst_beg,{$endif} last_hashed: PAnsiChar; // initial src and dst value
    src_end: PAnsiChar;
    CWbit: integer;
    CW, v, t, h, i: integer;
    offset: TOffsets; // 16KB hashing code
label nextCW;
begin
  {$ifopt C+}
  dst_beg := dst;
  {$endif}
  src_end := src+size;
  {$ifndef ISDELPHI102}
  {$ifndef CPU64}
  t := 0; // make compiler happy
  {$endif}
  {$endif}
  // 1. retrieve out_len
  result := PWord(src)^;
  if result=0 then exit;
  inc(src,2);
  if result and $8000<>0 then begin
    result := (result and $7fff) or (integer(PWord(src)^) shl 15);
    inc(src,2);
  end;
  // 2. decompress
  last_hashed := dst-1;
nextCW:
  CW := PCardinal(src)^;
  inc(src,4);
  CWbit := 1;
  if src<src_end then
  repeat
    if CW and CWbit=0 then begin
      dst^ := src^;
      inc(src);
      inc(dst);
      if src>=src_end then break;
      if last_hashed<dst-3 then begin
        inc(last_hashed);
        v := PCardinal(last_hashed)^;
        offset[((v shr 12) xor v) and 4095] := last_hashed;
      end;
      CWbit := CWbit shl 1;
      if CWbit<>0 then
        continue else
        goto nextCW;
    end else begin
      case ord(src^) and 15 of // get size
      0: begin // size=0 -> next byte=matchlen-15-2
        h := PWord(src)^ shr 4;
        t := ord(src[2])+(15+2);
        inc(src,3);
        if dst-offset[h]<t then
          movechars(offset[h],dst,t) else
          move(offset[h]^,dst^,t);
      end;
      15: begin
        for i := 1 to ord(src^) shr 4+9 do begin // size=15 -> tdiff
          inc(src);
          dst^ := src^;
          inc(dst);
        end;
        inc(src);
        if src>=src_end then break;
        while last_hashed<dst-3 do begin
          inc(last_hashed);
          v := PCardinal(last_hashed)^;
          offset[((v shr 12) xor v) and 4095] := last_hashed;
        end;
        CWbit := CWbit shl 1;
        if CWbit<>0 then
          continue else
          goto nextCW;
      end;
      else begin // size=1..14=matchlen-2
        h := PWord(src)^;
        inc(src,2);
        t := (h and 15)+2;
        h := h shr 4;
        if dst-offset[h]<t then
          movechars(offset[h],dst,t) else
          move(offset[h]^,dst^,t);
      end;
      end;
      if last_hashed<dst then
        repeat
          inc(last_hashed);
          v := PCardinal(last_hashed)^;
          offset[((v shr 12) xor v) and 4095] := last_hashed;
        until last_hashed>=dst;
      inc(dst,t);
      if src>=src_end then break;
      last_hashed := dst-1;
      CWbit := CWbit shl 1;
      if CWbit<>0 then
        continue else
        goto nextCW;
    end;
  until false;
  {$ifopt C+}
  assert(result=dst-dst_beg);
  {$endif}
end;

function Hash32(P: PIntegerArray; L: integer): cardinal;
// faster than Adler32, even asm version, because read DWORD aligned data
var s1,s2: cardinal;
    i: integer;
begin
  if P<>nil then begin
    s1 := 0;
    s2 := 0;
    for i := 1 to L shr 4 do begin // 16 bytes (4 DWORD) by loop - aligned read
      inc(s1,P^[0]);
      inc(s2,s1);
      inc(s1,P^[1]);
      inc(s2,s1);
      inc(s1,P^[2]);
      inc(s2,s1);
      inc(s1,P^[3]);
      inc(s2,s1);
      inc(PByte(P),16);
    end;
    for i := 1 to (L shr 2)and 3 do begin // 4 bytes (DWORD) by loop
      inc(s1,P^[0]);
      inc(s2,s1);
      inc(PInteger(P));
    end;
    case L and 3 of // remaining 0..3 bytes
    1: inc(s1,PByte(P)^);
    2: inc(s1,PWord(P)^);
    3: inc(s1,PWord(P)^ or (ord(PAnsiChar(P)[2]) shl 16));
    end;
    inc(s2,s1);
    result := s1 xor (s2 shl 16);
  end else
    result := 0;
end;


end.