fix: EDITCHANGE should not append; SELCHANGE should delay (2)

commit add56b5aed
Author: Tino Reichardt <milky-7zip@mcmilk.de>
Date:   Thu Nov 1 23:08:00 2018 +0100

    Add MD5 hash function

commit 36a17a5184
Author: Tino Reichardt <milky-7zip@mcmilk.de>
Date:   Sat Nov 3 00:18:33 2018 +0100

    Add some hash functions
    - new: md2, md4, md5, sha384, sha512, xxhash-32, xxhash-64
    - put Blake2sp hash stuff back to rar code
    - added the hashes to GUI and Explorer Menu code

commit 576c5df947
Author: Tino Reichardt <milky-7zip@mcmilk.de>
Date:   Tue Apr 6 19:35:46 2021 +0200

    Add BLAKE3 hash function

commit 6b2a151549
Author: Tino Reichardt <milky-7zip@mcmilk.de>
Date:   Tue Apr 6 19:51:01 2021 +0200

    Remove unneeded file HashesReg.cpp

commit dddf507557
Author: Tino Reichardt <milky-7zip@mcmilk.de>
Date:   Sun Jun 18 09:13:59 2023 +0200

    Add SHA3 hashing

    - added these variants: SHA3-256, SHA3-384, SHA3-512
    - reordered also the hashing id's
    - added some notes about them in DOC/Hashes.txt

    Signed-off-by: Tino Reichardt <milky-7zip@mcmilk.de>

The cherry-picking was a chaos; they're not applied in order, and some
commits even got cherry-picked twice (1->4->0->2->4->3). So subsequent fixes and
adjustments were applied to make it build.

.gitattributes

@@ -0,0 +1,13 @@
+# Set default behavior to automatically normalize line endings.
+* text=crlf
+
+# These files are text and should be normalized (Convert crlf => lf)
+
+*.txt text
+*.vcproj text
+*.cpp text
+*.h text
+*.def text
+*.rc text
+*.cmd
+

.github/workflows/build.yaml

@@ -0,0 +1,42 @@
+name: Build and Upload Artifact
+
+on:
+  push:
+    branches:
+      - easy7zip-sf
+      - e7z-sf-zstd
+      - e7z-sf-without-zstd
+  pull_request:
+    branches:
+      - easy7zip-sf
+      - e7z-sf-zstd
+      - e7z-sf-without-zstd
+
+jobs:
+  build:
+    runs-on: windows-latest
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v2
+
+      - name: Build project
+        shell: cmd
+        run: ${{ '"C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && build.cmd' }}
+
+      - name: Download and unpack Lang
+        shell: cmd
+        run: ${{ 'cd out && curl -L -o official.exe.7z "https://sourceforge.net/projects/sevenzip/files/latest/download" && .\7z x official.exe.7z Lang/ License.txt && DEL official.exe.7z' }}
+
+      - name: Pack
+        shell: cmd
+        run: ${{ 'cd out && COPY 7zipUninstall.exe Uninstall.exe && .\7z a -m0=LZMA -mx=9 install.7z Lang/ 7z.dll 7z.exe 7z.sfx 7zCon.sfx 7zFM.exe 7zG.exe 7-zip.dll Uninstall.exe License.txt && RENAME 7zipInstall.exe 7zipInstall.exe.bak && COPY /Y /B 7zipInstall.exe.bak + install.7z 7zipInstall.exe && DEL Uninstall.exe && DEL install.7z' }}
+
+      - name: Upload artifact
+        uses: actions/upload-artifact@v2
+        with:
+          name: Easy_7zip_Artifact
+          # Exclude directories
+          path: |
+            ./out/*
+            !./out/*/*

.gitignore

@@ -0,0 +1,7 @@
+*.o
+errorfile.txt
+*.user
+*.obj
+out/
+*.vcxproj
+*.db

Asm/arm64/7zAsm.S

@@ -0,0 +1,181 @@
+// 7zAsm.S -- ASM macros for arm64
+// 2021-04-25 : Igor Pavlov : Public domain
+
+#define  r0 x0
+#define  r1 x1
+#define  r2 x2
+#define  r3 x3
+#define  r4 x4
+#define  r5 x5
+#define  r6 x6
+#define  r7 x7
+#define  r8 x8
+#define  r9 x9
+#define  r10 x10
+#define  r11 x11
+#define  r12 x12
+#define  r13 x13
+#define  r14 x14
+#define  r15 x15
+#define  r16 x16
+#define  r17 x17
+#define  r18 x18
+#define  r19 x19
+#define  r20 x20
+#define  r21 x21
+#define  r22 x22
+#define  r23 x23
+#define  r24 x24
+#define  r25 x25
+#define  r26 x26
+#define  r27 x27
+#define  r28 x28
+#define  r29 x29
+#define  r30 x30
+
+#define  REG_ABI_PARAM_0 r0
+#define  REG_ABI_PARAM_1 r1
+#define  REG_ABI_PARAM_2 r2
+
+
+.macro p2_add reg:req, param:req
+        add     \reg, \reg, \param
+.endm
+
+.macro p2_sub reg:req, param:req
+        sub     \reg, \reg, \param
+.endm
+
+.macro p2_sub_s reg:req, param:req
+        subs    \reg, \reg, \param
+.endm
+
+.macro p2_and reg:req, param:req
+        and     \reg, \reg, \param
+.endm
+
+.macro xor reg:req, param:req
+        eor     \reg, \reg, \param
+.endm
+
+.macro or reg:req, param:req
+        orr     \reg, \reg, \param
+.endm
+
+.macro shl reg:req, param:req
+        lsl     \reg, \reg, \param
+.endm
+
+.macro shr reg:req, param:req
+        lsr     \reg, \reg, \param
+.endm
+
+.macro sar reg:req, param:req
+        asr     \reg, \reg, \param
+.endm
+
+.macro p1_neg reg:req
+        neg     \reg, \reg
+.endm
+
+.macro dec reg:req
+        sub     \reg, \reg, 1
+.endm
+
+.macro dec_s reg:req
+        subs    \reg, \reg, 1
+.endm
+
+.macro inc reg:req
+        add     \reg, \reg, 1
+.endm
+
+.macro inc_s reg:req
+        adds    \reg, \reg, 1
+.endm
+
+
+.macro imul reg:req, param:req
+        mul     \reg, \reg, \param
+.endm
+
+/*
+arm64 and arm use reverted c flag after subs/cmp instructions:
+  arm64-arm   :     x86
+ b.lo / b.cc  :  jb  / jc
+ b.hs / b.cs  :  jae / jnc
+*/ 
+
+.macro jmp lab:req
+        b       \lab
+.endm
+
+.macro je lab:req
+        b.eq    \lab
+.endm
+
+.macro jz lab:req
+        b.eq    \lab
+.endm
+
+.macro jnz lab:req
+        b.ne    \lab
+.endm
+
+.macro jne lab:req
+        b.ne    \lab
+.endm
+
+.macro jb lab:req
+        b.lo    \lab
+.endm
+
+.macro jbe lab:req
+        b.ls    \lab
+.endm
+
+.macro ja lab:req
+        b.hi    \lab
+.endm
+
+.macro jae lab:req
+        b.hs    \lab
+.endm
+
+
+.macro cmove dest:req, srcTrue:req
+        csel    \dest, \srcTrue, \dest, eq
+.endm
+
+.macro cmovne dest:req, srcTrue:req
+        csel    \dest, \srcTrue, \dest, ne
+.endm
+
+.macro cmovs dest:req, srcTrue:req
+        csel    \dest, \srcTrue, \dest, mi
+.endm
+
+.macro cmovns dest:req, srcTrue:req
+        csel    \dest, \srcTrue, \dest, pl
+.endm
+
+.macro cmovb dest:req, srcTrue:req
+        csel    \dest, \srcTrue, \dest, lo
+.endm
+
+.macro cmovae dest:req, srcTrue:req
+        csel    \dest, \srcTrue, \dest, hs
+.endm
+
+
+.macro MY_ALIGN_16 macro
+	.p2align 4,, (1 << 4) - 1
+.endm
+
+.macro MY_ALIGN_32 macro
+        .p2align 5,, (1 << 5) - 1
+.endm
+
+.macro MY_ALIGN_64 macro
+        .p2align 6,, (1 << 6) - 1
+.endm

Asm/x86/7zAsm.asm

@@ -1,5 +1,50 @@
 ; 7zAsm.asm -- ASM macros
-; 2009-12-12 : Igor Pavlov : Public domain
+; 2022-05-16 : Igor Pavlov : Public domain
+
+
+; UASM can require these changes
+; OPTION FRAMEPRESERVEFLAGS:ON
+; OPTION PROLOGUE:NONE
+; OPTION EPILOGUE:NONE
+
+ifdef @wordsize
+; @wordsize is defined only in JWASM and ASMC and is not defined in MASM
+; @wordsize eq 8 for 64-bit x64
+; @wordsize eq 2 for 32-bit x86
+if @wordsize eq 8
+  x64 equ 1
+endif
+else
+ifdef RAX
+  x64 equ 1
+endif
+endif
+
+
+ifdef x64
+  IS_X64 equ 1
+else
+  IS_X64 equ 0
+endif
+
+ifdef ABI_LINUX
+  IS_LINUX equ 1
+else
+  IS_LINUX equ 0
+endif
+
+ifndef x64
+; Use ABI_CDECL for x86 (32-bit) only
+; if ABI_CDECL is not defined, we use fastcall abi
+ifdef ABI_CDECL
+  IS_CDECL equ 1
+else
+  IS_CDECL equ 0
+endif
+endif
+
+OPTION PROLOGUE:NONE 
+OPTION EPILOGUE:NONE
 
 MY_ASM_START macro
   ifdef x64
@@ -13,30 +58,39 @@ endm
 
 MY_PROC macro name:req, numParams:req
   align 16
-  proc_numParams equ numParams
-  ifdef x64
+  proc_numParams = numParams
+  if (IS_X64 gt 0)
     proc_name equ name
-    name PROC
+  elseif (IS_LINUX gt 0)
+    proc_name equ name
+  elseif (IS_CDECL gt 0)
+    proc_name equ @CatStr(_,name)
   else
-    proc_fastcall_name equ @CatStr(@,name,@, %numParams * 4)
-    public proc_fastcall_name
-    proc_fastcall_name:
+    proc_name equ @CatStr(@,name,@, %numParams * 4)
   endif
+  proc_name PROC
 endm
 
 MY_ENDP macro
-  ifdef x64
-    ret
-    proc_name ENDP
-  else
-    ret (proc_numParams - 2) * 4
-  endif
+    if (IS_X64 gt 0)
+        ret
+    elseif (IS_CDECL gt 0)
+        ret
+    elseif (proc_numParams LT 3)
+        ret
+    else
+        ret (proc_numParams - 2) * 4
+    endif
+  proc_name ENDP
 endm
 
+
 ifdef x64
   REG_SIZE equ 8
+  REG_LOGAR_SIZE equ 3
 else
   REG_SIZE equ 4
+  REG_LOGAR_SIZE equ 2
 endif
 
   x0 equ EAX
@@ -48,6 +102,15 @@ endif
   x6 equ ESI
   x7 equ EDI
 
+  x0_W equ AX
+  x1_W equ CX
+  x2_W equ DX
+  x3_W equ BX
+
+  x5_W equ BP
+  x6_W equ SI
+  x7_W equ DI
+
   x0_L equ AL
   x1_L equ CL
   x2_L equ DL
@@ -59,6 +122,10 @@ endif
   x3_H equ BH
 
 ifdef x64
+  x5_L equ BPL
+  x6_L equ SIL
+  x7_L equ DIL
+
   r0 equ RAX
   r1 equ RCX
   r2 equ RDX
@@ -67,6 +134,14 @@ ifdef x64
   r5 equ RBP
   r6 equ RSI
   r7 equ RDI
+  x8 equ r8d
+  x9 equ r9d
+  x10 equ r10d
+  x11 equ r11d
+  x12 equ r12d
+  x13 equ r13d
+  x14 equ r14d
+  x15 equ r15d
 else
   r0 equ x0
   r1 equ x1
@@ -78,6 +153,24 @@ else
   r7 equ x7
 endif
 
+
+ifdef x64
+ifdef ABI_LINUX
+
+MY_PUSH_2_REGS macro
+    push    r3
+    push    r5
+endm
+
+MY_POP_2_REGS macro
+    pop     r5
+    pop     r3
+endm
+
+endif
+endif
+
+
 MY_PUSH_4_REGS macro
     push    r3
     push    r5
@@ -91,3 +184,106 @@ MY_POP_4_REGS macro
     pop     r5
     pop     r3
 endm
+
+
+; for fastcall and for WIN-x64
+REG_PARAM_0_x   equ x1
+REG_PARAM_0     equ r1
+REG_PARAM_1_x   equ x2
+REG_PARAM_1     equ r2
+
+ifndef x64
+; for x86-fastcall
+
+REG_ABI_PARAM_0_x equ REG_PARAM_0_x
+REG_ABI_PARAM_0   equ REG_PARAM_0
+REG_ABI_PARAM_1_x equ REG_PARAM_1_x
+REG_ABI_PARAM_1   equ REG_PARAM_1
+
+else
+; x64
+
+if  (IS_LINUX eq 0)
+
+; for WIN-x64:
+REG_PARAM_2_x   equ x8
+REG_PARAM_2     equ r8
+REG_PARAM_3     equ r9
+
+REG_ABI_PARAM_0_x equ REG_PARAM_0_x
+REG_ABI_PARAM_0   equ REG_PARAM_0
+REG_ABI_PARAM_1_x equ REG_PARAM_1_x
+REG_ABI_PARAM_1   equ REG_PARAM_1
+REG_ABI_PARAM_2_x equ REG_PARAM_2_x
+REG_ABI_PARAM_2   equ REG_PARAM_2
+REG_ABI_PARAM_3   equ REG_PARAM_3
+
+else
+; for LINUX-x64:
+REG_LINUX_PARAM_0_x equ x7
+REG_LINUX_PARAM_0   equ r7
+REG_LINUX_PARAM_1_x equ x6
+REG_LINUX_PARAM_1   equ r6
+REG_LINUX_PARAM_2   equ r2
+REG_LINUX_PARAM_3   equ r1
+REG_LINUX_PARAM_4_x equ x8
+REG_LINUX_PARAM_4   equ r8
+REG_LINUX_PARAM_5   equ r9
+
+REG_ABI_PARAM_0_x equ REG_LINUX_PARAM_0_x
+REG_ABI_PARAM_0   equ REG_LINUX_PARAM_0
+REG_ABI_PARAM_1_x equ REG_LINUX_PARAM_1_x
+REG_ABI_PARAM_1   equ REG_LINUX_PARAM_1
+REG_ABI_PARAM_2   equ REG_LINUX_PARAM_2
+REG_ABI_PARAM_3   equ REG_LINUX_PARAM_3
+REG_ABI_PARAM_4_x equ REG_LINUX_PARAM_4_x
+REG_ABI_PARAM_4   equ REG_LINUX_PARAM_4
+REG_ABI_PARAM_5   equ REG_LINUX_PARAM_5
+
+MY_ABI_LINUX_TO_WIN_2 macro
+        mov     r2, r6
+        mov     r1, r7
+endm
+
+MY_ABI_LINUX_TO_WIN_3 macro
+        mov     r8, r2
+        mov     r2, r6
+        mov     r1, r7
+endm
+
+MY_ABI_LINUX_TO_WIN_4 macro
+        mov     r9, r1
+        mov     r8, r2
+        mov     r2, r6
+        mov     r1, r7
+endm
+
+endif ; IS_LINUX
+
+
+MY_PUSH_PRESERVED_ABI_REGS macro
+    if  (IS_LINUX gt 0)
+        MY_PUSH_2_REGS
+    else
+        MY_PUSH_4_REGS
+    endif
+        push    r12
+        push    r13
+        push    r14
+        push    r15
+endm
+
+
+MY_POP_PRESERVED_ABI_REGS macro
+        pop     r15
+        pop     r14
+        pop     r13
+        pop     r12
+    if  (IS_LINUX gt 0)
+        MY_POP_2_REGS
+    else
+        MY_POP_4_REGS
+    endif
+endm
+
+endif ; x64

Asm/x86/7zCrcOpt.asm

@@ -1,5 +1,5 @@
 ; 7zCrcOpt.asm -- CRC32 calculation : optimized version
-; 2009-12-12 : Igor Pavlov : Public domain
+; 2021-02-07 : Igor Pavlov : Public domain
 
 include 7zAsm.asm
 
@@ -7,21 +7,28 @@ MY_ASM_START
 
 rD   equ  r2
 rN   equ  r7
+rT   equ  r5
 
 ifdef x64
     num_VAR     equ r8
     table_VAR   equ r9
 else
-    data_size   equ (REG_SIZE * 5)
-    crc_table   equ (REG_SIZE + data_size)
-    num_VAR     equ [r4 + data_size]
-    table_VAR   equ [r4 + crc_table]
+  if (IS_CDECL gt 0)
+    crc_OFFS    equ (REG_SIZE * 5)
+    data_OFFS   equ (REG_SIZE + crc_OFFS)
+    size_OFFS   equ (REG_SIZE + data_OFFS)
+  else
+    size_OFFS   equ (REG_SIZE * 5)
+  endif
+    table_OFFS  equ (REG_SIZE + size_OFFS)
+    num_VAR     equ [r4 + size_OFFS]
+    table_VAR   equ [r4 + table_OFFS]
 endif
 
-SRCDAT  equ  rN + rD + 4 *
+SRCDAT  equ  rD + rN * 1 + 4 *
 
 CRC macro op:req, dest:req, src:req, t:req
-    op      dest, DWORD PTR [r5 + src * 4 + 0400h * t]
+    op      dest, DWORD PTR [rT + src * 4 + 0400h * t]
 endm
 
 CRC_XOR macro dest:req, src:req, t:req
@@ -43,11 +50,33 @@ CRC1b macro
 endm
 
 MY_PROLOG macro crc_end:req
-    MY_PUSH_4_REGS
+
+    ifdef x64
+      if  (IS_LINUX gt 0)
+        MY_PUSH_2_REGS
+        mov     x0, REG_ABI_PARAM_0_x   ; x0 = x7
+        mov     rT, REG_ABI_PARAM_3     ; r5 = r1
+        mov     rN, REG_ABI_PARAM_2     ; r7 = r2
+        mov     rD, REG_ABI_PARAM_1     ; r2 = r6
+      else
+        MY_PUSH_4_REGS
+        mov     x0, REG_ABI_PARAM_0_x   ; x0 = x1
+        mov     rT, REG_ABI_PARAM_3     ; r5 = r9
+        mov     rN, REG_ABI_PARAM_2     ; r7 = r8
+        ; mov     rD, REG_ABI_PARAM_1     ; r2 = r2
+      endif
+    else
+        MY_PUSH_4_REGS
+      if  (IS_CDECL gt 0)
+        mov     x0, [r4 + crc_OFFS]
+        mov     rD, [r4 + data_OFFS]
+      else
+        mov     x0, REG_ABI_PARAM_0_x
+      endif
+        mov     rN, num_VAR
+        mov     rT, table_VAR
+    endif
     
-    mov     x0, x1
-    mov     rN, num_VAR
-    mov     r5, table_VAR
     test    rN, rN
     jz      crc_end
   @@:
@@ -77,7 +106,11 @@ MY_EPILOG macro crc_end:req
     CRC1b
     jmp     crc_end
   @@:
-    MY_POP_4_REGS
+      if (IS_X64 gt 0) and (IS_LINUX gt 0)
+        MY_POP_2_REGS
+      else
+        MY_POP_4_REGS
+      endif
 endm
 
 MY_PROC CrcUpdateT8, 4

Asm/x86/AesOpt.asm

@@ -1,237 +1,742 @@
-; AesOpt.asm -- Intel's AES.
-; 2009-12-12 : Igor Pavlov : Public domain
+; AesOpt.asm -- AES optimized code for x86 AES hardware instructions
+; 2021-12-25 : Igor Pavlov : Public domain
 
 include 7zAsm.asm
 
+ifdef __ASMC__
+  use_vaes_256 equ 1
+else
+ifdef ymm0
+  use_vaes_256 equ 1
+endif
+endif
+
+
+ifdef use_vaes_256
+  ECHO "++ VAES 256"
+else
+  ECHO "-- NO VAES 256"
+endif
+
+ifdef x64
+  ECHO "x86-64"
+else
+  ECHO "x86"
+if (IS_CDECL gt 0)
+  ECHO "ABI : CDECL"
+else
+  ECHO "ABI : no CDECL : FASTCALL"
+endif
+endif
+
+if (IS_LINUX gt 0)
+  ECHO "ABI : LINUX"
+else
+  ECHO "ABI : WINDOWS"
+endif
+
 MY_ASM_START
 
 ifndef x64
+    .686
     .xmm
 endif
 
-ifdef x64
-    num     equ r8
-else
-    num     equ [r4 + REG_SIZE * 4]
+
+; MY_ALIGN EQU ALIGN(64)
+MY_ALIGN EQU
+
+SEG_ALIGN EQU MY_ALIGN
+
+MY_SEG_PROC macro name:req, numParams:req
+    ; seg_name equ @CatStr(_TEXT$, name)
+    ; seg_name SEGMENT SEG_ALIGN 'CODE'
+    MY_PROC name, numParams
+endm
+
+MY_SEG_ENDP macro
+    ; seg_name ENDS
+endm
+
+
+NUM_AES_KEYS_MAX equ 15
+
+; the number of push operators in function PROLOG
+if (IS_LINUX eq 0) or (IS_X64 eq 0)
+num_regs_push   equ 2
+stack_param_offset equ (REG_SIZE * (1 + num_regs_push))
 endif
 
-rD equ r2
-rN equ r0
+ifdef x64
+    num_param   equ REG_ABI_PARAM_2
+else
+  if (IS_CDECL gt 0)
+    ;   size_t     size
+    ;   void *     data
+    ;   UInt32 *   aes
+    ;   ret-ip <- (r4)
+    aes_OFFS    equ (stack_param_offset)
+    data_OFFS   equ (REG_SIZE + aes_OFFS)
+    size_OFFS   equ (REG_SIZE + data_OFFS)
+    num_param   equ [r4 + size_OFFS]
+  else
+    num_param   equ [r4 + stack_param_offset]
+  endif
+endif
 
-MY_PROLOG macro reg:req
-    ifdef x64
-    movdqa  [r4 + 8], xmm6
-    movdqa  [r4 + 8 + 16], xmm7
-    endif
+keys    equ  REG_PARAM_0  ; r1
+rD      equ  REG_PARAM_1  ; r2
+rN      equ  r0
 
-    push    r3
-    push    r5
-    push    r6
+koffs_x equ  x7
+koffs_r equ  r7
 
-    mov     rN, num
-    mov     x6, [r1 + 16]
-    shl     x6, 5
+ksize_x equ  x6
+ksize_r equ  r6
 
-    movdqa  reg, [r1]
-    add     r1, 32
-endm
+keys2   equ  r3
 
-MY_EPILOG macro
-    pop     r6
-    pop     r5
-    pop     r3
+state   equ  xmm0
+key     equ  xmm0
+key_ymm equ  ymm0
+key_ymm_n equ   0
 
-    ifdef x64
-    movdqa  xmm6, [r4 + 8]
-    movdqa  xmm7, [r4 + 8 + 16]
-    endif
+ifdef x64
+        ways = 11
+else
+        ways = 4
+endif
 
-    MY_ENDP
-endm
+ways_start_reg equ 1
 
-ways equ 4
-ways16 equ (ways * 16)
+iv      equ     @CatStr(xmm, %(ways_start_reg + ways))
+iv_ymm  equ     @CatStr(ymm, %(ways_start_reg + ways))
 
-OP_W macro op, op2
+
+WOP macro op, op2
     i = 0
     rept ways
-    op @CatStr(xmm,%i), op2
-    i = i + 1
+        op      @CatStr(xmm, %(ways_start_reg + i)), op2
+        i = i + 1
     endm
 endm
 
-LOAD_OP macro op:req, offs:req
-    op      xmm0, [r1 + r3 offs]
-endm
+
+ifndef ABI_LINUX
+ifdef x64
+
+; we use 32 bytes of home space in stack in WIN64-x64
+NUM_HOME_MM_REGS   equ (32 / 16)
+; we preserve xmm registers starting from xmm6 in WIN64-x64
+MM_START_SAVE_REG  equ 6
+
+SAVE_XMM macro num_used_mm_regs:req
+  num_save_mm_regs = num_used_mm_regs - MM_START_SAVE_REG
+  if num_save_mm_regs GT 0
+    num_save_mm_regs2 = num_save_mm_regs - NUM_HOME_MM_REGS
+    ; RSP is (16*x + 8) after entering the function in WIN64-x64
+    stack_offset = 16 * num_save_mm_regs2 + (stack_param_offset mod 16)
   
-LOAD_OP_W macro op:req, offs:req
-    movdqa  xmm7, [r1 + r3 offs]
-    OP_W    op, xmm7
+    i = 0
+    rept num_save_mm_regs
+      
+      if i eq NUM_HOME_MM_REGS
+        sub  r4, stack_offset
+      endif
+      
+      if i lt NUM_HOME_MM_REGS
+        movdqa  [r4 + stack_param_offset + i * 16], @CatStr(xmm, %(MM_START_SAVE_REG + i))
+      else
+        movdqa  [r4 + (i - NUM_HOME_MM_REGS) * 16], @CatStr(xmm, %(MM_START_SAVE_REG + i))
+      endif
+      
+      i = i + 1
+    endm
+  endif
+endm
+
+RESTORE_XMM macro num_used_mm_regs:req
+  if num_save_mm_regs GT 0
+    i = 0
+    if num_save_mm_regs2 GT 0
+      rept num_save_mm_regs2
+        movdqa  @CatStr(xmm, %(MM_START_SAVE_REG + NUM_HOME_MM_REGS + i)), [r4 + i * 16]
+        i = i + 1
+      endm
+        add     r4, stack_offset
+    endif
+
+    num_low_regs = num_save_mm_regs - i
+    i = 0
+      rept num_low_regs
+        movdqa  @CatStr(xmm, %(MM_START_SAVE_REG + i)), [r4 + stack_param_offset + i * 16]
+        i = i + 1
+      endm
+  endif
+endm
+
+endif ; x64
+endif ; ABI_LINUX
+
+
+MY_PROLOG macro num_used_mm_regs:req
+        ; num_regs_push: must be equal to the number of push operators
+        ; push    r3
+        ; push    r5
+    if (IS_LINUX eq 0) or (IS_X64 eq 0)
+        push    r6
+        push    r7
+    endif
+
+        mov     rN, num_param  ; don't move it; num_param can use stack pointer (r4)
+
+    if (IS_X64 eq 0)
+      if (IS_CDECL gt 0)
+        mov     rD,   [r4 + data_OFFS]
+        mov     keys, [r4 + aes_OFFS]
+      endif
+    elseif (IS_LINUX gt 0)
+        MY_ABI_LINUX_TO_WIN_2
+    endif
+
+
+    ifndef ABI_LINUX
+    ifdef x64
+        SAVE_XMM num_used_mm_regs
+    endif
+    endif
+   
+        mov     ksize_x, [keys + 16]
+        shl     ksize_x, 5
+endm
+
+
+MY_EPILOG macro
+    ifndef ABI_LINUX
+    ifdef x64
+        RESTORE_XMM num_save_mm_regs
+    endif
+    endif
+    
+    if (IS_LINUX eq 0) or (IS_X64 eq 0)
+        pop     r7
+        pop     r6
+    endif
+        ; pop     r5
+        ; pop     r3
+    MY_ENDP
+endm
+
+
+OP_KEY macro op:req, offs:req
+        op      state, [keys + offs]
+endm
+
+ 
+WOP_KEY macro op:req, offs:req
+        movdqa  key, [keys + offs]
+        WOP     op, key
 endm
 
 
 ; ---------- AES-CBC Decode ----------
 
-CBC_DEC_UPDATE macro reg, offs
-    pxor    reg, xmm6
-    movdqa  xmm6, [rD + offs]
-    movdqa  [rD + offs], reg
+
+XOR_WITH_DATA macro reg, _ppp_
+        pxor    reg, [rD + i * 16]
 endm
 
-DECODE macro op:req
-    op      aesdec, +16
-  @@:
-    op      aesdec, +0
-    op      aesdec, -16
-    sub     x3, 32
-    jnz     @B
-    op      aesdeclast, +0
+WRITE_TO_DATA macro reg, _ppp_
+        movdqa  [rD + i * 16], reg
 endm
 
-MY_PROC AesCbc_Decode_Intel, 3
-    MY_PROLOG xmm6
 
-    sub     x6, 32
+; state0    equ  @CatStr(xmm, %(ways_start_reg))
 
-    jmp     check2
+key0            equ  @CatStr(xmm, %(ways_start_reg + ways + 1))
+key0_ymm        equ  @CatStr(ymm, %(ways_start_reg + ways + 1))
 
-  align 16
-  nextBlocks2:
-    mov     x3, x6
-    OP_W    movdqa, [rD + i * 16]
-    LOAD_OP_W  pxor, +32
-    DECODE  LOAD_OP_W
-    OP_W    CBC_DEC_UPDATE, i * 16
-    add     rD, ways16
-  check2:
-    sub     rN, ways
-    jnc     nextBlocks2
+key_last        equ  @CatStr(xmm, %(ways_start_reg + ways + 2))
+key_last_ymm    equ  @CatStr(ymm, %(ways_start_reg + ways + 2))
+key_last_ymm_n  equ                (ways_start_reg + ways + 2)
 
-    add     rN, ways
-    jmp     check
-
-  nextBlock:
-    mov     x3, x6
-    movdqa  xmm1, [rD]
-    LOAD_OP movdqa, +32
-    pxor    xmm0, xmm1
-    DECODE  LOAD_OP
-    pxor    xmm0, xmm6
-    movdqa  [rD], xmm0
-    movdqa  xmm6, xmm1
-    add     rD, 16
-  check:
-    sub     rN, 1
-    jnc     nextBlock
-
-    movdqa  [r1 - 32], xmm6
-    MY_EPILOG
+NUM_CBC_REGS    equ  (ways_start_reg + ways + 3)
 
 
-; ---------- AES-CBC Encode ----------
+MY_SEG_PROC AesCbc_Decode_HW, 3
 
-ENCODE macro op:req
-    op      aesenc, -16
-  @@:
-    op      aesenc, +0
-    op      aesenc, +16
-    add     r3, 32
-    jnz     @B
-    op      aesenclast, +0
-endm
+    AesCbc_Decode_HW_start::
+        MY_PROLOG NUM_CBC_REGS
+   
+    AesCbc_Decode_HW_start_2::
+        movdqa  iv, [keys]
+        add     keys, 32
 
-MY_PROC AesCbc_Encode_Intel, 3
-    MY_PROLOG xmm0
+        movdqa  key0, [keys + 1 * ksize_r]
+        movdqa  key_last, [keys]
+        sub     ksize_x, 16
 
-    add     r1, r6
-    neg     r6
-    add     r6, 32
+        jmp     check2
+    align 16
+    nextBlocks2:
+        WOP     movdqa, [rD + i * 16]
+        mov     koffs_x, ksize_x
+        ; WOP_KEY pxor, ksize_r + 16
+        WOP     pxor, key0
+    ; align 16
+    @@:
+        WOP_KEY aesdec, 1 * koffs_r
+        sub     koffs_r, 16
+        jnz     @B
+        ; WOP_KEY aesdeclast, 0
+        WOP     aesdeclast, key_last
+        
+        pxor    @CatStr(xmm, %(ways_start_reg)), iv
+    i = 1
+    rept ways - 1
+        pxor    @CatStr(xmm, %(ways_start_reg + i)), [rD + i * 16 - 16]
+        i = i + 1
+    endm
+        movdqa  iv, [rD + ways * 16 - 16]
+        WOP     WRITE_TO_DATA
 
-    jmp     check_e
+        add     rD, ways * 16
+    AesCbc_Decode_HW_start_3::
+    check2:
+        sub     rN, ways
+        jnc     nextBlocks2
+        add     rN, ways
 
-  align 16
-  nextBlock_e:
-    mov     r3, r6
-    pxor    xmm0, [rD]
-    pxor    xmm0, [r1 + r3 - 32]
-    ENCODE  LOAD_OP
-    movdqa  [rD], xmm0
-    add     rD, 16
-  check_e:
-    sub     rN, 1
-    jnc     nextBlock_e
+        sub     ksize_x, 16
 
-    movdqa  [r1 + r6 - 64], xmm0
-    MY_EPILOG
+        jmp     check
+    nextBlock:
+        movdqa  state, [rD]
+        mov     koffs_x, ksize_x
+        ; OP_KEY  pxor, 1 * ksize_r + 32
+        pxor    state, key0
+        ; movdqa  state0, [rD]
+        ; movdqa  state, key0
+        ; pxor    state, state0
+    @@:
+        OP_KEY  aesdec, 1 * koffs_r + 16
+        OP_KEY  aesdec, 1 * koffs_r
+        sub     koffs_r, 32
+        jnz     @B
+        OP_KEY  aesdec, 16
+        ; OP_KEY  aesdeclast, 0
+        aesdeclast state, key_last
+        
+        pxor    state, iv
+        movdqa  iv, [rD]
+        ; movdqa  iv, state0
+        movdqa  [rD], state
+        
+        add     rD, 16
+    check:
+        sub     rN, 1
+        jnc     nextBlock
+
+        movdqa  [keys - 32], iv
+MY_EPILOG
 
 
-; ---------- AES-CTR ----------
 
-XOR_UPD_1 macro reg, offs
-    pxor    reg, [rD + offs]
-endm
 
-XOR_UPD_2 macro reg, offs
-    movdqa  [rD + offs], reg
-endm
+; ---------- AVX ----------
 
-MY_PROC AesCtr_Code_Intel, 3
-    MY_PROLOG xmm6
-
-    mov     r5, r4
-    shr     r5, 4
-    dec     r5
-    shl     r5, 4
-
-    mov     DWORD PTR [r5], 1
-    mov     DWORD PTR [r5 + 4], 0
-    mov     DWORD PTR [r5 + 8], 0
-    mov     DWORD PTR [r5 + 12], 0
-    
-    add     r1, r6
-    neg     r6
-    add     r6, 32
-
-    jmp     check2_c
-
-  align 16
-  nextBlocks2_c:
-    movdqa  xmm7, [r5]
 
+AVX__WOP_n macro op
     i = 0
     rept ways
-    paddq   xmm6, xmm7
-    movdqa  @CatStr(xmm,%i), xmm6
-    i = i + 1
+        op      (ways_start_reg + i)
+        i = i + 1
+    endm
+endm
+
+AVX__WOP macro op
+    i = 0
+    rept ways
+        op      @CatStr(ymm, %(ways_start_reg + i))
+        i = i + 1
+    endm
+endm
+
+
+AVX__WOP_KEY macro op:req, offs:req
+        vmovdqa  key_ymm, ymmword ptr [keys2 + offs]
+        AVX__WOP_n op
+endm
+
+
+AVX__CBC_START macro reg
+        ; vpxor   reg, key_ymm, ymmword ptr [rD + 32 * i]
+        vpxor   reg, key0_ymm, ymmword ptr [rD + 32 * i]
+endm
+
+AVX__CBC_END macro reg
+    if i eq 0
+        vpxor   reg, reg, iv_ymm
+    else
+        vpxor   reg, reg, ymmword ptr [rD + i * 32 - 16]
+    endif
+endm
+
+
+AVX__WRITE_TO_DATA macro reg
+        vmovdqu ymmword ptr [rD + 32 * i], reg
+endm
+
+AVX__XOR_WITH_DATA macro reg
+        vpxor   reg, reg, ymmword ptr [rD + 32 * i]
+endm
+
+AVX__CTR_START macro reg
+        vpaddq  iv_ymm, iv_ymm, one_ymm
+        ; vpxor   reg, iv_ymm, key_ymm
+        vpxor   reg, iv_ymm, key0_ymm
+endm
+
+
+MY_VAES_INSTR_2 macro cmd, dest, a1, a2
+  db 0c4H
+  db 2 + 040H + 020h * (1 - (a2) / 8) + 080h * (1 - (dest) / 8)
+  db 5 + 8 * ((not (a1)) and 15)
+  db cmd
+  db 0c0H + 8 * ((dest) and 7) + ((a2) and 7)
+endm
+
+MY_VAES_INSTR macro cmd, dest, a
+        MY_VAES_INSTR_2  cmd, dest, dest, a
+endm
+
+MY_vaesenc macro dest, a
+        MY_VAES_INSTR  0dcH, dest, a
+endm
+MY_vaesenclast macro dest, a
+        MY_VAES_INSTR  0ddH, dest, a
+endm
+MY_vaesdec macro dest, a
+        MY_VAES_INSTR  0deH, dest, a
+endm
+MY_vaesdeclast macro dest, a
+        MY_VAES_INSTR  0dfH, dest, a
+endm
+
+
+AVX__VAES_DEC macro reg
+        MY_vaesdec reg, key_ymm_n
+endm
+
+AVX__VAES_DEC_LAST_key_last macro reg
+        ; MY_vaesdeclast reg, key_ymm_n
+        MY_vaesdeclast reg, key_last_ymm_n
+endm
+
+AVX__VAES_ENC macro reg
+        MY_vaesenc reg, key_ymm_n
+endm
+
+AVX__VAES_ENC_LAST macro reg
+        MY_vaesenclast reg, key_ymm_n
+endm
+
+AVX__vinserti128_TO_HIGH macro dest, src
+        vinserti128  dest, dest, src, 1
+endm
+
+
+MY_PROC AesCbc_Decode_HW_256, 3
+  ifdef use_vaes_256
+        MY_PROLOG NUM_CBC_REGS
+        
+        cmp    rN, ways * 2
+        jb     AesCbc_Decode_HW_start_2
+
+        vmovdqa iv, xmmword ptr [keys]
+        add     keys, 32
+
+        vbroadcasti128  key0_ymm, xmmword ptr [keys + 1 * ksize_r]
+        vbroadcasti128  key_last_ymm, xmmword ptr [keys]
+        sub     ksize_x, 16
+        mov     koffs_x, ksize_x
+        add     ksize_x, ksize_x
+        
+        AVX_STACK_SUB = ((NUM_AES_KEYS_MAX + 1 - 2) * 32)
+        push    keys2
+        sub     r4, AVX_STACK_SUB
+        ; sub     r4, 32
+        ; sub     r4, ksize_r
+        ; lea     keys2, [r4 + 32]
+        mov     keys2, r4
+        and     keys2, -32
+    broad:
+        vbroadcasti128  key_ymm, xmmword ptr [keys + 1 * koffs_r]
+        vmovdqa         ymmword ptr [keys2 + koffs_r * 2], key_ymm
+        sub     koffs_r, 16
+        ; jnc     broad
+        jnz     broad
+
+        sub     rN, ways * 2
+
+    align 16
+    avx_cbcdec_nextBlock2:
+        mov     koffs_x, ksize_x
+        ; AVX__WOP_KEY    AVX__CBC_START, 1 * koffs_r + 32
+        AVX__WOP    AVX__CBC_START
+    @@:
+        AVX__WOP_KEY    AVX__VAES_DEC, 1 * koffs_r
+        sub     koffs_r, 32
+        jnz     @B
+        ; AVX__WOP_KEY    AVX__VAES_DEC_LAST, 0
+        AVX__WOP_n   AVX__VAES_DEC_LAST_key_last
+
+        AVX__vinserti128_TO_HIGH  iv_ymm, xmmword ptr [rD]
+        AVX__WOP        AVX__CBC_END
+
+        vmovdqa         iv, xmmword ptr [rD + ways * 32 - 16]
+        AVX__WOP        AVX__WRITE_TO_DATA
+        
+        add     rD, ways * 32
+        sub     rN, ways * 2
+        jnc     avx_cbcdec_nextBlock2
+        add     rN, ways * 2
+
+        shr     ksize_x, 1
+        
+        ; lea     r4, [r4 + 1 * ksize_r + 32]
+        add     r4, AVX_STACK_SUB
+        pop     keys2
+
+        vzeroupper
+        jmp     AesCbc_Decode_HW_start_3
+  else
+        jmp     AesCbc_Decode_HW_start
+  endif
+MY_ENDP
+MY_SEG_ENDP
+
+
+
+    
+; ---------- AES-CBC Encode ----------
+
+e0  equ  xmm1
+
+CENC_START_KEY     equ 2
+CENC_NUM_REG_KEYS  equ (3 * 2)
+; last_key equ @CatStr(xmm, %(CENC_START_KEY + CENC_NUM_REG_KEYS))
+
+MY_SEG_PROC AesCbc_Encode_HW, 3
+        MY_PROLOG (CENC_START_KEY + CENC_NUM_REG_KEYS + 0)
+
+        movdqa  state, [keys]
+        add     keys, 32
+        
+    i = 0
+    rept CENC_NUM_REG_KEYS
+        movdqa  @CatStr(xmm, %(CENC_START_KEY + i)), [keys + i * 16]
+        i = i + 1
+    endm
+                                        
+        add     keys, ksize_r
+        neg     ksize_r
+        add     ksize_r, (16 * CENC_NUM_REG_KEYS)
+        ; movdqa  last_key, [keys]
+        jmp     check_e
+
+    align 16
+    nextBlock_e:
+        movdqa  e0, [rD]
+        mov     koffs_r, ksize_r
+        pxor    e0, @CatStr(xmm, %(CENC_START_KEY))
+        pxor    state, e0
+        
+    i = 1
+    rept (CENC_NUM_REG_KEYS - 1)
+        aesenc  state, @CatStr(xmm, %(CENC_START_KEY + i))
+        i = i + 1
     endm
 
-    mov     r3, r6
-    LOAD_OP_W  pxor, -32
-    ENCODE  LOAD_OP_W
-    OP_W    XOR_UPD_1, i * 16
-    OP_W    XOR_UPD_2, i * 16
-    add     rD, ways16
-  check2_c:
-    sub     rN, ways
-    jnc     nextBlocks2_c
+    @@:
+        OP_KEY  aesenc, 1 * koffs_r
+        OP_KEY  aesenc, 1 * koffs_r + 16
+        add     koffs_r, 32
+        jnz     @B
+        OP_KEY  aesenclast, 0
+        ; aesenclast state, last_key
+        
+        movdqa  [rD], state
+        add     rD, 16
+    check_e:
+        sub     rN, 1
+        jnc     nextBlock_e
 
-    add     rN, ways
-    jmp     check_c
+        ; movdqa  [keys - 32], state
+        movdqa  [keys + 1 * ksize_r - (16 * CENC_NUM_REG_KEYS) - 32], state
+MY_EPILOG
+MY_SEG_ENDP
 
-  nextBlock_c:
-    paddq   xmm6, [r5]
-    mov     r3, r6
-    movdqa  xmm0, [r1 + r3 - 32]
-    pxor    xmm0, xmm6
-    ENCODE  LOAD_OP
-    XOR_UPD_1 xmm0, 0
-    XOR_UPD_2 xmm0, 0
-    add     rD, 16
-  check_c:
-    sub     rN, 1
-    jnc     nextBlock_c
 
-    movdqa  [r1 + r6 - 64], xmm6
-    MY_EPILOG
+    
+; ---------- AES-CTR ----------
+
+ifdef x64
+        ; ways = 11
+endif
+
+       
+one             equ  @CatStr(xmm, %(ways_start_reg + ways + 1))
+one_ymm         equ  @CatStr(ymm, %(ways_start_reg + ways + 1))
+key0            equ  @CatStr(xmm, %(ways_start_reg + ways + 2))
+key0_ymm        equ  @CatStr(ymm, %(ways_start_reg + ways + 2))
+NUM_CTR_REGS    equ  (ways_start_reg + ways + 3)
+
+INIT_CTR macro reg, _ppp_
+        paddq   iv, one
+        movdqa  reg, iv
+endm
+
+
+MY_SEG_PROC AesCtr_Code_HW, 3
+    Ctr_start::
+        MY_PROLOG NUM_CTR_REGS
+
+    Ctr_start_2::
+        movdqa  iv, [keys]
+        add     keys, 32
+        movdqa  key0, [keys]
+
+        add     keys, ksize_r
+        neg     ksize_r
+        add     ksize_r, 16
+        
+    Ctr_start_3::
+        mov     koffs_x, 1
+        movd    one, koffs_x
+        jmp     check2_c
+
+    align 16
+    nextBlocks2_c:
+        WOP     INIT_CTR, 0
+        mov     koffs_r, ksize_r
+        ; WOP_KEY pxor, 1 * koffs_r -16
+        WOP     pxor, key0
+    @@:
+        WOP_KEY aesenc, 1 * koffs_r
+        add     koffs_r, 16
+        jnz     @B
+        WOP_KEY aesenclast, 0
+        
+        WOP     XOR_WITH_DATA
+        WOP     WRITE_TO_DATA
+        add     rD, ways * 16
+    check2_c:
+        sub     rN, ways
+        jnc     nextBlocks2_c
+        add     rN, ways
+
+        sub     keys, 16
+        add     ksize_r, 16
+        
+        jmp     check_c
+
+    ; align 16
+    nextBlock_c:
+        paddq   iv, one
+        ; movdqa  state, [keys + 1 * koffs_r - 16]
+        movdqa  state, key0
+        mov     koffs_r, ksize_r
+        pxor    state, iv
+        
+    @@:
+        OP_KEY  aesenc, 1 * koffs_r
+        OP_KEY  aesenc, 1 * koffs_r + 16
+        add     koffs_r, 32
+        jnz     @B
+        OP_KEY  aesenc, 0
+        OP_KEY  aesenclast, 16
+        
+        pxor    state, [rD]
+        movdqa  [rD], state
+        add     rD, 16
+    check_c:
+        sub     rN, 1
+        jnc     nextBlock_c
+
+        ; movdqa  [keys - 32], iv
+        movdqa  [keys + 1 * ksize_r - 16 - 32], iv
+MY_EPILOG
+
+
+MY_PROC AesCtr_Code_HW_256, 3
+  ifdef use_vaes_256
+        MY_PROLOG NUM_CTR_REGS
+
+        cmp    rN, ways * 2
+        jb     Ctr_start_2
+
+        vbroadcasti128  iv_ymm, xmmword ptr [keys]
+        add     keys, 32
+        vbroadcasti128  key0_ymm, xmmword ptr [keys]
+        mov     koffs_x, 1
+        vmovd           one, koffs_x
+        vpsubq  iv_ymm, iv_ymm, one_ymm
+        vpaddq  one, one, one
+        AVX__vinserti128_TO_HIGH     one_ymm, one
+        
+        add     keys, ksize_r
+        sub     ksize_x, 16
+        neg     ksize_r
+        mov     koffs_r, ksize_r
+        add     ksize_r, ksize_r
+
+        AVX_STACK_SUB = ((NUM_AES_KEYS_MAX + 1 - 1) * 32)
+        push    keys2
+        lea     keys2, [r4 - 32]
+        sub     r4, AVX_STACK_SUB
+        and     keys2, -32
+        vbroadcasti128  key_ymm, xmmword ptr [keys]
+        vmovdqa         ymmword ptr [keys2], key_ymm
+     @@:
+        vbroadcasti128  key_ymm, xmmword ptr [keys + 1 * koffs_r]
+        vmovdqa         ymmword ptr [keys2 + koffs_r * 2], key_ymm
+        add     koffs_r, 16
+        jnz     @B
+
+        sub     rN, ways * 2
+        
+    align 16
+    avx_ctr_nextBlock2:
+        mov             koffs_r, ksize_r
+        AVX__WOP        AVX__CTR_START
+        ; AVX__WOP_KEY    AVX__CTR_START, 1 * koffs_r - 32
+    @@:
+        AVX__WOP_KEY    AVX__VAES_ENC, 1 * koffs_r
+        add     koffs_r, 32
+        jnz     @B
+        AVX__WOP_KEY    AVX__VAES_ENC_LAST, 0
+       
+        AVX__WOP        AVX__XOR_WITH_DATA
+        AVX__WOP        AVX__WRITE_TO_DATA
+        
+        add     rD, ways * 32
+        sub     rN, ways * 2
+        jnc     avx_ctr_nextBlock2
+        add     rN, ways * 2
+        
+        vextracti128    iv, iv_ymm, 1
+        sar     ksize_r, 1
+       
+        add     r4, AVX_STACK_SUB
+        pop     keys2
+        
+        vzeroupper
+        jmp     Ctr_start_3
+  else
+        jmp     Ctr_start
+  endif
+MY_ENDP
+MY_SEG_ENDP
 
 end

Asm/x86/LzFindOpt.asm

@@ -0,0 +1,513 @@
+; LzFindOpt.asm -- ASM version of GetMatchesSpecN_2() function
+; 2021-07-21: Igor Pavlov : Public domain
+;
+
+ifndef x64
+; x64=1
+; .err <x64_IS_REQUIRED>
+endif
+
+include 7zAsm.asm
+
+MY_ASM_START
+
+_TEXT$LZFINDOPT SEGMENT ALIGN(64) 'CODE'
+
+MY_ALIGN macro num:req
+        align  num
+endm
+
+MY_ALIGN_32 macro
+        MY_ALIGN 32
+endm
+
+MY_ALIGN_64 macro
+        MY_ALIGN 64
+endm
+
+
+t0_L    equ x0_L
+t0_x    equ x0
+t0      equ r0
+t1_x    equ x3
+t1      equ r3
+
+cp_x    equ t1_x
+cp_r    equ t1
+m       equ x5
+m_r     equ r5
+len_x   equ x6
+len     equ r6
+diff_x  equ x7
+diff    equ r7
+len0    equ r10
+len1_x  equ x11
+len1    equ r11
+maxLen_x equ x12
+maxLen  equ r12
+d       equ r13
+ptr0    equ r14
+ptr1    equ r15
+
+d_lim       equ m_r
+cycSize     equ len_x
+hash_lim    equ len0
+delta1_x    equ len1_x
+delta1_r    equ len1
+delta_x     equ maxLen_x
+delta_r     equ maxLen
+hash        equ ptr0
+src         equ ptr1
+
+
+
+if (IS_LINUX gt 0)
+
+; r1 r2  r8 r9        : win32
+; r7 r6  r2 r1  r8 r9 : linux
+
+lenLimit        equ r8
+lenLimit_x      equ x8
+; pos_r           equ r2
+pos             equ x2
+cur             equ r1
+son             equ r9
+
+else
+
+lenLimit        equ REG_ABI_PARAM_2
+lenLimit_x      equ REG_ABI_PARAM_2_x
+pos             equ REG_ABI_PARAM_1_x
+cur             equ REG_ABI_PARAM_0
+son             equ REG_ABI_PARAM_3
+
+endif
+
+
+if (IS_LINUX gt 0)
+    maxLen_OFFS         equ  (REG_SIZE * (6 + 1))
+else
+    cutValue_OFFS       equ  (REG_SIZE * (8 + 1 + 4))
+    d_OFFS              equ  (REG_SIZE + cutValue_OFFS)
+    maxLen_OFFS         equ  (REG_SIZE + d_OFFS)
+endif
+    hash_OFFS           equ  (REG_SIZE + maxLen_OFFS)
+    limit_OFFS          equ  (REG_SIZE + hash_OFFS)
+    size_OFFS           equ  (REG_SIZE + limit_OFFS)
+    cycPos_OFFS         equ  (REG_SIZE + size_OFFS)
+    cycSize_OFFS        equ  (REG_SIZE + cycPos_OFFS)
+    posRes_OFFS         equ  (REG_SIZE + cycSize_OFFS)
+    
+if (IS_LINUX gt 0)
+else
+    cutValue_PAR        equ  [r0 + cutValue_OFFS]
+    d_PAR               equ  [r0 + d_OFFS]
+endif
+    maxLen_PAR          equ  [r0 + maxLen_OFFS]
+    hash_PAR            equ  [r0 + hash_OFFS]
+    limit_PAR           equ  [r0 + limit_OFFS]
+    size_PAR            equ  [r0 + size_OFFS]
+    cycPos_PAR          equ  [r0 + cycPos_OFFS]
+    cycSize_PAR         equ  [r0 + cycSize_OFFS]
+    posRes_PAR          equ  [r0 + posRes_OFFS]
+
+
+    cutValue_VAR        equ  DWORD PTR [r4 + 8 * 0]
+    cutValueCur_VAR     equ  DWORD PTR [r4 + 8 * 0 + 4]
+    cycPos_VAR          equ  DWORD PTR [r4 + 8 * 1 + 0]
+    cycSize_VAR         equ  DWORD PTR [r4 + 8 * 1 + 4]
+    hash_VAR            equ  QWORD PTR [r4 + 8 * 2]
+    limit_VAR           equ  QWORD PTR [r4 + 8 * 3]
+    size_VAR            equ  QWORD PTR [r4 + 8 * 4]
+    distances           equ  QWORD PTR [r4 + 8 * 5]
+    maxLen_VAR          equ  QWORD PTR [r4 + 8 * 6]
+
+    Old_RSP             equ  QWORD PTR [r4 + 8 * 7]
+    LOCAL_SIZE          equ  8 * 8
+
+COPY_VAR_32 macro dest_var, src_var
+        mov     x3, src_var
+        mov     dest_var, x3
+endm
+
+COPY_VAR_64 macro dest_var, src_var
+        mov     r3, src_var
+        mov     dest_var, r3
+endm
+
+
+; MY_ALIGN_64
+MY_PROC GetMatchesSpecN_2, 13
+MY_PUSH_PRESERVED_ABI_REGS
+        mov     r0, RSP
+        lea     r3, [r0 - LOCAL_SIZE]
+        and     r3, -64
+        mov     RSP, r3
+        mov     Old_RSP, r0
+
+if (IS_LINUX gt 0)
+        mov     d,            REG_ABI_PARAM_5       ; r13 = r9
+        mov     cutValue_VAR, REG_ABI_PARAM_4_x     ;     = r8
+        mov     son,          REG_ABI_PARAM_3       ;  r9 = r1
+        mov     r8,           REG_ABI_PARAM_2       ;  r8 = r2
+        mov     pos,          REG_ABI_PARAM_1_x     ;  r2 = x6
+        mov     r1,           REG_ABI_PARAM_0       ;  r1 = r7
+else
+        COPY_VAR_32 cutValue_VAR, cutValue_PAR
+        mov     d, d_PAR
+endif
+
+        COPY_VAR_64 limit_VAR, limit_PAR
+        
+        mov     hash_lim, size_PAR
+        mov     size_VAR, hash_lim
+        
+        mov     cp_x, cycPos_PAR
+        mov     hash, hash_PAR
+
+        mov     cycSize, cycSize_PAR
+        mov     cycSize_VAR, cycSize
+        
+        ; we want cur in (rcx). So we change the cur and lenLimit variables
+        sub     lenLimit, cur
+        neg     lenLimit_x
+        inc     lenLimit_x
+        
+        mov     t0_x, maxLen_PAR
+        sub     t0, lenLimit
+        mov     maxLen_VAR, t0
+
+        jmp     main_loop
+
+MY_ALIGN_64
+fill_empty:
+        ; ptr0 = *ptr1 = kEmptyHashValue;
+        mov     QWORD PTR [ptr1], 0
+        inc     pos
+        inc     cp_x
+        mov     DWORD PTR [d - 4], 0
+        cmp     d, limit_VAR
+        jae     fin
+        cmp     hash, hash_lim
+        je      fin
+
+; MY_ALIGN_64
+main_loop:
+        ; UInt32 delta = *hash++;
+        mov     diff_x, [hash]  ; delta
+        add     hash, 4
+        ; mov     cycPos_VAR, cp_x
+       
+        inc     cur
+        add     d, 4
+        mov     m, pos
+        sub     m, diff_x;      ; matchPos
+        
+        ; CLzRef *ptr1 = son + ((size_t)(pos) << 1) - CYC_TO_POS_OFFSET * 2;
+        lea     ptr1, [son + 8 * cp_r]
+        ; mov     cycSize, cycSize_VAR
+        cmp     pos, cycSize
+        jb      directMode      ; if (pos < cycSize_VAR)
+        
+        ; CYC MODE
+
+        cmp     diff_x, cycSize
+        jae     fill_empty      ; if (delta >= cycSize_VAR)
+        
+        xor     t0_x, t0_x
+        mov     cycPos_VAR, cp_x
+        sub     cp_x, diff_x
+        ; jae     prepare_for_tree_loop
+        ; add     cp_x, cycSize
+        cmovb   t0_x, cycSize
+        add     cp_x, t0_x      ; cp_x +=  (cycPos < delta ? cycSize : 0)
+        jmp     prepare_for_tree_loop
+        
+        
+directMode:
+        cmp     diff_x,  pos
+        je      fill_empty      ; if (delta == pos)
+        jae     fin_error       ; if (delta >= pos)
+        
+        mov     cycPos_VAR, cp_x
+        mov     cp_x, m
+        
+prepare_for_tree_loop:
+        mov     len0, lenLimit
+        mov     hash_VAR, hash
+        ; CLzRef *ptr0 = son + ((size_t)(pos) << 1) - CYC_TO_POS_OFFSET * 2 + 1;
+        lea     ptr0, [ptr1 + 4]
+        ; UInt32 *_distances = ++d;
+        mov     distances, d
+
+        neg     len0
+        mov     len1, len0
+
+        mov     t0_x, cutValue_VAR
+        mov     maxLen, maxLen_VAR
+        mov     cutValueCur_VAR, t0_x
+
+MY_ALIGN_32
+tree_loop:
+        neg     diff
+        mov     len, len0
+        cmp     len1, len0
+        cmovb   len, len1       ; len = (len1 < len0 ? len1 : len0);
+        add     diff, cur
+
+        mov     t0_x, [son + cp_r * 8]  ; prefetch
+        movzx   t0_x, BYTE PTR [diff + 1 * len]
+        lea     cp_r, [son + cp_r * 8]
+        cmp     [cur + 1 * len], t0_L
+        je      matched_1
+        
+        jb      left_0
+
+        mov     [ptr1], m
+        mov        m, [cp_r + 4]
+        lea     ptr1, [cp_r + 4]
+        sub     diff, cur ; FIX32
+        jmp     next_node
+
+MY_ALIGN_32
+left_0:
+        mov     [ptr0], m
+        mov        m, [cp_r]
+        mov     ptr0, cp_r
+        sub     diff, cur ; FIX32
+        ; jmp     next_node
+
+; ------------ NEXT NODE ------------
+; MY_ALIGN_32
+next_node:
+        mov     cycSize, cycSize_VAR
+        dec     cutValueCur_VAR
+        je      finish_tree
+        
+        add     diff_x, pos     ; prev_match = pos + diff
+        cmp     m, diff_x
+        jae     fin_error       ; if (new_match >= prev_match)
+        
+        mov     diff_x, pos
+        sub     diff_x, m       ; delta = pos - new_match
+        cmp     pos, cycSize
+        jae     cyc_mode_2      ; if (pos >= cycSize)
+
+        mov     cp_x, m
+        test    m, m
+        jne     tree_loop       ; if (m != 0)
+        
+finish_tree:
+        ; ptr0 = *ptr1 = kEmptyHashValue;
+        mov     DWORD PTR [ptr0], 0
+        mov     DWORD PTR [ptr1], 0
+
+        inc     pos
+        
+        ; _distances[-1] = (UInt32)(d - _distances);
+        mov     t0, distances
+        mov     t1, d
+        sub     t1, t0
+        shr     t1_x, 2
+        mov     [t0 - 4], t1_x
+
+        cmp     d, limit_VAR
+        jae     fin             ; if (d >= limit)
+   
+        mov     cp_x, cycPos_VAR
+        mov     hash, hash_VAR
+        mov     hash_lim, size_VAR
+        inc     cp_x
+        cmp     hash, hash_lim
+        jne     main_loop       ; if (hash != size)
+        jmp     fin
+        
+
+MY_ALIGN_32
+cyc_mode_2:
+        cmp     diff_x, cycSize
+        jae     finish_tree     ; if (delta >= cycSize)
+
+        mov     cp_x, cycPos_VAR
+        xor     t0_x, t0_x
+        sub     cp_x, diff_x    ; cp_x = cycPos - delta
+        cmovb   t0_x, cycSize
+        add     cp_x, t0_x      ; cp_x += (cycPos < delta ? cycSize : 0)
+        jmp     tree_loop
+
+        
+MY_ALIGN_32
+matched_1:
+
+        inc     len
+        ; cmp     len_x, lenLimit_x
+        je      short lenLimit_reach
+        movzx   t0_x, BYTE PTR [diff + 1 * len]
+        cmp     [cur + 1 * len], t0_L
+        jne     mismatch
+
+        
+MY_ALIGN_32
+match_loop:
+        ;  while (++len != lenLimit)  (len[diff] != len[0]) ;
+
+        inc     len
+        ; cmp     len_x, lenLimit_x
+        je      short lenLimit_reach
+        movzx   t0_x, BYTE PTR [diff + 1 * len]
+        cmp     BYTE PTR [cur + 1 * len], t0_L
+        je      match_loop
+
+mismatch:
+        jb      left_2
+
+        mov     [ptr1], m
+        mov        m, [cp_r + 4]
+        lea     ptr1, [cp_r + 4]
+        mov     len1, len
+
+        jmp     max_update
+        
+MY_ALIGN_32
+left_2:
+        mov     [ptr0], m
+        mov        m, [cp_r]
+        mov     ptr0, cp_r
+        mov     len0, len
+
+max_update:
+        sub     diff, cur       ; restore diff
+
+        cmp     maxLen, len
+        jae     next_node
+        
+        mov     maxLen, len
+        add     len, lenLimit
+        mov     [d], len_x
+        mov     t0_x, diff_x
+        not     t0_x
+        mov     [d + 4], t0_x
+        add     d, 8
+       
+        jmp     next_node
+
+
+        
+MY_ALIGN_32
+lenLimit_reach:
+
+        mov     delta_r, cur
+        sub     delta_r, diff
+        lea     delta1_r, [delta_r - 1]
+
+        mov     t0_x, [cp_r]
+        mov     [ptr1], t0_x
+        mov     t0_x, [cp_r + 4]
+        mov     [ptr0], t0_x
+
+        mov     [d], lenLimit_x
+        mov     [d + 4], delta1_x
+        add     d, 8
+
+        ; _distances[-1] = (UInt32)(d - _distances);
+        mov     t0, distances
+        mov     t1, d
+        sub     t1, t0
+        shr     t1_x, 2
+        mov     [t0 - 4], t1_x
+
+        mov     hash, hash_VAR
+        mov     hash_lim, size_VAR
+
+        inc     pos
+        mov     cp_x, cycPos_VAR
+        inc     cp_x
+
+        mov     d_lim, limit_VAR
+        mov     cycSize, cycSize_VAR
+        ; if (hash == size || *hash != delta || lenLimit[diff] != lenLimit[0] || d >= limit)
+        ;    break;
+        cmp     hash, hash_lim
+        je      fin
+        cmp     d, d_lim
+        jae     fin
+        cmp     delta_x, [hash]
+        jne     main_loop
+        movzx   t0_x, BYTE PTR [diff]
+        cmp     [cur], t0_L
+        jne     main_loop
+
+        ; jmp     main_loop     ; bypass for debug
+        
+        mov     cycPos_VAR, cp_x
+        shl     len, 3          ; cycSize * 8
+        sub     diff, cur       ; restore diff
+        xor     t0_x, t0_x
+        cmp     cp_x, delta_x   ; cmp (cycPos_VAR, delta)
+        lea     cp_r, [son + 8 * cp_r]  ; dest
+        lea     src, [cp_r + 8 * diff]
+        cmovb   t0, len         ; t0 =  (cycPos_VAR < delta ? cycSize * 8 : 0)
+        add     src, t0
+        add     len, son        ; len = son + cycSize * 8
+
+       
+MY_ALIGN_32
+long_loop:
+        add     hash, 4
+        
+        ; *(UInt64 *)(void *)ptr = ((const UInt64 *)(const void *)ptr)[diff];
+        
+        mov     t0, [src]
+        add     src, 8
+        mov     [cp_r], t0
+        add     cp_r, 8
+        cmp     src, len
+        cmove   src, son       ; if end of (son) buffer is reached, we wrap to begin
+
+        mov     DWORD PTR [d], 2
+        mov     [d + 4], lenLimit_x
+        mov     [d + 8], delta1_x
+        add     d, 12
+
+        inc     cur
+
+        cmp     hash, hash_lim
+        je      long_footer
+        cmp     delta_x, [hash]
+        jne     long_footer
+        movzx   t0_x, BYTE PTR [diff + 1 * cur]
+        cmp     [cur], t0_L
+        jne     long_footer
+        cmp     d, d_lim
+        jb      long_loop
+
+long_footer:
+        sub     cp_r, son
+        shr     cp_r, 3
+        add     pos, cp_x
+        sub     pos, cycPos_VAR
+        mov     cycSize, cycSize_VAR
+        
+        cmp     d, d_lim
+        jae     fin
+        cmp     hash, hash_lim
+        jne     main_loop
+        jmp     fin
+
+
+
+fin_error:
+        xor     d, d
+        
+fin:
+        mov     RSP, Old_RSP
+        mov     t0, [r4 + posRes_OFFS]
+        mov     [t0], pos
+        mov     r0, d
+
+MY_POP_PRESERVED_ABI_REGS
+MY_ENDP
+
+_TEXT$LZFINDOPT ENDS
+
+end

Asm/x86/Sha1Opt.asm

@@ -0,0 +1,263 @@
+; Sha1Opt.asm -- SHA-1 optimized code for SHA-1 x86 hardware instructions
+; 2021-03-10 : Igor Pavlov : Public domain
+
+include 7zAsm.asm
+
+MY_ASM_START
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+CONST   SEGMENT
+
+align 16
+Reverse_Endian_Mask db 15,14,13,12, 11,10,9,8, 7,6,5,4, 3,2,1,0
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+CONST   ENDS
+
+; _TEXT$SHA1OPT SEGMENT 'CODE'
+
+ifndef x64
+    .686
+    .xmm
+endif
+
+ifdef x64
+        rNum    equ REG_ABI_PARAM_2
+    if (IS_LINUX eq 0)
+        LOCAL_SIZE equ (16 * 2)
+    endif
+else
+        rNum    equ r0
+        LOCAL_SIZE equ (16 * 1)
+endif
+
+rState equ REG_ABI_PARAM_0
+rData  equ REG_ABI_PARAM_1
+
+
+MY_sha1rnds4 macro a1, a2, imm
+        db 0fH, 03aH, 0ccH, (0c0H + a1 * 8 + a2), imm
+endm
+
+MY_SHA_INSTR macro cmd, a1, a2
+        db 0fH, 038H, cmd, (0c0H + a1 * 8 + a2)
+endm
+
+cmd_sha1nexte   equ 0c8H
+cmd_sha1msg1    equ 0c9H
+cmd_sha1msg2    equ 0caH
+
+MY_sha1nexte macro a1, a2
+        MY_SHA_INSTR  cmd_sha1nexte, a1, a2
+endm
+
+MY_sha1msg1 macro a1, a2
+        MY_SHA_INSTR  cmd_sha1msg1, a1, a2
+endm
+
+MY_sha1msg2 macro a1, a2
+        MY_SHA_INSTR  cmd_sha1msg2, a1, a2
+endm
+
+MY_PROLOG macro
+    ifdef x64
+      if (IS_LINUX eq 0)
+        movdqa  [r4 + 8], xmm6
+        movdqa  [r4 + 8 + 16], xmm7
+        sub     r4, LOCAL_SIZE + 8
+        movdqa  [r4     ], xmm8
+        movdqa  [r4 + 16], xmm9
+      endif
+    else ; x86
+      if (IS_CDECL gt 0)
+        mov     rState, [r4 + REG_SIZE * 1]
+        mov     rData,  [r4 + REG_SIZE * 2]
+        mov     rNum,   [r4 + REG_SIZE * 3]
+      else ; fastcall
+        mov     rNum,   [r4 + REG_SIZE * 1]
+      endif
+        push    r5
+        mov     r5, r4
+        and     r4, -16
+        sub     r4, LOCAL_SIZE
+    endif
+endm
+
+MY_EPILOG macro
+    ifdef x64
+      if (IS_LINUX eq 0)
+        movdqa  xmm8, [r4]
+        movdqa  xmm9, [r4 + 16]
+        add     r4, LOCAL_SIZE + 8
+        movdqa  xmm6, [r4 + 8]
+        movdqa  xmm7, [r4 + 8 + 16]
+      endif
+    else ; x86
+        mov     r4, r5
+        pop     r5
+    endif
+    MY_ENDP
+endm
+
+
+e0_N       equ 0
+e1_N       equ 1
+abcd_N     equ 2
+e0_save_N  equ 3
+w_regs     equ 4
+
+e0      equ @CatStr(xmm, %e0_N)
+e1      equ @CatStr(xmm, %e1_N)
+abcd    equ @CatStr(xmm, %abcd_N)
+e0_save equ @CatStr(xmm, %e0_save_N)
+
+
+ifdef x64
+        abcd_save    equ  xmm8
+        mask2        equ  xmm9
+else
+        abcd_save    equ  [r4]
+        mask2        equ  e1
+endif
+
+LOAD_MASK macro
+        movdqa  mask2, XMMWORD PTR Reverse_Endian_Mask
+endm
+
+LOAD_W macro k:req
+        movdqu  @CatStr(xmm, %(w_regs + k)), [rData + (16 * (k))]
+        pshufb  @CatStr(xmm, %(w_regs + k)), mask2
+endm
+
+
+; pre2 can be 2 or 3 (recommended)
+pre2 equ 3
+pre1 equ (pre2 + 1)
+
+NUM_ROUNDS4 equ 20
+   
+RND4 macro k
+        movdqa  @CatStr(xmm, %(e0_N + ((k + 1) mod 2))), abcd
+        MY_sha1rnds4 abcd_N, (e0_N + (k mod 2)), k / 5
+
+        nextM = (w_regs + ((k + 1) mod 4))
+
+    if (k EQ NUM_ROUNDS4 - 1)
+        nextM = e0_save_N
+    endif
+        
+        MY_sha1nexte (e0_N + ((k + 1) mod 2)), nextM
+        
+    if (k GE (4 - pre2)) AND (k LT (NUM_ROUNDS4 - pre2))
+        pxor @CatStr(xmm, %(w_regs + ((k + pre2) mod 4))), @CatStr(xmm, %(w_regs + ((k + pre2 - 2) mod 4)))
+    endif
+
+    if (k GE (4 - pre1)) AND (k LT (NUM_ROUNDS4 - pre1))
+        MY_sha1msg1 (w_regs + ((k + pre1) mod 4)), (w_regs + ((k + pre1 - 3) mod 4))
+    endif
+    
+    if (k GE (4 - pre2)) AND (k LT (NUM_ROUNDS4 - pre2))
+        MY_sha1msg2 (w_regs + ((k + pre2) mod 4)), (w_regs + ((k + pre2 - 1) mod 4))
+    endif
+endm
+
+
+REVERSE_STATE macro
+                               ; abcd   ; dcba
+                               ; e0     ; 000e
+        pshufd  abcd, abcd, 01bH        ; abcd
+        pshufd    e0,   e0, 01bH        ; e000
+endm
+
+
+
+
+
+MY_PROC Sha1_UpdateBlocks_HW, 3
+    MY_PROLOG
+
+        cmp     rNum, 0
+        je      end_c
+
+        movdqu   abcd, [rState]               ; dcba
+        movd     e0, dword ptr [rState + 16]  ; 000e
+
+        REVERSE_STATE
+       
+        ifdef x64
+        LOAD_MASK
+        endif
+
+    align 16
+    nextBlock:
+        movdqa  abcd_save, abcd
+        movdqa  e0_save, e0
+        
+        ifndef x64
+        LOAD_MASK
+        endif
+        
+        LOAD_W 0
+        LOAD_W 1
+        LOAD_W 2
+        LOAD_W 3
+
+        paddd   e0, @CatStr(xmm, %(w_regs))
+        k = 0
+        rept NUM_ROUNDS4
+          RND4 k
+          k = k + 1
+        endm
+
+        paddd   abcd, abcd_save
+
+
+        add     rData, 64
+        sub     rNum, 1
+        jnz     nextBlock
+        
+        REVERSE_STATE
+
+        movdqu  [rState], abcd
+        movd    dword ptr [rState + 16], e0
+       
+  end_c:
+MY_EPILOG
+
+; _TEXT$SHA1OPT ENDS
+
+end

Asm/x86/Sha256Opt.asm

@@ -0,0 +1,275 @@
+; Sha256Opt.asm -- SHA-256 optimized code for SHA-256 x86 hardware instructions
+; 2022-04-17 : Igor Pavlov : Public domain
+
+include 7zAsm.asm
+
+MY_ASM_START
+
+; .data
+; public K
+
+; we can use external SHA256_K_ARRAY defined in Sha256.c
+; but we must guarantee that SHA256_K_ARRAY is aligned for 16-bytes
+
+COMMENT @
+ifdef x64
+K_CONST equ SHA256_K_ARRAY
+else
+K_CONST equ _SHA256_K_ARRAY
+endif
+EXTRN   K_CONST:xmmword
+@
+
+CONST   SEGMENT
+
+align 16
+Reverse_Endian_Mask db 3,2,1,0, 7,6,5,4, 11,10,9,8, 15,14,13,12
+
+; COMMENT @
+align 16
+K_CONST \
+DD 0428a2f98H, 071374491H, 0b5c0fbcfH, 0e9b5dba5H
+DD 03956c25bH, 059f111f1H, 0923f82a4H, 0ab1c5ed5H
+DD 0d807aa98H, 012835b01H, 0243185beH, 0550c7dc3H
+DD 072be5d74H, 080deb1feH, 09bdc06a7H, 0c19bf174H
+DD 0e49b69c1H, 0efbe4786H, 00fc19dc6H, 0240ca1ccH
+DD 02de92c6fH, 04a7484aaH, 05cb0a9dcH, 076f988daH
+DD 0983e5152H, 0a831c66dH, 0b00327c8H, 0bf597fc7H
+DD 0c6e00bf3H, 0d5a79147H, 006ca6351H, 014292967H
+DD 027b70a85H, 02e1b2138H, 04d2c6dfcH, 053380d13H
+DD 0650a7354H, 0766a0abbH, 081c2c92eH, 092722c85H
+DD 0a2bfe8a1H, 0a81a664bH, 0c24b8b70H, 0c76c51a3H
+DD 0d192e819H, 0d6990624H, 0f40e3585H, 0106aa070H
+DD 019a4c116H, 01e376c08H, 02748774cH, 034b0bcb5H
+DD 0391c0cb3H, 04ed8aa4aH, 05b9cca4fH, 0682e6ff3H
+DD 0748f82eeH, 078a5636fH, 084c87814H, 08cc70208H
+DD 090befffaH, 0a4506cebH, 0bef9a3f7H, 0c67178f2H
+; @
+
+CONST   ENDS
+
+; _TEXT$SHA256OPT SEGMENT 'CODE'
+
+ifndef x64
+    .686
+    .xmm
+endif
+        
+; jwasm-based assemblers for linux and linker from new versions of binutils
+; can generate incorrect code for load [ARRAY + offset] instructions.
+; 22.00: we load K_CONST offset to (rTable) register to avoid jwasm+binutils problem 
+        rTable  equ r0
+        ; rTable  equ K_CONST
+        
+ifdef x64
+        rNum    equ REG_ABI_PARAM_2
+    if (IS_LINUX eq 0)
+        LOCAL_SIZE equ (16 * 2)
+    endif
+else
+        rNum    equ r3
+        LOCAL_SIZE equ (16 * 1)
+endif
+
+rState equ REG_ABI_PARAM_0
+rData  equ REG_ABI_PARAM_1
+
+
+
+
+
+
+MY_SHA_INSTR macro cmd, a1, a2
+        db 0fH, 038H, cmd, (0c0H + a1 * 8 + a2)
+endm
+
+cmd_sha256rnds2 equ 0cbH
+cmd_sha256msg1  equ 0ccH
+cmd_sha256msg2  equ 0cdH
+
+MY_sha256rnds2 macro a1, a2
+        MY_SHA_INSTR  cmd_sha256rnds2, a1, a2
+endm
+
+MY_sha256msg1 macro a1, a2
+        MY_SHA_INSTR  cmd_sha256msg1, a1, a2
+endm
+
+MY_sha256msg2 macro a1, a2
+        MY_SHA_INSTR  cmd_sha256msg2, a1, a2
+endm
+
+MY_PROLOG macro
+    ifdef x64
+      if (IS_LINUX eq 0)
+        movdqa  [r4 + 8], xmm6
+        movdqa  [r4 + 8 + 16], xmm7
+        sub     r4, LOCAL_SIZE + 8
+        movdqa  [r4     ], xmm8
+        movdqa  [r4 + 16], xmm9
+      endif
+    else ; x86
+        push    r3
+        push    r5
+        mov     r5, r4
+        NUM_PUSH_REGS   equ 2
+        PARAM_OFFSET    equ (REG_SIZE * (1 + NUM_PUSH_REGS))
+      if (IS_CDECL gt 0)
+        mov     rState, [r4 + PARAM_OFFSET]
+        mov     rData,  [r4 + PARAM_OFFSET + REG_SIZE * 1]
+        mov     rNum,   [r4 + PARAM_OFFSET + REG_SIZE * 2]
+      else ; fastcall
+        mov     rNum,   [r4 + PARAM_OFFSET]
+      endif
+        and     r4, -16
+        sub     r4, LOCAL_SIZE
+    endif
+endm
+
+MY_EPILOG macro
+    ifdef x64
+      if (IS_LINUX eq 0)
+        movdqa  xmm8, [r4]
+        movdqa  xmm9, [r4 + 16]
+        add     r4, LOCAL_SIZE + 8
+        movdqa  xmm6, [r4 + 8]
+        movdqa  xmm7, [r4 + 8 + 16]
+      endif
+    else ; x86
+        mov     r4, r5
+        pop     r5
+        pop     r3
+    endif
+    MY_ENDP
+endm
+
+
+msg        equ xmm0
+tmp        equ xmm0
+state0_N   equ 2
+state1_N   equ 3
+w_regs     equ 4
+
+
+state1_save equ xmm1
+state0  equ @CatStr(xmm, %state0_N)
+state1  equ @CatStr(xmm, %state1_N)
+
+
+ifdef x64
+        state0_save  equ  xmm8
+        mask2        equ  xmm9
+else
+        state0_save  equ  [r4]
+        mask2        equ  xmm0
+endif
+
+LOAD_MASK macro
+        movdqa  mask2, XMMWORD PTR Reverse_Endian_Mask
+endm
+
+LOAD_W macro k:req
+        movdqu  @CatStr(xmm, %(w_regs + k)), [rData + (16 * (k))]
+        pshufb  @CatStr(xmm, %(w_regs + k)), mask2
+endm
+
+
+; pre1 <= 4 && pre2 >= 1 && pre1 > pre2 && (pre1 - pre2) <= 1
+pre1 equ 3
+pre2 equ 2
+   
+
+
+RND4 macro k
+        movdqa  msg, xmmword ptr [rTable + (k) * 16]
+        paddd   msg, @CatStr(xmm, %(w_regs + ((k + 0) mod 4)))
+        MY_sha256rnds2 state0_N, state1_N
+        pshufd   msg, msg, 0eH
+        
+    if (k GE (4 - pre1)) AND (k LT (16 - pre1))
+        ; w4[0] = msg1(w4[-4], w4[-3])
+        MY_sha256msg1 (w_regs + ((k + pre1) mod 4)), (w_regs + ((k + pre1 - 3) mod 4))
+    endif
+        
+        MY_sha256rnds2 state1_N, state0_N
+
+    if (k GE (4 - pre2)) AND (k LT (16 - pre2))
+        movdqa  tmp, @CatStr(xmm, %(w_regs + ((k + pre2 - 1) mod 4)))
+        palignr tmp, @CatStr(xmm, %(w_regs + ((k + pre2 - 2) mod 4))), 4
+        paddd   @CatStr(xmm, %(w_regs + ((k + pre2) mod 4))), tmp
+        ; w4[0] = msg2(w4[0], w4[-1])
+        MY_sha256msg2 %(w_regs + ((k + pre2) mod 4)), %(w_regs + ((k + pre2 - 1) mod 4))
+    endif
+endm
+
+
+
+
+
+REVERSE_STATE macro
+                               ; state0 ; dcba
+                               ; state1 ; hgfe
+        pshufd      tmp, state0, 01bH   ; abcd
+        pshufd   state0, state1, 01bH   ; efgh
+        movdqa   state1, state0         ; efgh
+        punpcklqdq  state0, tmp         ; cdgh
+        punpckhqdq  state1, tmp         ; abef
+endm
+
+
+MY_PROC Sha256_UpdateBlocks_HW, 3
+    MY_PROLOG
+
+        lea     rTable, [K_CONST]
+
+        cmp     rNum, 0
+        je      end_c
+
+        movdqu   state0, [rState]       ; dcba
+        movdqu   state1, [rState + 16]  ; hgfe
+
+        REVERSE_STATE
+       
+        ifdef x64
+        LOAD_MASK
+        endif
+
+    align 16
+    nextBlock:
+        movdqa  state0_save, state0
+        movdqa  state1_save, state1
+        
+        ifndef x64
+        LOAD_MASK
+        endif
+        
+        LOAD_W 0
+        LOAD_W 1
+        LOAD_W 2
+        LOAD_W 3
+
+        
+        k = 0
+        rept 16
+          RND4 k
+          k = k + 1
+        endm
+
+        paddd   state0, state0_save
+        paddd   state1, state1_save
+
+        add     rData, 64
+        sub     rNum, 1
+        jnz     nextBlock
+        
+        REVERSE_STATE
+
+        movdqu  [rState], state0
+        movdqu  [rState + 16], state1
+       
+  end_c:
+MY_EPILOG
+
+; _TEXT$SHA256OPT ENDS
+
+end

Asm/x86/XzCrc64Opt.asm

@@ -0,0 +1,239 @@
+; XzCrc64Opt.asm -- CRC64 calculation : optimized version
+; 2021-02-06 : Igor Pavlov : Public domain
+
+include 7zAsm.asm
+
+MY_ASM_START
+
+ifdef x64
+
+rD      equ  r9
+rN      equ  r10
+rT      equ  r5
+num_VAR equ  r8
+
+SRCDAT4 equ  dword ptr [rD + rN * 1]
+    
+CRC_XOR macro dest:req, src:req, t:req
+    xor     dest, QWORD PTR [rT + src * 8 + 0800h * t]
+endm
+
+CRC1b macro
+    movzx   x6, BYTE PTR [rD]
+    inc     rD
+    movzx   x3, x0_L
+    xor     x6, x3
+    shr     r0, 8
+    CRC_XOR r0, r6, 0
+    dec     rN
+endm
+
+MY_PROLOG macro crc_end:req
+  ifdef ABI_LINUX
+    MY_PUSH_2_REGS
+  else
+    MY_PUSH_4_REGS
+  endif
+    mov     r0, REG_ABI_PARAM_0
+    mov     rN, REG_ABI_PARAM_2
+    mov     rT, REG_ABI_PARAM_3
+    mov     rD, REG_ABI_PARAM_1
+    test    rN, rN
+    jz      crc_end
+  @@:
+    test    rD, 3
+    jz      @F
+    CRC1b
+    jnz     @B
+  @@:
+    cmp     rN, 8
+    jb      crc_end
+    add     rN, rD
+    mov     num_VAR, rN
+    sub     rN, 4
+    and     rN, NOT 3
+    sub     rD, rN
+    mov     x1, SRCDAT4
+    xor     r0, r1
+    add     rN, 4
+endm
+
+MY_EPILOG macro crc_end:req
+    sub     rN, 4
+    mov     x1, SRCDAT4
+    xor     r0, r1
+    mov     rD, rN
+    mov     rN, num_VAR
+    sub     rN, rD
+  crc_end:
+    test    rN, rN
+    jz      @F
+    CRC1b
+    jmp     crc_end
+  @@:
+  ifdef ABI_LINUX
+    MY_POP_2_REGS
+  else
+    MY_POP_4_REGS
+  endif
+endm
+
+MY_PROC XzCrc64UpdateT4, 4
+    MY_PROLOG crc_end_4
+    align 16
+  main_loop_4:
+    mov     x1, SRCDAT4
+    movzx   x2, x0_L
+    movzx   x3, x0_H
+    shr     r0, 16
+    movzx   x6, x0_L
+    movzx   x7, x0_H
+    shr     r0, 16
+    CRC_XOR r1, r2, 3
+    CRC_XOR r0, r3, 2
+    CRC_XOR r1, r6, 1
+    CRC_XOR r0, r7, 0
+    xor     r0, r1
+
+    add     rD, 4
+    jnz     main_loop_4
+
+    MY_EPILOG crc_end_4
+MY_ENDP
+
+else
+; x86 (32-bit)
+
+rD      equ  r1
+rN      equ  r7
+rT      equ  r5
+
+crc_OFFS  equ  (REG_SIZE * 5)
+
+if (IS_CDECL gt 0) or (IS_LINUX gt 0)
+    ; cdecl or (GNU fastcall) stack:
+    ;   (UInt32 *) table
+    ;   size_t     size
+    ;   void *     data
+    ;   (UInt64)   crc
+    ;   ret-ip <-(r4)
+    data_OFFS   equ  (8 + crc_OFFS)
+    size_OFFS   equ  (REG_SIZE + data_OFFS)
+    table_OFFS  equ  (REG_SIZE + size_OFFS)
+    num_VAR     equ  [r4 + size_OFFS]
+    table_VAR   equ  [r4 + table_OFFS]
+else
+    ; Windows fastcall:
+    ;   r1 = data, r2 = size
+    ; stack:
+    ;   (UInt32 *) table
+    ;   (UInt64)   crc
+    ;   ret-ip <-(r4)
+    table_OFFS  equ  (8 + crc_OFFS)
+    table_VAR   equ  [r4 + table_OFFS]
+    num_VAR     equ  table_VAR
+endif
+
+SRCDAT4 equ  dword ptr [rD + rN * 1]
+
+CRC macro op0:req, op1:req, dest0:req, dest1:req, src:req, t:req
+    op0     dest0, DWORD PTR [rT + src * 8 + 0800h * t]
+    op1     dest1, DWORD PTR [rT + src * 8 + 0800h * t + 4]
+endm
+
+CRC_XOR macro dest0:req, dest1:req, src:req, t:req
+    CRC xor, xor, dest0, dest1, src, t
+endm
+
+
+CRC1b macro
+    movzx   x6, BYTE PTR [rD]
+    inc     rD
+    movzx   x3, x0_L
+    xor     x6, x3
+    shrd    r0, r2, 8
+    shr     r2, 8
+    CRC_XOR r0, r2, r6, 0
+    dec     rN
+endm
+
+MY_PROLOG macro crc_end:req
+    MY_PUSH_4_REGS
+
+  if (IS_CDECL gt 0) or (IS_LINUX gt 0)
+    proc_numParams = proc_numParams + 2 ; for ABI_LINUX
+    mov     rN, [r4 + size_OFFS]
+    mov     rD, [r4 + data_OFFS]
+  else
+    mov     rN, r2
+  endif
+
+    mov     x0, [r4 + crc_OFFS]
+    mov     x2, [r4 + crc_OFFS + 4]
+    mov     rT, table_VAR
+    test    rN, rN
+    jz      crc_end
+  @@:
+    test    rD, 3
+    jz      @F
+    CRC1b
+    jnz     @B
+  @@:
+    cmp     rN, 8
+    jb      crc_end
+    add     rN, rD
+
+    mov     num_VAR, rN
+
+    sub     rN, 4
+    and     rN, NOT 3
+    sub     rD, rN
+    xor     r0, SRCDAT4
+    add     rN, 4
+endm
+
+MY_EPILOG macro crc_end:req
+    sub     rN, 4
+    xor     r0, SRCDAT4
+
+    mov     rD, rN
+    mov     rN, num_VAR
+    sub     rN, rD
+  crc_end:
+    test    rN, rN
+    jz      @F
+    CRC1b
+    jmp     crc_end
+  @@:
+    MY_POP_4_REGS
+endm
+
+MY_PROC XzCrc64UpdateT4, 5
+    MY_PROLOG crc_end_4
+    movzx   x6, x0_L
+    align 16
+  main_loop_4:
+    mov     r3, SRCDAT4
+    xor     r3, r2
+
+    CRC xor, mov, r3, r2, r6, 3
+    movzx   x6, x0_H
+    shr     r0, 16
+    CRC_XOR r3, r2, r6, 2
+
+    movzx   x6, x0_L
+    movzx   x0, x0_H
+    CRC_XOR r3, r2, r6, 1
+    CRC_XOR r3, r2, r0, 0
+    movzx   x6, x3_L
+    mov     r0, r3
+
+    add     rD, 4
+    jnz     main_loop_4
+
+    MY_EPILOG crc_end_4
+MY_ENDP
+
+endif ; ! x64
+
+end

C/7z.h

@@ -1,89 +1,57 @@
 /* 7z.h -- 7z interface
-2010-03-11 : Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
 
-#ifndef __7Z_H
-#define __7Z_H
+#ifndef ZIP7_INC_7Z_H
+#define ZIP7_INC_7Z_H
 
-#include "7zBuf.h"
+#include "7zTypes.h"
 
 EXTERN_C_BEGIN
 
 #define k7zStartHeaderSize 0x20
 #define k7zSignatureSize 6
-extern Byte k7zSignature[k7zSignatureSize];
-#define k7zMajorVersion 0
 
-enum EIdEnum
-{
-  k7zIdEnd,
-  k7zIdHeader,
-  k7zIdArchiveProperties,
-  k7zIdAdditionalStreamsInfo,
-  k7zIdMainStreamsInfo,
-  k7zIdFilesInfo,
-  k7zIdPackInfo,
-  k7zIdUnpackInfo,
-  k7zIdSubStreamsInfo,
-  k7zIdSize,
-  k7zIdCRC,
-  k7zIdFolder,
-  k7zIdCodersUnpackSize,
-  k7zIdNumUnpackStream,
-  k7zIdEmptyStream,
-  k7zIdEmptyFile,
-  k7zIdAnti,
-  k7zIdName,
-  k7zIdCTime,
-  k7zIdATime,
-  k7zIdMTime,
-  k7zIdWinAttributes,
-  k7zIdComment,
-  k7zIdEncodedHeader,
-  k7zIdStartPos,
-  k7zIdDummy
-};
+extern const Byte k7zSignature[k7zSignatureSize];
 
 typedef struct
 {
-  UInt32 NumInStreams;
-  UInt32 NumOutStreams;
-  UInt64 MethodID;
-  CBuf Props;
-} CSzCoderInfo;
+  const Byte *Data;
+  size_t Size;
+} CSzData;
 
-void SzCoderInfo_Init(CSzCoderInfo *p);
-void SzCoderInfo_Free(CSzCoderInfo *p, ISzAlloc *alloc);
+/* CSzCoderInfo & CSzFolder support only default methods */
+
+typedef struct
+{
+  size_t PropsOffset;
+  UInt32 MethodID;
+  Byte NumStreams;
+  Byte PropsSize;
+} CSzCoderInfo;
 
 typedef struct
 {
   UInt32 InIndex;
   UInt32 OutIndex;
-} CSzBindPair;
+} CSzBond;
+
+#define SZ_NUM_CODERS_IN_FOLDER_MAX 4
+#define SZ_NUM_BONDS_IN_FOLDER_MAX 3
+#define SZ_NUM_PACK_STREAMS_IN_FOLDER_MAX 4
 
 typedef struct
 {
-  CSzCoderInfo *Coders;
-  CSzBindPair *BindPairs;
-  UInt32 *PackStreams;
-  UInt64 *UnpackSizes;
   UInt32 NumCoders;
-  UInt32 NumBindPairs;
+  UInt32 NumBonds;
   UInt32 NumPackStreams;
-  int UnpackCRCDefined;
-  UInt32 UnpackCRC;
-
-  UInt32 NumUnpackStreams;
+  UInt32 UnpackStream;
+  UInt32 PackStreams[SZ_NUM_PACK_STREAMS_IN_FOLDER_MAX];
+  CSzBond Bonds[SZ_NUM_BONDS_IN_FOLDER_MAX];
+  CSzCoderInfo Coders[SZ_NUM_CODERS_IN_FOLDER_MAX];
 } CSzFolder;
 
-void SzFolder_Init(CSzFolder *p);
-UInt64 SzFolder_GetUnpackSize(CSzFolder *p);
-int SzFolder_FindBindPairForInStream(CSzFolder *p, UInt32 inStreamIndex);
-UInt32 SzFolder_GetNumOutStreams(CSzFolder *p);
-UInt64 SzFolder_GetUnpackSize(CSzFolder *p);
 
-SRes SzFolder_Decode(const CSzFolder *folder, const UInt64 *packSizes,
-    ILookInStream *stream, UInt64 startPos,
-    Byte *outBuffer, size_t outSize, ISzAlloc *allocMain);
+SRes SzGetNextFolderItem(CSzFolder *f, CSzData *sd);
 
 typedef struct
 {
@@ -93,38 +61,99 @@ typedef struct
 
 typedef struct
 {
-  CNtfsFileTime MTime;
-  UInt64 Size;
-  UInt32 Crc;
-  UInt32 Attrib;
-  Byte HasStream;
-  Byte IsDir;
-  Byte IsAnti;
-  Byte CrcDefined;
-  Byte MTimeDefined;
-  Byte AttribDefined;
-} CSzFileItem;
-
-void SzFile_Init(CSzFileItem *p);
+  Byte *Defs; /* MSB 0 bit numbering */
+  UInt32 *Vals;
+} CSzBitUi32s;
+
+typedef struct
+{
+  Byte *Defs; /* MSB 0 bit numbering */
+  // UInt64 *Vals;
+  CNtfsFileTime *Vals;
+} CSzBitUi64s;
+
+#define SzBitArray_Check(p, i) (((p)[(i) >> 3] & (0x80 >> ((i) & 7))) != 0)
+
+#define SzBitWithVals_Check(p, i) ((p)->Defs && ((p)->Defs[(i) >> 3] & (0x80 >> ((i) & 7))) != 0)
 
 typedef struct
 {
-  UInt64 *PackSizes;
-  Byte *PackCRCsDefined;
-  UInt32 *PackCRCs;
-  CSzFolder *Folders;
-  CSzFileItem *Files;
   UInt32 NumPackStreams;
   UInt32 NumFolders;
-  UInt32 NumFiles;
+
+  UInt64 *PackPositions;          // NumPackStreams + 1
+  CSzBitUi32s FolderCRCs;         // NumFolders
+
+  size_t *FoCodersOffsets;        // NumFolders + 1
+  UInt32 *FoStartPackStreamIndex; // NumFolders + 1
+  UInt32 *FoToCoderUnpackSizes;   // NumFolders + 1
+  Byte *FoToMainUnpackSizeIndex;  // NumFolders
+  UInt64 *CoderUnpackSizes;       // for all coders in all folders
+
+  Byte *CodersData;
+
+  UInt64 RangeLimit;
 } CSzAr;
 
-void SzAr_Init(CSzAr *p);
-void SzAr_Free(CSzAr *p, ISzAlloc *alloc);
+UInt64 SzAr_GetFolderUnpackSize(const CSzAr *p, UInt32 folderIndex);
+
+SRes SzAr_DecodeFolder(const CSzAr *p, UInt32 folderIndex,
+    ILookInStreamPtr stream, UInt64 startPos,
+    Byte *outBuffer, size_t outSize,
+    ISzAllocPtr allocMain);
+
+typedef struct
+{
+  CSzAr db;
+
+  UInt64 startPosAfterHeader;
+  UInt64 dataPos;
+  
+  UInt32 NumFiles;
+
+  UInt64 *UnpackPositions;  // NumFiles + 1
+  // Byte *IsEmptyFiles;
+  Byte *IsDirs;
+  CSzBitUi32s CRCs;
+
+  CSzBitUi32s Attribs;
+  // CSzBitUi32s Parents;
+  CSzBitUi64s MTime;
+  CSzBitUi64s CTime;
+
+  UInt32 *FolderToFile;   // NumFolders + 1
+  UInt32 *FileToFolder;   // NumFiles
+
+  size_t *FileNameOffsets; /* in 2-byte steps */
+  Byte *FileNames;  /* UTF-16-LE */
+} CSzArEx;
+
+#define SzArEx_IsDir(p, i) (SzBitArray_Check((p)->IsDirs, i))
+
+#define SzArEx_GetFileSize(p, i) ((p)->UnpackPositions[(i) + 1] - (p)->UnpackPositions[i])
+
+void SzArEx_Init(CSzArEx *p);
+void SzArEx_Free(CSzArEx *p, ISzAllocPtr alloc);
+UInt64 SzArEx_GetFolderStreamPos(const CSzArEx *p, UInt32 folderIndex, UInt32 indexInFolder);
+int SzArEx_GetFolderFullPackSize(const CSzArEx *p, UInt32 folderIndex, UInt64 *resSize);
+
+/*
+if dest == NULL, the return value specifies the required size of the buffer,
+  in 16-bit characters, including the null-terminating character.
+if dest != NULL, the return value specifies the number of 16-bit characters that
+  are written to the dest, including the null-terminating character. */
+
+size_t SzArEx_GetFileNameUtf16(const CSzArEx *p, size_t fileIndex, UInt16 *dest);
+
+/*
+size_t SzArEx_GetFullNameLen(const CSzArEx *p, size_t fileIndex);
+UInt16 *SzArEx_GetFullNameUtf16_Back(const CSzArEx *p, size_t fileIndex, UInt16 *dest);
+*/
+
 
 
 /*
-  SzExtract extracts file from archive
+  SzArEx_Extract extracts file from archive
 
   *outBuffer must be 0 before first call for each new archive.
 
@@ -143,46 +172,17 @@ void SzAr_Free(CSzAr *p, ISzAlloc *alloc);
     Free *outBuffer and set *outBuffer to 0, if you want to flush cache.
 */
 
-typedef struct
-{
-  CSzAr db;
-  
-  UInt64 startPosAfterHeader;
-  UInt64 dataPos;
-
-  UInt32 *FolderStartPackStreamIndex;
-  UInt64 *PackStreamStartPositions;
-  UInt32 *FolderStartFileIndex;
-  UInt32 *FileIndexToFolderIndexMap;
-
-  size_t *FileNameOffsets; /* in 2-byte steps */
-  CBuf FileNames;  /* UTF-16-LE */
-} CSzArEx;
-
-void SzArEx_Init(CSzArEx *p);
-void SzArEx_Free(CSzArEx *p, ISzAlloc *alloc);
-UInt64 SzArEx_GetFolderStreamPos(const CSzArEx *p, UInt32 folderIndex, UInt32 indexInFolder);
-int SzArEx_GetFolderFullPackSize(const CSzArEx *p, UInt32 folderIndex, UInt64 *resSize);
-
-/*
-if dest == NULL, the return value specifies the required size of the buffer,
-  in 16-bit characters, including the null-terminating character.
-if dest != NULL, the return value specifies the number of 16-bit characters that
-  are written to the dest, including the null-terminating character. */
-
-size_t SzArEx_GetFileNameUtf16(const CSzArEx *p, size_t fileIndex, UInt16 *dest);
-
 SRes SzArEx_Extract(
     const CSzArEx *db,
-    ILookInStream *inStream,
+    ILookInStreamPtr inStream,
     UInt32 fileIndex,         /* index of file */
     UInt32 *blockIndex,       /* index of solid block */
     Byte **outBuffer,         /* pointer to pointer to output buffer (allocated with allocMain) */
     size_t *outBufferSize,    /* buffer size for output buffer */
     size_t *offset,           /* offset of stream for required file in *outBuffer */
     size_t *outSizeProcessed, /* size of file in *outBuffer */
-    ISzAlloc *allocMain,
-    ISzAlloc *allocTemp);
+    ISzAllocPtr allocMain,
+    ISzAllocPtr allocTemp);
 
 
 /*
@@ -196,7 +196,8 @@ SZ_ERROR_INPUT_EOF
 SZ_ERROR_FAIL
 */
 
-SRes SzArEx_Open(CSzArEx *p, ILookInStream *inStream, ISzAlloc *allocMain, ISzAlloc *allocTemp);
+SRes SzArEx_Open(CSzArEx *p, ILookInStreamPtr inStream,
+    ISzAllocPtr allocMain, ISzAllocPtr allocTemp);
 
 EXTERN_C_END
 

C/7zAlloc.c

@@ -0,0 +1,89 @@
+/* 7zAlloc.c -- Allocation functions for 7z processing
+2023-03-04 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
+
+#include <stdlib.h>
+
+#include "7zAlloc.h"
+
+/* #define SZ_ALLOC_DEBUG */
+/* use SZ_ALLOC_DEBUG to debug alloc/free operations */
+
+#ifdef SZ_ALLOC_DEBUG
+
+/*
+#ifdef _WIN32
+#include "7zWindows.h"
+#endif
+*/
+
+#include <stdio.h>
+static int g_allocCount = 0;
+static int g_allocCountTemp = 0;
+
+static void Print_Alloc(const char *s, size_t size, int *counter)
+{
+  const unsigned size2 = (unsigned)size;
+  fprintf(stderr, "\n%s count = %10d : %10u bytes; ", s, *counter, size2);
+  (*counter)++;
+}
+static void Print_Free(const char *s, int *counter)
+{
+  (*counter)--;
+  fprintf(stderr, "\n%s count = %10d", s, *counter);
+}
+#endif
+
+void *SzAlloc(ISzAllocPtr p, size_t size)
+{
+  UNUSED_VAR(p)
+  if (size == 0)
+    return 0;
+  #ifdef SZ_ALLOC_DEBUG
+  Print_Alloc("Alloc", size, &g_allocCount);
+  #endif
+  return malloc(size);
+}
+
+void SzFree(ISzAllocPtr p, void *address)
+{
+  UNUSED_VAR(p)
+  #ifdef SZ_ALLOC_DEBUG
+  if (address)
+    Print_Free("Free ", &g_allocCount);
+  #endif
+  free(address);
+}
+
+void *SzAllocTemp(ISzAllocPtr p, size_t size)
+{
+  UNUSED_VAR(p)
+  if (size == 0)
+    return 0;
+  #ifdef SZ_ALLOC_DEBUG
+  Print_Alloc("Alloc_temp", size, &g_allocCountTemp);
+  /*
+  #ifdef _WIN32
+  return HeapAlloc(GetProcessHeap(), 0, size);
+  #endif
+  */
+  #endif
+  return malloc(size);
+}
+
+void SzFreeTemp(ISzAllocPtr p, void *address)
+{
+  UNUSED_VAR(p)
+  #ifdef SZ_ALLOC_DEBUG
+  if (address)
+    Print_Free("Free_temp ", &g_allocCountTemp);
+  /*
+  #ifdef _WIN32
+  HeapFree(GetProcessHeap(), 0, address);
+  return;
+  #endif
+  */
+  #endif
+  free(address);
+}

C/7zAlloc.h

@@ -0,0 +1,19 @@
+/* 7zAlloc.h -- Allocation functions
+2023-03-04 : Igor Pavlov : Public domain */
+
+#ifndef ZIP7_INC_7Z_ALLOC_H
+#define ZIP7_INC_7Z_ALLOC_H
+
+#include "7zTypes.h"
+
+EXTERN_C_BEGIN
+
+void *SzAlloc(ISzAllocPtr p, size_t size);
+void SzFree(ISzAllocPtr p, void *address);
+
+void *SzAllocTemp(ISzAllocPtr p, size_t size);
+void SzFreeTemp(ISzAllocPtr p, void *address);
+
+EXTERN_C_END
+
+#endif

C/7zBuf.c

@@ -1,7 +1,7 @@
 /* 7zBuf.c -- Byte Buffer
-2008-03-28
-Igor Pavlov
-Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "7zBuf.h"
 
@@ -11,7 +11,7 @@ void Buf_Init(CBuf *p)
   p->size = 0;
 }
 
-int Buf_Create(CBuf *p, size_t size, ISzAlloc *alloc)
+int Buf_Create(CBuf *p, size_t size, ISzAllocPtr alloc)
 {
   p->size = 0;
   if (size == 0)
@@ -19,8 +19,8 @@ int Buf_Create(CBuf *p, size_t size, ISzAlloc *alloc)
     p->data = 0;
     return 1;
   }
-  p->data = (Byte *)alloc->Alloc(alloc, size);
-  if (p->data != 0)
+  p->data = (Byte *)ISzAlloc_Alloc(alloc, size);
+  if (p->data)
   {
     p->size = size;
     return 1;
@@ -28,9 +28,9 @@ int Buf_Create(CBuf *p, size_t size, ISzAlloc *alloc)
   return 0;
 }
 
-void Buf_Free(CBuf *p, ISzAlloc *alloc)
+void Buf_Free(CBuf *p, ISzAllocPtr alloc)
 {
-  alloc->Free(alloc, p->data);
+  ISzAlloc_Free(alloc, p->data);
   p->data = 0;
   p->size = 0;
 }

C/7zBuf.h

@@ -1,14 +1,12 @@
 /* 7zBuf.h -- Byte Buffer
-2009-02-07 : Igor Pavlov : Public domain */
+2023-03-04 : Igor Pavlov : Public domain */
 
-#ifndef __7Z_BUF_H
-#define __7Z_BUF_H
+#ifndef ZIP7_INC_7Z_BUF_H
+#define ZIP7_INC_7Z_BUF_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
 typedef struct
 {
@@ -17,8 +15,8 @@ typedef struct
 } CBuf;
 
 void Buf_Init(CBuf *p);
-int Buf_Create(CBuf *p, size_t size, ISzAlloc *alloc);
-void Buf_Free(CBuf *p, ISzAlloc *alloc);
+int Buf_Create(CBuf *p, size_t size, ISzAllocPtr alloc);
+void Buf_Free(CBuf *p, ISzAllocPtr alloc);
 
 typedef struct
 {
@@ -29,11 +27,9 @@ typedef struct
 
 void DynBuf_Construct(CDynBuf *p);
 void DynBuf_SeekToBeg(CDynBuf *p);
-int DynBuf_Write(CDynBuf *p, const Byte *buf, size_t size, ISzAlloc *alloc);
-void DynBuf_Free(CDynBuf *p, ISzAlloc *alloc);
+int DynBuf_Write(CDynBuf *p, const Byte *buf, size_t size, ISzAllocPtr alloc);
+void DynBuf_Free(CDynBuf *p, ISzAllocPtr alloc);
 
-#ifdef __cplusplus
-}
-#endif
+EXTERN_C_END
 
 #endif

C/7zBuf2.c

@@ -1,7 +1,10 @@
 /* 7zBuf2.c -- Byte Buffer
-2008-10-04 : Igor Pavlov : Public domain */
+2017-04-03 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include <string.h>
+
 #include "7zBuf.h"
 
 void DynBuf_Construct(CDynBuf *p)
@@ -16,29 +19,33 @@ void DynBuf_SeekToBeg(CDynBuf *p)
   p->pos = 0;
 }
 
-int DynBuf_Write(CDynBuf *p, const Byte *buf, size_t size, ISzAlloc *alloc)
+int DynBuf_Write(CDynBuf *p, const Byte *buf, size_t size, ISzAllocPtr alloc)
 {
   if (size > p->size - p->pos)
   {
     size_t newSize = p->pos + size;
     Byte *data;
     newSize += newSize / 4;
-    data = (Byte *)alloc->Alloc(alloc, newSize);
-    if (data == 0)
+    data = (Byte *)ISzAlloc_Alloc(alloc, newSize);
+    if (!data)
       return 0;
     p->size = newSize;
-    memcpy(data, p->data, p->pos);
-    alloc->Free(alloc, p->data);
+    if (p->pos != 0)
+      memcpy(data, p->data, p->pos);
+    ISzAlloc_Free(alloc, p->data);
     p->data = data;
   }
-  memcpy(p->data + p->pos, buf, size);
-  p->pos += size;
+  if (size != 0)
+  {
+    memcpy(p->data + p->pos, buf, size);
+    p->pos += size;
+  }
   return 1;
 }
 
-void DynBuf_Free(CDynBuf *p, ISzAlloc *alloc)
+void DynBuf_Free(CDynBuf *p, ISzAllocPtr alloc)
 {
-  alloc->Free(alloc, p->data);
+  ISzAlloc_Free(alloc, p->data);
   p->data = 0;
   p->size = 0;
   p->pos = 0;

C/7zCrc.c

@@ -1,5 +1,7 @@
-/* 7zCrc.c -- CRC32 calculation
-2009-11-23 : Igor Pavlov : Public domain */
+/* 7zCrc.c -- CRC32 calculation and init
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "7zCrc.h"
 #include "CpuArch.h"
@@ -7,46 +9,240 @@
 #define kCrcPoly 0xEDB88320
 
 #ifdef MY_CPU_LE
-#define CRC_NUM_TABLES 8
+  #define CRC_NUM_TABLES 8
 #else
-#define CRC_NUM_TABLES 1
+  #define CRC_NUM_TABLES 9
+
+  UInt32 Z7_FASTCALL CrcUpdateT1_BeT4(UInt32 v, const void *data, size_t size, const UInt32 *table);
+  UInt32 Z7_FASTCALL CrcUpdateT1_BeT8(UInt32 v, const void *data, size_t size, const UInt32 *table);
 #endif
 
-typedef UInt32 (MY_FAST_CALL *CRC_FUNC)(UInt32 v, const void *data, size_t size, const UInt32 *table);
+#ifndef MY_CPU_BE
+  UInt32 Z7_FASTCALL CrcUpdateT4(UInt32 v, const void *data, size_t size, const UInt32 *table);
+  UInt32 Z7_FASTCALL CrcUpdateT8(UInt32 v, const void *data, size_t size, const UInt32 *table);
+#endif
+
+/*
+extern
+CRC_FUNC g_CrcUpdateT4;
+CRC_FUNC g_CrcUpdateT4;
+*/
+extern
+CRC_FUNC g_CrcUpdateT8;
+CRC_FUNC g_CrcUpdateT8;
+extern
+CRC_FUNC g_CrcUpdateT0_32;
+CRC_FUNC g_CrcUpdateT0_32;
+extern
+CRC_FUNC g_CrcUpdateT0_64;
+CRC_FUNC g_CrcUpdateT0_64;
+extern
+CRC_FUNC g_CrcUpdate;
+CRC_FUNC g_CrcUpdate;
 
-static CRC_FUNC g_CrcUpdate;
 UInt32 g_CrcTable[256 * CRC_NUM_TABLES];
 
-#if CRC_NUM_TABLES == 1
-
-#define CRC_UPDATE_BYTE_2(crc, b) (table[((crc) ^ (b)) & 0xFF] ^ ((crc) >> 8))
-
-static UInt32 MY_FAST_CALL CrcUpdateT1(UInt32 v, const void *data, size_t size, const UInt32 *table)
-{
-  const Byte *p = (const Byte *)data;
-  for (; size > 0; size--, p++)
-    v = CRC_UPDATE_BYTE_2(v, *p);
-  return v;
-}
-
-#else
-
-UInt32 MY_FAST_CALL CrcUpdateT4(UInt32 v, const void *data, size_t size, const UInt32 *table);
-UInt32 MY_FAST_CALL CrcUpdateT8(UInt32 v, const void *data, size_t size, const UInt32 *table);
-
-#endif
-
-UInt32 MY_FAST_CALL CrcUpdate(UInt32 v, const void *data, size_t size)
+UInt32 Z7_FASTCALL CrcUpdate(UInt32 v, const void *data, size_t size)
 {
   return g_CrcUpdate(v, data, size, g_CrcTable);
 }
 
-UInt32 MY_FAST_CALL CrcCalc(const void *data, size_t size)
+UInt32 Z7_FASTCALL CrcCalc(const void *data, size_t size)
 {
   return g_CrcUpdate(CRC_INIT_VAL, data, size, g_CrcTable) ^ CRC_INIT_VAL;
 }
 
-void MY_FAST_CALL CrcGenerateTable()
+#if CRC_NUM_TABLES < 4 \
+   || (CRC_NUM_TABLES == 4 && defined(MY_CPU_BE)) \
+   || (!defined(MY_CPU_LE) && !defined(MY_CPU_BE))
+#define CRC_UPDATE_BYTE_2(crc, b) (table[((crc) ^ (b)) & 0xFF] ^ ((crc) >> 8))
+UInt32 Z7_FASTCALL CrcUpdateT1(UInt32 v, const void *data, size_t size, const UInt32 *table);
+UInt32 Z7_FASTCALL CrcUpdateT1(UInt32 v, const void *data, size_t size, const UInt32 *table)
+{
+  const Byte *p = (const Byte *)data;
+  const Byte *pEnd = p + size;
+  for (; p != pEnd; p++)
+    v = CRC_UPDATE_BYTE_2(v, *p);
+  return v;
+}
+#endif
+
+/* ---------- hardware CRC ---------- */
+
+#ifdef MY_CPU_LE
+
+#if defined(MY_CPU_ARM_OR_ARM64)
+
+// #pragma message("ARM*")
+
+  #if defined(_MSC_VER)
+    #if defined(MY_CPU_ARM64)
+    #if (_MSC_VER >= 1910)
+    #ifndef __clang__
+        #define USE_ARM64_CRC
+        #include <intrin.h>
+    #endif
+    #endif
+    #endif
+  #elif (defined(__clang__) && (__clang_major__ >= 3)) \
+     || (defined(__GNUC__) && (__GNUC__ > 4))
+      #if !defined(__ARM_FEATURE_CRC32)
+        #define __ARM_FEATURE_CRC32 1
+        #if defined(__clang__)
+          #if defined(MY_CPU_ARM64)
+            #define ATTRIB_CRC __attribute__((__target__("crc")))
+          #else
+            #define ATTRIB_CRC __attribute__((__target__("armv8-a,crc")))
+          #endif
+        #else
+          #if defined(MY_CPU_ARM64)
+            #define ATTRIB_CRC __attribute__((__target__("+crc")))
+          #else
+            #define ATTRIB_CRC __attribute__((__target__("arch=armv8-a+crc")))
+          #endif
+        #endif
+      #endif
+      #if defined(__ARM_FEATURE_CRC32)
+        #define USE_ARM64_CRC
+        #include <arm_acle.h>
+      #endif
+  #endif
+
+#else
+
+// no hardware CRC
+
+// #define USE_CRC_EMU
+
+#ifdef USE_CRC_EMU
+
+#pragma message("ARM64 CRC emulation")
+
+Z7_FORCE_INLINE
+UInt32 __crc32b(UInt32 v, UInt32 data)
+{
+  const UInt32 *table = g_CrcTable;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data);
+  return v;
+}
+
+Z7_FORCE_INLINE
+UInt32 __crc32w(UInt32 v, UInt32 data)
+{
+  const UInt32 *table = g_CrcTable;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  return v;
+}
+
+Z7_FORCE_INLINE
+UInt32 __crc32d(UInt32 v, UInt64 data)
+{
+  const UInt32 *table = g_CrcTable;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  v = CRC_UPDATE_BYTE_2(v, (Byte)data); data >>= 8;
+  return v;
+}
+
+#endif // USE_CRC_EMU
+
+#endif // defined(MY_CPU_ARM64) && defined(MY_CPU_LE)
+
+
+
+#if defined(USE_ARM64_CRC) || defined(USE_CRC_EMU)
+
+#define T0_32_UNROLL_BYTES (4 * 4)
+#define T0_64_UNROLL_BYTES (4 * 8)
+
+#ifndef ATTRIB_CRC
+#define ATTRIB_CRC
+#endif
+// #pragma message("USE ARM HW CRC")
+
+ATTRIB_CRC
+UInt32 Z7_FASTCALL CrcUpdateT0_32(UInt32 v, const void *data, size_t size, const UInt32 *table);
+ATTRIB_CRC
+UInt32 Z7_FASTCALL CrcUpdateT0_32(UInt32 v, const void *data, size_t size, const UInt32 *table)
+{
+  const Byte *p = (const Byte *)data;
+  UNUSED_VAR(table);
+
+  for (; size != 0 && ((unsigned)(ptrdiff_t)p & (T0_32_UNROLL_BYTES - 1)) != 0; size--)
+    v = __crc32b(v, *p++);
+
+  if (size >= T0_32_UNROLL_BYTES)
+  {
+    const Byte *lim = p + size;
+    size &= (T0_32_UNROLL_BYTES - 1);
+    lim -= size;
+    do
+    {
+      v = __crc32w(v, *(const UInt32 *)(const void *)(p));
+      v = __crc32w(v, *(const UInt32 *)(const void *)(p + 4)); p += 2 * 4;
+      v = __crc32w(v, *(const UInt32 *)(const void *)(p));
+      v = __crc32w(v, *(const UInt32 *)(const void *)(p + 4)); p += 2 * 4;
+    }
+    while (p != lim);
+  }
+  
+  for (; size != 0; size--)
+    v = __crc32b(v, *p++);
+
+  return v;
+}
+
+ATTRIB_CRC
+UInt32 Z7_FASTCALL CrcUpdateT0_64(UInt32 v, const void *data, size_t size, const UInt32 *table);
+ATTRIB_CRC
+UInt32 Z7_FASTCALL CrcUpdateT0_64(UInt32 v, const void *data, size_t size, const UInt32 *table)
+{
+  const Byte *p = (const Byte *)data;
+  UNUSED_VAR(table);
+
+  for (; size != 0 && ((unsigned)(ptrdiff_t)p & (T0_64_UNROLL_BYTES - 1)) != 0; size--)
+    v = __crc32b(v, *p++);
+
+  if (size >= T0_64_UNROLL_BYTES)
+  {
+    const Byte *lim = p + size;
+    size &= (T0_64_UNROLL_BYTES - 1);
+    lim -= size;
+    do
+    {
+      v = __crc32d(v, *(const UInt64 *)(const void *)(p));
+      v = __crc32d(v, *(const UInt64 *)(const void *)(p + 8)); p += 2 * 8;
+      v = __crc32d(v, *(const UInt64 *)(const void *)(p));
+      v = __crc32d(v, *(const UInt64 *)(const void *)(p + 8)); p += 2 * 8;
+    }
+    while (p != lim);
+  }
+  
+  for (; size != 0; size--)
+    v = __crc32b(v, *p++);
+
+  return v;
+}
+
+#undef T0_32_UNROLL_BYTES
+#undef T0_64_UNROLL_BYTES
+
+#endif // defined(USE_ARM64_CRC) || defined(USE_CRC_EMU)
+
+#endif // MY_CPU_LE
+
+
+
+
+void Z7_FASTCALL CrcGenerateTable(void)
 {
   UInt32 i;
   for (i = 0; i < 256; i++)
@@ -54,21 +250,91 @@ void MY_FAST_CALL CrcGenerateTable()
     UInt32 r = i;
     unsigned j;
     for (j = 0; j < 8; j++)
-      r = (r >> 1) ^ (kCrcPoly & ~((r & 1) - 1));
+      r = (r >> 1) ^ (kCrcPoly & ((UInt32)0 - (r & 1)));
     g_CrcTable[i] = r;
   }
-  #if CRC_NUM_TABLES == 1
-  g_CrcUpdate = CrcUpdateT1;
-  #else
-  for (; i < 256 * CRC_NUM_TABLES; i++)
+  for (i = 256; i < 256 * CRC_NUM_TABLES; i++)
   {
-    UInt32 r = g_CrcTable[i - 256];
+    const UInt32 r = g_CrcTable[(size_t)i - 256];
     g_CrcTable[i] = g_CrcTable[r & 0xFF] ^ (r >> 8);
   }
-  g_CrcUpdate = CrcUpdateT4;
-  #ifdef MY_CPU_X86_OR_AMD64
-  if (!CPU_Is_InOrder())
-    g_CrcUpdate = CrcUpdateT8;
-  #endif
+
+  #if CRC_NUM_TABLES < 4
+    g_CrcUpdate = CrcUpdateT1;
+  #elif defined(MY_CPU_LE)
+    // g_CrcUpdateT4 = CrcUpdateT4;
+    #if CRC_NUM_TABLES < 8
+      g_CrcUpdate = CrcUpdateT4;
+    #else // CRC_NUM_TABLES >= 8
+      g_CrcUpdateT8 = CrcUpdateT8;
+      /*
+      #ifdef MY_CPU_X86_OR_AMD64
+      if (!CPU_Is_InOrder())
+      #endif
+      */
+      g_CrcUpdate = CrcUpdateT8;
+    #endif
+  #else
+  {
+   #ifndef MY_CPU_BE
+    UInt32 k = 0x01020304;
+    const Byte *p = (const Byte *)&k;
+    if (p[0] == 4 && p[1] == 3)
+    {
+      #if CRC_NUM_TABLES < 8
+        // g_CrcUpdateT4 = CrcUpdateT4;
+        g_CrcUpdate   = CrcUpdateT4;
+      #else  // CRC_NUM_TABLES >= 8
+        g_CrcUpdateT8 = CrcUpdateT8;
+        g_CrcUpdate   = CrcUpdateT8;
+      #endif
+    }
+    else if (p[0] != 1 || p[1] != 2)
+      g_CrcUpdate = CrcUpdateT1;
+    else
+   #endif // MY_CPU_BE
+    {
+      for (i = 256 * CRC_NUM_TABLES - 1; i >= 256; i--)
+      {
+        const UInt32 x = g_CrcTable[(size_t)i - 256];
+        g_CrcTable[i] = Z7_BSWAP32(x);
+      }
+      #if CRC_NUM_TABLES <= 4
+        g_CrcUpdate = CrcUpdateT1;
+      #elif CRC_NUM_TABLES <= 8
+        // g_CrcUpdateT4 = CrcUpdateT1_BeT4;
+        g_CrcUpdate   = CrcUpdateT1_BeT4;
+      #else  // CRC_NUM_TABLES > 8
+        g_CrcUpdateT8 = CrcUpdateT1_BeT8;
+        g_CrcUpdate   = CrcUpdateT1_BeT8;
+      #endif
+    }
+  }
+  #endif // CRC_NUM_TABLES < 4
+
+  #ifdef MY_CPU_LE
+    #ifdef USE_ARM64_CRC
+      if (CPU_IsSupported_CRC32())
+      {
+        g_CrcUpdateT0_32 = CrcUpdateT0_32;
+        g_CrcUpdateT0_64 = CrcUpdateT0_64;
+        g_CrcUpdate =
+          #if defined(MY_CPU_ARM)
+            CrcUpdateT0_32;
+          #else
+            CrcUpdateT0_64;
+          #endif
+      }
+    #endif
+    
+    #ifdef USE_CRC_EMU
+      g_CrcUpdateT0_32 = CrcUpdateT0_32;
+      g_CrcUpdateT0_64 = CrcUpdateT0_64;
+      g_CrcUpdate = CrcUpdateT0_64;
+    #endif
   #endif
 }
+
+#undef kCrcPoly
+#undef CRC64_NUM_TABLES
+#undef CRC_UPDATE_BYTE_2

C/7zCrc.h

@@ -1,24 +1,26 @@
 /* 7zCrc.h -- CRC32 calculation
-2009-11-21 : Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
 
-#ifndef __7Z_CRC_H
-#define __7Z_CRC_H
+#ifndef ZIP7_INC_7Z_CRC_H
+#define ZIP7_INC_7Z_CRC_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
 EXTERN_C_BEGIN
 
 extern UInt32 g_CrcTable[];
 
 /* Call CrcGenerateTable one time before other CRC functions */
-void MY_FAST_CALL CrcGenerateTable(void);
+void Z7_FASTCALL CrcGenerateTable(void);
 
 #define CRC_INIT_VAL 0xFFFFFFFF
 #define CRC_GET_DIGEST(crc) ((crc) ^ CRC_INIT_VAL)
 #define CRC_UPDATE_BYTE(crc, b) (g_CrcTable[((crc) ^ (b)) & 0xFF] ^ ((crc) >> 8))
 
-UInt32 MY_FAST_CALL CrcUpdate(UInt32 crc, const void *data, size_t size);
-UInt32 MY_FAST_CALL CrcCalc(const void *data, size_t size);
+UInt32 Z7_FASTCALL CrcUpdate(UInt32 crc, const void *data, size_t size);
+UInt32 Z7_FASTCALL CrcCalc(const void *data, size_t size);
+
+typedef UInt32 (Z7_FASTCALL *CRC_FUNC)(UInt32 v, const void *data, size_t size, const UInt32 *table);
 
 EXTERN_C_END
 

C/7zCrcOpt.c

@@ -1,34 +1,117 @@
-/* 7zCrcOpt.c -- CRC32 calculation : optimized version
-2009-11-23 : Igor Pavlov : Public domain */
+/* 7zCrcOpt.c -- CRC32 calculation
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "CpuArch.h"
 
-#ifdef MY_CPU_LE
+#ifndef MY_CPU_BE
 
 #define CRC_UPDATE_BYTE_2(crc, b) (table[((crc) ^ (b)) & 0xFF] ^ ((crc) >> 8))
 
-UInt32 MY_FAST_CALL CrcUpdateT4(UInt32 v, const void *data, size_t size, const UInt32 *table)
+UInt32 Z7_FASTCALL CrcUpdateT4(UInt32 v, const void *data, size_t size, const UInt32 *table);
+UInt32 Z7_FASTCALL CrcUpdateT4(UInt32 v, const void *data, size_t size, const UInt32 *table)
 {
   const Byte *p = (const Byte *)data;
   for (; size > 0 && ((unsigned)(ptrdiff_t)p & 3) != 0; size--, p++)
     v = CRC_UPDATE_BYTE_2(v, *p);
   for (; size >= 4; size -= 4, p += 4)
   {
-    v ^= *(const UInt32 *)p;
+    v ^= *(const UInt32 *)(const void *)p;
     v =
-      table[0x300 + (v & 0xFF)] ^
-      table[0x200 + ((v >> 8) & 0xFF)] ^
-      table[0x100 + ((v >> 16) & 0xFF)] ^
-      table[0x000 + ((v >> 24))];
+          (table + 0x300)[((v      ) & 0xFF)]
+        ^ (table + 0x200)[((v >>  8) & 0xFF)]
+        ^ (table + 0x100)[((v >> 16) & 0xFF)]
+        ^ (table + 0x000)[((v >> 24))];
   }
   for (; size > 0; size--, p++)
     v = CRC_UPDATE_BYTE_2(v, *p);
   return v;
 }
 
-UInt32 MY_FAST_CALL CrcUpdateT8(UInt32 v, const void *data, size_t size, const UInt32 *table)
+UInt32 Z7_FASTCALL CrcUpdateT8(UInt32 v, const void *data, size_t size, const UInt32 *table);
+UInt32 Z7_FASTCALL CrcUpdateT8(UInt32 v, const void *data, size_t size, const UInt32 *table)
 {
-  return CrcUpdateT4(v, data, size, table);
+  const Byte *p = (const Byte *)data;
+  for (; size > 0 && ((unsigned)(ptrdiff_t)p & 7) != 0; size--, p++)
+    v = CRC_UPDATE_BYTE_2(v, *p);
+  for (; size >= 8; size -= 8, p += 8)
+  {
+    UInt32 d;
+    v ^= *(const UInt32 *)(const void *)p;
+    v =
+          (table + 0x700)[((v      ) & 0xFF)]
+        ^ (table + 0x600)[((v >>  8) & 0xFF)]
+        ^ (table + 0x500)[((v >> 16) & 0xFF)]
+        ^ (table + 0x400)[((v >> 24))];
+    d = *((const UInt32 *)(const void *)p + 1);
+    v ^=
+          (table + 0x300)[((d      ) & 0xFF)]
+        ^ (table + 0x200)[((d >>  8) & 0xFF)]
+        ^ (table + 0x100)[((d >> 16) & 0xFF)]
+        ^ (table + 0x000)[((d >> 24))];
+  }
+  for (; size > 0; size--, p++)
+    v = CRC_UPDATE_BYTE_2(v, *p);
+  return v;
+}
+
+#endif
+
+
+#ifndef MY_CPU_LE
+
+#define CRC_UINT32_SWAP(v)  Z7_BSWAP32(v)
+
+#define CRC_UPDATE_BYTE_2_BE(crc, b) (table[(((crc) >> 24) ^ (b))] ^ ((crc) << 8))
+
+UInt32 Z7_FASTCALL CrcUpdateT1_BeT4(UInt32 v, const void *data, size_t size, const UInt32 *table)
+{
+  const Byte *p = (const Byte *)data;
+  table += 0x100;
+  v = CRC_UINT32_SWAP(v);
+  for (; size > 0 && ((unsigned)(ptrdiff_t)p & 3) != 0; size--, p++)
+    v = CRC_UPDATE_BYTE_2_BE(v, *p);
+  for (; size >= 4; size -= 4, p += 4)
+  {
+    v ^= *(const UInt32 *)(const void *)p;
+    v =
+          (table + 0x000)[((v      ) & 0xFF)]
+        ^ (table + 0x100)[((v >>  8) & 0xFF)]
+        ^ (table + 0x200)[((v >> 16) & 0xFF)]
+        ^ (table + 0x300)[((v >> 24))];
+  }
+  for (; size > 0; size--, p++)
+    v = CRC_UPDATE_BYTE_2_BE(v, *p);
+  return CRC_UINT32_SWAP(v);
+}
+
+UInt32 Z7_FASTCALL CrcUpdateT1_BeT8(UInt32 v, const void *data, size_t size, const UInt32 *table)
+{
+  const Byte *p = (const Byte *)data;
+  table += 0x100;
+  v = CRC_UINT32_SWAP(v);
+  for (; size > 0 && ((unsigned)(ptrdiff_t)p & 7) != 0; size--, p++)
+    v = CRC_UPDATE_BYTE_2_BE(v, *p);
+  for (; size >= 8; size -= 8, p += 8)
+  {
+    UInt32 d;
+    v ^= *(const UInt32 *)(const void *)p;
+    v =
+          (table + 0x400)[((v      ) & 0xFF)]
+        ^ (table + 0x500)[((v >>  8) & 0xFF)]
+        ^ (table + 0x600)[((v >> 16) & 0xFF)]
+        ^ (table + 0x700)[((v >> 24))];
+    d = *((const UInt32 *)(const void *)p + 1);
+    v ^=
+          (table + 0x000)[((d      ) & 0xFF)]
+        ^ (table + 0x100)[((d >>  8) & 0xFF)]
+        ^ (table + 0x200)[((d >> 16) & 0xFF)]
+        ^ (table + 0x300)[((d >> 24))];
+  }
+  for (; size > 0; size--, p++)
+    v = CRC_UPDATE_BYTE_2_BE(v, *p);
+  return CRC_UINT32_SWAP(v);
 }
 
 #endif

C/7zDec.c

@@ -1,84 +1,120 @@
 /* 7zDec.c -- Decoding from 7z folder
-2010-03-15 : Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include <string.h>
 
-/* #define _7ZIP_PPMD_SUPPPORT */
+/* #define Z7_PPMD_SUPPORT */
 
 #include "7z.h"
+#include "7zCrc.h"
 
 #include "Bcj2.h"
 #include "Bra.h"
 #include "CpuArch.h"
+#include "Delta.h"
 #include "LzmaDec.h"
 #include "Lzma2Dec.h"
-#ifdef _7ZIP_PPMD_SUPPPORT
+#ifdef Z7_PPMD_SUPPORT
 #include "Ppmd7.h"
 #endif
 
 #define k_Copy 0
+#ifndef Z7_NO_METHOD_LZMA2
 #define k_LZMA2 0x21
-#define k_LZMA 0x30101
-#define k_BCJ 0x03030103
-#define k_BCJ2 0x0303011B
+#endif
+#define k_LZMA  0x30101
+#define k_BCJ2  0x303011B
 
-#ifdef _7ZIP_PPMD_SUPPPORT
+#if !defined(Z7_NO_METHODS_FILTERS)
+#define Z7_USE_BRANCH_FILTER
+#endif
+
+#if !defined(Z7_NO_METHODS_FILTERS) || \
+     defined(Z7_USE_NATIVE_BRANCH_FILTER) && defined(MY_CPU_ARM64)
+#define Z7_USE_FILTER_ARM64
+#ifndef Z7_USE_BRANCH_FILTER
+#define Z7_USE_BRANCH_FILTER
+#endif
+#define k_ARM64 0xa
+#endif
+
+#if !defined(Z7_NO_METHODS_FILTERS) || \
+     defined(Z7_USE_NATIVE_BRANCH_FILTER) && defined(MY_CPU_ARMT)
+#define Z7_USE_FILTER_ARMT
+#ifndef Z7_USE_BRANCH_FILTER
+#define Z7_USE_BRANCH_FILTER
+#endif
+#define k_ARMT  0x3030701
+#endif
+
+#ifndef Z7_NO_METHODS_FILTERS
+#define k_Delta 3
+#define k_BCJ   0x3030103
+#define k_PPC   0x3030205
+#define k_IA64  0x3030401
+#define k_ARM   0x3030501
+#define k_SPARC 0x3030805
+#endif
+
+#ifdef Z7_PPMD_SUPPORT
 
 #define k_PPMD 0x30401
 
 typedef struct
 {
-  IByteIn p;
+  IByteIn vt;
   const Byte *cur;
   const Byte *end;
   const Byte *begin;
   UInt64 processed;
-  Bool extra;
+  BoolInt extra;
   SRes res;
-  ILookInStream *inStream;
+  ILookInStreamPtr inStream;
 } CByteInToLook;
 
-static Byte ReadByte(void *pp)
+static Byte ReadByte(IByteInPtr pp)
 {
-  CByteInToLook *p = (CByteInToLook *)pp;
+  Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(CByteInToLook)
   if (p->cur != p->end)
     return *p->cur++;
   if (p->res == SZ_OK)
   {
-    size_t size = p->cur - p->begin;
+    size_t size = (size_t)(p->cur - p->begin);
     p->processed += size;
-    p->res = p->inStream->Skip(p->inStream, size);
+    p->res = ILookInStream_Skip(p->inStream, size);
     size = (1 << 25);
-    p->res = p->inStream->Look(p->inStream, (const void **)&p->begin, &size);
+    p->res = ILookInStream_Look(p->inStream, (const void **)&p->begin, &size);
     p->cur = p->begin;
     p->end = p->begin + size;
     if (size != 0)
-      return *p->cur++;;
+      return *p->cur++;
   }
   p->extra = True;
   return 0;
 }
 
-static SRes SzDecodePpmd(CSzCoderInfo *coder, UInt64 inSize, ILookInStream *inStream,
-    Byte *outBuffer, SizeT outSize, ISzAlloc *allocMain)
+static SRes SzDecodePpmd(const Byte *props, unsigned propsSize, UInt64 inSize, ILookInStreamPtr inStream,
+    Byte *outBuffer, SizeT outSize, ISzAllocPtr allocMain)
 {
   CPpmd7 ppmd;
   CByteInToLook s;
   SRes res = SZ_OK;
 
-  s.p.Read = ReadByte;
+  s.vt.Read = ReadByte;
   s.inStream = inStream;
   s.begin = s.end = s.cur = NULL;
   s.extra = False;
   s.res = SZ_OK;
   s.processed = 0;
 
-  if (coder->Props.size != 5)
+  if (propsSize != 5)
     return SZ_ERROR_UNSUPPORTED;
 
   {
-    unsigned order = coder->Props.data[0];
-    UInt32 memSize = GetUi32(coder->Props.data + 1);
+    unsigned order = props[0];
+    UInt32 memSize = GetUi32(props + 1);
     if (order < PPMD7_MIN_ORDER ||
         order > PPMD7_MAX_ORDER ||
         memSize < PPMD7_MIN_MEM_SIZE ||
@@ -90,28 +126,32 @@ static SRes SzDecodePpmd(CSzCoderInfo *coder, UInt64 inSize, ILookInStream *inSt
     Ppmd7_Init(&ppmd, order);
   }
   {
-    CPpmd7z_RangeDec rc;
-    Ppmd7z_RangeDec_CreateVTable(&rc);
-    rc.Stream = &s.p;
-    if (!Ppmd7z_RangeDec_Init(&rc))
+    ppmd.rc.dec.Stream = &s.vt;
+    if (!Ppmd7z_RangeDec_Init(&ppmd.rc.dec))
       res = SZ_ERROR_DATA;
-    else if (s.extra)
-      res = (s.res != SZ_OK ? s.res : SZ_ERROR_DATA);
-    else
+    else if (!s.extra)
     {
-      SizeT i;
-      for (i = 0; i < outSize; i++)
+      Byte *buf = outBuffer;
+      const Byte *lim = buf + outSize;
+      for (; buf != lim; buf++)
       {
-        int sym = Ppmd7_DecodeSymbol(&ppmd, &rc.p);
+        int sym = Ppmd7z_DecodeSymbol(&ppmd);
         if (s.extra || sym < 0)
           break;
-        outBuffer[i] = (Byte)sym;
+        *buf = (Byte)sym;
       }
-      if (i != outSize)
-        res = (s.res != SZ_OK ? s.res : SZ_ERROR_DATA);
-      else if (s.processed + (s.cur - s.begin) != inSize || !Ppmd7z_RangeDec_IsFinishedOK(&rc))
+      if (buf != lim)
         res = SZ_ERROR_DATA;
+      else if (!Ppmd7z_RangeDec_IsFinishedOK(&ppmd.rc.dec))
+      {
+        /* if (Ppmd7z_DecodeSymbol(&ppmd) != PPMD7_SYM_END || !Ppmd7z_RangeDec_IsFinishedOK(&ppmd.rc.dec)) */
+        res = SZ_ERROR_DATA;
+      }
     }
+    if (s.extra)
+      res = (s.res != SZ_OK ? s.res : SZ_ERROR_DATA);
+    else if (s.processed + (size_t)(s.cur - s.begin) != inSize)
+      res = SZ_ERROR_DATA;
   }
   Ppmd7_Free(&ppmd, allocMain);
   return res;
@@ -120,45 +160,54 @@ static SRes SzDecodePpmd(CSzCoderInfo *coder, UInt64 inSize, ILookInStream *inSt
 #endif
 
 
-static SRes SzDecodeLzma(CSzCoderInfo *coder, UInt64 inSize, ILookInStream *inStream,
-    Byte *outBuffer, SizeT outSize, ISzAlloc *allocMain)
+static SRes SzDecodeLzma(const Byte *props, unsigned propsSize, UInt64 inSize, ILookInStreamPtr inStream,
+    Byte *outBuffer, SizeT outSize, ISzAllocPtr allocMain)
 {
   CLzmaDec state;
   SRes res = SZ_OK;
 
-  LzmaDec_Construct(&state);
-  RINOK(LzmaDec_AllocateProbs(&state, coder->Props.data, (unsigned)coder->Props.size, allocMain));
+  LzmaDec_CONSTRUCT(&state)
+  RINOK(LzmaDec_AllocateProbs(&state, props, propsSize, allocMain))
   state.dic = outBuffer;
   state.dicBufSize = outSize;
   LzmaDec_Init(&state);
 
   for (;;)
   {
-    Byte *inBuf = NULL;
+    const void *inBuf = NULL;
     size_t lookahead = (1 << 18);
     if (lookahead > inSize)
       lookahead = (size_t)inSize;
-    res = inStream->Look((void *)inStream, (const void **)&inBuf, &lookahead);
+    res = ILookInStream_Look(inStream, &inBuf, &lookahead);
     if (res != SZ_OK)
       break;
 
     {
       SizeT inProcessed = (SizeT)lookahead, dicPos = state.dicPos;
       ELzmaStatus status;
-      res = LzmaDec_DecodeToDic(&state, outSize, inBuf, &inProcessed, LZMA_FINISH_END, &status);
+      res = LzmaDec_DecodeToDic(&state, outSize, (const Byte *)inBuf, &inProcessed, LZMA_FINISH_END, &status);
       lookahead -= inProcessed;
       inSize -= inProcessed;
       if (res != SZ_OK)
         break;
-      if (state.dicPos == state.dicBufSize || (inProcessed == 0 && dicPos == state.dicPos))
+
+      if (status == LZMA_STATUS_FINISHED_WITH_MARK)
       {
-        if (state.dicBufSize != outSize || lookahead != 0 ||
-            (status != LZMA_STATUS_FINISHED_WITH_MARK &&
-             status != LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK))
+        if (outSize != state.dicPos || inSize != 0)
           res = SZ_ERROR_DATA;
         break;
       }
-      res = inStream->Skip((void *)inStream, inProcessed);
+
+      if (outSize == state.dicPos && inSize == 0 && status == LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK)
+        break;
+
+      if (inProcessed == 0 && dicPos == state.dicPos)
+      {
+        res = SZ_ERROR_DATA;
+        break;
+      }
+
+      res = ILookInStream_Skip(inStream, inProcessed);
       if (res != SZ_OK)
         break;
     }
@@ -168,46 +217,56 @@ static SRes SzDecodeLzma(CSzCoderInfo *coder, UInt64 inSize, ILookInStream *inSt
   return res;
 }
 
-static SRes SzDecodeLzma2(CSzCoderInfo *coder, UInt64 inSize, ILookInStream *inStream,
-    Byte *outBuffer, SizeT outSize, ISzAlloc *allocMain)
+
+#ifndef Z7_NO_METHOD_LZMA2
+
+static SRes SzDecodeLzma2(const Byte *props, unsigned propsSize, UInt64 inSize, ILookInStreamPtr inStream,
+    Byte *outBuffer, SizeT outSize, ISzAllocPtr allocMain)
 {
   CLzma2Dec state;
   SRes res = SZ_OK;
 
-  Lzma2Dec_Construct(&state);
-  if (coder->Props.size != 1)
+  Lzma2Dec_CONSTRUCT(&state)
+  if (propsSize != 1)
     return SZ_ERROR_DATA;
-  RINOK(Lzma2Dec_AllocateProbs(&state, coder->Props.data[0], allocMain));
+  RINOK(Lzma2Dec_AllocateProbs(&state, props[0], allocMain))
   state.decoder.dic = outBuffer;
   state.decoder.dicBufSize = outSize;
   Lzma2Dec_Init(&state);
 
   for (;;)
   {
-    Byte *inBuf = NULL;
+    const void *inBuf = NULL;
     size_t lookahead = (1 << 18);
     if (lookahead > inSize)
       lookahead = (size_t)inSize;
-    res = inStream->Look((void *)inStream, (const void **)&inBuf, &lookahead);
+    res = ILookInStream_Look(inStream, &inBuf, &lookahead);
     if (res != SZ_OK)
       break;
 
     {
       SizeT inProcessed = (SizeT)lookahead, dicPos = state.decoder.dicPos;
       ELzmaStatus status;
-      res = Lzma2Dec_DecodeToDic(&state, outSize, inBuf, &inProcessed, LZMA_FINISH_END, &status);
+      res = Lzma2Dec_DecodeToDic(&state, outSize, (const Byte *)inBuf, &inProcessed, LZMA_FINISH_END, &status);
       lookahead -= inProcessed;
       inSize -= inProcessed;
       if (res != SZ_OK)
         break;
-      if (state.decoder.dicPos == state.decoder.dicBufSize || (inProcessed == 0 && dicPos == state.decoder.dicPos))
+
+      if (status == LZMA_STATUS_FINISHED_WITH_MARK)
       {
-        if (state.decoder.dicBufSize != outSize || lookahead != 0 ||
-            (status != LZMA_STATUS_FINISHED_WITH_MARK))
+        if (outSize != state.decoder.dicPos || inSize != 0)
           res = SZ_ERROR_DATA;
         break;
       }
-      res = inStream->Skip((void *)inStream, inProcessed);
+
+      if (inProcessed == 0 && dicPos == state.decoder.dicPos)
+      {
+        res = SZ_ERROR_DATA;
+        break;
+      }
+
+      res = ILookInStream_Skip(inStream, inProcessed);
       if (res != SZ_OK)
         break;
     }
@@ -217,51 +276,54 @@ static SRes SzDecodeLzma2(CSzCoderInfo *coder, UInt64 inSize, ILookInStream *inS
   return res;
 }
 
-static SRes SzDecodeCopy(UInt64 inSize, ILookInStream *inStream, Byte *outBuffer)
+#endif
+
+
+static SRes SzDecodeCopy(UInt64 inSize, ILookInStreamPtr inStream, Byte *outBuffer)
 {
   while (inSize > 0)
   {
-    void *inBuf;
+    const void *inBuf;
     size_t curSize = (1 << 18);
     if (curSize > inSize)
       curSize = (size_t)inSize;
-    RINOK(inStream->Look((void *)inStream, (const void **)&inBuf, &curSize));
+    RINOK(ILookInStream_Look(inStream, &inBuf, &curSize))
     if (curSize == 0)
       return SZ_ERROR_INPUT_EOF;
     memcpy(outBuffer, inBuf, curSize);
     outBuffer += curSize;
     inSize -= curSize;
-    RINOK(inStream->Skip((void *)inStream, curSize));
+    RINOK(ILookInStream_Skip(inStream, curSize))
   }
   return SZ_OK;
 }
 
-static Bool IS_MAIN_METHOD(UInt32 m)
+static BoolInt IS_MAIN_METHOD(UInt32 m)
 {
-  switch(m)
+  switch (m)
   {
     case k_Copy:
     case k_LZMA:
+  #ifndef Z7_NO_METHOD_LZMA2
     case k_LZMA2:
-    #ifdef _7ZIP_PPMD_SUPPPORT
+  #endif
+  #ifdef Z7_PPMD_SUPPORT
     case k_PPMD:
-    #endif
+  #endif
       return True;
   }
   return False;
 }
 
-static Bool IS_SUPPORTED_CODER(const CSzCoderInfo *c)
+static BoolInt IS_SUPPORTED_CODER(const CSzCoderInfo *c)
 {
   return
-      c->NumInStreams == 1 &&
-      c->NumOutStreams == 1 &&
-      c->MethodID <= (UInt32)0xFFFFFFFF &&
-      IS_MAIN_METHOD((UInt32)c->MethodID);
+      c->NumStreams == 1
+      /* && c->MethodID <= (UInt32)0xFFFFFFFF */
+      && IS_MAIN_METHOD((UInt32)c->MethodID);
 }
 
-#define IS_BCJ(c) ((c)->MethodID == k_BCJ && (c)->NumInStreams == 1 && (c)->NumOutStreams == 1)
-#define IS_BCJ2(c) ((c)->MethodID == k_BCJ2 && (c)->NumInStreams == 4 && (c)->NumOutStreams == 1)
+#define IS_BCJ2(c) ((c)->MethodID == k_BCJ2 && (c)->NumStreams == 4)
 
 static SRes CheckSupportedFolder(const CSzFolder *f)
 {
@@ -271,52 +333,85 @@ static SRes CheckSupportedFolder(const CSzFolder *f)
     return SZ_ERROR_UNSUPPORTED;
   if (f->NumCoders == 1)
   {
-    if (f->NumPackStreams != 1 || f->PackStreams[0] != 0 || f->NumBindPairs != 0)
+    if (f->NumPackStreams != 1 || f->PackStreams[0] != 0 || f->NumBonds != 0)
       return SZ_ERROR_UNSUPPORTED;
     return SZ_OK;
   }
+  
+  
+  #if defined(Z7_USE_BRANCH_FILTER)
+
   if (f->NumCoders == 2)
   {
-    if (!IS_BCJ(&f->Coders[1]) ||
-        f->NumPackStreams != 1 || f->PackStreams[0] != 0 ||
-        f->NumBindPairs != 1 ||
-        f->BindPairs[0].InIndex != 1 || f->BindPairs[0].OutIndex != 0)
+    const CSzCoderInfo *c = &f->Coders[1];
+    if (
+        /* c->MethodID > (UInt32)0xFFFFFFFF || */
+        c->NumStreams != 1
+        || f->NumPackStreams != 1
+        || f->PackStreams[0] != 0
+        || f->NumBonds != 1
+        || f->Bonds[0].InIndex != 1
+        || f->Bonds[0].OutIndex != 0)
       return SZ_ERROR_UNSUPPORTED;
+    switch ((UInt32)c->MethodID)
+    {
+    #if !defined(Z7_NO_METHODS_FILTERS)
+      case k_Delta:
+      case k_BCJ:
+      case k_PPC:
+      case k_IA64:
+      case k_SPARC:
+      case k_ARM:
+    #endif
+    #ifdef Z7_USE_FILTER_ARM64
+      case k_ARM64:
+    #endif
+    #ifdef Z7_USE_FILTER_ARMT
+      case k_ARMT:
+    #endif
+        break;
+      default:
+        return SZ_ERROR_UNSUPPORTED;
+    }
     return SZ_OK;
   }
+
+  #endif
+
+  
   if (f->NumCoders == 4)
   {
-    if (!IS_SUPPORTED_CODER(&f->Coders[1]) ||
-        !IS_SUPPORTED_CODER(&f->Coders[2]) ||
-        !IS_BCJ2(&f->Coders[3]))
+    if (!IS_SUPPORTED_CODER(&f->Coders[1])
+        || !IS_SUPPORTED_CODER(&f->Coders[2])
+        || !IS_BCJ2(&f->Coders[3]))
       return SZ_ERROR_UNSUPPORTED;
-    if (f->NumPackStreams != 4 ||
-        f->PackStreams[0] != 2 ||
-        f->PackStreams[1] != 6 ||
-        f->PackStreams[2] != 1 ||
-        f->PackStreams[3] != 0 ||
-        f->NumBindPairs != 3 ||
-        f->BindPairs[0].InIndex != 5 || f->BindPairs[0].OutIndex != 0 ||
-        f->BindPairs[1].InIndex != 4 || f->BindPairs[1].OutIndex != 1 ||
-        f->BindPairs[2].InIndex != 3 || f->BindPairs[2].OutIndex != 2)
+    if (f->NumPackStreams != 4
+        || f->PackStreams[0] != 2
+        || f->PackStreams[1] != 6
+        || f->PackStreams[2] != 1
+        || f->PackStreams[3] != 0
+        || f->NumBonds != 3
+        || f->Bonds[0].InIndex != 5 || f->Bonds[0].OutIndex != 0
+        || f->Bonds[1].InIndex != 4 || f->Bonds[1].OutIndex != 1
+        || f->Bonds[2].InIndex != 3 || f->Bonds[2].OutIndex != 2)
       return SZ_ERROR_UNSUPPORTED;
     return SZ_OK;
   }
+  
   return SZ_ERROR_UNSUPPORTED;
 }
 
-static UInt64 GetSum(const UInt64 *values, UInt32 index)
-{
-  UInt64 sum = 0;
-  UInt32 i;
-  for (i = 0; i < index; i++)
-    sum += values[i];
-  return sum;
-}
 
-static SRes SzFolder_Decode2(const CSzFolder *folder, const UInt64 *packSizes,
-    ILookInStream *inStream, UInt64 startPos,
-    Byte *outBuffer, SizeT outSize, ISzAlloc *allocMain,
+
+
+
+
+static SRes SzFolder_Decode2(const CSzFolder *folder,
+    const Byte *propsData,
+    const UInt64 *unpackSizes,
+    const UInt64 *packPositions,
+    ILookInStreamPtr inStream, UInt64 startPos,
+    Byte *outBuffer, SizeT outSize, ISzAllocPtr allocMain,
     Byte *tempBuf[])
 {
   UInt32 ci;
@@ -324,11 +419,11 @@ static SRes SzFolder_Decode2(const CSzFolder *folder, const UInt64 *packSizes,
   SizeT tempSize3 = 0;
   Byte *tempBuf3 = 0;
 
-  RINOK(CheckSupportedFolder(folder));
+  RINOK(CheckSupportedFolder(folder))
 
   for (ci = 0; ci < folder->NumCoders; ci++)
   {
-    CSzCoderInfo *coder = &folder->Coders[ci];
+    const CSzCoderInfo *coder = &folder->Coders[ci];
 
     if (IS_MAIN_METHOD((UInt32)coder->MethodID))
     {
@@ -339,8 +434,8 @@ static SRes SzFolder_Decode2(const CSzFolder *folder, const UInt64 *packSizes,
       SizeT outSizeCur = outSize;
       if (folder->NumCoders == 4)
       {
-        UInt32 indices[] = { 3, 2, 0 };
-        UInt64 unpackSize = folder->UnpackSizes[ci];
+        const UInt32 indices[] = { 3, 2, 0 };
+        const UInt64 unpackSize = unpackSizes[ci];
         si = indices[ci];
         if (ci < 2)
         {
@@ -348,8 +443,8 @@ static SRes SzFolder_Decode2(const CSzFolder *folder, const UInt64 *packSizes,
           outSizeCur = (SizeT)unpackSize;
           if (outSizeCur != unpackSize)
             return SZ_ERROR_MEM;
-          temp = (Byte *)IAlloc_Alloc(allocMain, outSizeCur);
-          if (temp == 0 && outSizeCur != 0)
+          temp = (Byte *)ISzAlloc_Alloc(allocMain, outSizeCur);
+          if (!temp && outSizeCur != 0)
             return SZ_ERROR_MEM;
           outBufCur = tempBuf[1 - ci] = temp;
           tempSizes[1 - ci] = outSizeCur;
@@ -364,81 +459,190 @@ static SRes SzFolder_Decode2(const CSzFolder *folder, const UInt64 *packSizes,
         else
           return SZ_ERROR_UNSUPPORTED;
       }
-      offset = GetSum(packSizes, si);
-      inSize = packSizes[si];
-      RINOK(LookInStream_SeekTo(inStream, startPos + offset));
+      offset = packPositions[si];
+      inSize = packPositions[(size_t)si + 1] - offset;
+      RINOK(LookInStream_SeekTo(inStream, startPos + offset))
 
       if (coder->MethodID == k_Copy)
       {
         if (inSize != outSizeCur) /* check it */
           return SZ_ERROR_DATA;
-        RINOK(SzDecodeCopy(inSize, inStream, outBufCur));
+        RINOK(SzDecodeCopy(inSize, inStream, outBufCur))
       }
       else if (coder->MethodID == k_LZMA)
       {
-        RINOK(SzDecodeLzma(coder, inSize, inStream, outBufCur, outSizeCur, allocMain));
+        RINOK(SzDecodeLzma(propsData + coder->PropsOffset, coder->PropsSize, inSize, inStream, outBufCur, outSizeCur, allocMain))
       }
+    #ifndef Z7_NO_METHOD_LZMA2
       else if (coder->MethodID == k_LZMA2)
       {
-        RINOK(SzDecodeLzma2(coder, inSize, inStream, outBufCur, outSizeCur, allocMain));
+        RINOK(SzDecodeLzma2(propsData + coder->PropsOffset, coder->PropsSize, inSize, inStream, outBufCur, outSizeCur, allocMain))
       }
-      else
+    #endif
+    #ifdef Z7_PPMD_SUPPORT
+      else if (coder->MethodID == k_PPMD)
       {
-        #ifdef _7ZIP_PPMD_SUPPPORT
-        RINOK(SzDecodePpmd(coder, inSize, inStream, outBufCur, outSizeCur, allocMain));
-        #else
-        return SZ_ERROR_UNSUPPORTED;
-        #endif
+        RINOK(SzDecodePpmd(propsData + coder->PropsOffset, coder->PropsSize, inSize, inStream, outBufCur, outSizeCur, allocMain))
       }
-    }
-    else if (coder->MethodID == k_BCJ)
-    {
-      UInt32 state;
-      if (ci != 1)
+    #endif
+      else
         return SZ_ERROR_UNSUPPORTED;
-      x86_Convert_Init(state);
-      x86_Convert(outBuffer, outSize, 0, &state, 0);
     }
     else if (coder->MethodID == k_BCJ2)
     {
-      UInt64 offset = GetSum(packSizes, 1);
-      UInt64 s3Size = packSizes[1];
-      SRes res;
+      const UInt64 offset = packPositions[1];
+      const UInt64 s3Size = packPositions[2] - offset;
+      
       if (ci != 3)
         return SZ_ERROR_UNSUPPORTED;
-      RINOK(LookInStream_SeekTo(inStream, startPos + offset));
+      
       tempSizes[2] = (SizeT)s3Size;
       if (tempSizes[2] != s3Size)
         return SZ_ERROR_MEM;
-      tempBuf[2] = (Byte *)IAlloc_Alloc(allocMain, tempSizes[2]);
-      if (tempBuf[2] == 0 && tempSizes[2] != 0)
+      tempBuf[2] = (Byte *)ISzAlloc_Alloc(allocMain, tempSizes[2]);
+      if (!tempBuf[2] && tempSizes[2] != 0)
         return SZ_ERROR_MEM;
-      res = SzDecodeCopy(s3Size, inStream, tempBuf[2]);
-      RINOK(res)
+      
+      RINOK(LookInStream_SeekTo(inStream, startPos + offset))
+      RINOK(SzDecodeCopy(s3Size, inStream, tempBuf[2]))
 
-      res = Bcj2_Decode(
-          tempBuf3, tempSize3,
-          tempBuf[0], tempSizes[0],
-          tempBuf[1], tempSizes[1],
-          tempBuf[2], tempSizes[2],
-          outBuffer, outSize);
-      RINOK(res)
+      if ((tempSizes[0] & 3) != 0 ||
+          (tempSizes[1] & 3) != 0 ||
+          tempSize3 + tempSizes[0] + tempSizes[1] != outSize)
+        return SZ_ERROR_DATA;
+
+      {
+        CBcj2Dec p;
+        
+        p.bufs[0] = tempBuf3;   p.lims[0] = tempBuf3 + tempSize3;
+        p.bufs[1] = tempBuf[0]; p.lims[1] = tempBuf[0] + tempSizes[0];
+        p.bufs[2] = tempBuf[1]; p.lims[2] = tempBuf[1] + tempSizes[1];
+        p.bufs[3] = tempBuf[2]; p.lims[3] = tempBuf[2] + tempSizes[2];
+        
+        p.dest = outBuffer;
+        p.destLim = outBuffer + outSize;
+        
+        Bcj2Dec_Init(&p);
+        RINOK(Bcj2Dec_Decode(&p))
+
+        {
+          unsigned i;
+          for (i = 0; i < 4; i++)
+            if (p.bufs[i] != p.lims[i])
+              return SZ_ERROR_DATA;
+          if (p.dest != p.destLim || !Bcj2Dec_IsMaybeFinished(&p))
+            return SZ_ERROR_DATA;
+        }
+      }
     }
+  #if defined(Z7_USE_BRANCH_FILTER)
+    else if (ci == 1)
+    {
+     #if !defined(Z7_NO_METHODS_FILTERS)
+      if (coder->MethodID == k_Delta)
+      {
+        if (coder->PropsSize != 1)
+          return SZ_ERROR_UNSUPPORTED;
+        {
+          Byte state[DELTA_STATE_SIZE];
+          Delta_Init(state);
+          Delta_Decode(state, (unsigned)(propsData[coder->PropsOffset]) + 1, outBuffer, outSize);
+        }
+        continue;
+      }
+     #endif
+     
+     #ifdef Z7_USE_FILTER_ARM64
+      if (coder->MethodID == k_ARM64)
+      {
+        UInt32 pc = 0;
+        if (coder->PropsSize == 4)
+          pc = GetUi32(propsData + coder->PropsOffset);
+        else if (coder->PropsSize != 0)
+          return SZ_ERROR_UNSUPPORTED;
+        z7_BranchConv_ARM64_Dec(outBuffer, outSize, pc);
+        continue;
+      }
+     #endif
+     
+     #if !defined(Z7_NO_METHODS_FILTERS) || defined(Z7_USE_FILTER_ARMT)
+      {
+        if (coder->PropsSize != 0)
+          return SZ_ERROR_UNSUPPORTED;
+       #define CASE_BRA_CONV(isa) case k_ ## isa: Z7_BRANCH_CONV_DEC(isa)(outBuffer, outSize, 0); break; // pc = 0;
+        switch (coder->MethodID)
+        {
+         #if !defined(Z7_NO_METHODS_FILTERS)
+          case k_BCJ:
+          {
+            UInt32 state = Z7_BRANCH_CONV_ST_X86_STATE_INIT_VAL;
+            z7_BranchConvSt_X86_Dec(outBuffer, outSize, 0, &state); // pc = 0
+            break;
+          }
+          CASE_BRA_CONV(PPC)
+          CASE_BRA_CONV(IA64)
+          CASE_BRA_CONV(SPARC)
+          CASE_BRA_CONV(ARM)
+         #endif
+         #if !defined(Z7_NO_METHODS_FILTERS) || defined(Z7_USE_FILTER_ARMT)
+          CASE_BRA_CONV(ARMT)
+         #endif
+          default:
+            return SZ_ERROR_UNSUPPORTED;
+        }
+        continue;
+      }
+     #endif
+    } // (c == 1)
+  #endif
     else
       return SZ_ERROR_UNSUPPORTED;
   }
+
   return SZ_OK;
 }
 
-SRes SzFolder_Decode(const CSzFolder *folder, const UInt64 *packSizes,
-    ILookInStream *inStream, UInt64 startPos,
-    Byte *outBuffer, size_t outSize, ISzAlloc *allocMain)
+
+SRes SzAr_DecodeFolder(const CSzAr *p, UInt32 folderIndex,
+    ILookInStreamPtr inStream, UInt64 startPos,
+    Byte *outBuffer, size_t outSize,
+    ISzAllocPtr allocMain)
 {
-  Byte *tempBuf[3] = { 0, 0, 0};
-  int i;
-  SRes res = SzFolder_Decode2(folder, packSizes, inStream, startPos,
-      outBuffer, (SizeT)outSize, allocMain, tempBuf);
-  for (i = 0; i < 3; i++)
-    IAlloc_Free(allocMain, tempBuf[i]);
-  return res;
+  SRes res;
+  CSzFolder folder;
+  CSzData sd;
+  
+  const Byte *data = p->CodersData + p->FoCodersOffsets[folderIndex];
+  sd.Data = data;
+  sd.Size = p->FoCodersOffsets[(size_t)folderIndex + 1] - p->FoCodersOffsets[folderIndex];
+  
+  res = SzGetNextFolderItem(&folder, &sd);
+  
+  if (res != SZ_OK)
+    return res;
+
+  if (sd.Size != 0
+      || folder.UnpackStream != p->FoToMainUnpackSizeIndex[folderIndex]
+      || outSize != SzAr_GetFolderUnpackSize(p, folderIndex))
+    return SZ_ERROR_FAIL;
+  {
+    unsigned i;
+    Byte *tempBuf[3] = { 0, 0, 0};
+
+    res = SzFolder_Decode2(&folder, data,
+        &p->CoderUnpackSizes[p->FoToCoderUnpackSizes[folderIndex]],
+        p->PackPositions + p->FoStartPackStreamIndex[folderIndex],
+        inStream, startPos,
+        outBuffer, (SizeT)outSize, allocMain, tempBuf);
+    
+    for (i = 0; i < 3; i++)
+      ISzAlloc_Free(allocMain, tempBuf[i]);
+
+    if (res == SZ_OK)
+      if (SzBitWithVals_Check(&p->FolderCRCs, folderIndex))
+        if (CrcCalc(outBuffer, outSize) != p->FolderCRCs.Vals[folderIndex])
+          res = SZ_ERROR_CRC;
+
+    return res;
+  }
 }

C/7zFile.c

@@ -1,13 +1,25 @@
 /* 7zFile.c -- File IO
-2009-11-24 : Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "7zFile.h"
 
 #ifndef USE_WINDOWS_FILE
 
-#ifndef UNDER_CE
-#include <errno.h>
-#endif
+  #include <errno.h>
+
+  #ifndef USE_FOPEN
+    #include <stdio.h>
+    #include <fcntl.h>
+    #ifdef _WIN32
+      #include <io.h>
+      typedef int ssize_t;
+      typedef int off_t;
+    #else
+      #include <unistd.h>
+    #endif
+  #endif
 
 #else
 
@@ -21,30 +33,36 @@
    And message can be "Network connection was lost"
 */
 
-#define kChunkSizeMax (1 << 22)
-
 #endif
 
+#define kChunkSizeMax (1 << 22)
+
 void File_Construct(CSzFile *p)
 {
   #ifdef USE_WINDOWS_FILE
   p->handle = INVALID_HANDLE_VALUE;
-  #else
+  #elif defined(USE_FOPEN)
   p->file = NULL;
+  #else
+  p->fd = -1;
   #endif
 }
 
 #if !defined(UNDER_CE) || !defined(USE_WINDOWS_FILE)
+
 static WRes File_Open(CSzFile *p, const char *name, int writeMode)
 {
   #ifdef USE_WINDOWS_FILE
+  
   p->handle = CreateFileA(name,
       writeMode ? GENERIC_WRITE : GENERIC_READ,
       FILE_SHARE_READ, NULL,
       writeMode ? CREATE_ALWAYS : OPEN_EXISTING,
       FILE_ATTRIBUTE_NORMAL, NULL);
   return (p->handle != INVALID_HANDLE_VALUE) ? 0 : GetLastError();
-  #else
+  
+  #elif defined(USE_FOPEN)
+  
   p->file = fopen(name, writeMode ? "wb+" : "rb");
   return (p->file != 0) ? 0 :
     #ifdef UNDER_CE
@@ -52,13 +70,34 @@ static WRes File_Open(CSzFile *p, const char *name, int writeMode)
     #else
     errno;
     #endif
+  
+  #else
+
+  int flags = (writeMode ? (O_CREAT | O_EXCL | O_WRONLY) : O_RDONLY);
+  #ifdef O_BINARY
+  flags |= O_BINARY;
+  #endif
+  p->fd = open(name, flags, 0666);
+  return (p->fd != -1) ? 0 : errno;
+
   #endif
 }
 
 WRes InFile_Open(CSzFile *p, const char *name) { return File_Open(p, name, 0); }
-WRes OutFile_Open(CSzFile *p, const char *name) { return File_Open(p, name, 1); }
+
+WRes OutFile_Open(CSzFile *p, const char *name)
+{
+  #if defined(USE_WINDOWS_FILE) || defined(USE_FOPEN)
+  return File_Open(p, name, 1);
+  #else
+  p->fd = creat(name, 0666);
+  return (p->fd != -1) ? 0 : errno;
+  #endif
+}
+
 #endif
 
+
 #ifdef USE_WINDOWS_FILE
 static WRes File_OpenW(CSzFile *p, const WCHAR *name, int writeMode)
 {
@@ -76,74 +115,124 @@ WRes OutFile_OpenW(CSzFile *p, const WCHAR *name) { return File_OpenW(p, name, 1
 WRes File_Close(CSzFile *p)
 {
   #ifdef USE_WINDOWS_FILE
+  
   if (p->handle != INVALID_HANDLE_VALUE)
   {
     if (!CloseHandle(p->handle))
       return GetLastError();
     p->handle = INVALID_HANDLE_VALUE;
   }
-  #else
+  
+  #elif defined(USE_FOPEN)
+
   if (p->file != NULL)
   {
     int res = fclose(p->file);
     if (res != 0)
+    {
+      if (res == EOF)
+        return errno;
       return res;
+    }
     p->file = NULL;
   }
+
+  #else
+
+  if (p->fd != -1)
+  {
+    if (close(p->fd) != 0)
+      return errno;
+    p->fd = -1;
+  }
+
   #endif
+
   return 0;
 }
 
+
 WRes File_Read(CSzFile *p, void *data, size_t *size)
 {
   size_t originalSize = *size;
+  *size = 0;
   if (originalSize == 0)
     return 0;
 
   #ifdef USE_WINDOWS_FILE
 
-  *size = 0;
   do
   {
-    DWORD curSize = (originalSize > kChunkSizeMax) ? kChunkSizeMax : (DWORD)originalSize;
+    const DWORD curSize = (originalSize > kChunkSizeMax) ? kChunkSizeMax : (DWORD)originalSize;
     DWORD processed = 0;
-    BOOL res = ReadFile(p->handle, data, curSize, &processed, NULL);
+    const BOOL res = ReadFile(p->handle, data, curSize, &processed, NULL);
     data = (void *)((Byte *)data + processed);
     originalSize -= processed;
     *size += processed;
     if (!res)
       return GetLastError();
+    // debug : we can break here for partial reading mode
+    if (processed == 0)
+      break;
+  }
+  while (originalSize > 0);
+
+  #elif defined(USE_FOPEN)
+
+  do
+  {
+    const size_t curSize = (originalSize > kChunkSizeMax) ? kChunkSizeMax : originalSize;
+    const size_t processed = fread(data, 1, curSize, p->file);
+    data = (void *)((Byte *)data + (size_t)processed);
+    originalSize -= processed;
+    *size += processed;
+    if (processed != curSize)
+      return ferror(p->file);
+    // debug : we can break here for partial reading mode
     if (processed == 0)
       break;
   }
   while (originalSize > 0);
-  return 0;
 
   #else
-  
-  *size = fread(data, 1, originalSize, p->file);
-  if (*size == originalSize)
-    return 0;
-  return ferror(p->file);
-  
+
+  do
+  {
+    const size_t curSize = (originalSize > kChunkSizeMax) ? kChunkSizeMax : originalSize;
+    const ssize_t processed = read(p->fd, data, curSize);
+    if (processed == -1)
+      return errno;
+    if (processed == 0)
+      break;
+    data = (void *)((Byte *)data + (size_t)processed);
+    originalSize -= (size_t)processed;
+    *size += (size_t)processed;
+    // debug : we can break here for partial reading mode
+    // break;
+  }
+  while (originalSize > 0);
+
   #endif
+
+  return 0;
 }
 
+
 WRes File_Write(CSzFile *p, const void *data, size_t *size)
 {
   size_t originalSize = *size;
+  *size = 0;
   if (originalSize == 0)
     return 0;
   
   #ifdef USE_WINDOWS_FILE
 
-  *size = 0;
   do
   {
-    DWORD curSize = (originalSize > kChunkSizeMax) ? kChunkSizeMax : (DWORD)originalSize;
+    const DWORD curSize = (originalSize > kChunkSizeMax) ? kChunkSizeMax : (DWORD)originalSize;
     DWORD processed = 0;
-    BOOL res = WriteFile(p->handle, data, curSize, &processed, NULL);
-    data = (void *)((Byte *)data + processed);
+    const BOOL res = WriteFile(p->handle, data, curSize, &processed, NULL);
+    data = (const void *)((const Byte *)data + processed);
     originalSize -= processed;
     *size += processed;
     if (!res)
@@ -152,61 +241,106 @@ WRes File_Write(CSzFile *p, const void *data, size_t *size)
       break;
   }
   while (originalSize > 0);
-  return 0;
+
+  #elif defined(USE_FOPEN)
+
+  do
+  {
+    const size_t curSize = (originalSize > kChunkSizeMax) ? kChunkSizeMax : originalSize;
+    const size_t processed = fwrite(data, 1, curSize, p->file);
+    data = (void *)((Byte *)data + (size_t)processed);
+    originalSize -= processed;
+    *size += processed;
+    if (processed != curSize)
+      return ferror(p->file);
+    if (processed == 0)
+      break;
+  }
+  while (originalSize > 0);
 
   #else
 
-  *size = fwrite(data, 1, originalSize, p->file);
-  if (*size == originalSize)
-    return 0;
-  return ferror(p->file);
-  
+  do
+  {
+    const size_t curSize = (originalSize > kChunkSizeMax) ? kChunkSizeMax : originalSize;
+    const ssize_t processed = write(p->fd, data, curSize);
+    if (processed == -1)
+      return errno;
+    if (processed == 0)
+      break;
+    data = (const void *)((const Byte *)data + (size_t)processed);
+    originalSize -= (size_t)processed;
+    *size += (size_t)processed;
+  }
+  while (originalSize > 0);
+
   #endif
+
+  return 0;
 }
 
+
 WRes File_Seek(CSzFile *p, Int64 *pos, ESzSeek origin)
 {
   #ifdef USE_WINDOWS_FILE
 
-  LARGE_INTEGER value;
   DWORD moveMethod;
-  value.LowPart = (DWORD)*pos;
-  value.HighPart = (LONG)((UInt64)*pos >> 16 >> 16); /* for case when UInt64 is 32-bit only */
-  switch (origin)
+  UInt32 low = (UInt32)*pos;
+  LONG high = (LONG)((UInt64)*pos >> 16 >> 16); /* for case when UInt64 is 32-bit only */
+  // (int) to eliminate clang warning
+  switch ((int)origin)
   {
     case SZ_SEEK_SET: moveMethod = FILE_BEGIN; break;
     case SZ_SEEK_CUR: moveMethod = FILE_CURRENT; break;
     case SZ_SEEK_END: moveMethod = FILE_END; break;
     default: return ERROR_INVALID_PARAMETER;
   }
-  value.LowPart = SetFilePointer(p->handle, value.LowPart, &value.HighPart, moveMethod);
-  if (value.LowPart == 0xFFFFFFFF)
+  low = SetFilePointer(p->handle, (LONG)low, &high, moveMethod);
+  if (low == (UInt32)0xFFFFFFFF)
   {
     WRes res = GetLastError();
     if (res != NO_ERROR)
       return res;
   }
-  *pos = ((Int64)value.HighPart << 32) | value.LowPart;
+  *pos = ((Int64)high << 32) | low;
   return 0;
 
   #else
   
-  int moveMethod;
-  int res;
-  switch (origin)
+  int moveMethod; // = origin;
+
+  switch ((int)origin)
   {
     case SZ_SEEK_SET: moveMethod = SEEK_SET; break;
     case SZ_SEEK_CUR: moveMethod = SEEK_CUR; break;
     case SZ_SEEK_END: moveMethod = SEEK_END; break;
-    default: return 1;
+    default: return EINVAL;
   }
-  res = fseek(p->file, (long)*pos, moveMethod);
-  *pos = ftell(p->file);
-  return res;
   
-  #endif
+  #if defined(USE_FOPEN)
+  {
+    int res = fseek(p->file, (long)*pos, moveMethod);
+    if (res == -1)
+      return errno;
+    *pos = ftell(p->file);
+    if (*pos == -1)
+      return errno;
+    return 0;
+  }
+  #else
+  {
+    off_t res = lseek(p->fd, (off_t)*pos, moveMethod);
+    if (res == -1)
+      return errno;
+    *pos = res;
+    return 0;
+  }
+  
+  #endif // USE_FOPEN
+  #endif // USE_WINDOWS_FILE
 }
 
+
 WRes File_GetLength(CSzFile *p, UInt64 *length)
 {
   #ifdef USE_WINDOWS_FILE
@@ -222,13 +356,31 @@ WRes File_GetLength(CSzFile *p, UInt64 *length)
   *length = (((UInt64)sizeHigh) << 32) + sizeLow;
   return 0;
   
-  #else
+  #elif defined(USE_FOPEN)
   
   long pos = ftell(p->file);
   int res = fseek(p->file, 0, SEEK_END);
   *length = ftell(p->file);
   fseek(p->file, pos, SEEK_SET);
   return res;
+
+  #else
+
+  off_t pos;
+  *length = 0;
+  pos = lseek(p->fd, 0, SEEK_CUR);
+  if (pos != -1)
+  {
+    const off_t len2 = lseek(p->fd, 0, SEEK_END);
+    const off_t res2 = lseek(p->fd, pos, SEEK_SET);
+    if (len2 != -1)
+    {
+      *length = (UInt64)len2;
+      if (res2 != -1)
+        return 0;
+    }
+  }
+  return errno;
   
   #endif
 }
@@ -236,49 +388,56 @@ WRes File_GetLength(CSzFile *p, UInt64 *length)
 
 /* ---------- FileSeqInStream ---------- */
 
-static SRes FileSeqInStream_Read(void *pp, void *buf, size_t *size)
+static SRes FileSeqInStream_Read(ISeqInStreamPtr pp, void *buf, size_t *size)
 {
-  CFileSeqInStream *p = (CFileSeqInStream *)pp;
-  return File_Read(&p->file, buf, size) == 0 ? SZ_OK : SZ_ERROR_READ;
+  Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(CFileSeqInStream)
+  const WRes wres = File_Read(&p->file, buf, size);
+  p->wres = wres;
+  return (wres == 0) ? SZ_OK : SZ_ERROR_READ;
 }
 
 void FileSeqInStream_CreateVTable(CFileSeqInStream *p)
 {
-  p->s.Read = FileSeqInStream_Read;
+  p->vt.Read = FileSeqInStream_Read;
 }
 
 
 /* ---------- FileInStream ---------- */
 
-static SRes FileInStream_Read(void *pp, void *buf, size_t *size)
+static SRes FileInStream_Read(ISeekInStreamPtr pp, void *buf, size_t *size)
 {
-  CFileInStream *p = (CFileInStream *)pp;
-  return (File_Read(&p->file, buf, size) == 0) ? SZ_OK : SZ_ERROR_READ;
+  Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(CFileInStream)
+  const WRes wres = File_Read(&p->file, buf, size);
+  p->wres = wres;
+  return (wres == 0) ? SZ_OK : SZ_ERROR_READ;
 }
 
-static SRes FileInStream_Seek(void *pp, Int64 *pos, ESzSeek origin)
+static SRes FileInStream_Seek(ISeekInStreamPtr pp, Int64 *pos, ESzSeek origin)
 {
-  CFileInStream *p = (CFileInStream *)pp;
-  return File_Seek(&p->file, pos, origin);
+  Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(CFileInStream)
+  const WRes wres = File_Seek(&p->file, pos, origin);
+  p->wres = wres;
+  return (wres == 0) ? SZ_OK : SZ_ERROR_READ;
 }
 
 void FileInStream_CreateVTable(CFileInStream *p)
 {
-  p->s.Read = FileInStream_Read;
-  p->s.Seek = FileInStream_Seek;
+  p->vt.Read = FileInStream_Read;
+  p->vt.Seek = FileInStream_Seek;
 }
 
 
 /* ---------- FileOutStream ---------- */
 
-static size_t FileOutStream_Write(void *pp, const void *data, size_t size)
+static size_t FileOutStream_Write(ISeqOutStreamPtr pp, const void *data, size_t size)
 {
-  CFileOutStream *p = (CFileOutStream *)pp;
-  File_Write(&p->file, data, &size);
+  Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(CFileOutStream)
+  const WRes wres = File_Write(&p->file, data, &size);
+  p->wres = wres;
   return size;
 }
 
 void FileOutStream_CreateVTable(CFileOutStream *p)
 {
-  p->s.Write = FileOutStream_Write;
+  p->vt.Write = FileOutStream_Write;
 }

C/7zFile.h

@@ -1,20 +1,24 @@
 /* 7zFile.h -- File IO
-2009-11-24 : Igor Pavlov : Public domain */
+2023-03-05 : Igor Pavlov : Public domain */
 
-#ifndef __7Z_FILE_H
-#define __7Z_FILE_H
+#ifndef ZIP7_INC_FILE_H
+#define ZIP7_INC_FILE_H
 
 #ifdef _WIN32
 #define USE_WINDOWS_FILE
+// #include <windows.h>
 #endif
 
 #ifdef USE_WINDOWS_FILE
-#include <windows.h>
+#include "7zWindows.h"
+
 #else
-#include <stdio.h>
+// note: USE_FOPEN mode is limited to 32-bit file size
+// #define USE_FOPEN
+// #include <stdio.h>
 #endif
 
-#include "Types.h"
+#include "7zTypes.h"
 
 EXTERN_C_BEGIN
 
@@ -24,8 +28,10 @@ typedef struct
 {
   #ifdef USE_WINDOWS_FILE
   HANDLE handle;
-  #else
+  #elif defined(USE_FOPEN)
   FILE *file;
+  #else
+  int fd;
   #endif
 } CSzFile;
 
@@ -54,8 +60,9 @@ WRes File_GetLength(CSzFile *p, UInt64 *length);
 
 typedef struct
 {
-  ISeqInStream s;
+  ISeqInStream vt;
   CSzFile file;
+  WRes wres;
 } CFileSeqInStream;
 
 void FileSeqInStream_CreateVTable(CFileSeqInStream *p);
@@ -63,8 +70,9 @@ void FileSeqInStream_CreateVTable(CFileSeqInStream *p);
 
 typedef struct
 {
-  ISeekInStream s;
+  ISeekInStream vt;
   CSzFile file;
+  WRes wres;
 } CFileInStream;
 
 void FileInStream_CreateVTable(CFileInStream *p);
@@ -72,8 +80,9 @@ void FileInStream_CreateVTable(CFileInStream *p);
 
 typedef struct
 {
-  ISeqOutStream s;
+  ISeqOutStream vt;
   CSzFile file;
+  WRes wres;
 } CFileOutStream;
 
 void FileOutStream_CreateVTable(CFileOutStream *p);

C/7zStream.c

@@ -1,16 +1,39 @@
 /* 7zStream.c -- 7z Stream functions
-2010-03-11 : Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include <string.h>
 
-#include "Types.h"
+#include "7zTypes.h"
 
-SRes SeqInStream_Read2(ISeqInStream *stream, void *buf, size_t size, SRes errorType)
+
+SRes SeqInStream_ReadMax(ISeqInStreamPtr stream, void *buf, size_t *processedSize)
+{
+  size_t size = *processedSize;
+  *processedSize = 0;
+  while (size != 0)
+  {
+    size_t cur = size;
+    const SRes res = ISeqInStream_Read(stream, buf, &cur);
+    *processedSize += cur;
+    buf = (void *)((Byte *)buf + cur);
+    size -= cur;
+    if (res != SZ_OK)
+      return res;
+    if (cur == 0)
+      return SZ_OK;
+  }
+  return SZ_OK;
+}
+
+/*
+SRes SeqInStream_Read2(ISeqInStreamPtr stream, void *buf, size_t size, SRes errorType)
 {
   while (size != 0)
   {
     size_t processed = size;
-    RINOK(stream->Read(stream, buf, &processed));
+    RINOK(ISeqInStream_Read(stream, buf, &processed))
     if (processed == 0)
       return errorType;
     buf = (void *)((Byte *)buf + processed);
@@ -19,40 +42,44 @@ SRes SeqInStream_Read2(ISeqInStream *stream, void *buf, size_t size, SRes errorT
   return SZ_OK;
 }
 
-SRes SeqInStream_Read(ISeqInStream *stream, void *buf, size_t size)
+SRes SeqInStream_Read(ISeqInStreamPtr stream, void *buf, size_t size)
 {
   return SeqInStream_Read2(stream, buf, size, SZ_ERROR_INPUT_EOF);
 }
+*/
 
-SRes SeqInStream_ReadByte(ISeqInStream *stream, Byte *buf)
+
+SRes SeqInStream_ReadByte(ISeqInStreamPtr stream, Byte *buf)
 {
   size_t processed = 1;
-  RINOK(stream->Read(stream, buf, &processed));
+  RINOK(ISeqInStream_Read(stream, buf, &processed))
   return (processed == 1) ? SZ_OK : SZ_ERROR_INPUT_EOF;
 }
 
-SRes LookInStream_SeekTo(ILookInStream *stream, UInt64 offset)
+
+
+SRes LookInStream_SeekTo(ILookInStreamPtr stream, UInt64 offset)
 {
-  Int64 t = offset;
-  return stream->Seek(stream, &t, SZ_SEEK_SET);
+  Int64 t = (Int64)offset;
+  return ILookInStream_Seek(stream, &t, SZ_SEEK_SET);
 }
 
-SRes LookInStream_LookRead(ILookInStream *stream, void *buf, size_t *size)
+SRes LookInStream_LookRead(ILookInStreamPtr stream, void *buf, size_t *size)
 {
   const void *lookBuf;
   if (*size == 0)
     return SZ_OK;
-  RINOK(stream->Look(stream, &lookBuf, size));
+  RINOK(ILookInStream_Look(stream, &lookBuf, size))
   memcpy(buf, lookBuf, *size);
-  return stream->Skip(stream, *size);
+  return ILookInStream_Skip(stream, *size);
 }
 
-SRes LookInStream_Read2(ILookInStream *stream, void *buf, size_t size, SRes errorType)
+SRes LookInStream_Read2(ILookInStreamPtr stream, void *buf, size_t size, SRes errorType)
 {
   while (size != 0)
   {
     size_t processed = size;
-    RINOK(stream->Read(stream, buf, &processed));
+    RINOK(ILookInStream_Read(stream, buf, &processed))
     if (processed == 0)
       return errorType;
     buf = (void *)((Byte *)buf + processed);
@@ -61,61 +88,67 @@ SRes LookInStream_Read2(ILookInStream *stream, void *buf, size_t size, SRes erro
   return SZ_OK;
 }
 
-SRes LookInStream_Read(ILookInStream *stream, void *buf, size_t size)
+SRes LookInStream_Read(ILookInStreamPtr stream, void *buf, size_t size)
 {
   return LookInStream_Read2(stream, buf, size, SZ_ERROR_INPUT_EOF);
 }
 
-static SRes LookToRead_Look_Lookahead(void *pp, const void **buf, size_t *size)
+
+
+#define GET_LookToRead2  Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(CLookToRead2)
+
+static SRes LookToRead2_Look_Lookahead(ILookInStreamPtr pp, const void **buf, size_t *size)
 {
   SRes res = SZ_OK;
-  CLookToRead *p = (CLookToRead *)pp;
+  GET_LookToRead2
   size_t size2 = p->size - p->pos;
-  if (size2 == 0 && *size > 0)
+  if (size2 == 0 && *size != 0)
   {
     p->pos = 0;
-    size2 = LookToRead_BUF_SIZE;
-    res = p->realStream->Read(p->realStream, p->buf, &size2);
+    p->size = 0;
+    size2 = p->bufSize;
+    res = ISeekInStream_Read(p->realStream, p->buf, &size2);
     p->size = size2;
   }
-  if (size2 < *size)
+  if (*size > size2)
     *size = size2;
   *buf = p->buf + p->pos;
   return res;
 }
 
-static SRes LookToRead_Look_Exact(void *pp, const void **buf, size_t *size)
+static SRes LookToRead2_Look_Exact(ILookInStreamPtr pp, const void **buf, size_t *size)
 {
   SRes res = SZ_OK;
-  CLookToRead *p = (CLookToRead *)pp;
+  GET_LookToRead2
   size_t size2 = p->size - p->pos;
-  if (size2 == 0 && *size > 0)
+  if (size2 == 0 && *size != 0)
   {
     p->pos = 0;
-    if (*size > LookToRead_BUF_SIZE)
-      *size = LookToRead_BUF_SIZE;
-    res = p->realStream->Read(p->realStream, p->buf, size);
+    p->size = 0;
+    if (*size > p->bufSize)
+      *size = p->bufSize;
+    res = ISeekInStream_Read(p->realStream, p->buf, size);
     size2 = p->size = *size;
   }
-  if (size2 < *size)
+  if (*size > size2)
     *size = size2;
   *buf = p->buf + p->pos;
   return res;
 }
 
-static SRes LookToRead_Skip(void *pp, size_t offset)
+static SRes LookToRead2_Skip(ILookInStreamPtr pp, size_t offset)
 {
-  CLookToRead *p = (CLookToRead *)pp;
+  GET_LookToRead2
   p->pos += offset;
   return SZ_OK;
 }
 
-static SRes LookToRead_Read(void *pp, void *buf, size_t *size)
+static SRes LookToRead2_Read(ILookInStreamPtr pp, void *buf, size_t *size)
 {
-  CLookToRead *p = (CLookToRead *)pp;
+  GET_LookToRead2
   size_t rem = p->size - p->pos;
   if (rem == 0)
-    return p->realStream->Read(p->realStream, buf, size);
+    return ISeekInStream_Read(p->realStream, buf, size);
   if (rem > *size)
     rem = *size;
   memcpy(buf, p->buf + p->pos, rem);
@@ -124,46 +157,43 @@ static SRes LookToRead_Read(void *pp, void *buf, size_t *size)
   return SZ_OK;
 }
 
-static SRes LookToRead_Seek(void *pp, Int64 *pos, ESzSeek origin)
+static SRes LookToRead2_Seek(ILookInStreamPtr pp, Int64 *pos, ESzSeek origin)
 {
-  CLookToRead *p = (CLookToRead *)pp;
+  GET_LookToRead2
   p->pos = p->size = 0;
-  return p->realStream->Seek(p->realStream, pos, origin);
+  return ISeekInStream_Seek(p->realStream, pos, origin);
 }
 
-void LookToRead_CreateVTable(CLookToRead *p, int lookahead)
+void LookToRead2_CreateVTable(CLookToRead2 *p, int lookahead)
 {
-  p->s.Look = lookahead ?
-      LookToRead_Look_Lookahead :
-      LookToRead_Look_Exact;
-  p->s.Skip = LookToRead_Skip;
-  p->s.Read = LookToRead_Read;
-  p->s.Seek = LookToRead_Seek;
+  p->vt.Look = lookahead ?
+      LookToRead2_Look_Lookahead :
+      LookToRead2_Look_Exact;
+  p->vt.Skip = LookToRead2_Skip;
+  p->vt.Read = LookToRead2_Read;
+  p->vt.Seek = LookToRead2_Seek;
 }
 
-void LookToRead_Init(CLookToRead *p)
-{
-  p->pos = p->size = 0;
-}
 
-static SRes SecToLook_Read(void *pp, void *buf, size_t *size)
+
+static SRes SecToLook_Read(ISeqInStreamPtr pp, void *buf, size_t *size)
 {
-  CSecToLook *p = (CSecToLook *)pp;
+  Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(CSecToLook)
   return LookInStream_LookRead(p->realStream, buf, size);
 }
 
 void SecToLook_CreateVTable(CSecToLook *p)
 {
-  p->s.Read = SecToLook_Read;
+  p->vt.Read = SecToLook_Read;
 }
 
-static SRes SecToRead_Read(void *pp, void *buf, size_t *size)
+static SRes SecToRead_Read(ISeqInStreamPtr pp, void *buf, size_t *size)
 {
-  CSecToRead *p = (CSecToRead *)pp;
-  return p->realStream->Read(p->realStream, buf, size);
+  Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(CSecToRead)
+  return ILookInStream_Read(p->realStream, buf, size);
 }
 
 void SecToRead_CreateVTable(CSecToRead *p)
 {
-  p->s.Read = SecToRead_Read;
+  p->vt.Read = SecToRead_Read;
 }

C/7zTypes.h

@@ -0,0 +1,597 @@
+/* 7zTypes.h -- Basic types
+2023-04-02 : Igor Pavlov : Public domain */
+
+#ifndef ZIP7_7Z_TYPES_H
+#define ZIP7_7Z_TYPES_H
+
+#ifdef _WIN32
+/* #include <windows.h> */
+#else
+#include <errno.h>
+#endif
+
+#include <stddef.h>
+
+#ifndef EXTERN_C_BEGIN
+#ifdef __cplusplus
+#define EXTERN_C_BEGIN extern "C" {
+#define EXTERN_C_END }
+#else
+#define EXTERN_C_BEGIN
+#define EXTERN_C_END
+#endif
+#endif
+
+EXTERN_C_BEGIN
+
+#define SZ_OK 0
+
+#define SZ_ERROR_DATA 1
+#define SZ_ERROR_MEM 2
+#define SZ_ERROR_CRC 3
+#define SZ_ERROR_UNSUPPORTED 4
+#define SZ_ERROR_PARAM 5
+#define SZ_ERROR_INPUT_EOF 6
+#define SZ_ERROR_OUTPUT_EOF 7
+#define SZ_ERROR_READ 8
+#define SZ_ERROR_WRITE 9
+#define SZ_ERROR_PROGRESS 10
+#define SZ_ERROR_FAIL 11
+#define SZ_ERROR_THREAD 12
+
+#define SZ_ERROR_ARCHIVE 16
+#define SZ_ERROR_NO_ARCHIVE 17
+
+typedef int SRes;
+
+
+#ifdef _MSC_VER
+  #if _MSC_VER > 1200
+    #define MY_ALIGN(n) __declspec(align(n))
+  #else
+    #define MY_ALIGN(n)
+  #endif
+#else
+  /*
+  // C11/C++11:
+  #include <stdalign.h>
+  #define MY_ALIGN(n) alignas(n)
+  */
+  #define MY_ALIGN(n) __attribute__ ((aligned(n)))
+#endif
+
+
+#ifdef _WIN32
+
+/* typedef DWORD WRes; */
+typedef unsigned WRes;
+#define MY_SRes_HRESULT_FROM_WRes(x) HRESULT_FROM_WIN32(x)
+
+// #define MY_HRES_ERROR_INTERNAL_ERROR  MY_SRes_HRESULT_FROM_WRes(ERROR_INTERNAL_ERROR)
+
+#else // _WIN32
+
+// #define ENV_HAVE_LSTAT
+typedef int WRes;
+
+// (FACILITY_ERRNO = 0x800) is 7zip's FACILITY constant to represent (errno) errors in HRESULT
+#define MY_FACILITY_ERRNO  0x800
+#define MY_FACILITY_WIN32  7
+#define MY_FACILITY_WRes  MY_FACILITY_ERRNO
+
+#define MY_HRESULT_FROM_errno_CONST_ERROR(x) ((HRESULT)( \
+          ( (HRESULT)(x) & 0x0000FFFF) \
+          | (MY_FACILITY_WRes << 16)  \
+          | (HRESULT)0x80000000 ))
+
+#define MY_SRes_HRESULT_FROM_WRes(x) \
+  ((HRESULT)(x) <= 0 ? ((HRESULT)(x)) : MY_HRESULT_FROM_errno_CONST_ERROR(x))
+
+// we call macro HRESULT_FROM_WIN32 for system errors (WRes) that are (errno)
+#define HRESULT_FROM_WIN32(x) MY_SRes_HRESULT_FROM_WRes(x)
+
+/*
+#define ERROR_FILE_NOT_FOUND             2L
+#define ERROR_ACCESS_DENIED              5L
+#define ERROR_NO_MORE_FILES              18L
+#define ERROR_LOCK_VIOLATION             33L
+#define ERROR_FILE_EXISTS                80L
+#define ERROR_DISK_FULL                  112L
+#define ERROR_NEGATIVE_SEEK              131L
+#define ERROR_ALREADY_EXISTS             183L
+#define ERROR_DIRECTORY                  267L
+#define ERROR_TOO_MANY_POSTS             298L
+
+#define ERROR_INTERNAL_ERROR             1359L
+#define ERROR_INVALID_REPARSE_DATA       4392L
+#define ERROR_REPARSE_TAG_INVALID        4393L
+#define ERROR_REPARSE_TAG_MISMATCH       4394L
+*/
+
+// we use errno equivalents for some WIN32 errors:
+
+#define ERROR_INVALID_PARAMETER     EINVAL
+#define ERROR_INVALID_FUNCTION      EINVAL
+#define ERROR_ALREADY_EXISTS        EEXIST
+#define ERROR_FILE_EXISTS           EEXIST
+#define ERROR_PATH_NOT_FOUND        ENOENT
+#define ERROR_FILE_NOT_FOUND        ENOENT
+#define ERROR_DISK_FULL             ENOSPC
+// #define ERROR_INVALID_HANDLE        EBADF
+
+// we use FACILITY_WIN32 for errors that has no errno equivalent
+// Too many posts were made to a semaphore.
+#define ERROR_TOO_MANY_POSTS        ((HRESULT)0x8007012AL)
+#define ERROR_INVALID_REPARSE_DATA  ((HRESULT)0x80071128L)
+#define ERROR_REPARSE_TAG_INVALID   ((HRESULT)0x80071129L)
+
+// if (MY_FACILITY_WRes != FACILITY_WIN32),
+// we use FACILITY_WIN32 for COM errors:
+#define E_OUTOFMEMORY               ((HRESULT)0x8007000EL)
+#define E_INVALIDARG                ((HRESULT)0x80070057L)
+#define MY_E_ERROR_NEGATIVE_SEEK    ((HRESULT)0x80070083L)
+
+/*
+// we can use FACILITY_ERRNO for some COM errors, that have errno equivalents:
+#define E_OUTOFMEMORY             MY_HRESULT_FROM_errno_CONST_ERROR(ENOMEM)
+#define E_INVALIDARG              MY_HRESULT_FROM_errno_CONST_ERROR(EINVAL)
+#define MY_E_ERROR_NEGATIVE_SEEK  MY_HRESULT_FROM_errno_CONST_ERROR(EINVAL)
+*/
+
+#define TEXT(quote) quote
+
+#define FILE_ATTRIBUTE_READONLY       0x0001
+#define FILE_ATTRIBUTE_HIDDEN         0x0002
+#define FILE_ATTRIBUTE_SYSTEM         0x0004
+#define FILE_ATTRIBUTE_DIRECTORY      0x0010
+#define FILE_ATTRIBUTE_ARCHIVE        0x0020
+#define FILE_ATTRIBUTE_DEVICE         0x0040
+#define FILE_ATTRIBUTE_NORMAL         0x0080
+#define FILE_ATTRIBUTE_TEMPORARY      0x0100
+#define FILE_ATTRIBUTE_SPARSE_FILE    0x0200
+#define FILE_ATTRIBUTE_REPARSE_POINT  0x0400
+#define FILE_ATTRIBUTE_COMPRESSED     0x0800
+#define FILE_ATTRIBUTE_OFFLINE        0x1000
+#define FILE_ATTRIBUTE_NOT_CONTENT_INDEXED 0x2000
+#define FILE_ATTRIBUTE_ENCRYPTED      0x4000
+
+#define FILE_ATTRIBUTE_UNIX_EXTENSION 0x8000   /* trick for Unix */
+
+#endif
+
+
+#ifndef RINOK
+#define RINOK(x) { const int _result_ = (x); if (_result_ != 0) return _result_; }
+#endif
+
+#ifndef RINOK_WRes
+#define RINOK_WRes(x) { const WRes _result_ = (x); if (_result_ != 0) return _result_; }
+#endif
+
+typedef unsigned char Byte;
+typedef short Int16;
+typedef unsigned short UInt16;
+
+#ifdef Z7_DECL_Int32_AS_long
+typedef long Int32;
+typedef unsigned long UInt32;
+#else
+typedef int Int32;
+typedef unsigned int UInt32;
+#endif
+
+
+#ifndef _WIN32
+
+typedef int INT;
+typedef Int32 INT32;
+typedef unsigned int UINT;
+typedef UInt32 UINT32;
+typedef INT32 LONG;   // LONG, ULONG and DWORD must be 32-bit for _WIN32 compatibility
+typedef UINT32 ULONG;
+
+#undef DWORD
+typedef UINT32 DWORD;
+
+#define VOID void
+
+#define HRESULT LONG
+
+typedef void *LPVOID;
+// typedef void VOID;
+// typedef ULONG_PTR DWORD_PTR, *PDWORD_PTR;
+// gcc / clang on Unix  : sizeof(long==sizeof(void*) in 32 or 64 bits)
+typedef          long  INT_PTR;
+typedef unsigned long  UINT_PTR;
+typedef          long  LONG_PTR;
+typedef unsigned long  DWORD_PTR;
+
+typedef size_t SIZE_T;
+
+#endif //  _WIN32
+
+
+#define MY_HRES_ERROR_INTERNAL_ERROR  ((HRESULT)0x8007054FL)
+
+
+#ifdef Z7_DECL_Int64_AS_long
+
+typedef long Int64;
+typedef unsigned long UInt64;
+
+#else
+
+#if (defined(_MSC_VER) || defined(__BORLANDC__)) && !defined(__clang__)
+typedef __int64 Int64;
+typedef unsigned __int64 UInt64;
+#else
+#if defined(__clang__) || defined(__GNUC__)
+#include <stdint.h>
+typedef int64_t Int64;
+typedef uint64_t UInt64;
+#else
+typedef long long int Int64;
+typedef unsigned long long int UInt64;
+// #define UINT64_CONST(n) n ## ULL
+#endif
+#endif
+
+#endif
+
+#define UINT64_CONST(n) n
+
+
+#ifdef Z7_DECL_SizeT_AS_unsigned_int
+typedef unsigned int SizeT;
+#else
+typedef size_t SizeT;
+#endif
+
+/*
+#if (defined(_MSC_VER) && _MSC_VER <= 1200)
+typedef size_t MY_uintptr_t;
+#else
+#include <stdint.h>
+typedef uintptr_t MY_uintptr_t;
+#endif
+*/
+
+typedef int BoolInt;
+/* typedef BoolInt Bool; */
+#define True 1
+#define False 0
+
+
+#ifdef _WIN32
+#define Z7_STDCALL __stdcall
+#else
+#define Z7_STDCALL
+#endif
+
+#ifdef _MSC_VER
+
+#if _MSC_VER >= 1300
+#define Z7_NO_INLINE __declspec(noinline)
+#else
+#define Z7_NO_INLINE
+#endif
+
+#define Z7_FORCE_INLINE __forceinline
+
+#define Z7_CDECL      __cdecl
+#define Z7_FASTCALL  __fastcall
+
+#else //  _MSC_VER
+
+#if (defined(__GNUC__) && (__GNUC__ >= 4)) \
+    || (defined(__clang__) && (__clang_major__ >= 4)) \
+    || defined(__INTEL_COMPILER) \
+    || defined(__xlC__)
+#define Z7_NO_INLINE      __attribute__((noinline))
+#define Z7_FORCE_INLINE   __attribute__((always_inline)) inline
+#else
+#define Z7_NO_INLINE
+#define Z7_FORCE_INLINE
+#endif
+
+#define Z7_CDECL
+
+#if  defined(_M_IX86) \
+  || defined(__i386__)
+// #define Z7_FASTCALL __attribute__((fastcall))
+// #define Z7_FASTCALL __attribute__((cdecl))
+#define Z7_FASTCALL
+#elif defined(MY_CPU_AMD64)
+// #define Z7_FASTCALL __attribute__((ms_abi))
+#define Z7_FASTCALL
+#else
+#define Z7_FASTCALL
+#endif
+
+#endif //  _MSC_VER
+
+
+/* The following interfaces use first parameter as pointer to structure */
+
+// #define Z7_C_IFACE_CONST_QUAL
+#define Z7_C_IFACE_CONST_QUAL const
+
+#define Z7_C_IFACE_DECL(a) \
+  struct a ## _; \
+  typedef Z7_C_IFACE_CONST_QUAL struct a ## _ * a ## Ptr; \
+  typedef struct a ## _ a; \
+  struct a ## _
+
+
+Z7_C_IFACE_DECL (IByteIn)
+{
+  Byte (*Read)(IByteInPtr p); /* reads one byte, returns 0 in case of EOF or error */
+};
+#define IByteIn_Read(p) (p)->Read(p)
+
+
+Z7_C_IFACE_DECL (IByteOut)
+{
+  void (*Write)(IByteOutPtr p, Byte b);
+};
+#define IByteOut_Write(p, b) (p)->Write(p, b)
+
+
+Z7_C_IFACE_DECL (ISeqInStream)
+{
+  SRes (*Read)(ISeqInStreamPtr p, void *buf, size_t *size);
+    /* if (input(*size) != 0 && output(*size) == 0) means end_of_stream.
+       (output(*size) < input(*size)) is allowed */
+};
+#define ISeqInStream_Read(p, buf, size) (p)->Read(p, buf, size)
+
+/* try to read as much as avail in stream and limited by (*processedSize) */
+SRes SeqInStream_ReadMax(ISeqInStreamPtr stream, void *buf, size_t *processedSize);
+/* it can return SZ_ERROR_INPUT_EOF */
+// SRes SeqInStream_Read(ISeqInStreamPtr stream, void *buf, size_t size);
+// SRes SeqInStream_Read2(ISeqInStreamPtr stream, void *buf, size_t size, SRes errorType);
+SRes SeqInStream_ReadByte(ISeqInStreamPtr stream, Byte *buf);
+
+
+Z7_C_IFACE_DECL (ISeqOutStream)
+{
+  size_t (*Write)(ISeqOutStreamPtr p, const void *buf, size_t size);
+    /* Returns: result - the number of actually written bytes.
+       (result < size) means error */
+};
+#define ISeqOutStream_Write(p, buf, size) (p)->Write(p, buf, size)
+
+typedef enum
+{
+  SZ_SEEK_SET = 0,
+  SZ_SEEK_CUR = 1,
+  SZ_SEEK_END = 2
+} ESzSeek;
+
+
+Z7_C_IFACE_DECL (ISeekInStream)
+{
+  SRes (*Read)(ISeekInStreamPtr p, void *buf, size_t *size);  /* same as ISeqInStream::Read */
+  SRes (*Seek)(ISeekInStreamPtr p, Int64 *pos, ESzSeek origin);
+};
+#define ISeekInStream_Read(p, buf, size)   (p)->Read(p, buf, size)
+#define ISeekInStream_Seek(p, pos, origin) (p)->Seek(p, pos, origin)
+
+
+Z7_C_IFACE_DECL (ILookInStream)
+{
+  SRes (*Look)(ILookInStreamPtr p, const void **buf, size_t *size);
+    /* if (input(*size) != 0 && output(*size) == 0) means end_of_stream.
+       (output(*size) > input(*size)) is not allowed
+       (output(*size) < input(*size)) is allowed */
+  SRes (*Skip)(ILookInStreamPtr p, size_t offset);
+    /* offset must be <= output(*size) of Look */
+  SRes (*Read)(ILookInStreamPtr p, void *buf, size_t *size);
+    /* reads directly (without buffer). It's same as ISeqInStream::Read */
+  SRes (*Seek)(ILookInStreamPtr p, Int64 *pos, ESzSeek origin);
+};
+
+#define ILookInStream_Look(p, buf, size)   (p)->Look(p, buf, size)
+#define ILookInStream_Skip(p, offset)      (p)->Skip(p, offset)
+#define ILookInStream_Read(p, buf, size)   (p)->Read(p, buf, size)
+#define ILookInStream_Seek(p, pos, origin) (p)->Seek(p, pos, origin)
+
+
+SRes LookInStream_LookRead(ILookInStreamPtr stream, void *buf, size_t *size);
+SRes LookInStream_SeekTo(ILookInStreamPtr stream, UInt64 offset);
+
+/* reads via ILookInStream::Read */
+SRes LookInStream_Read2(ILookInStreamPtr stream, void *buf, size_t size, SRes errorType);
+SRes LookInStream_Read(ILookInStreamPtr stream, void *buf, size_t size);
+
+
+typedef struct
+{
+  ILookInStream vt;
+  ISeekInStreamPtr realStream;
+ 
+  size_t pos;
+  size_t size; /* it's data size */
+  
+  /* the following variables must be set outside */
+  Byte *buf;
+  size_t bufSize;
+} CLookToRead2;
+
+void LookToRead2_CreateVTable(CLookToRead2 *p, int lookahead);
+
+#define LookToRead2_INIT(p) { (p)->pos = (p)->size = 0; }
+
+
+typedef struct
+{
+  ISeqInStream vt;
+  ILookInStreamPtr realStream;
+} CSecToLook;
+
+void SecToLook_CreateVTable(CSecToLook *p);
+
+
+
+typedef struct
+{
+  ISeqInStream vt;
+  ILookInStreamPtr realStream;
+} CSecToRead;
+
+void SecToRead_CreateVTable(CSecToRead *p);
+
+
+Z7_C_IFACE_DECL (ICompressProgress)
+{
+  SRes (*Progress)(ICompressProgressPtr p, UInt64 inSize, UInt64 outSize);
+    /* Returns: result. (result != SZ_OK) means break.
+       Value (UInt64)(Int64)-1 for size means unknown value. */
+};
+
+#define ICompressProgress_Progress(p, inSize, outSize) (p)->Progress(p, inSize, outSize)
+
+
+
+typedef struct ISzAlloc ISzAlloc;
+typedef const ISzAlloc * ISzAllocPtr;
+
+struct ISzAlloc
+{
+  void *(*Alloc)(ISzAllocPtr p, size_t size);
+  void (*Free)(ISzAllocPtr p, void *address); /* address can be 0 */
+};
+
+#define ISzAlloc_Alloc(p, size) (p)->Alloc(p, size)
+#define ISzAlloc_Free(p, a) (p)->Free(p, a)
+
+/* deprecated */
+#define IAlloc_Alloc(p, size) ISzAlloc_Alloc(p, size)
+#define IAlloc_Free(p, a) ISzAlloc_Free(p, a)
+
+
+
+
+
+#ifndef MY_offsetof
+  #ifdef offsetof
+    #define MY_offsetof(type, m) offsetof(type, m)
+    /*
+    #define MY_offsetof(type, m) FIELD_OFFSET(type, m)
+    */
+  #else
+    #define MY_offsetof(type, m) ((size_t)&(((type *)0)->m))
+  #endif
+#endif
+
+
+
+#ifndef Z7_container_of
+
+/*
+#define Z7_container_of(ptr, type, m) container_of(ptr, type, m)
+#define Z7_container_of(ptr, type, m) CONTAINING_RECORD(ptr, type, m)
+#define Z7_container_of(ptr, type, m) ((type *)((char *)(ptr) - offsetof(type, m)))
+#define Z7_container_of(ptr, type, m) (&((type *)0)->m == (ptr), ((type *)(((char *)(ptr)) - MY_offsetof(type, m))))
+*/
+
+/*
+  GCC shows warning: "perhaps the 'offsetof' macro was used incorrectly"
+    GCC 3.4.4 : classes with constructor
+    GCC 4.8.1 : classes with non-public variable members"
+*/
+
+#define Z7_container_of(ptr, type, m) \
+  ((type *)(void *)((char *)(void *) \
+  (1 ? (ptr) : &((type *)NULL)->m) - MY_offsetof(type, m)))
+
+#define Z7_container_of_CONST(ptr, type, m) \
+  ((const type *)(const void *)((const char *)(const void *) \
+  (1 ? (ptr) : &((type *)NULL)->m) - MY_offsetof(type, m)))
+
+/*
+#define Z7_container_of_NON_CONST_FROM_CONST(ptr, type, m) \
+  ((type *)(void *)(const void *)((const char *)(const void *) \
+  (1 ? (ptr) : &((type *)NULL)->m) - MY_offsetof(type, m)))
+*/
+
+#endif
+
+#define Z7_CONTAINER_FROM_VTBL_SIMPLE(ptr, type, m) ((type *)(void *)(ptr))
+
+// #define Z7_CONTAINER_FROM_VTBL(ptr, type, m) Z7_CONTAINER_FROM_VTBL_SIMPLE(ptr, type, m)
+#define Z7_CONTAINER_FROM_VTBL(ptr, type, m) Z7_container_of(ptr, type, m)
+// #define Z7_CONTAINER_FROM_VTBL(ptr, type, m) Z7_container_of_NON_CONST_FROM_CONST(ptr, type, m)
+
+#define Z7_CONTAINER_FROM_VTBL_CONST(ptr, type, m) Z7_container_of_CONST(ptr, type, m)
+
+#define Z7_CONTAINER_FROM_VTBL_CLS(ptr, type, m) Z7_CONTAINER_FROM_VTBL_SIMPLE(ptr, type, m)
+/*
+#define Z7_CONTAINER_FROM_VTBL_CLS(ptr, type, m) Z7_CONTAINER_FROM_VTBL(ptr, type, m)
+*/
+#if defined (__clang__) || defined(__GNUC__)
+#define Z7_DIAGNOSCTIC_IGNORE_BEGIN_CAST_QUAL \
+  _Pragma("GCC diagnostic push") \
+  _Pragma("GCC diagnostic ignored \"-Wcast-qual\"")
+#define Z7_DIAGNOSCTIC_IGNORE_END_CAST_QUAL \
+  _Pragma("GCC diagnostic pop")
+#else
+#define Z7_DIAGNOSCTIC_IGNORE_BEGIN_CAST_QUAL
+#define Z7_DIAGNOSCTIC_IGNORE_END_CAST_QUAL
+#endif
+
+#define Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR(ptr, type, m, p) \
+  Z7_DIAGNOSCTIC_IGNORE_BEGIN_CAST_QUAL \
+  type *p = Z7_CONTAINER_FROM_VTBL(ptr, type, m); \
+  Z7_DIAGNOSCTIC_IGNORE_END_CAST_QUAL
+
+#define Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(type) \
+  Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR(pp, type, vt, p)
+
+
+// #define ZIP7_DECLARE_HANDLE(name)  typedef void *name;
+#define Z7_DECLARE_HANDLE(name)  struct name##_dummy{int unused;}; typedef struct name##_dummy *name;
+
+
+#define Z7_memset_0_ARRAY(a)  memset((a), 0, sizeof(a))
+
+#ifndef Z7_ARRAY_SIZE
+#define Z7_ARRAY_SIZE(a) (sizeof(a) / sizeof((a)[0]))
+#endif
+
+
+#ifdef _WIN32
+
+#define CHAR_PATH_SEPARATOR '\\'
+#define WCHAR_PATH_SEPARATOR L'\\'
+#define STRING_PATH_SEPARATOR "\\"
+#define WSTRING_PATH_SEPARATOR L"\\"
+
+#else
+
+#define CHAR_PATH_SEPARATOR '/'
+#define WCHAR_PATH_SEPARATOR L'/'
+#define STRING_PATH_SEPARATOR "/"
+#define WSTRING_PATH_SEPARATOR L"/"
+
+#endif
+
+#define k_PropVar_TimePrec_0        0
+#define k_PropVar_TimePrec_Unix     1
+#define k_PropVar_TimePrec_DOS      2
+#define k_PropVar_TimePrec_HighPrec 3
+#define k_PropVar_TimePrec_Base     16
+#define k_PropVar_TimePrec_100ns (k_PropVar_TimePrec_Base + 7)
+#define k_PropVar_TimePrec_1ns   (k_PropVar_TimePrec_Base + 9)
+
+EXTERN_C_END
+
+#endif
+
+/*
+#ifndef Z7_ST
+#ifdef _7ZIP_ST
+#define Z7_ST
+#endif
+#endif
+*/

C/7zVersion.h

@@ -1,7 +1,42 @@
-#define MY_VER_MAJOR 9
-#define MY_VER_MINOR 11
+#define MY_VER_MAJOR 23
+#define MY_VER_MINOR 01
 #define MY_VER_BUILD 0
-#define MY_VERSION "9.11 beta"
-#define MY_DATE "2010-03-15"
-#define MY_COPYRIGHT ": Igor Pavlov : Public domain"
-#define MY_VERSION_COPYRIGHT_DATE MY_VERSION " " MY_COPYRIGHT " : " MY_DATE
+#define MY_VERSION_NUMBERS "23.01"
+#define MY_VERSION MY_VERSION_NUMBERS
+
+#ifdef MY_CPU_NAME
+  #define MY_VERSION_CPU MY_VERSION " (" MY_CPU_NAME ")"
+#else
+  #define MY_VERSION_CPU MY_VERSION
+#endif
+
+#define MY_DATE "2023-06-20"
+#undef MY_COPYRIGHT
+#undef MY_VERSION_COPYRIGHT_DATE
+#define MY_AUTHOR_NAME "Igor Pavlov"
+#define MY_COPYRIGHT_PD "Igor Pavlov : Public domain"
+#define MY_COPYRIGHT_CR "Copyright (c) 1999-2023 Igor Pavlov"
+
+#ifdef USE_COPYRIGHT_CR
+  #define MY_COPYRIGHT MY_COPYRIGHT_CR
+#else
+  #define MY_COPYRIGHT MY_COPYRIGHT_PD
+#endif
+
+#define MY_COPYRIGHT_DATE MY_COPYRIGHT " : " MY_DATE
+#define MY_VERSION_COPYRIGHT_DATE MY_VERSION_CPU " : " MY_COPYRIGHT " : " MY_DATE
+
+#define MY_EASY7ZIP_VER_MAJOR 0
+#define MY_EASY7ZIP_VER_MINOR 1
+
+#define MY_EASY7ZIP_7ZIP "Easy 7-Zip"
+
+#define MY_EASY7ZIP_VERSION "0.1.6-shunf4-2"
+
+#define MY_EASY7ZIP_7ZIP_VERSION "Easy 7-Zip v0.1.6-shunf4-2"
+
+#define MY_EASY7ZIP_COPYRIGHT "Portions Copyright (C) 2013-2016 James Hoo"
+
+#define MY_EASY7ZIP_AUTHOR "James Hoo"
+#define MY_EASY7ZIP_HOMEPAGE "e7z.org"
+#define MY_EASY7ZIP_SPECIAL_BUILD MY_EASY7ZIP_7ZIP_VERSION " (www." MY_EASY7ZIP_HOMEPAGE ") made by " MY_EASY7ZIP_AUTHOR

C/7zVersion.rc

@@ -0,0 +1,55 @@
+#define MY_VS_FFI_FILEFLAGSMASK  0x0000003FL
+#define MY_VOS_NT_WINDOWS32  0x00040004L
+#define MY_VOS_CE_WINDOWS32  0x00050004L
+
+#define MY_VFT_APP  0x00000001L
+#define MY_VFT_DLL  0x00000002L
+
+// #include <WinVer.h>
+
+#ifndef MY_VERSION
+#include "7zVersion.h"
+#endif
+
+#define MY_VER MY_VER_MAJOR,MY_VER_MINOR,MY_VER_BUILD,0
+
+#ifdef DEBUG
+#define DBG_FL VS_FF_DEBUG
+#else
+#define DBG_FL 0
+#endif
+
+#define MY_VERSION_INFO(fileType, descr, intName, origName)  \
+LANGUAGE 9, 1 \
+1 VERSIONINFO \
+  FILEVERSION MY_VER \
+  PRODUCTVERSION MY_VER \
+  FILEFLAGSMASK MY_VS_FFI_FILEFLAGSMASK \
+  FILEFLAGS DBG_FL \
+  FILEOS MY_VOS_NT_WINDOWS32 \
+  FILETYPE fileType \
+  FILESUBTYPE 0x0L \
+BEGIN \
+    BLOCK "StringFileInfo" \
+    BEGIN  \
+        BLOCK "040904b0" \
+        BEGIN \
+            VALUE "CompanyName", "Igor Pavlov" \
+            VALUE "FileDescription", descr \
+            VALUE "FileVersion", MY_VERSION  \
+            VALUE "InternalName", intName \
+            VALUE "LegalCopyright", MY_COPYRIGHT \
+            VALUE "OriginalFilename", origName \
+            VALUE "ProductName", "7-Zip" \
+            VALUE "ProductVersion", MY_VERSION \
+        END \
+    END \
+    BLOCK "VarFileInfo" \
+    BEGIN \
+        VALUE "Translation", 0x409, 1200 \
+    END \
+END
+
+#define MY_VERSION_INFO_APP(descr, intName) MY_VERSION_INFO(MY_VFT_APP, descr, intName, intName ".exe")
+
+#define MY_VERSION_INFO_DLL(descr, intName) MY_VERSION_INFO(MY_VFT_DLL, descr, intName, intName ".dll")

C/7zWindows.h

@@ -0,0 +1,103 @@
+/* 7zWindows.h -- StdAfx
+2023-04-02 : Igor Pavlov : Public domain */
+
+#ifndef ZIP7_INC_7Z_WINDOWS_H
+#define ZIP7_INC_7Z_WINDOWS_H
+
+#ifdef _WIN32
+
+#if defined(__clang__)
+# pragma clang diagnostic push
+#endif
+
+#pragma warning(disable : 4255)
+
+#if defined(_MSC_VER)
+
+#pragma warning(push)
+#pragma warning(disable : 4668) // '_WIN32_WINNT' is not defined as a preprocessor macro, replacing with '0' for '#if/#elif'
+
+#if _MSC_VER == 1900
+// for old kit10 versions
+#pragma warning(disable : 4255) // winuser.h(13979): warning C4255: 'GetThreadDpiAwarenessContext':
+#endif
+// win10 Windows Kit:
+#endif // _MSC_VER
+
+#if defined(_MSC_VER) && _MSC_VER <= 1200 && !defined(_WIN64)
+// for msvc6 without sdk2003
+#define RPC_NO_WINDOWS_H
+#endif
+
+#if defined(__MINGW32__) || defined(__MINGW64__)
+// #if defined(__GNUC__) && !defined(__clang__)
+#include <windows.h>
+#else
+#include <Windows.h>
+#endif
+// #include <basetsd.h>
+// #include <wtypes.h>
+
+// but if precompiled with clang-cl then we need
+// #include <windows.h>
+#if defined(_MSC_VER)
+#pragma warning(pop)
+#endif
+
+#if defined(__clang__)
+# pragma clang diagnostic pop
+#endif
+
+#if defined(_MSC_VER) && _MSC_VER <= 1200 && !defined(_WIN64)
+#ifndef _W64
+
+typedef long LONG_PTR, *PLONG_PTR;
+typedef unsigned long ULONG_PTR, *PULONG_PTR;
+typedef ULONG_PTR DWORD_PTR, *PDWORD_PTR;
+
+#define Z7_OLD_WIN_SDK
+#endif // _W64
+#endif // _MSC_VER == 1200
+
+#ifdef Z7_OLD_WIN_SDK
+
+#ifndef INVALID_FILE_ATTRIBUTES
+#define INVALID_FILE_ATTRIBUTES ((DWORD)-1)
+#endif
+#ifndef INVALID_SET_FILE_POINTER
+#define INVALID_SET_FILE_POINTER ((DWORD)-1)
+#endif
+#ifndef FILE_SPECIAL_ACCESS
+#define FILE_SPECIAL_ACCESS    (FILE_ANY_ACCESS)
+#endif
+
+// ShlObj.h:
+// #define BIF_NEWDIALOGSTYLE     0x0040
+
+#pragma warning(disable : 4201)
+// #pragma warning(disable : 4115)
+
+#undef  VARIANT_TRUE
+#define VARIANT_TRUE ((VARIANT_BOOL)-1)
+#endif
+
+#endif // Z7_OLD_WIN_SDK
+
+#ifdef UNDER_CE
+#undef  VARIANT_TRUE
+#define VARIANT_TRUE ((VARIANT_BOOL)-1)
+#endif
+
+
+#if defined(_MSC_VER)
+#if _MSC_VER >= 1400 && _MSC_VER <= 1600
+  // BaseTsd.h(148) : 'HandleToULong' : unreferenced inline function has been removed
+  // string.h
+  // #pragma warning(disable : 4514)
+#endif
+#endif
+
+
+/* #include "7zTypes.h" */
+
+#endif

C/7zip_gcc_c.mak

@@ -0,0 +1,360 @@
+
+MY_ARCH_2 = $(MY_ARCH)
+
+MY_ASM = jwasm
+MY_ASM = asmc
+
+ifndef RC
+#RC=windres.exe --target=pe-x86-64
+#RC=windres.exe -F pe-i386
+RC=windres.exe
+endif
+
+PROGPATH = $(O)/$(PROG)
+PROGPATH_STATIC = $(O)/$(PROG)s
+
+ifneq ($(CC), xlc)
+CFLAGS_WARN_WALL = -Wall -Werror -Wextra
+endif
+
+# for object file
+CFLAGS_BASE_LIST = -c
+# for ASM file
+# CFLAGS_BASE_LIST = -S
+
+FLAGS_FLTO =
+FLAGS_FLTO = -flto
+
+CFLAGS_BASE = $(MY_ARCH_2) -O2 $(CFLAGS_BASE_LIST) $(CFLAGS_WARN_WALL) $(CFLAGS_WARN) \
+ -DNDEBUG -D_REENTRANT -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE
+
+
+ifdef SystemDrive
+IS_MINGW = 1
+else
+ifdef SYSTEMDRIVE
+# ifdef OS
+IS_MINGW = 1
+endif
+endif
+
+ifdef IS_MINGW
+LDFLAGS_STATIC_2 = -static
+else
+ifndef DEF_FILE
+ifndef IS_NOT_STANDALONE
+ifndef MY_DYNAMIC_LINK
+ifneq ($(CC), clang)
+LDFLAGS_STATIC_2 =
+# -static
+# -static-libstdc++ -static-libgcc
+endif
+endif
+endif
+endif
+endif
+
+LDFLAGS_STATIC = -DNDEBUG $(LDFLAGS_STATIC_2)
+
+ifdef DEF_FILE
+
+
+ifdef IS_MINGW
+SHARED_EXT=.dll
+LDFLAGS = -shared -DEF $(DEF_FILE) $(LDFLAGS_STATIC)
+else
+SHARED_EXT=.so
+LDFLAGS = -shared -fPIC  $(LDFLAGS_STATIC)
+CC_SHARED=-fPIC
+endif
+
+
+else
+
+LDFLAGS = $(LDFLAGS_STATIC)
+# -s is not required for clang, do we need it for GGC ???
+# -s
+
+#-static -static-libgcc -static-libstdc++
+
+ifdef IS_MINGW
+SHARED_EXT=.exe
+else
+SHARED_EXT=
+endif
+
+endif
+
+
+PROGPATH = $(O)/$(PROG)$(SHARED_EXT)
+PROGPATH_STATIC = $(O)/$(PROG)s$(SHARED_EXT)
+	
+ifndef O
+O=_o
+endif
+
+ifdef IS_MINGW
+
+ifdef MSYSTEM
+RM = rm -f
+MY_MKDIR=mkdir -p
+DEL_OBJ_EXE = -$(RM) $(PROGPATH) $(PROGPATH_STATIC) $(OBJS)
+else
+RM = del
+MY_MKDIR=mkdir
+DEL_OBJ_EXE = -$(RM) $(O)\*.o $(O)\$(PROG).exe $(O)\$(PROG).dll
+endif
+
+
+LIB2 = -lOle32 -loleaut32 -luuid -ladvapi32 -lUser32 -lShell32
+
+CFLAGS_EXTRA = -DUNICODE -D_UNICODE
+# -Wno-delete-non-virtual-dtor
+
+ 
+else
+
+RM = rm -f
+MY_MKDIR=mkdir -p
+# CFLAGS_BASE := $(CFLAGS_BASE) -DZ7_ST
+# CFLAGS_EXTRA = -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE
+
+# LOCAL_LIBS=-lpthread
+# LOCAL_LIBS_DLL=$(LOCAL_LIBS) -ldl
+LIB2 = -lpthread -ldl
+
+DEL_OBJ_EXE = -$(RM) $(PROGPATH) $(PROGPATH_STATIC) $(OBJS)
+
+endif
+
+
+ifdef IS_X64
+AFLAGS_ABI = -elf64 -DABI_LINUX
+else
+AFLAGS_ABI = -elf -DABI_LINUX -DABI_CDECL
+# -DABI_CDECL
+# -DABI_LINUX
+# -DABI_CDECL
+endif
+AFLAGS = $(AFLAGS_ABI) -Fo$(O)/
+
+C_WARN_FLAGS =
+
+CFLAGS = $(LOCAL_FLAGS) $(CFLAGS_BASE2) $(CFLAGS_BASE) $(CFLAGS_EXTRA) $(C_WARN_FLAGS) $(FLAGS_FLTO) $(CC_SHARED) -o $@
+
+STATIC_TARGET=
+ifdef COMPL_STATIC
+STATIC_TARGET=$(PROGPATH_STATIC)
+endif
+
+
+all: $(O) $(PROGPATH) $(STATIC_TARGET)
+
+$(O):
+	$(MY_MKDIR) $(O)
+
+ifneq ($(CC), $(CROSS_COMPILE)clang)
+LFLAGS_STRIP = -s
+endif
+
+LFLAGS_ALL = $(LFLAGS_STRIP) $(MY_ARCH_2) $(LDFLAGS) $(FLAGS_FLTO) $(LD_arch) $(OBJS) $(MY_LIBS) $(LIB2)
+$(PROGPATH): $(OBJS)
+	$(CC) -o $(PROGPATH) $(LFLAGS_ALL)
+
+$(PROGPATH_STATIC): $(OBJS)
+	$(CC) -static -o $(PROGPATH_STATIC) $(LFLAGS_ALL)
+
+
+ifndef NO_DEFAULT_RES
+# old mingw without -FO
+# windres.exe $(RFLAGS) resource.rc $O/resource.o
+$O/resource.o: resource.rc
+	$(RC) $(RFLAGS) resource.rc $(O)/resource.o
+endif
+# windres.exe $(RFLAGS) resource.rc $(O)\resource.o
+# windres.exe $(RFLAGS) resource.rc -FO $(O)/resource.o
+# $(RC) $(RFLAGS) resource.rc -FO $(O)/resource.o
+
+
+
+$O/7zAlloc.o: ../../../C/7zAlloc.c
+	$(CC) $(CFLAGS) $<
+$O/7zArcIn.o: ../../../C/7zArcIn.c
+	$(CC) $(CFLAGS) $<
+$O/7zBuf.o: ../../../C/7zBuf.c
+	$(CC) $(CFLAGS) $<
+$O/7zBuf2.o: ../../../C/7zBuf2.c
+	$(CC) $(CFLAGS) $<
+$O/7zCrc.o: ../../../C/7zCrc.c
+	$(CC) $(CFLAGS) $<
+$O/7zDec.o: ../../../C/7zDec.c
+	$(CC) $(CFLAGS) $<
+$O/7zFile.o: ../../../C/7zFile.c
+	$(CC) $(CFLAGS) $<
+$O/7zStream.o: ../../../C/7zStream.c
+	$(CC) $(CFLAGS) $<
+$O/Aes.o: ../../../C/Aes.c
+	$(CC) $(CFLAGS) $<
+$O/Alloc.o: ../../../C/Alloc.c
+	$(CC) $(CFLAGS) $<
+$O/Bcj2.o: ../../../C/Bcj2.c
+	$(CC) $(CFLAGS) $<
+$O/Bcj2Enc.o: ../../../C/Bcj2Enc.c
+	$(CC) $(CFLAGS) $<
+$O/Blake2s.o: ../../../C/Blake2s.c
+	$(CC) $(CFLAGS) $<
+$O/Bra.o: ../../../C/Bra.c
+	$(CC) $(CFLAGS) $<
+$O/Bra86.o: ../../../C/Bra86.c
+	$(CC) $(CFLAGS) $<
+$O/BraIA64.o: ../../../C/BraIA64.c
+	$(CC) $(CFLAGS) $<
+$O/BwtSort.o: ../../../C/BwtSort.c
+	$(CC) $(CFLAGS) $<
+
+$O/CpuArch.o: ../../../C/CpuArch.c
+	$(CC) $(CFLAGS) $<
+$O/Delta.o: ../../../C/Delta.c
+	$(CC) $(CFLAGS) $<
+$O/DllSecur.o: ../../../C/DllSecur.c
+	$(CC) $(CFLAGS) $<
+$O/HuffEnc.o: ../../../C/HuffEnc.c
+	$(CC) $(CFLAGS) $<
+$O/LzFind.o: ../../../C/LzFind.c
+	$(CC) $(CFLAGS) $<
+
+# ifdef MT_FILES
+$O/LzFindMt.o: ../../../C/LzFindMt.c
+	$(CC) $(CFLAGS) $<
+$O/LzFindOpt.o: ../../../C/LzFindOpt.c
+	$(CC) $(CFLAGS) $<
+
+$O/Threads.o: ../../../C/Threads.c
+	$(CC) $(CFLAGS) $<
+# endif
+
+$O/LzmaEnc.o: ../../../C/LzmaEnc.c
+	$(CC) $(CFLAGS) $<
+$O/Lzma86Dec.o: ../../../C/Lzma86Dec.c
+	$(CC) $(CFLAGS) $<
+$O/Lzma86Enc.o: ../../../C/Lzma86Enc.c
+	$(CC) $(CFLAGS) $<
+$O/Lzma2Dec.o: ../../../C/Lzma2Dec.c
+	$(CC) $(CFLAGS) $<
+$O/Lzma2DecMt.o: ../../../C/Lzma2DecMt.c
+	$(CC) $(CFLAGS) $<
+$O/Lzma2Enc.o: ../../../C/Lzma2Enc.c
+	$(CC) $(CFLAGS) $<
+$O/LzmaLib.o: ../../../C/LzmaLib.c
+	$(CC) $(CFLAGS) $<
+$O/MtCoder.o: ../../../C/MtCoder.c
+	$(CC) $(CFLAGS) $<
+$O/MtDec.o: ../../../C/MtDec.c
+	$(CC) $(CFLAGS) $<
+$O/Ppmd7.o: ../../../C/Ppmd7.c
+	$(CC) $(CFLAGS) $<
+$O/Ppmd7aDec.o: ../../../C/Ppmd7aDec.c
+	$(CC) $(CFLAGS) $<
+$O/Ppmd7Dec.o: ../../../C/Ppmd7Dec.c
+	$(CC) $(CFLAGS) $<
+$O/Ppmd7Enc.o: ../../../C/Ppmd7Enc.c
+	$(CC) $(CFLAGS) $<
+$O/Ppmd8.o: ../../../C/Ppmd8.c
+	$(CC) $(CFLAGS) $<
+$O/Ppmd8Dec.o: ../../../C/Ppmd8Dec.c
+	$(CC) $(CFLAGS) $<
+$O/Ppmd8Enc.o: ../../../C/Ppmd8Enc.c
+	$(CC) $(CFLAGS) $<
+$O/Sha1.o: ../../../C/Sha1.c
+	$(CC) $(CFLAGS) $<
+$O/Sha256.o: ../../../C/Sha256.c
+	$(CC) $(CFLAGS) $<
+$O/Sort.o: ../../../C/Sort.c
+	$(CC) $(CFLAGS) $<
+$O/SwapBytes.o: ../../../C/SwapBytes.c
+	$(CC) $(CFLAGS) $<
+$O/Xz.o: ../../../C/Xz.c
+	$(CC) $(CFLAGS) $<
+$O/XzCrc64.o: ../../../C/XzCrc64.c
+	$(CC) $(CFLAGS) $<
+$O/XzDec.o: ../../../C/XzDec.c
+	$(CC) $(CFLAGS) $<
+$O/XzEnc.o: ../../../C/XzEnc.c
+	$(CC) $(CFLAGS) $<
+$O/XzIn.o: ../../../C/XzIn.c
+	$(CC) $(CFLAGS) $<
+
+
+ifdef USE_ASM
+ifdef IS_X64
+USE_X86_ASM=1
+else
+ifdef IS_X86
+USE_X86_ASM=1
+endif
+endif
+endif
+
+ifdef USE_X86_ASM
+$O/7zCrcOpt.o: ../../../Asm/x86/7zCrcOpt.asm
+	$(MY_ASM) $(AFLAGS) $<
+$O/XzCrc64Opt.o: ../../../Asm/x86/XzCrc64Opt.asm
+	$(MY_ASM) $(AFLAGS) $<
+$O/AesOpt.o: ../../../Asm/x86/AesOpt.asm
+	$(MY_ASM) $(AFLAGS) $<
+$O/Sha1Opt.o: ../../../Asm/x86/Sha1Opt.asm
+	$(MY_ASM) $(AFLAGS) $<
+$O/Sha256Opt.o: ../../../Asm/x86/Sha256Opt.asm
+	$(MY_ASM) $(AFLAGS) $<
+else
+$O/7zCrcOpt.o: ../../7zCrcOpt.c
+	$(CC) $(CFLAGS) $<
+$O/XzCrc64Opt.o: ../../XzCrc64Opt.c
+	$(CC) $(CFLAGS) $<
+$O/Sha1Opt.o: ../../Sha1Opt.c
+	$(CC) $(CFLAGS) $<
+$O/Sha256Opt.o: ../../Sha256Opt.c
+	$(CC) $(CFLAGS) $<
+$O/AesOpt.o: ../../AesOpt.c
+	$(CC) $(CFLAGS) $<
+endif
+
+
+ifdef USE_LZMA_DEC_ASM
+
+ifdef IS_X64
+$O/LzmaDecOpt.o: ../../../Asm/x86/LzmaDecOpt.asm
+	$(MY_ASM) $(AFLAGS) $<
+endif
+
+ifdef IS_ARM64
+$O/LzmaDecOpt.o: ../../../Asm/arm64/LzmaDecOpt.S ../../../Asm/arm64/7zAsm.S
+	$(CC) $(CFLAGS) $<
+endif
+
+$O/LzmaDec.o: ../../LzmaDec.c
+	$(CC) $(CFLAGS) -DZ7_LZMA_DEC_OPT $<
+
+else
+
+$O/LzmaDec.o: ../../LzmaDec.c
+	$(CC) $(CFLAGS) $<
+
+endif
+
+
+
+$O/7zMain.o: ../../../C/Util/7z/7zMain.c
+	$(CC) $(CFLAGS) $<
+$O/7zipInstall.o: ../../../C/Util/7zipInstall/7zipInstall.c
+	$(CC) $(CFLAGS) $<
+$O/7zipUninstall.o: ../../../C/Util/7zipUninstall/7zipUninstall.c
+	$(CC) $(CFLAGS) $<
+$O/LzmaUtil.o: ../../../C/Util/Lzma/LzmaUtil.c
+	$(CC) $(CFLAGS) $<
+$O/XzUtil.o: ../../../C/Util/Xz/XzUtil.c
+	$(CC) $(CFLAGS) $<
+
+
+clean:
+	-$(DEL_OBJ_EXE)

C/Aes.c

@@ -1,11 +1,20 @@
 /* Aes.c -- AES encryption / decryption
-2009-11-23 : Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
-#include "Aes.h"
 #include "CpuArch.h"
+#include "Aes.h"
+
+AES_CODE_FUNC g_AesCbc_Decode;
+#ifndef Z7_SFX
+AES_CODE_FUNC g_AesCbc_Encode;
+AES_CODE_FUNC g_AesCtr_Code;
+UInt32 g_Aes_SupportedFunctions_Flags;
+#endif
 
 static UInt32 T[256 * 4];
-static Byte Sbox[256] = {
+static const Byte Sbox[256] = {
   0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
   0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
   0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15,
@@ -23,23 +32,10 @@ static Byte Sbox[256] = {
   0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94, 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf,
   0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16};
 
-void MY_FAST_CALL AesCbc_Encode(UInt32 *ivAes, Byte *data, size_t numBlocks);
-void MY_FAST_CALL AesCbc_Decode(UInt32 *ivAes, Byte *data, size_t numBlocks);
-void MY_FAST_CALL AesCtr_Code(UInt32 *ivAes, Byte *data, size_t numBlocks);
-
-void MY_FAST_CALL AesCbc_Encode_Intel(UInt32 *ivAes, Byte *data, size_t numBlocks);
-void MY_FAST_CALL AesCbc_Decode_Intel(UInt32 *ivAes, Byte *data, size_t numBlocks);
-void MY_FAST_CALL AesCtr_Code_Intel(UInt32 *ivAes, Byte *data, size_t numBlocks);
-
-AES_CODE_FUNC g_AesCbc_Encode;
-AES_CODE_FUNC g_AesCbc_Decode;
-AES_CODE_FUNC g_AesCtr_Code;
 
 static UInt32 D[256 * 4];
 static Byte InvS[256];
 
-static Byte Rcon[11] = { 0x00, 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1b, 0x36 };
-
 #define xtime(x) ((((x) << 1) ^ (((x) & 0x80) != 0 ? 0x1B : 0)) & 0xFF)
 
 #define Ui32(a0, a1, a2, a3) ((UInt32)(a0) | ((UInt32)(a1) << 8) | ((UInt32)(a2) << 16) | ((UInt32)(a3) << 24))
@@ -47,88 +43,164 @@ static Byte Rcon[11] = { 0x00, 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0
 #define gb0(x) ( (x)          & 0xFF)
 #define gb1(x) (((x) >> ( 8)) & 0xFF)
 #define gb2(x) (((x) >> (16)) & 0xFF)
-#define gb3(x) (((x) >> (24)) & 0xFF)
+#define gb3(x) (((x) >> (24)))
+
+#define gb(n, x) gb ## n(x)
+
+#define TT(x) (T + (x << 8))
+#define DD(x) (D + (x << 8))
+
+
+// #define Z7_SHOW_AES_STATUS
+
+#ifdef MY_CPU_X86_OR_AMD64
+  #define USE_HW_AES
+#elif defined(MY_CPU_ARM_OR_ARM64) && defined(MY_CPU_LE)
+  #if defined(__clang__)
+    #if (__clang_major__ >= 8) // fix that check
+      #define USE_HW_AES
+    #endif
+  #elif defined(__GNUC__)
+    #if (__GNUC__ >= 6) // fix that check
+      #define USE_HW_AES
+    #endif
+  #elif defined(_MSC_VER)
+    #if _MSC_VER >= 1910
+      #define USE_HW_AES
+    #endif
+  #endif
+#endif
+
+#ifdef USE_HW_AES
+#ifdef Z7_SHOW_AES_STATUS
+#include <stdio.h>
+#define PRF(x) x
+#else
+#define PRF(x)
+#endif
+#endif
+
 
 void AesGenTables(void)
 {
   unsigned i;
   for (i = 0; i < 256; i++)
     InvS[Sbox[i]] = (Byte)i;
+  
   for (i = 0; i < 256; i++)
   {
     {
-      UInt32 a1 = Sbox[i];
-      UInt32 a2 = xtime(a1);
-      UInt32 a3 = a2 ^ a1;
-      T[        i] = Ui32(a2, a1, a1, a3);
-      T[0x100 + i] = Ui32(a3, a2, a1, a1);
-      T[0x200 + i] = Ui32(a1, a3, a2, a1);
-      T[0x300 + i] = Ui32(a1, a1, a3, a2);
+      const UInt32 a1 = Sbox[i];
+      const UInt32 a2 = xtime(a1);
+      const UInt32 a3 = a2 ^ a1;
+      TT(0)[i] = Ui32(a2, a1, a1, a3);
+      TT(1)[i] = Ui32(a3, a2, a1, a1);
+      TT(2)[i] = Ui32(a1, a3, a2, a1);
+      TT(3)[i] = Ui32(a1, a1, a3, a2);
     }
     {
-      UInt32 a1 = InvS[i];
-      UInt32 a2 = xtime(a1);
-      UInt32 a4 = xtime(a2);
-      UInt32 a8 = xtime(a4);
-      UInt32 a9 = a8 ^ a1;
-      UInt32 aB = a8 ^ a2 ^ a1;
-      UInt32 aD = a8 ^ a4 ^ a1;
-      UInt32 aE = a8 ^ a4 ^ a2;
-      D[        i] = Ui32(aE, a9, aD, aB);
-      D[0x100 + i] = Ui32(aB, aE, a9, aD);
-      D[0x200 + i] = Ui32(aD, aB, aE, a9);
-      D[0x300 + i] = Ui32(a9, aD, aB, aE);
+      const UInt32 a1 = InvS[i];
+      const UInt32 a2 = xtime(a1);
+      const UInt32 a4 = xtime(a2);
+      const UInt32 a8 = xtime(a4);
+      const UInt32 a9 = a8 ^ a1;
+      const UInt32 aB = a8 ^ a2 ^ a1;
+      const UInt32 aD = a8 ^ a4 ^ a1;
+      const UInt32 aE = a8 ^ a4 ^ a2;
+      DD(0)[i] = Ui32(aE, a9, aD, aB);
+      DD(1)[i] = Ui32(aB, aE, a9, aD);
+      DD(2)[i] = Ui32(aD, aB, aE, a9);
+      DD(3)[i] = Ui32(a9, aD, aB, aE);
     }
   }
-  g_AesCbc_Encode = AesCbc_Encode;
-  g_AesCbc_Decode = AesCbc_Decode;
-  g_AesCtr_Code = AesCtr_Code;
-  #ifdef MY_CPU_X86_OR_AMD64
-  if (CPU_Is_Aes_Supported())
+  
   {
-    g_AesCbc_Encode = AesCbc_Encode_Intel;
-    g_AesCbc_Decode = AesCbc_Decode_Intel;
-    g_AesCtr_Code = AesCtr_Code_Intel;
+  AES_CODE_FUNC d = AesCbc_Decode;
+  #ifndef Z7_SFX
+  AES_CODE_FUNC e = AesCbc_Encode;
+  AES_CODE_FUNC c = AesCtr_Code;
+  UInt32 flags = 0;
+  #endif
+  
+  #ifdef USE_HW_AES
+  if (CPU_IsSupported_AES())
+  {
+    // #pragma message ("AES HW")
+    PRF(printf("\n===AES HW\n"));
+    d = AesCbc_Decode_HW;
+
+    #ifndef Z7_SFX
+    e = AesCbc_Encode_HW;
+    c = AesCtr_Code_HW;
+    flags = k_Aes_SupportedFunctions_HW;
+    #endif
+
+    #ifdef MY_CPU_X86_OR_AMD64
+    if (CPU_IsSupported_VAES_AVX2())
+    {
+      PRF(printf("\n===vaes avx2\n"));
+      d = AesCbc_Decode_HW_256;
+      #ifndef Z7_SFX
+      c = AesCtr_Code_HW_256;
+      flags |= k_Aes_SupportedFunctions_HW_256;
+      #endif
+    }
+    #endif
   }
   #endif
+
+  g_AesCbc_Decode = d;
+  #ifndef Z7_SFX
+  g_AesCbc_Encode = e;
+  g_AesCtr_Code = c;
+  g_Aes_SupportedFunctions_Flags = flags;
+  #endif
+  }
 }
 
-#define HT(i, x, s) (T + (x << 8))[gb ## x(s[(i + x) & 3])]
+
+#define HT(i, x, s) TT(x)[gb(x, s[(i + x) & 3])]
+
 #define HT4(m, i, s, p) m[i] = \
     HT(i, 0, s) ^ \
     HT(i, 1, s) ^ \
     HT(i, 2, s) ^ \
     HT(i, 3, s) ^ w[p + i]
-/* such order (2031) in HT16 is for VC6/K8 speed optimization) */
-#define HT16(m, s, p) \
-    HT4(m, 2, s, p); \
-    HT4(m, 0, s, p); \
-    HT4(m, 3, s, p); \
-    HT4(m, 1, s, p); \
 
-#define FT(i, x) Sbox[gb ## x(m[(i + x) & 3])]
+#define HT16(m, s, p) \
+    HT4(m, 0, s, p); \
+    HT4(m, 1, s, p); \
+    HT4(m, 2, s, p); \
+    HT4(m, 3, s, p); \
+
+#define FT(i, x) Sbox[gb(x, m[(i + x) & 3])]
 #define FT4(i) dest[i] = Ui32(FT(i, 0), FT(i, 1), FT(i, 2), FT(i, 3)) ^ w[i];
 
-#define HD(i, x, s) (D + (x << 8))[gb ## x(s[(i - x) & 3])]
+
+#define HD(i, x, s) DD(x)[gb(x, s[(i - x) & 3])]
+
 #define HD4(m, i, s, p) m[i] = \
     HD(i, 0, s) ^ \
     HD(i, 1, s) ^ \
     HD(i, 2, s) ^ \
     HD(i, 3, s) ^ w[p + i];
-/* such order (0231) in HD16 is for VC6/K8 speed optimization) */
+
 #define HD16(m, s, p) \
     HD4(m, 0, s, p); \
+    HD4(m, 1, s, p); \
     HD4(m, 2, s, p); \
     HD4(m, 3, s, p); \
-    HD4(m, 1, s, p); \
 
-#define FD(i, x) InvS[gb ## x(m[(i - x) & 3])]
+#define FD(i, x) InvS[gb(x, m[(i - x) & 3])]
 #define FD4(i) dest[i] = Ui32(FD(i, 0), FD(i, 1), FD(i, 2), FD(i, 3)) ^ w[i];
 
-void MY_FAST_CALL Aes_SetKey_Enc(UInt32 *w, const Byte *key, unsigned keySize)
+void Z7_FASTCALL Aes_SetKey_Enc(UInt32 *w, const Byte *key, unsigned keySize)
 {
-  unsigned i, wSize;
-  wSize = keySize + 28;
+  unsigned i, m;
+  const UInt32 *wLim;
+  UInt32 t;
+  UInt32 rcon = 1;
+  
   keySize /= 4;
   w[0] = ((UInt32)keySize / 2) + 3;
   w += 4;
@@ -136,19 +208,29 @@ void MY_FAST_CALL Aes_SetKey_Enc(UInt32 *w, const Byte *key, unsigned keySize)
   for (i = 0; i < keySize; i++, key += 4)
     w[i] = GetUi32(key);
 
-  for (; i < wSize; i++)
+  t = w[(size_t)keySize - 1];
+  wLim = w + (size_t)keySize * 3 + 28;
+  m = 0;
+  do
   {
-    UInt32 t = w[i - 1];
-    unsigned rem = i % keySize;
-    if (rem == 0)
-      t = Ui32(Sbox[gb1(t)] ^ Rcon[i / keySize], Sbox[gb2(t)], Sbox[gb3(t)], Sbox[gb0(t)]);
-    else if (keySize > 6 && rem == 4)
+    if (m == 0)
+    {
+      t = Ui32(Sbox[gb1(t)] ^ rcon, Sbox[gb2(t)], Sbox[gb3(t)], Sbox[gb0(t)]);
+      rcon <<= 1;
+      if (rcon & 0x100)
+        rcon = 0x1b;
+      m = keySize;
+    }
+    else if (m == 4 && keySize > 6)
       t = Ui32(Sbox[gb0(t)], Sbox[gb1(t)], Sbox[gb2(t)], Sbox[gb3(t)]);
-    w[i] = w[i - keySize] ^ t;
+    m--;
+    t ^= w[0];
+    w[keySize] = t;
   }
+  while (++w != wLim);
 }
 
-void MY_FAST_CALL Aes_SetKey_Dec(UInt32 *w, const Byte *key, unsigned keySize)
+void Z7_FASTCALL Aes_SetKey_Dec(UInt32 *w, const Byte *key, unsigned keySize)
 {
   unsigned i, num;
   Aes_SetKey_Enc(w, key, keySize);
@@ -158,17 +240,18 @@ void MY_FAST_CALL Aes_SetKey_Dec(UInt32 *w, const Byte *key, unsigned keySize)
   {
     UInt32 r = w[i];
     w[i] =
-      D[        Sbox[gb0(r)]] ^
-      D[0x100 + Sbox[gb1(r)]] ^
-      D[0x200 + Sbox[gb2(r)]] ^
-      D[0x300 + Sbox[gb3(r)]];
+      DD(0)[Sbox[gb0(r)]] ^
+      DD(1)[Sbox[gb1(r)]] ^
+      DD(2)[Sbox[gb2(r)]] ^
+      DD(3)[Sbox[gb3(r)]];
   }
 }
 
 /* Aes_Encode and Aes_Decode functions work with little-endian words.
   src and dest are pointers to 4 UInt32 words.
-  arc and dest can point to same block */
+  src and dest can point to same block */
 
+// Z7_FORCE_INLINE
 static void Aes_Encode(const UInt32 *w, UInt32 *dest, const UInt32 *src)
 {
   UInt32 s[4];
@@ -182,16 +265,20 @@ static void Aes_Encode(const UInt32 *w, UInt32 *dest, const UInt32 *src)
   w += 4;
   for (;;)
   {
-    HT16(m, s, 0);
+    HT16(m, s, 0)
     if (--numRounds2 == 0)
       break;
-    HT16(s, m, 4);
+    HT16(s, m, 4)
     w += 8;
   }
   w += 4;
-  FT4(0); FT4(1); FT4(2); FT4(3);
+  FT4(0)
+  FT4(1)
+  FT4(2)
+  FT4(3)
 }
 
+Z7_FORCE_INLINE
 static void Aes_Decode(const UInt32 *w, UInt32 *dest, const UInt32 *src)
 {
   UInt32 s[4];
@@ -205,12 +292,15 @@ static void Aes_Decode(const UInt32 *w, UInt32 *dest, const UInt32 *src)
   for (;;)
   {
     w -= 8;
-    HD16(m, s, 4);
+    HD16(m, s, 4)
     if (--numRounds2 == 0)
       break;
-    HD16(s, m, 0);
+    HD16(s, m, 0)
   }
-  FD4(0); FD4(1); FD4(2); FD4(3);
+  FD4(0)
+  FD4(1)
+  FD4(2)
+  FD4(3)
 }
 
 void AesCbc_Init(UInt32 *p, const Byte *iv)
@@ -220,7 +310,7 @@ void AesCbc_Init(UInt32 *p, const Byte *iv)
     p[i] = GetUi32(iv + i * 4);
 }
 
-void MY_FAST_CALL AesCbc_Encode(UInt32 *p, Byte *data, size_t numBlocks)
+void Z7_FASTCALL AesCbc_Encode(UInt32 *p, Byte *data, size_t numBlocks)
 {
   for (; numBlocks != 0; numBlocks--, data += AES_BLOCK_SIZE)
   {
@@ -231,14 +321,14 @@ void MY_FAST_CALL AesCbc_Encode(UInt32 *p, Byte *data, size_t numBlocks)
     
     Aes_Encode(p + 4, p, p);
     
-    SetUi32(data,      p[0]);
-    SetUi32(data + 4,  p[1]);
-    SetUi32(data + 8,  p[2]);
-    SetUi32(data + 12, p[3]);
+    SetUi32(data,      p[0])
+    SetUi32(data + 4,  p[1])
+    SetUi32(data + 8,  p[2])
+    SetUi32(data + 12, p[3])
   }
 }
 
-void MY_FAST_CALL AesCbc_Decode(UInt32 *p, Byte *data, size_t numBlocks)
+void Z7_FASTCALL AesCbc_Decode(UInt32 *p, Byte *data, size_t numBlocks)
 {
   UInt32 in[4], out[4];
   for (; numBlocks != 0; numBlocks--, data += AES_BLOCK_SIZE)
@@ -250,10 +340,10 @@ void MY_FAST_CALL AesCbc_Decode(UInt32 *p, Byte *data, size_t numBlocks)
 
     Aes_Decode(p + 4, out, in);
 
-    SetUi32(data,      p[0] ^ out[0]);
-    SetUi32(data + 4,  p[1] ^ out[1]);
-    SetUi32(data + 8,  p[2] ^ out[2]);
-    SetUi32(data + 12, p[3] ^ out[3]);
+    SetUi32(data,      p[0] ^ out[0])
+    SetUi32(data + 4,  p[1] ^ out[1])
+    SetUi32(data + 8,  p[2] ^ out[2])
+    SetUi32(data + 12, p[3] ^ out[3])
     
     p[0] = in[0];
     p[1] = in[1];
@@ -262,21 +352,42 @@ void MY_FAST_CALL AesCbc_Decode(UInt32 *p, Byte *data, size_t numBlocks)
   }
 }
 
-void MY_FAST_CALL AesCtr_Code(UInt32 *p, Byte *data, size_t numBlocks)
+void Z7_FASTCALL AesCtr_Code(UInt32 *p, Byte *data, size_t numBlocks)
 {
   for (; numBlocks != 0; numBlocks--)
   {
     UInt32 temp[4];
-    Byte buf[16];
-    int i;
+    unsigned i;
+
     if (++p[0] == 0)
       p[1]++;
+    
     Aes_Encode(p + 4, temp, p);
-    SetUi32(buf,      temp[0]);
-    SetUi32(buf + 4,  temp[1]);
-    SetUi32(buf + 8,  temp[2]);
-    SetUi32(buf + 12, temp[3]);
-    for (i = 0; i < 16; i++)
-      *data++ ^= buf[i];
+    
+    for (i = 0; i < 4; i++, data += 4)
+    {
+      const UInt32 t = temp[i];
+
+      #ifdef MY_CPU_LE_UNALIGN
+        *((UInt32 *)(void *)data) ^= t;
+      #else
+        data[0] = (Byte)(data[0] ^ (t & 0xFF));
+        data[1] = (Byte)(data[1] ^ ((t >> 8) & 0xFF));
+        data[2] = (Byte)(data[2] ^ ((t >> 16) & 0xFF));
+        data[3] = (Byte)(data[3] ^ ((t >> 24)));
+      #endif
+    }
   }
 }
+
+#undef xtime
+#undef Ui32
+#undef gb0
+#undef gb1
+#undef gb2
+#undef gb3
+#undef gb
+#undef TT
+#undef DD
+#undef USE_HW_AES
+#undef PRF

C/Aes.h

@@ -1,10 +1,10 @@
 /* Aes.h -- AES encryption / decryption
-2009-11-23 : Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
 
-#ifndef __AES_H
-#define __AES_H
+#ifndef ZIP7_INC_AES_H
+#define ZIP7_INC_AES_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
 EXTERN_C_BEGIN
 
@@ -20,18 +20,40 @@ void AesGenTables(void);
 
 /* aes - 16-byte aligned pointer to keyMode+roundKeys sequence */
 /* keySize = 16 or 24 or 32 (bytes) */
-typedef void (MY_FAST_CALL *AES_SET_KEY_FUNC)(UInt32 *aes, const Byte *key, unsigned keySize);
-void MY_FAST_CALL Aes_SetKey_Enc(UInt32 *aes, const Byte *key, unsigned keySize);
-void MY_FAST_CALL Aes_SetKey_Dec(UInt32 *aes, const Byte *key, unsigned keySize);
+typedef void (Z7_FASTCALL *AES_SET_KEY_FUNC)(UInt32 *aes, const Byte *key, unsigned keySize);
+void Z7_FASTCALL Aes_SetKey_Enc(UInt32 *aes, const Byte *key, unsigned keySize);
+void Z7_FASTCALL Aes_SetKey_Dec(UInt32 *aes, const Byte *key, unsigned keySize);
 
 /* ivAes - 16-byte aligned pointer to iv+keyMode+roundKeys sequence: UInt32[AES_NUM_IVMRK_WORDS] */
 void AesCbc_Init(UInt32 *ivAes, const Byte *iv); /* iv size is AES_BLOCK_SIZE */
+
 /* data - 16-byte aligned pointer to data */
 /* numBlocks - the number of 16-byte blocks in data array */
-typedef void (MY_FAST_CALL *AES_CODE_FUNC)(UInt32 *ivAes, Byte *data, size_t numBlocks);
-extern AES_CODE_FUNC g_AesCbc_Encode;
+typedef void (Z7_FASTCALL *AES_CODE_FUNC)(UInt32 *ivAes, Byte *data, size_t numBlocks);
+
 extern AES_CODE_FUNC g_AesCbc_Decode;
+#ifndef Z7_SFX
+extern AES_CODE_FUNC g_AesCbc_Encode;
 extern AES_CODE_FUNC g_AesCtr_Code;
+#define k_Aes_SupportedFunctions_HW     (1 << 2)
+#define k_Aes_SupportedFunctions_HW_256 (1 << 3)
+extern UInt32 g_Aes_SupportedFunctions_Flags;
+#endif
+
+
+#define Z7_DECLARE_AES_CODE_FUNC(funcName) \
+    void Z7_FASTCALL funcName(UInt32 *ivAes, Byte *data, size_t numBlocks);
+
+Z7_DECLARE_AES_CODE_FUNC (AesCbc_Encode)
+Z7_DECLARE_AES_CODE_FUNC (AesCbc_Decode)
+Z7_DECLARE_AES_CODE_FUNC (AesCtr_Code)
+
+Z7_DECLARE_AES_CODE_FUNC (AesCbc_Encode_HW)
+Z7_DECLARE_AES_CODE_FUNC (AesCbc_Decode_HW)
+Z7_DECLARE_AES_CODE_FUNC (AesCtr_Code_HW)
+
+Z7_DECLARE_AES_CODE_FUNC (AesCbc_Decode_HW_256)
+Z7_DECLARE_AES_CODE_FUNC (AesCtr_Code_HW_256)
 
 EXTERN_C_END
 

C/AesOpt.c

@@ -1,182 +1,840 @@
-/* AesOpt.c -- Intel's AES
-2009-11-23 : Igor Pavlov : Public domain */
+/* AesOpt.c -- AES optimized code for x86 AES hardware instructions
+2023-04-02 : Igor Pavlov : Public domain */
 
+#include "Precomp.h"
+
+#include "Aes.h"
 #include "CpuArch.h"
 
 #ifdef MY_CPU_X86_OR_AMD64
-#if _MSC_VER >= 1500
-#define USE_INTEL_AES
+
+  #if defined(__INTEL_COMPILER)
+    #if (__INTEL_COMPILER >= 1110)
+      #define USE_INTEL_AES
+      #if (__INTEL_COMPILER >= 1900)
+        #define USE_INTEL_VAES
+      #endif
+    #endif
+  #elif defined(__clang__) && (__clang_major__ > 3 || __clang_major__ == 3 && __clang_minor__ >= 8) \
+       || defined(__GNUC__) && (__GNUC__ > 4 || __GNUC__ == 4 && __GNUC_MINOR__ >= 4)
+        #define USE_INTEL_AES
+        #if !defined(__AES__)
+          #define ATTRIB_AES __attribute__((__target__("aes")))
+        #endif
+      #if defined(__clang__) && (__clang_major__ >= 8) \
+          || defined(__GNUC__) && (__GNUC__ >= 8)
+        #define USE_INTEL_VAES
+        #if !defined(__AES__) || !defined(__VAES__) || !defined(__AVX__) || !defined(__AVX2__)
+          #define ATTRIB_VAES __attribute__((__target__("aes,vaes,avx,avx2")))
+        #endif
+      #endif
+  #elif defined(_MSC_VER)
+    #if (_MSC_VER > 1500) || (_MSC_FULL_VER >= 150030729)
+      #define USE_INTEL_AES
+      #if (_MSC_VER >= 1910)
+        #define USE_INTEL_VAES
+      #endif
+    #endif
+  #endif
+
+#ifndef ATTRIB_AES
+  #define ATTRIB_AES
 #endif
+#ifndef ATTRIB_VAES
+  #define ATTRIB_VAES
 #endif
 
+
 #ifdef USE_INTEL_AES
 
 #include <wmmintrin.h>
 
-void MY_FAST_CALL AesCbc_Encode_Intel(__m128i *p, __m128i *data, size_t numBlocks)
+#ifndef USE_INTEL_VAES
+#define AES_TYPE_keys UInt32
+#define AES_TYPE_data Byte
+// #define AES_TYPE_keys __m128i
+// #define AES_TYPE_data __m128i
+#endif
+
+#define AES_FUNC_START(name) \
+    void Z7_FASTCALL name(UInt32 *ivAes, Byte *data8, size_t numBlocks)
+    // void Z7_FASTCALL name(__m128i *p, __m128i *data, size_t numBlocks)
+
+#define AES_FUNC_START2(name) \
+AES_FUNC_START (name); \
+ATTRIB_AES \
+AES_FUNC_START (name)
+
+#define MM_OP(op, dest, src)  dest = op(dest, src);
+#define MM_OP_m(op, src)      MM_OP(op, m, src)
+
+#define MM_XOR( dest, src)    MM_OP(_mm_xor_si128,    dest, src)
+#define AVX_XOR(dest, src)    MM_OP(_mm256_xor_si256, dest, src)
+
+
+AES_FUNC_START2 (AesCbc_Encode_HW)
 {
+  __m128i *p = (__m128i *)(void *)ivAes;
+  __m128i *data = (__m128i *)(void *)data8;
   __m128i m = *p;
+  const __m128i k0 = p[2];
+  const __m128i k1 = p[3];
+  const UInt32 numRounds2 = *(const UInt32 *)(p + 1) - 1;
   for (; numBlocks != 0; numBlocks--, data++)
   {
-    UInt32 numRounds2 = *(const UInt32 *)(p + 1) - 1;
-    const __m128i *w = p + 3;
-    m = _mm_xor_si128(m, *data);
-    m = _mm_xor_si128(m, p[2]);
+    UInt32 r = numRounds2;
+    const __m128i *w = p + 4;
+    __m128i temp = *data;
+    MM_XOR (temp, k0)
+    MM_XOR (m, temp)
+    MM_OP_m (_mm_aesenc_si128, k1)
     do
     {
-      m = _mm_aesenc_si128(m, w[0]);
-      m = _mm_aesenc_si128(m, w[1]);
+      MM_OP_m (_mm_aesenc_si128, w[0])
+      MM_OP_m (_mm_aesenc_si128, w[1])
       w += 2;
     }
-    while (--numRounds2 != 0);
-    m = _mm_aesenc_si128(m, w[0]);
-    m = _mm_aesenclast_si128(m, w[1]);
+    while (--r);
+    MM_OP_m (_mm_aesenclast_si128, w[0])
     *data = m;
   }
   *p = m;
 }
 
-#define NUM_WAYS 3
 
-#define AES_OP_W(op, n) { \
-    const __m128i t = w[n]; \
-    m0 = op(m0, t); \
-    m1 = op(m1, t); \
-    m2 = op(m2, t); \
-    }
+#define WOP_1(op)
+#define WOP_2(op)   WOP_1 (op)  op (m1, 1)
+#define WOP_3(op)   WOP_2 (op)  op (m2, 2)
+#define WOP_4(op)   WOP_3 (op)  op (m3, 3)
+#ifdef MY_CPU_AMD64
+#define WOP_5(op)   WOP_4 (op)  op (m4, 4)
+#define WOP_6(op)   WOP_5 (op)  op (m5, 5)
+#define WOP_7(op)   WOP_6 (op)  op (m6, 6)
+#define WOP_8(op)   WOP_7 (op)  op (m7, 7)
+#endif
+/*
+#define WOP_9(op)   WOP_8 (op)  op (m8, 8);
+#define WOP_10(op)  WOP_9 (op)  op (m9, 9);
+#define WOP_11(op)  WOP_10(op)  op (m10, 10);
+#define WOP_12(op)  WOP_11(op)  op (m11, 11);
+#define WOP_13(op)  WOP_12(op)  op (m12, 12);
+#define WOP_14(op)  WOP_13(op)  op (m13, 13);
+*/
 
-#define AES_DEC(n) AES_OP_W(_mm_aesdec_si128, n)
-#define AES_DEC_LAST(n) AES_OP_W(_mm_aesdeclast_si128, n)
-#define AES_ENC(n) AES_OP_W(_mm_aesenc_si128, n)
-#define AES_ENC_LAST(n) AES_OP_W(_mm_aesenclast_si128, n)
+#ifdef MY_CPU_AMD64
+  #define NUM_WAYS      8
+  #define WOP_M1    WOP_8
+#else
+  #define NUM_WAYS      4
+  #define WOP_M1    WOP_4
+#endif
 
-void MY_FAST_CALL AesCbc_Decode_Intel(__m128i *p, __m128i *data, size_t numBlocks)
+#define WOP(op)  op (m0, 0)  WOP_M1(op)
+
+
+#define DECLARE_VAR(reg, ii)  __m128i reg;
+#define LOAD_data(  reg, ii)  reg = data[ii];
+#define STORE_data( reg, ii)  data[ii] = reg;
+#if (NUM_WAYS > 1)
+#define XOR_data_M1(reg, ii)  MM_XOR (reg, data[ii- 1])
+#endif
+
+#define AVX_DECLARE_VAR(reg, ii)  __m256i reg;
+#define AVX_LOAD_data(  reg, ii)  reg = ((const __m256i *)(const void *)data)[ii];
+#define AVX_STORE_data( reg, ii)  ((__m256i *)(void *)data)[ii] = reg;
+#define AVX_XOR_data_M1(reg, ii)  AVX_XOR (reg, (((const __m256i *)(const void *)(data - 1))[ii]))
+
+#define MM_OP_key(op, reg)  MM_OP(op, reg, key);
+
+#define AES_DEC(      reg, ii)   MM_OP_key (_mm_aesdec_si128,     reg)
+#define AES_DEC_LAST( reg, ii)   MM_OP_key (_mm_aesdeclast_si128, reg)
+#define AES_ENC(      reg, ii)   MM_OP_key (_mm_aesenc_si128,     reg)
+#define AES_ENC_LAST( reg, ii)   MM_OP_key (_mm_aesenclast_si128, reg)
+#define AES_XOR(      reg, ii)   MM_OP_key (_mm_xor_si128,        reg)
+
+
+#define AVX_AES_DEC(      reg, ii)   MM_OP_key (_mm256_aesdec_epi128,     reg)
+#define AVX_AES_DEC_LAST( reg, ii)   MM_OP_key (_mm256_aesdeclast_epi128, reg)
+#define AVX_AES_ENC(      reg, ii)   MM_OP_key (_mm256_aesenc_epi128,     reg)
+#define AVX_AES_ENC_LAST( reg, ii)   MM_OP_key (_mm256_aesenclast_epi128, reg)
+#define AVX_AES_XOR(      reg, ii)   MM_OP_key (_mm256_xor_si256,         reg)
+
+#define CTR_START(reg, ii)  MM_OP (_mm_add_epi64, ctr, one)  reg = ctr;
+#define CTR_END(  reg, ii)  MM_XOR (data[ii], reg)
+
+#define AVX_CTR_START(reg, ii)  MM_OP (_mm256_add_epi64, ctr2, two)  reg = _mm256_xor_si256(ctr2, key);
+#define AVX_CTR_END(  reg, ii)  AVX_XOR (((__m256i *)(void *)data)[ii], reg)
+
+#define WOP_KEY(op, n) { \
+    const __m128i key = w[n]; \
+    WOP(op); }
+
+#define AVX_WOP_KEY(op, n) { \
+    const __m256i key = w[n]; \
+    WOP(op); }
+
+
+#define WIDE_LOOP_START  \
+    dataEnd = data + numBlocks;  \
+    if (numBlocks >= NUM_WAYS)  \
+    { dataEnd -= NUM_WAYS; do {  \
+
+
+#define WIDE_LOOP_END  \
+    data += NUM_WAYS;  \
+    } while (data <= dataEnd);  \
+    dataEnd += NUM_WAYS; }  \
+
+
+#define SINGLE_LOOP  \
+    for (; data < dataEnd; data++)
+
+
+#define NUM_AES_KEYS_MAX 15
+
+#define WIDE_LOOP_START_AVX(OP)  \
+    dataEnd = data + numBlocks;  \
+    if (numBlocks >= NUM_WAYS * 2)  \
+    { __m256i keys[NUM_AES_KEYS_MAX]; \
+    UInt32 ii; \
+    OP \
+    for (ii = 0; ii < numRounds; ii++) \
+      keys[ii] = _mm256_broadcastsi128_si256(p[ii]); \
+    dataEnd -= NUM_WAYS * 2; do {  \
+
+
+#define WIDE_LOOP_END_AVX(OP)  \
+    data += NUM_WAYS * 2;  \
+    } while (data <= dataEnd);  \
+    dataEnd += NUM_WAYS * 2;  \
+    OP  \
+    _mm256_zeroupper();  \
+    }  \
+
+/* MSVC for x86: If we don't call _mm256_zeroupper(), and -arch:IA32 is not specified,
+   MSVC still can insert vzeroupper instruction. */
+
+
+AES_FUNC_START2 (AesCbc_Decode_HW)
 {
+  __m128i *p = (__m128i *)(void *)ivAes;
+  __m128i *data = (__m128i *)(void *)data8;
   __m128i iv = *p;
-  for (; numBlocks >= NUM_WAYS; numBlocks -= NUM_WAYS, data += NUM_WAYS)
+  const __m128i *wStart = p + *(const UInt32 *)(p + 1) * 2 + 2 - 1;
+  const __m128i *dataEnd;
+  p += 2;
+  
+  WIDE_LOOP_START
   {
-    UInt32 numRounds2 = *(const UInt32 *)(p + 1);
-    const __m128i *w = p + numRounds2 * 2;
-    __m128i m0, m1, m2;
-    {
-      const __m128i t = w[2];
-      m0 = _mm_xor_si128(t, data[0]);
-      m1 = _mm_xor_si128(t, data[1]);
-      m2 = _mm_xor_si128(t, data[2]);
-    }
-    numRounds2--;
+    const __m128i *w = wStart;
+    
+    WOP (DECLARE_VAR)
+    WOP (LOAD_data)
+    WOP_KEY (AES_XOR, 1)
+
     do
     {
-      AES_DEC(1)
-      AES_DEC(0)
-      w -= 2;
+      WOP_KEY (AES_DEC, 0)
+      w--;
     }
-    while (--numRounds2 != 0);
-    AES_DEC(1)
-    AES_DEC_LAST(0)
+    while (w != p);
+    WOP_KEY (AES_DEC_LAST, 0)
 
-    {
-      __m128i t;
-      t = _mm_xor_si128(m0, iv); iv = data[0]; data[0] = t;
-      t = _mm_xor_si128(m1, iv); iv = data[1]; data[1] = t;
-      t = _mm_xor_si128(m2, iv); iv = data[2]; data[2] = t;
-    }
+    MM_XOR (m0, iv)
+    WOP_M1 (XOR_data_M1)
+    iv = data[NUM_WAYS - 1];
+    WOP (STORE_data)
   }
-  for (; numBlocks != 0; numBlocks--, data++)
+  WIDE_LOOP_END
+
+  SINGLE_LOOP
   {
-    UInt32 numRounds2 = *(const UInt32 *)(p + 1);
-    const __m128i *w = p + numRounds2 * 2;
-    __m128i m = _mm_xor_si128(w[2], *data);
-    numRounds2--;
+    const __m128i *w = wStart - 1;
+    __m128i m = _mm_xor_si128 (w[2], *data);
     do
     {
-      m = _mm_aesdec_si128(m, w[1]);
-      m = _mm_aesdec_si128(m, w[0]);
+      MM_OP_m (_mm_aesdec_si128, w[1])
+      MM_OP_m (_mm_aesdec_si128, w[0])
       w -= 2;
     }
-    while (--numRounds2 != 0);
-    m = _mm_aesdec_si128(m, w[1]);
-    m = _mm_aesdeclast_si128(m, w[0]);
+    while (w != p);
+    MM_OP_m (_mm_aesdec_si128,     w[1])
+    MM_OP_m (_mm_aesdeclast_si128, w[0])
 
-    m = _mm_xor_si128(m, iv);
+    MM_XOR (m, iv)
     iv = *data;
     *data = m;
   }
-  *p = iv;
+  
+  p[-2] = iv;
 }
 
-void MY_FAST_CALL AesCtr_Code_Intel(__m128i *p, __m128i *data, size_t numBlocks)
+
+AES_FUNC_START2 (AesCtr_Code_HW)
 {
+  __m128i *p = (__m128i *)(void *)ivAes;
+  __m128i *data = (__m128i *)(void *)data8;
   __m128i ctr = *p;
-  __m128i one;
-  one.m128i_u64[0] = 1;
-  one.m128i_u64[1] = 0;
-  for (; numBlocks >= NUM_WAYS; numBlocks -= NUM_WAYS, data += NUM_WAYS)
+  UInt32 numRoundsMinus2 = *(const UInt32 *)(p + 1) * 2 - 1;
+  const __m128i *dataEnd;
+  __m128i one = _mm_cvtsi32_si128(1);
+
+  p += 2;
+  
+  WIDE_LOOP_START
   {
-    UInt32 numRounds2 = *(const UInt32 *)(p + 1) - 1;
     const __m128i *w = p;
-    __m128i m0, m1, m2;
-    {
-      const __m128i t = w[2];
-      ctr = _mm_add_epi64(ctr, one); m0 = _mm_xor_si128(ctr, t);
-      ctr = _mm_add_epi64(ctr, one); m1 = _mm_xor_si128(ctr, t);
-      ctr = _mm_add_epi64(ctr, one); m2 = _mm_xor_si128(ctr, t);
-    }
-    w += 3;
+    UInt32 r = numRoundsMinus2;
+    WOP (DECLARE_VAR)
+    WOP (CTR_START)
+    WOP_KEY (AES_XOR, 0)
+    w += 1;
     do
     {
-      AES_ENC(0)
-      AES_ENC(1)
-      w += 2;
+      WOP_KEY (AES_ENC, 0)
+      w += 1;
     }
-    while (--numRounds2 != 0);
-    AES_ENC(0)
-    AES_ENC_LAST(1)
-    data[0] = _mm_xor_si128(data[0], m0);
-    data[1] = _mm_xor_si128(data[1], m1);
-    data[2] = _mm_xor_si128(data[2], m2);
+    while (--r);
+    WOP_KEY (AES_ENC_LAST, 0)
+   
+    WOP (CTR_END)
   }
-  for (; numBlocks != 0; numBlocks--, data++)
+  WIDE_LOOP_END
+
+  SINGLE_LOOP
   {
-    UInt32 numRounds2 = *(const UInt32 *)(p + 1) - 1;
+    UInt32 numRounds2 = *(const UInt32 *)(p - 2 + 1) - 1;
     const __m128i *w = p;
     __m128i m;
-    ctr = _mm_add_epi64(ctr, one);
-    m = _mm_xor_si128(ctr, p[2]);
-    w += 3;
+    MM_OP (_mm_add_epi64, ctr, one)
+    m = _mm_xor_si128 (ctr, p[0]);
+    w += 1;
     do
     {
-      m = _mm_aesenc_si128(m, w[0]);
-      m = _mm_aesenc_si128(m, w[1]);
+      MM_OP_m (_mm_aesenc_si128, w[0])
+      MM_OP_m (_mm_aesenc_si128, w[1])
       w += 2;
     }
-    while (--numRounds2 != 0);
-    m = _mm_aesenc_si128(m, w[0]);
-    m = _mm_aesenclast_si128(m, w[1]);
-    *data = _mm_xor_si128(*data, m);
+    while (--numRounds2);
+    MM_OP_m (_mm_aesenc_si128,     w[0])
+    MM_OP_m (_mm_aesenclast_si128, w[1])
+    MM_XOR (*data, m)
   }
-  *p = ctr;
+  
+  p[-2] = ctr;
 }
 
-#else
 
-void MY_FAST_CALL AesCbc_Encode(UInt32 *ivAes, Byte *data, size_t numBlocks);
-void MY_FAST_CALL AesCbc_Decode(UInt32 *ivAes, Byte *data, size_t numBlocks);
-void MY_FAST_CALL AesCtr_Code(UInt32 *ivAes, Byte *data, size_t numBlocks);
 
-void MY_FAST_CALL AesCbc_Encode_Intel(UInt32 *p, Byte *data, size_t numBlocks)
+#ifdef USE_INTEL_VAES
+
+/*
+GCC before 2013-Jun:
+  <immintrin.h>:
+    #ifdef __AVX__
+     #include <avxintrin.h>
+    #endif
+GCC after 2013-Jun:
+  <immintrin.h>:
+    #include <avxintrin.h>
+CLANG 3.8+:
 {
-  AesCbc_Encode(p, data, numBlocks);
+  <immintrin.h>:
+    #if !defined(_MSC_VER) || defined(__AVX__)
+      #include <avxintrin.h>
+    #endif
+
+  if (the compiler is clang for Windows and if global arch is not set for __AVX__)
+    [ if (defined(_MSC_VER) && !defined(__AVX__)) ]
+  {
+    <immintrin.h> doesn't include <avxintrin.h>
+    and we have 2 ways to fix it:
+      1) we can define required __AVX__ before <immintrin.h>
+      or
+      2) we can include <avxintrin.h> after <immintrin.h>
+  }
 }
 
-void MY_FAST_CALL AesCbc_Decode_Intel(UInt32 *p, Byte *data, size_t numBlocks)
-{
-  AesCbc_Decode(p, data, numBlocks);
-}
-
-void MY_FAST_CALL AesCtr_Code_Intel(UInt32 *p, Byte *data, size_t numBlocks)
-{
-  AesCtr_Code(p, data, numBlocks);
-}
+If we include <avxintrin.h> manually for GCC/CLANG, it's
+required that <immintrin.h> must be included before <avxintrin.h>.
+*/
 
+/*
+#if defined(__clang__) && defined(_MSC_VER)
+#define __AVX__
+#define __AVX2__
+#define __VAES__
 #endif
+*/
+
+#include <immintrin.h>
+#if defined(__clang__) && defined(_MSC_VER)
+  #if !defined(__AVX__)
+    #include <avxintrin.h>
+  #endif
+  #if !defined(__AVX2__)
+    #include <avx2intrin.h>
+  #endif
+  #if !defined(__VAES__)
+    #include <vaesintrin.h>
+  #endif
+#endif  // __clang__ && _MSC_VER
+
+
+#define VAES_FUNC_START2(name) \
+AES_FUNC_START (name); \
+ATTRIB_VAES \
+AES_FUNC_START (name)
+
+VAES_FUNC_START2 (AesCbc_Decode_HW_256)
+{
+  __m128i *p = (__m128i *)(void *)ivAes;
+  __m128i *data = (__m128i *)(void *)data8;
+  __m128i iv = *p;
+  const __m128i *dataEnd;
+  UInt32 numRounds = *(const UInt32 *)(p + 1) * 2 + 1;
+  p += 2;
+  
+  WIDE_LOOP_START_AVX(;)
+  {
+    const __m256i *w = keys + numRounds - 2;
+    
+    WOP (AVX_DECLARE_VAR)
+    WOP (AVX_LOAD_data)
+    AVX_WOP_KEY (AVX_AES_XOR, 1)
+
+    do
+    {
+      AVX_WOP_KEY (AVX_AES_DEC, 0)
+      w--;
+    }
+    while (w != keys);
+    AVX_WOP_KEY (AVX_AES_DEC_LAST, 0)
+
+    AVX_XOR (m0, _mm256_setr_m128i(iv, data[0]))
+    WOP_M1 (AVX_XOR_data_M1)
+    iv = data[NUM_WAYS * 2 - 1];
+    WOP (AVX_STORE_data)
+  }
+  WIDE_LOOP_END_AVX(;)
+
+  SINGLE_LOOP
+  {
+    const __m128i *w = p + *(const UInt32 *)(p + 1 - 2) * 2 + 1 - 3;
+    __m128i m = _mm_xor_si128 (w[2], *data);
+    do
+    {
+      MM_OP_m (_mm_aesdec_si128, w[1])
+      MM_OP_m (_mm_aesdec_si128, w[0])
+      w -= 2;
+    }
+    while (w != p);
+    MM_OP_m (_mm_aesdec_si128,     w[1])
+    MM_OP_m (_mm_aesdeclast_si128, w[0])
+
+    MM_XOR (m, iv)
+    iv = *data;
+    *data = m;
+  }
+  
+  p[-2] = iv;
+}
+
+
+/*
+SSE2: _mm_cvtsi32_si128 : movd
+AVX:  _mm256_setr_m128i            : vinsertf128
+AVX2: _mm256_add_epi64             : vpaddq ymm, ymm, ymm
+      _mm256_extracti128_si256     : vextracti128
+      _mm256_broadcastsi128_si256  : vbroadcasti128
+*/
+
+#define AVX_CTR_LOOP_START  \
+    ctr2 = _mm256_setr_m128i(_mm_sub_epi64(ctr, one), ctr); \
+    two = _mm256_setr_m128i(one, one); \
+    two = _mm256_add_epi64(two, two); \
+
+// two = _mm256_setr_epi64x(2, 0, 2, 0);
+  
+#define AVX_CTR_LOOP_ENC  \
+    ctr = _mm256_extracti128_si256 (ctr2, 1); \
+ 
+VAES_FUNC_START2 (AesCtr_Code_HW_256)
+{
+  __m128i *p = (__m128i *)(void *)ivAes;
+  __m128i *data = (__m128i *)(void *)data8;
+  __m128i ctr = *p;
+  UInt32 numRounds = *(const UInt32 *)(p + 1) * 2 + 1;
+  const __m128i *dataEnd;
+  __m128i one = _mm_cvtsi32_si128(1);
+  __m256i ctr2, two;
+  p += 2;
+  
+  WIDE_LOOP_START_AVX (AVX_CTR_LOOP_START)
+  {
+    const __m256i *w = keys;
+    UInt32 r = numRounds - 2;
+    WOP (AVX_DECLARE_VAR)
+    AVX_WOP_KEY (AVX_CTR_START, 0)
+
+    w += 1;
+    do
+    {
+      AVX_WOP_KEY (AVX_AES_ENC, 0)
+      w += 1;
+    }
+    while (--r);
+    AVX_WOP_KEY (AVX_AES_ENC_LAST, 0)
+   
+    WOP (AVX_CTR_END)
+  }
+  WIDE_LOOP_END_AVX (AVX_CTR_LOOP_ENC)
+  
+  SINGLE_LOOP
+  {
+    UInt32 numRounds2 = *(const UInt32 *)(p - 2 + 1) - 1;
+    const __m128i *w = p;
+    __m128i m;
+    MM_OP (_mm_add_epi64, ctr, one)
+    m = _mm_xor_si128 (ctr, p[0]);
+    w += 1;
+    do
+    {
+      MM_OP_m (_mm_aesenc_si128, w[0])
+      MM_OP_m (_mm_aesenc_si128, w[1])
+      w += 2;
+    }
+    while (--numRounds2);
+    MM_OP_m (_mm_aesenc_si128,     w[0])
+    MM_OP_m (_mm_aesenclast_si128, w[1])
+    MM_XOR (*data, m)
+  }
+
+  p[-2] = ctr;
+}
+
+#endif // USE_INTEL_VAES
+
+#else // USE_INTEL_AES
+
+/* no USE_INTEL_AES */
+
+#pragma message("AES  HW_SW stub was used")
+
+#define AES_TYPE_keys UInt32
+#define AES_TYPE_data Byte
+
+#define AES_FUNC_START(name) \
+    void Z7_FASTCALL name(UInt32 *p, Byte *data, size_t numBlocks) \
+
+#define AES_COMPAT_STUB(name) \
+    AES_FUNC_START(name); \
+    AES_FUNC_START(name ## _HW) \
+    { name(p, data, numBlocks); }
+
+AES_COMPAT_STUB (AesCbc_Encode)
+AES_COMPAT_STUB (AesCbc_Decode)
+AES_COMPAT_STUB (AesCtr_Code)
+
+#endif // USE_INTEL_AES
+
+
+#ifndef USE_INTEL_VAES
+
+#pragma message("VAES HW_SW stub was used")
+
+#define VAES_COMPAT_STUB(name) \
+    void Z7_FASTCALL name ## _256(UInt32 *p, Byte *data, size_t numBlocks); \
+    void Z7_FASTCALL name ## _256(UInt32 *p, Byte *data, size_t numBlocks) \
+    { name((AES_TYPE_keys *)(void *)p, (AES_TYPE_data *)(void *)data, numBlocks); }
+
+VAES_COMPAT_STUB (AesCbc_Decode_HW)
+VAES_COMPAT_STUB (AesCtr_Code_HW)
+
+#endif // ! USE_INTEL_VAES
+
+
+#elif defined(MY_CPU_ARM_OR_ARM64) && defined(MY_CPU_LE)
+
+  #if defined(__clang__)
+    #if (__clang_major__ >= 8) // fix that check
+      #define USE_HW_AES
+    #endif
+  #elif defined(__GNUC__)
+    #if (__GNUC__ >= 6) // fix that check
+      #define USE_HW_AES
+    #endif
+  #elif defined(_MSC_VER)
+    #if _MSC_VER >= 1910
+      #define USE_HW_AES
+    #endif
+  #endif
+
+#ifdef USE_HW_AES
+
+// #pragma message("=== AES HW === ")
+
+#if defined(__clang__) || defined(__GNUC__)
+  #ifdef MY_CPU_ARM64
+    #define ATTRIB_AES __attribute__((__target__("+crypto")))
+  #else
+    #define ATTRIB_AES __attribute__((__target__("fpu=crypto-neon-fp-armv8")))
+  #endif
+#else
+  // _MSC_VER
+  // for arm32
+  #define _ARM_USE_NEW_NEON_INTRINSICS
+#endif
+
+#ifndef ATTRIB_AES
+  #define ATTRIB_AES
+#endif
+
+#if defined(_MSC_VER) && defined(MY_CPU_ARM64)
+#include <arm64_neon.h>
+#else
+#include <arm_neon.h>
+#endif
+
+typedef uint8x16_t v128;
+
+#define AES_FUNC_START(name) \
+    void Z7_FASTCALL name(UInt32 *ivAes, Byte *data8, size_t numBlocks)
+    // void Z7_FASTCALL name(v128 *p, v128 *data, size_t numBlocks)
+
+#define AES_FUNC_START2(name) \
+AES_FUNC_START (name); \
+ATTRIB_AES \
+AES_FUNC_START (name)
+
+#define MM_OP(op, dest, src)  dest = op(dest, src);
+#define MM_OP_m(op, src)      MM_OP(op, m, src)
+#define MM_OP1_m(op)          m = op(m);
+
+#define MM_XOR( dest, src)    MM_OP(veorq_u8, dest, src)
+#define MM_XOR_m( src)        MM_XOR(m, src)
+
+#define AES_E_m(k)     MM_OP_m (vaeseq_u8, k)
+#define AES_E_MC_m(k)  AES_E_m (k)  MM_OP1_m(vaesmcq_u8)
+
+
+AES_FUNC_START2 (AesCbc_Encode_HW)
+{
+  v128 *p = (v128*)(void*)ivAes;
+  v128 *data = (v128*)(void*)data8;
+  v128 m = *p;
+  const v128 k0 = p[2];
+  const v128 k1 = p[3];
+  const v128 k2 = p[4];
+  const v128 k3 = p[5];
+  const v128 k4 = p[6];
+  const v128 k5 = p[7];
+  const v128 k6 = p[8];
+  const v128 k7 = p[9];
+  const v128 k8 = p[10];
+  const v128 k9 = p[11];
+  const UInt32 numRounds2 = *(const UInt32 *)(p + 1);
+  const v128 *w = p + ((size_t)numRounds2 * 2);
+  const v128 k_z1 = w[1];
+  const v128 k_z0 = w[2];
+  for (; numBlocks != 0; numBlocks--, data++)
+  {
+    MM_XOR_m (*data);
+    AES_E_MC_m (k0)
+    AES_E_MC_m (k1)
+    AES_E_MC_m (k2)
+    AES_E_MC_m (k3)
+    AES_E_MC_m (k4)
+    AES_E_MC_m (k5)
+    AES_E_MC_m (k6)
+    AES_E_MC_m (k7)
+    AES_E_MC_m (k8)
+    if (numRounds2 >= 6)
+    {
+      AES_E_MC_m (k9)
+      AES_E_MC_m (p[12])
+      if (numRounds2 != 6)
+      {
+        AES_E_MC_m (p[13])
+        AES_E_MC_m (p[14])
+      }
+    }
+    AES_E_m  (k_z1)
+    MM_XOR_m (k_z0);
+    *data = m;
+  }
+  *p = m;
+}
+
+
+#define WOP_1(op)
+#define WOP_2(op)   WOP_1 (op)  op (m1, 1)
+#define WOP_3(op)   WOP_2 (op)  op (m2, 2)
+#define WOP_4(op)   WOP_3 (op)  op (m3, 3)
+#define WOP_5(op)   WOP_4 (op)  op (m4, 4)
+#define WOP_6(op)   WOP_5 (op)  op (m5, 5)
+#define WOP_7(op)   WOP_6 (op)  op (m6, 6)
+#define WOP_8(op)   WOP_7 (op)  op (m7, 7)
+
+  #define NUM_WAYS      8
+  #define WOP_M1    WOP_8
+
+#define WOP(op)  op (m0, 0)   WOP_M1(op)
+
+#define DECLARE_VAR(reg, ii)  v128 reg;
+#define LOAD_data(  reg, ii)  reg = data[ii];
+#define STORE_data( reg, ii)  data[ii] = reg;
+#if (NUM_WAYS > 1)
+#define XOR_data_M1(reg, ii)  MM_XOR (reg, data[ii- 1])
+#endif
+
+#define MM_OP_key(op, reg)  MM_OP (op, reg, key)
+
+#define AES_D_m(k)      MM_OP_m (vaesdq_u8, k)
+#define AES_D_IMC_m(k)  AES_D_m (k)  MM_OP1_m (vaesimcq_u8)
+
+#define AES_XOR(   reg, ii)  MM_OP_key (veorq_u8,  reg)
+#define AES_D(     reg, ii)  MM_OP_key (vaesdq_u8, reg)
+#define AES_E(     reg, ii)  MM_OP_key (vaeseq_u8, reg)
+
+#define AES_D_IMC( reg, ii)  AES_D (reg, ii)  reg = vaesimcq_u8(reg);
+#define AES_E_MC(  reg, ii)  AES_E (reg, ii)  reg = vaesmcq_u8(reg);
+
+#define CTR_START(reg, ii)  MM_OP (vaddq_u64, ctr, one)  reg = vreinterpretq_u8_u64(ctr);
+#define CTR_END(  reg, ii)  MM_XOR (data[ii], reg)
+
+#define WOP_KEY(op, n) { \
+    const v128 key = w[n]; \
+    WOP(op) }
+
+#define WIDE_LOOP_START  \
+    dataEnd = data + numBlocks;  \
+    if (numBlocks >= NUM_WAYS)  \
+    { dataEnd -= NUM_WAYS; do {  \
+
+#define WIDE_LOOP_END  \
+    data += NUM_WAYS;  \
+    } while (data <= dataEnd);  \
+    dataEnd += NUM_WAYS; }  \
+
+#define SINGLE_LOOP  \
+    for (; data < dataEnd; data++)
+
+
+AES_FUNC_START2 (AesCbc_Decode_HW)
+{
+  v128 *p = (v128*)(void*)ivAes;
+  v128 *data = (v128*)(void*)data8;
+  v128 iv = *p;
+  const v128 *wStart = p + ((size_t)*(const UInt32 *)(p + 1)) * 2;
+  const v128 *dataEnd;
+  p += 2;
+  
+  WIDE_LOOP_START
+  {
+    const v128 *w = wStart;
+    WOP (DECLARE_VAR)
+    WOP (LOAD_data)
+    WOP_KEY (AES_D_IMC, 2)
+    do
+    {
+      WOP_KEY (AES_D_IMC, 1)
+      WOP_KEY (AES_D_IMC, 0)
+      w -= 2;
+    }
+    while (w != p);
+    WOP_KEY (AES_D,   1)
+    WOP_KEY (AES_XOR, 0)
+    MM_XOR (m0, iv);
+    WOP_M1 (XOR_data_M1)
+    iv = data[NUM_WAYS - 1];
+    WOP (STORE_data)
+  }
+  WIDE_LOOP_END
+
+  SINGLE_LOOP
+  {
+    const v128 *w = wStart;
+    v128 m = *data;
+    AES_D_IMC_m (w[2])
+    do
+    {
+      AES_D_IMC_m (w[1]);
+      AES_D_IMC_m (w[0]);
+      w -= 2;
+    }
+    while (w != p);
+    AES_D_m  (w[1]);
+    MM_XOR_m (w[0]);
+    MM_XOR_m (iv);
+    iv = *data;
+    *data = m;
+  }
+  
+  p[-2] = iv;
+}
+
+
+AES_FUNC_START2 (AesCtr_Code_HW)
+{
+  v128 *p = (v128*)(void*)ivAes;
+  v128 *data = (v128*)(void*)data8;
+  uint64x2_t ctr = vreinterpretq_u64_u8(*p);
+  const v128 *wEnd = p + ((size_t)*(const UInt32 *)(p + 1)) * 2;
+  const v128 *dataEnd;
+  uint64x2_t one = vdupq_n_u64(0);
+  one = vsetq_lane_u64(1, one, 0);
+  p += 2;
+  
+  WIDE_LOOP_START
+  {
+    const v128 *w = p;
+    WOP (DECLARE_VAR)
+    WOP (CTR_START)
+    do
+    {
+      WOP_KEY (AES_E_MC, 0)
+      WOP_KEY (AES_E_MC, 1)
+      w += 2;
+    }
+    while (w != wEnd);
+    WOP_KEY (AES_E_MC, 0)
+    WOP_KEY (AES_E,    1)
+    WOP_KEY (AES_XOR,  2)
+    WOP (CTR_END)
+  }
+  WIDE_LOOP_END
+
+  SINGLE_LOOP
+  {
+    const v128 *w = p;
+    v128 m;
+    CTR_START (m, 0);
+    do
+    {
+      AES_E_MC_m (w[0]);
+      AES_E_MC_m (w[1]);
+      w += 2;
+    }
+    while (w != wEnd);
+    AES_E_MC_m (w[0])
+    AES_E_m    (w[1])
+    MM_XOR_m   (w[2])
+    CTR_END (m, 0)
+  }
+  
+  p[-2] = vreinterpretq_u8_u64(ctr);
+}
+
+#endif // USE_HW_AES
+
+#endif // MY_CPU_ARM_OR_ARM64
+
+#undef NUM_WAYS
+#undef WOP_M1
+#undef WOP
+#undef DECLARE_VAR
+#undef LOAD_data
+#undef STORE_data
+#undef USE_INTEL_AES
+#undef USE_HW_AES

C/Alloc.c

@@ -1,33 +1,182 @@
 /* Alloc.c -- Memory allocation functions
-2008-09-24
-Igor Pavlov
-Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #ifdef _WIN32
-#include <windows.h>
+#include "7zWindows.h"
 #endif
 #include <stdlib.h>
 
 #include "Alloc.h"
 
-/* #define _SZ_ALLOC_DEBUG */
-
-/* use _SZ_ALLOC_DEBUG to debug alloc/free operations */
-#ifdef _SZ_ALLOC_DEBUG
-#include <stdio.h>
-int g_allocCount = 0;
-int g_allocCountMid = 0;
-int g_allocCountBig = 0;
+#ifdef _WIN32
+#ifdef Z7_LARGE_PAGES
+#if defined(__clang__) || defined(__GNUC__)
+typedef void (*Z7_voidFunction)(void);
+#define MY_CAST_FUNC (Z7_voidFunction)
+#elif defined(_MSC_VER) && _MSC_VER > 1920
+#define MY_CAST_FUNC  (void *)
+// #pragma warning(disable : 4191) // 'type cast': unsafe conversion from 'FARPROC' to 'void (__cdecl *)()'
+#else
+#define MY_CAST_FUNC
 #endif
+#endif // Z7_LARGE_PAGES
+#endif // _WIN32
+
+// #define SZ_ALLOC_DEBUG
+/* #define SZ_ALLOC_DEBUG */
+
+/* use SZ_ALLOC_DEBUG to debug alloc/free operations */
+#ifdef SZ_ALLOC_DEBUG
+
+#include <string.h>
+#include <stdio.h>
+static int g_allocCount = 0;
+#ifdef _WIN32
+static int g_allocCountMid = 0;
+static int g_allocCountBig = 0;
+#endif
+
+
+#define CONVERT_INT_TO_STR(charType, tempSize) \
+  char temp[tempSize]; unsigned i = 0; \
+  while (val >= 10) { temp[i++] = (char)('0' + (unsigned)(val % 10)); val /= 10; } \
+  *s++ = (charType)('0' + (unsigned)val); \
+  while (i != 0) { i--; *s++ = temp[i]; } \
+  *s = 0;
+
+static void ConvertUInt64ToString(UInt64 val, char *s)
+{
+  CONVERT_INT_TO_STR(char, 24)
+}
+
+#define GET_HEX_CHAR(t) ((char)(((t < 10) ? ('0' + t) : ('A' + (t - 10)))))
+
+static void ConvertUInt64ToHex(UInt64 val, char *s)
+{
+  UInt64 v = val;
+  unsigned i;
+  for (i = 1;; i++)
+  {
+    v >>= 4;
+    if (v == 0)
+      break;
+  }
+  s[i] = 0;
+  do
+  {
+    unsigned t = (unsigned)(val & 0xF);
+    val >>= 4;
+    s[--i] = GET_HEX_CHAR(t);
+  }
+  while (i);
+}
+
+#define DEBUG_OUT_STREAM stderr
+
+static void Print(const char *s)
+{
+  fputs(s, DEBUG_OUT_STREAM);
+}
+
+static void PrintAligned(const char *s, size_t align)
+{
+  size_t len = strlen(s);
+  for(;;)
+  {
+    fputc(' ', DEBUG_OUT_STREAM);
+    if (len >= align)
+      break;
+    ++len;
+  }
+  Print(s);
+}
+
+static void PrintLn(void)
+{
+  Print("\n");
+}
+
+static void PrintHex(UInt64 v, size_t align)
+{
+  char s[32];
+  ConvertUInt64ToHex(v, s);
+  PrintAligned(s, align);
+}
+
+static void PrintDec(int v, size_t align)
+{
+  char s[32];
+  ConvertUInt64ToString((unsigned)v, s);
+  PrintAligned(s, align);
+}
+
+static void PrintAddr(void *p)
+{
+  PrintHex((UInt64)(size_t)(ptrdiff_t)p, 12);
+}
+
+
+#define PRINT_REALLOC(name, cnt, size, ptr) { \
+    Print(name " "); \
+    if (!ptr) PrintDec(cnt++, 10); \
+    PrintHex(size, 10); \
+    PrintAddr(ptr); \
+    PrintLn(); }
+
+#define PRINT_ALLOC(name, cnt, size, ptr) { \
+    Print(name " "); \
+    PrintDec(cnt++, 10); \
+    PrintHex(size, 10); \
+    PrintAddr(ptr); \
+    PrintLn(); }
+ 
+#define PRINT_FREE(name, cnt, ptr) if (ptr) { \
+    Print(name " "); \
+    PrintDec(--cnt, 10); \
+    PrintAddr(ptr); \
+    PrintLn(); }
+ 
+#else
+
+#ifdef _WIN32
+#define PRINT_ALLOC(name, cnt, size, ptr)
+#endif
+#define PRINT_FREE(name, cnt, ptr)
+#define Print(s)
+#define PrintLn()
+#define PrintHex(v, align)
+#define PrintAddr(p)
+
+#endif
+
+
+/*
+by specification:
+  malloc(non_NULL, 0)   : returns NULL or a unique pointer value that can later be successfully passed to free()
+  realloc(NULL, size)   : the call is equivalent to malloc(size)
+  realloc(non_NULL, 0)  : the call is equivalent to free(ptr)
+
+in main compilers:
+  malloc(0)             : returns non_NULL
+  realloc(NULL,     0)  : returns non_NULL
+  realloc(non_NULL, 0)  : returns NULL
+*/
+
 
 void *MyAlloc(size_t size)
 {
   if (size == 0)
-    return 0;
-  #ifdef _SZ_ALLOC_DEBUG
+    return NULL;
+  // PRINT_ALLOC("Alloc    ", g_allocCount, size, NULL)
+  #ifdef SZ_ALLOC_DEBUG
   {
     void *p = malloc(size);
-    fprintf(stderr, "\nAlloc %10d bytes, count = %10d,  addr = %8X", size, g_allocCount++, (unsigned)p);
+    if (p)
+    {
+      PRINT_ALLOC("Alloc    ", g_allocCount, size, p)
+    }
     return p;
   }
   #else
@@ -37,91 +186,350 @@ void *MyAlloc(size_t size)
 
 void MyFree(void *address)
 {
-  #ifdef _SZ_ALLOC_DEBUG
-  if (address != 0)
-    fprintf(stderr, "\nFree; count = %10d,  addr = %8X", --g_allocCount, (unsigned)address);
-  #endif
+  PRINT_FREE("Free    ", g_allocCount, address)
+  
   free(address);
 }
 
+void *MyRealloc(void *address, size_t size)
+{
+  if (size == 0)
+  {
+    MyFree(address);
+    return NULL;
+  }
+  // PRINT_REALLOC("Realloc  ", g_allocCount, size, address)
+  #ifdef SZ_ALLOC_DEBUG
+  {
+    void *p = realloc(address, size);
+    if (p)
+    {
+      PRINT_REALLOC("Realloc    ", g_allocCount, size, address)
+    }
+    return p;
+  }
+  #else
+  return realloc(address, size);
+  #endif
+}
+
+
 #ifdef _WIN32
 
 void *MidAlloc(size_t size)
 {
   if (size == 0)
-    return 0;
-  #ifdef _SZ_ALLOC_DEBUG
-  fprintf(stderr, "\nAlloc_Mid %10d bytes;  count = %10d", size, g_allocCountMid++);
+    return NULL;
+  #ifdef SZ_ALLOC_DEBUG
+  {
+    void *p = VirtualAlloc(NULL, size, MEM_COMMIT, PAGE_READWRITE);
+    if (p)
+    {
+      PRINT_ALLOC("Alloc-Mid", g_allocCountMid, size, p)
+    }
+    return p;
+  }
+  #else
+  return VirtualAlloc(NULL, size, MEM_COMMIT, PAGE_READWRITE);
   #endif
-  return VirtualAlloc(0, size, MEM_COMMIT, PAGE_READWRITE);
 }
 
 void MidFree(void *address)
 {
-  #ifdef _SZ_ALLOC_DEBUG
-  if (address != 0)
-    fprintf(stderr, "\nFree_Mid; count = %10d", --g_allocCountMid);
-  #endif
-  if (address == 0)
+  PRINT_FREE("Free-Mid", g_allocCountMid, address)
+
+  if (!address)
     return;
   VirtualFree(address, 0, MEM_RELEASE);
 }
 
-#ifndef MEM_LARGE_PAGES
-#undef _7ZIP_LARGE_PAGES
+#ifdef Z7_LARGE_PAGES
+
+#ifdef MEM_LARGE_PAGES
+  #define MY__MEM_LARGE_PAGES  MEM_LARGE_PAGES
+#else
+  #define MY__MEM_LARGE_PAGES  0x20000000
 #endif
 
-#ifdef _7ZIP_LARGE_PAGES
+extern
+SIZE_T g_LargePageSize;
 SIZE_T g_LargePageSize = 0;
-typedef SIZE_T (WINAPI *GetLargePageMinimumP)();
-#endif
+typedef SIZE_T (WINAPI *Func_GetLargePageMinimum)(VOID);
 
-void SetLargePageSize()
+void SetLargePageSize(void)
 {
-  #ifdef _7ZIP_LARGE_PAGES
-  SIZE_T size = 0;
-  GetLargePageMinimumP largePageMinimum = (GetLargePageMinimumP)
-        GetProcAddress(GetModuleHandle(TEXT("kernel32.dll")), "GetLargePageMinimum");
-  if (largePageMinimum == 0)
+  #ifdef Z7_LARGE_PAGES
+  SIZE_T size;
+  const
+   Func_GetLargePageMinimum fn =
+  (Func_GetLargePageMinimum) MY_CAST_FUNC GetProcAddress(GetModuleHandle(TEXT("kernel32.dll")),
+       "GetLargePageMinimum");
+  if (!fn)
     return;
-  size = largePageMinimum();
+  size = fn();
   if (size == 0 || (size & (size - 1)) != 0)
     return;
   g_LargePageSize = size;
   #endif
 }
 
+#endif // Z7_LARGE_PAGES
 
 void *BigAlloc(size_t size)
 {
   if (size == 0)
-    return 0;
-  #ifdef _SZ_ALLOC_DEBUG
-  fprintf(stderr, "\nAlloc_Big %10d bytes;  count = %10d", size, g_allocCountBig++);
-  #endif
-  
-  #ifdef _7ZIP_LARGE_PAGES
-  if (g_LargePageSize != 0 && g_LargePageSize <= (1 << 30) && size >= (1 << 18))
+    return NULL;
+
+  PRINT_ALLOC("Alloc-Big", g_allocCountBig, size, NULL)
+
+  #ifdef Z7_LARGE_PAGES
   {
-    void *res = VirtualAlloc(0, (size + g_LargePageSize - 1) & (~(g_LargePageSize - 1)),
-        MEM_COMMIT | MEM_LARGE_PAGES, PAGE_READWRITE);
-    if (res != 0)
-      return res;
+    SIZE_T ps = g_LargePageSize;
+    if (ps != 0 && ps <= (1 << 30) && size > (ps / 2))
+    {
+      size_t size2;
+      ps--;
+      size2 = (size + ps) & ~ps;
+      if (size2 >= size)
+      {
+        void *p = VirtualAlloc(NULL, size2, MEM_COMMIT | MY__MEM_LARGE_PAGES, PAGE_READWRITE);
+        if (p)
+        {
+          PRINT_ALLOC("Alloc-BM ", g_allocCountMid, size2, p)
+          return p;
+        }
+      }
+    }
   }
   #endif
-  return VirtualAlloc(0, size, MEM_COMMIT, PAGE_READWRITE);
+
+  return MidAlloc(size);
 }
 
 void BigFree(void *address)
 {
-  #ifdef _SZ_ALLOC_DEBUG
-  if (address != 0)
-    fprintf(stderr, "\nFree_Big; count = %10d", --g_allocCountBig);
-  #endif
-  
-  if (address == 0)
-    return;
-  VirtualFree(address, 0, MEM_RELEASE);
+  PRINT_FREE("Free-Big", g_allocCountBig, address)
+  MidFree(address);
 }
 
+#endif // _WIN32
+
+
+static void *SzAlloc(ISzAllocPtr p, size_t size) { UNUSED_VAR(p)  return MyAlloc(size); }
+static void SzFree(ISzAllocPtr p, void *address) { UNUSED_VAR(p)  MyFree(address); }
+const ISzAlloc g_Alloc = { SzAlloc, SzFree };
+
+#ifdef _WIN32
+static void *SzMidAlloc(ISzAllocPtr p, size_t size) { UNUSED_VAR(p)  return MidAlloc(size); }
+static void SzMidFree(ISzAllocPtr p, void *address) { UNUSED_VAR(p)  MidFree(address); }
+static void *SzBigAlloc(ISzAllocPtr p, size_t size) { UNUSED_VAR(p)  return BigAlloc(size); }
+static void SzBigFree(ISzAllocPtr p, void *address) { UNUSED_VAR(p)  BigFree(address); }
+const ISzAlloc g_MidAlloc = { SzMidAlloc, SzMidFree };
+const ISzAlloc g_BigAlloc = { SzBigAlloc, SzBigFree };
 #endif
+
+/*
+  uintptr_t : <stdint.h> C99 (optional)
+            : unsupported in VS6
+*/
+
+#ifdef _WIN32
+  typedef UINT_PTR UIntPtr;
+#else
+  /*
+  typedef uintptr_t UIntPtr;
+  */
+  typedef ptrdiff_t UIntPtr;
+#endif
+
+
+#define ADJUST_ALLOC_SIZE 0
+/*
+#define ADJUST_ALLOC_SIZE (sizeof(void *) - 1)
+*/
+/*
+  Use (ADJUST_ALLOC_SIZE = (sizeof(void *) - 1)), if
+     MyAlloc() can return address that is NOT multiple of sizeof(void *).
+*/
+
+
+/*
+#define MY_ALIGN_PTR_DOWN(p, align) ((void *)((char *)(p) - ((size_t)(UIntPtr)(p) & ((align) - 1))))
+*/
+#define MY_ALIGN_PTR_DOWN(p, align) ((void *)((((UIntPtr)(p)) & ~((UIntPtr)(align) - 1))))
+
+
+#if !defined(_WIN32) && defined(_POSIX_C_SOURCE) && (_POSIX_C_SOURCE >= 200112L)
+  #define USE_posix_memalign
+#endif
+
+#ifndef USE_posix_memalign
+#define MY_ALIGN_PTR_UP_PLUS(p, align) MY_ALIGN_PTR_DOWN(((char *)(p) + (align) + ADJUST_ALLOC_SIZE), align)
+#endif
+
+/*
+  This posix_memalign() is for test purposes only.
+  We also need special Free() function instead of free(),
+  if this posix_memalign() is used.
+*/
+
+/*
+static int posix_memalign(void **ptr, size_t align, size_t size)
+{
+  size_t newSize = size + align;
+  void *p;
+  void *pAligned;
+  *ptr = NULL;
+  if (newSize < size)
+    return 12; // ENOMEM
+  p = MyAlloc(newSize);
+  if (!p)
+    return 12; // ENOMEM
+  pAligned = MY_ALIGN_PTR_UP_PLUS(p, align);
+  ((void **)pAligned)[-1] = p;
+  *ptr = pAligned;
+  return 0;
+}
+*/
+
+/*
+  ALLOC_ALIGN_SIZE >= sizeof(void *)
+  ALLOC_ALIGN_SIZE >= cache_line_size
+*/
+
+#define ALLOC_ALIGN_SIZE ((size_t)1 << 7)
+
+static void *SzAlignedAlloc(ISzAllocPtr pp, size_t size)
+{
+  #ifndef USE_posix_memalign
+  
+  void *p;
+  void *pAligned;
+  size_t newSize;
+  UNUSED_VAR(pp)
+
+  /* also we can allocate additional dummy ALLOC_ALIGN_SIZE bytes after aligned
+     block to prevent cache line sharing with another allocated blocks */
+
+  newSize = size + ALLOC_ALIGN_SIZE * 1 + ADJUST_ALLOC_SIZE;
+  if (newSize < size)
+    return NULL;
+
+  p = MyAlloc(newSize);
+  
+  if (!p)
+    return NULL;
+  pAligned = MY_ALIGN_PTR_UP_PLUS(p, ALLOC_ALIGN_SIZE);
+
+  Print(" size="); PrintHex(size, 8);
+  Print(" a_size="); PrintHex(newSize, 8);
+  Print(" ptr="); PrintAddr(p);
+  Print(" a_ptr="); PrintAddr(pAligned);
+  PrintLn();
+
+  ((void **)pAligned)[-1] = p;
+
+  return pAligned;
+
+  #else
+
+  void *p;
+  UNUSED_VAR(pp)
+  if (posix_memalign(&p, ALLOC_ALIGN_SIZE, size))
+    return NULL;
+
+  Print(" posix_memalign="); PrintAddr(p);
+  PrintLn();
+
+  return p;
+
+  #endif
+}
+
+
+static void SzAlignedFree(ISzAllocPtr pp, void *address)
+{
+  UNUSED_VAR(pp)
+  #ifndef USE_posix_memalign
+  if (address)
+    MyFree(((void **)address)[-1]);
+  #else
+  free(address);
+  #endif
+}
+
+
+const ISzAlloc g_AlignedAlloc = { SzAlignedAlloc, SzAlignedFree };
+
+
+
+#define MY_ALIGN_PTR_DOWN_1(p) MY_ALIGN_PTR_DOWN(p, sizeof(void *))
+
+/* we align ptr to support cases where CAlignOffsetAlloc::offset is not multiply of sizeof(void *) */
+#define REAL_BLOCK_PTR_VAR(p) ((void **)MY_ALIGN_PTR_DOWN_1(p))[-1]
+/*
+#define REAL_BLOCK_PTR_VAR(p) ((void **)(p))[-1]
+*/
+
+static void *AlignOffsetAlloc_Alloc(ISzAllocPtr pp, size_t size)
+{
+  const CAlignOffsetAlloc *p = Z7_CONTAINER_FROM_VTBL_CONST(pp, CAlignOffsetAlloc, vt);
+  void *adr;
+  void *pAligned;
+  size_t newSize;
+  size_t extra;
+  size_t alignSize = (size_t)1 << p->numAlignBits;
+
+  if (alignSize < sizeof(void *))
+    alignSize = sizeof(void *);
+  
+  if (p->offset >= alignSize)
+    return NULL;
+
+  /* also we can allocate additional dummy ALLOC_ALIGN_SIZE bytes after aligned
+     block to prevent cache line sharing with another allocated blocks */
+  extra = p->offset & (sizeof(void *) - 1);
+  newSize = size + alignSize + extra + ADJUST_ALLOC_SIZE;
+  if (newSize < size)
+    return NULL;
+
+  adr = ISzAlloc_Alloc(p->baseAlloc, newSize);
+  
+  if (!adr)
+    return NULL;
+
+  pAligned = (char *)MY_ALIGN_PTR_DOWN((char *)adr +
+      alignSize - p->offset + extra + ADJUST_ALLOC_SIZE, alignSize) + p->offset;
+
+  PrintLn();
+  Print("- Aligned: ");
+  Print(" size="); PrintHex(size, 8);
+  Print(" a_size="); PrintHex(newSize, 8);
+  Print(" ptr="); PrintAddr(adr);
+  Print(" a_ptr="); PrintAddr(pAligned);
+  PrintLn();
+
+  REAL_BLOCK_PTR_VAR(pAligned) = adr;
+
+  return pAligned;
+}
+
+
+static void AlignOffsetAlloc_Free(ISzAllocPtr pp, void *address)
+{
+  if (address)
+  {
+    const CAlignOffsetAlloc *p = Z7_CONTAINER_FROM_VTBL_CONST(pp, CAlignOffsetAlloc, vt);
+    PrintLn();
+    Print("- Aligned Free: ");
+    PrintLn();
+    ISzAlloc_Free(p->baseAlloc, REAL_BLOCK_PTR_VAR(address));
+  }
+}
+
+
+void AlignOffsetAlloc_CreateVTable(CAlignOffsetAlloc *p)
+{
+  p->vt.Alloc = AlignOffsetAlloc_Alloc;
+  p->vt.Free = AlignOffsetAlloc_Free;
+}

C/Alloc.h

@@ -1,21 +1,32 @@
 /* Alloc.h -- Memory allocation functions
-2009-02-07 : Igor Pavlov : Public domain */
+2023-03-04 : Igor Pavlov : Public domain */
 
-#ifndef __COMMON_ALLOC_H
-#define __COMMON_ALLOC_H
+#ifndef ZIP7_INC_ALLOC_H
+#define ZIP7_INC_ALLOC_H
 
-#include <stddef.h>
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
+
+/*
+  MyFree(NULL)        : is allowed, as free(NULL)
+  MyAlloc(0)          : returns NULL : but malloc(0)        is allowed to return NULL or non_NULL
+  MyRealloc(NULL, 0)  : returns NULL : but realloc(NULL, 0) is allowed to return NULL or non_NULL
+MyRealloc() is similar to realloc() for the following cases:
+  MyRealloc(non_NULL, 0)         : returns NULL and always calls MyFree(ptr)
+  MyRealloc(NULL, non_ZERO)      : returns NULL, if allocation failed
+  MyRealloc(non_NULL, non_ZERO)  : returns NULL, if reallocation failed
+*/
 
 void *MyAlloc(size_t size);
 void MyFree(void *address);
+void *MyRealloc(void *address, size_t size);
 
 #ifdef _WIN32
 
-void SetLargePageSize();
+#ifdef Z7_LARGE_PAGES
+void SetLargePageSize(void);
+#endif
 
 void *MidAlloc(size_t size);
 void MidFree(void *address);
@@ -31,8 +42,30 @@ void BigFree(void *address);
 
 #endif
 
-#ifdef __cplusplus
-}
+extern const ISzAlloc g_Alloc;
+
+#ifdef _WIN32
+extern const ISzAlloc g_BigAlloc;
+extern const ISzAlloc g_MidAlloc;
+#else
+#define g_BigAlloc g_AlignedAlloc
+#define g_MidAlloc g_AlignedAlloc
 #endif
 
+extern const ISzAlloc g_AlignedAlloc;
+
+
+typedef struct
+{
+  ISzAlloc vt;
+  ISzAllocPtr baseAlloc;
+  unsigned numAlignBits; /* ((1 << numAlignBits) >= sizeof(void *)) */
+  size_t offset;         /* (offset == (k * sizeof(void *)) && offset < (1 << numAlignBits) */
+} CAlignOffsetAlloc;
+
+void AlignOffsetAlloc_CreateVTable(CAlignOffsetAlloc *p);
+
+
+EXTERN_C_END
+
 #endif

C/Bcj2.c

@@ -1,132 +1,290 @@
-/* Bcj2.c -- Converter for x86 code (BCJ2)
-2008-10-04 : Igor Pavlov : Public domain */
+/* Bcj2.c -- BCJ2 Decoder (Converter for x86 code)
+2023-03-01 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "Bcj2.h"
+#include "CpuArch.h"
 
-#ifdef _LZMA_PROB32
-#define CProb UInt32
-#else
-#define CProb UInt16
-#endif
-
-#define IsJcc(b0, b1) ((b0) == 0x0F && ((b1) & 0xF0) == 0x80)
-#define IsJ(b0, b1) ((b1 & 0xFE) == 0xE8 || IsJcc(b0, b1))
-
-#define kNumTopBits 24
-#define kTopValue ((UInt32)1 << kNumTopBits)
-
+#define kTopValue ((UInt32)1 << 24)
 #define kNumBitModelTotalBits 11
 #define kBitModelTotal (1 << kNumBitModelTotalBits)
 #define kNumMoveBits 5
 
-#define RC_READ_BYTE (*buffer++)
-#define RC_TEST { if (buffer == bufferLim) return SZ_ERROR_DATA; }
-#define RC_INIT2 code = 0; range = 0xFFFFFFFF; \
-  { int i; for (i = 0; i < 5; i++) { RC_TEST; code = (code << 8) | RC_READ_BYTE; }}
+// UInt32 bcj2_stats[256 + 2][2];
 
-#define NORMALIZE if (range < kTopValue) { RC_TEST; range <<= 8; code = (code << 8) | RC_READ_BYTE; }
-
-#define IF_BIT_0(p) ttt = *(p); bound = (range >> kNumBitModelTotalBits) * ttt; if (code < bound)
-#define UPDATE_0(p) range = bound; *(p) = (CProb)(ttt + ((kBitModelTotal - ttt) >> kNumMoveBits)); NORMALIZE;
-#define UPDATE_1(p) range -= bound; code -= bound; *(p) = (CProb)(ttt - (ttt >> kNumMoveBits)); NORMALIZE;
-
-int Bcj2_Decode(
-    const Byte *buf0, SizeT size0,
-    const Byte *buf1, SizeT size1,
-    const Byte *buf2, SizeT size2,
-    const Byte *buf3, SizeT size3,
-    Byte *outBuf, SizeT outSize)
+void Bcj2Dec_Init(CBcj2Dec *p)
 {
-  CProb p[256 + 2];
-  SizeT inPos = 0, outPos = 0;
+  unsigned i;
+  p->state = BCJ2_STREAM_RC; // BCJ2_DEC_STATE_OK;
+  p->ip = 0;
+  p->temp = 0;
+  p->range = 0;
+  p->code = 0;
+  for (i = 0; i < sizeof(p->probs) / sizeof(p->probs[0]); i++)
+    p->probs[i] = kBitModelTotal >> 1;
+}
 
-  const Byte *buffer, *bufferLim;
-  UInt32 range, code;
-  Byte prevByte = 0;
-
-  unsigned int i;
-  for (i = 0; i < sizeof(p) / sizeof(p[0]); i++)
-    p[i] = kBitModelTotal >> 1;
-
-  buffer = buf3;
-  bufferLim = buffer + size3;
-  RC_INIT2
-
-  if (outSize == 0)
-    return SZ_OK;
-
-  for (;;)
+SRes Bcj2Dec_Decode(CBcj2Dec *p)
+{
+  UInt32 v = p->temp;
+  // const Byte *src;
+  if (p->range <= 5)
   {
-    Byte b;
-    CProb *prob;
-    UInt32 bound;
-    UInt32 ttt;
-
-    SizeT limit = size0 - inPos;
-    if (outSize - outPos < limit)
-      limit = outSize - outPos;
-    while (limit != 0)
+    UInt32 code = p->code;
+    p->state = BCJ2_DEC_STATE_ERROR; /* for case if we return SZ_ERROR_DATA; */
+    for (; p->range != 5; p->range++)
     {
-      Byte b = buf0[inPos];
-      outBuf[outPos++] = b;
-      if (IsJ(prevByte, b))
-        break;
-      inPos++;
-      prevByte = b;
-      limit--;
-    }
-
-    if (limit == 0 || outPos == outSize)
-      break;
-
-    b = buf0[inPos++];
-
-    if (b == 0xE8)
-      prob = p + prevByte;
-    else if (b == 0xE9)
-      prob = p + 256;
-    else
-      prob = p + 257;
-
-    IF_BIT_0(prob)
-    {
-      UPDATE_0(prob)
-      prevByte = b;
-    }
-    else
-    {
-      UInt32 dest;
-      const Byte *v;
-      UPDATE_1(prob)
-      if (b == 0xE8)
+      if (p->range == 1 && code != 0)
+        return SZ_ERROR_DATA;
+      if (p->bufs[BCJ2_STREAM_RC] == p->lims[BCJ2_STREAM_RC])
       {
-        v = buf1;
-        if (size1 < 4)
-          return SZ_ERROR_DATA;
-        buf1 += 4;
-        size1 -= 4;
+        p->state = BCJ2_STREAM_RC;
+        return SZ_OK;
       }
-      else
+      code = (code << 8) | *(p->bufs[BCJ2_STREAM_RC])++;
+      p->code = code;
+    }
+    if (code == 0xffffffff)
+      return SZ_ERROR_DATA;
+    p->range = 0xffffffff;
+  }
+  // else
+  {
+    unsigned state = p->state;
+    // we check BCJ2_IS_32BIT_STREAM() here instead of check in the main loop
+    if (BCJ2_IS_32BIT_STREAM(state))
+    {
+      const Byte *cur = p->bufs[state];
+      if (cur == p->lims[state])
+        return SZ_OK;
+      p->bufs[state] = cur + 4;
       {
-        v = buf2;
-        if (size2 < 4)
-          return SZ_ERROR_DATA;
-        buf2 += 4;
-        size2 -= 4;
+        const UInt32 ip = p->ip + 4;
+        v = GetBe32a(cur) - ip;
+        p->ip = ip;
+      }
+      state = BCJ2_DEC_STATE_ORIG_0;
+    }
+    if ((unsigned)(state - BCJ2_DEC_STATE_ORIG_0) < 4)
+    {
+      Byte *dest = p->dest;
+      for (;;)
+      {
+        if (dest == p->destLim)
+        {
+          p->state = state;
+          p->temp = v;
+          return SZ_OK;
+        }
+        *dest++ = (Byte)v;
+        p->dest = dest;
+        if (++state == BCJ2_DEC_STATE_ORIG_3 + 1)
+          break;
+        v >>= 8;
       }
-      dest = (((UInt32)v[0] << 24) | ((UInt32)v[1] << 16) |
-          ((UInt32)v[2] << 8) | ((UInt32)v[3])) - ((UInt32)outPos + 4);
-      outBuf[outPos++] = (Byte)dest;
-      if (outPos == outSize)
-        break;
-      outBuf[outPos++] = (Byte)(dest >> 8);
-      if (outPos == outSize)
-        break;
-      outBuf[outPos++] = (Byte)(dest >> 16);
-      if (outPos == outSize)
-        break;
-      outBuf[outPos++] = prevByte = (Byte)(dest >> 24);
     }
   }
-  return (outPos == outSize) ? SZ_OK : SZ_ERROR_DATA;
+
+  // src = p->bufs[BCJ2_STREAM_MAIN];
+  for (;;)
+  {
+    /*
+    if (BCJ2_IS_32BIT_STREAM(p->state))
+      p->state = BCJ2_DEC_STATE_OK;
+    else
+    */
+    {
+      if (p->range < kTopValue)
+      {
+        if (p->bufs[BCJ2_STREAM_RC] == p->lims[BCJ2_STREAM_RC])
+        {
+          p->state = BCJ2_STREAM_RC;
+          p->temp = v;
+          return SZ_OK;
+        }
+        p->range <<= 8;
+        p->code = (p->code << 8) | *(p->bufs[BCJ2_STREAM_RC])++;
+      }
+      {
+        const Byte *src = p->bufs[BCJ2_STREAM_MAIN];
+        const Byte *srcLim;
+        Byte *dest = p->dest;
+        {
+          const SizeT rem = (SizeT)(p->lims[BCJ2_STREAM_MAIN] - src);
+          SizeT num = (SizeT)(p->destLim - dest);
+          if (num >= rem)
+            num = rem;
+        #define NUM_ITERS 4
+        #if (NUM_ITERS & (NUM_ITERS - 1)) == 0
+          num &= ~((SizeT)NUM_ITERS - 1);   // if (NUM_ITERS == (1 << x))
+        #else
+          num -= num % NUM_ITERS; // if (NUM_ITERS != (1 << x))
+        #endif
+          srcLim = src + num;
+        }
+
+        #define NUM_SHIFT_BITS  24
+        #define ONE_ITER(indx) { \
+          const unsigned b = src[indx]; \
+          *dest++ = (Byte)b; \
+          v = (v << NUM_SHIFT_BITS) | b; \
+          if (((b + (0x100 - 0xe8)) & 0xfe) == 0) break; \
+          if (((v - (((UInt32)0x0f << (NUM_SHIFT_BITS)) + 0x80)) & \
+              ((((UInt32)1 << (4 + NUM_SHIFT_BITS)) - 0x1) << 4)) == 0) break; \
+            /* ++dest */; /* v = b; */ }
+          
+        if (src != srcLim)
+        for (;;)
+        {
+            /* The dependency chain of 2-cycle for (v) calculation is not big problem here.
+               But we can remove dependency chain with v = b in the end of loop. */
+          ONE_ITER(0)
+          #if (NUM_ITERS > 1)
+            ONE_ITER(1)
+          #if (NUM_ITERS > 2)
+            ONE_ITER(2)
+          #if (NUM_ITERS > 3)
+            ONE_ITER(3)
+          #if (NUM_ITERS > 4)
+            ONE_ITER(4)
+          #if (NUM_ITERS > 5)
+            ONE_ITER(5)
+          #if (NUM_ITERS > 6)
+            ONE_ITER(6)
+          #if (NUM_ITERS > 7)
+            ONE_ITER(7)
+          #endif
+          #endif
+          #endif
+          #endif
+          #endif
+          #endif
+          #endif
+          
+          src += NUM_ITERS;
+          if (src == srcLim)
+            break;
+        }
+
+        if (src == srcLim)
+      #if (NUM_ITERS > 1)
+        for (;;)
+      #endif
+        {
+        #if (NUM_ITERS > 1)
+          if (src == p->lims[BCJ2_STREAM_MAIN] || dest == p->destLim)
+        #endif
+          {
+            const SizeT num = (SizeT)(src - p->bufs[BCJ2_STREAM_MAIN]);
+            p->bufs[BCJ2_STREAM_MAIN] = src;
+            p->dest = dest;
+            p->ip += (UInt32)num;
+            /* state BCJ2_STREAM_MAIN has more priority than BCJ2_STATE_ORIG */
+            p->state =
+              src == p->lims[BCJ2_STREAM_MAIN] ?
+                (unsigned)BCJ2_STREAM_MAIN :
+                (unsigned)BCJ2_DEC_STATE_ORIG;
+            p->temp = v;
+            return SZ_OK;
+          }
+        #if (NUM_ITERS > 1)
+          ONE_ITER(0)
+          src++;
+        #endif
+        }
+
+        {
+          const SizeT num = (SizeT)(dest - p->dest);
+          p->dest = dest; // p->dest += num;
+          p->bufs[BCJ2_STREAM_MAIN] += num; // = src;
+          p->ip += (UInt32)num;
+        }
+        {
+          UInt32 bound, ttt;
+          CBcj2Prob *prob; // unsigned index;
+          /*
+          prob = p->probs + (unsigned)((Byte)v == 0xe8 ?
+              2 + (Byte)(v >> 8) :
+              ((v >> 5) & 1));  // ((Byte)v < 0xe8 ? 0 : 1));
+          */
+          {
+            const unsigned c = ((v + 0x17) >> 6) & 1;
+            prob = p->probs + (unsigned)
+                (((0 - c) & (Byte)(v >> NUM_SHIFT_BITS)) + c + ((v >> 5) & 1));
+                // (Byte)
+                // 8x->0     : e9->1     : xxe8->xx+2
+                // 8x->0x100 : e9->0x101 : xxe8->xx
+                // (((0x100 - (e & ~v)) & (0x100 | (v >> 8))) + (e & v));
+                // (((0x101 + (~e | v)) & (0x100 | (v >> 8))) + (e & v));
+          }
+          ttt = *prob;
+          bound = (p->range >> kNumBitModelTotalBits) * ttt;
+          if (p->code < bound)
+          {
+            // bcj2_stats[prob - p->probs][0]++;
+            p->range = bound;
+            *prob = (CBcj2Prob)(ttt + ((kBitModelTotal - ttt) >> kNumMoveBits));
+            continue;
+          }
+          {
+            // bcj2_stats[prob - p->probs][1]++;
+            p->range -= bound;
+            p->code -= bound;
+            *prob = (CBcj2Prob)(ttt - (ttt >> kNumMoveBits));
+          }
+        }
+      }
+    }
+    {
+      /* (v == 0xe8 ? 0 : 1) uses setcc instruction with additional zero register usage in x64 MSVC. */
+      // const unsigned cj = ((Byte)v == 0xe8) ? BCJ2_STREAM_CALL : BCJ2_STREAM_JUMP;
+      const unsigned cj = (((v + 0x57) >> 6) & 1) + BCJ2_STREAM_CALL;
+      const Byte *cur = p->bufs[cj];
+      Byte *dest;
+      SizeT rem;
+      if (cur == p->lims[cj])
+      {
+        p->state = cj;
+        break;
+      }
+      v = GetBe32a(cur);
+      p->bufs[cj] = cur + 4;
+      {
+        const UInt32 ip = p->ip + 4;
+        v -= ip;
+        p->ip = ip;
+      }
+      dest = p->dest;
+      rem = (SizeT)(p->destLim - dest);
+      if (rem < 4)
+      {
+        if ((unsigned)rem > 0) { dest[0] = (Byte)v;  v >>= 8;
+        if ((unsigned)rem > 1) { dest[1] = (Byte)v;  v >>= 8;
+        if ((unsigned)rem > 2) { dest[2] = (Byte)v;  v >>= 8; }}}
+        p->temp = v;
+        p->dest = dest + rem;
+        p->state = BCJ2_DEC_STATE_ORIG_0 + (unsigned)rem;
+        break;
+      }
+      SetUi32(dest, v)
+      v >>= 24;
+      p->dest = dest + 4;
+    }
+  }
+
+  if (p->range < kTopValue && p->bufs[BCJ2_STREAM_RC] != p->lims[BCJ2_STREAM_RC])
+  {
+    p->range <<= 8;
+    p->code = (p->code << 8) | *(p->bufs[BCJ2_STREAM_RC])++;
+  }
+  return SZ_OK;
 }
+
+#undef NUM_ITERS
+#undef ONE_ITER
+#undef NUM_SHIFT_BITS
+#undef kTopValue
+#undef kNumBitModelTotalBits
+#undef kBitModelTotal
+#undef kNumMoveBits

C/Bcj2.h

@@ -1,38 +1,332 @@
-/* Bcj2.h -- Converter for x86 code (BCJ2)
-2009-02-07 : Igor Pavlov : Public domain */
+/* Bcj2.h -- BCJ2 converter for x86 code (Branch CALL/JUMP variant2)
+2023-03-02 : Igor Pavlov : Public domain */
 
-#ifndef __BCJ2_H
-#define __BCJ2_H
+#ifndef ZIP7_INC_BCJ2_H
+#define ZIP7_INC_BCJ2_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
+
+#define BCJ2_NUM_STREAMS 4
+
+enum
+{
+  BCJ2_STREAM_MAIN,
+  BCJ2_STREAM_CALL,
+  BCJ2_STREAM_JUMP,
+  BCJ2_STREAM_RC
+};
+
+enum
+{
+  BCJ2_DEC_STATE_ORIG_0 = BCJ2_NUM_STREAMS,
+  BCJ2_DEC_STATE_ORIG_1,
+  BCJ2_DEC_STATE_ORIG_2,
+  BCJ2_DEC_STATE_ORIG_3,
+  
+  BCJ2_DEC_STATE_ORIG,
+  BCJ2_DEC_STATE_ERROR     /* after detected data error */
+};
+
+enum
+{
+  BCJ2_ENC_STATE_ORIG = BCJ2_NUM_STREAMS,
+  BCJ2_ENC_STATE_FINISHED  /* it's state after fully encoded stream */
+};
+
+
+/* #define BCJ2_IS_32BIT_STREAM(s) ((s) == BCJ2_STREAM_CALL || (s) == BCJ2_STREAM_JUMP) */
+#define BCJ2_IS_32BIT_STREAM(s) ((unsigned)((unsigned)(s) - (unsigned)BCJ2_STREAM_CALL) < 2)
 
 /*
-Conditions:
-  outSize <= FullOutputSize,
-  where FullOutputSize is full size of output stream of x86_2 filter.
-
-If buf0 overlaps outBuf, there are two required conditions:
-  1) (buf0 >= outBuf)
-  2) (buf0 + size0 >= outBuf + FullOutputSize).
-
-Returns:
-  SZ_OK
-  SZ_ERROR_DATA - Data error
+CBcj2Dec / CBcj2Enc
+bufs sizes:
+  BUF_SIZE(n) = lims[n] - bufs[n]
+bufs sizes for BCJ2_STREAM_CALL and BCJ2_STREAM_JUMP must be multiply of 4:
+    (BUF_SIZE(BCJ2_STREAM_CALL) & 3) == 0
+    (BUF_SIZE(BCJ2_STREAM_JUMP) & 3) == 0
 */
 
-int Bcj2_Decode(
-    const Byte *buf0, SizeT size0,
-    const Byte *buf1, SizeT size1,
-    const Byte *buf2, SizeT size2,
-    const Byte *buf3, SizeT size3,
-    Byte *outBuf, SizeT outSize);
+// typedef UInt32 CBcj2Prob;
+typedef UInt16 CBcj2Prob;
 
-#ifdef __cplusplus
-}
-#endif
+/*
+BCJ2 encoder / decoder internal requirements:
+  - If last bytes of stream contain marker (e8/e8/0f8x), then
+    there is also encoded symbol (0 : no conversion) in RC stream.
+  - One case of overlapped instructions is supported,
+    if last byte of converted instruction is (0f) and next byte is (8x):
+      marker [xx xx xx 0f] 8x
+    then the pair (0f 8x) is treated as marker.
+*/
+
+/* ---------- BCJ2 Decoder ---------- */
+
+/*
+CBcj2Dec:
+(dest) is allowed to overlap with bufs[BCJ2_STREAM_MAIN], with the following conditions:
+  bufs[BCJ2_STREAM_MAIN] >= dest &&
+  bufs[BCJ2_STREAM_MAIN] - dest >=
+        BUF_SIZE(BCJ2_STREAM_CALL) +
+        BUF_SIZE(BCJ2_STREAM_JUMP)
+  reserve = bufs[BCJ2_STREAM_MAIN] - dest -
+      ( BUF_SIZE(BCJ2_STREAM_CALL) +
+        BUF_SIZE(BCJ2_STREAM_JUMP) )
+  and additional conditions:
+  if (it's first call of Bcj2Dec_Decode() after Bcj2Dec_Init())
+  {
+    (reserve != 1) : if (ver <  v23.00)
+  }
+  else // if there are more than one calls of Bcj2Dec_Decode() after Bcj2Dec_Init())
+  {
+    (reserve >= 6) : if (ver <  v23.00)
+    (reserve >= 4) : if (ver >= v23.00)
+    We need that (reserve) because after first call of Bcj2Dec_Decode(),
+    CBcj2Dec::temp can contain up to 4 bytes for writing to (dest).
+  }
+  (reserve == 0) is allowed, if we decode full stream via single call of Bcj2Dec_Decode().
+  (reserve == 0) also is allowed in case of multi-call, if we use fixed buffers,
+     and (reserve) is calculated from full (final) sizes of all streams before first call.
+*/
+
+typedef struct
+{
+  const Byte *bufs[BCJ2_NUM_STREAMS];
+  const Byte *lims[BCJ2_NUM_STREAMS];
+  Byte *dest;
+  const Byte *destLim;
+
+  unsigned state; /* BCJ2_STREAM_MAIN has more priority than BCJ2_STATE_ORIG */
+
+  UInt32 ip;      /* property of starting base for decoding */
+  UInt32 temp;    /* Byte temp[4]; */
+  UInt32 range;
+  UInt32 code;
+  CBcj2Prob probs[2 + 256];
+} CBcj2Dec;
+
+
+/* Note:
+   Bcj2Dec_Init() sets (CBcj2Dec::ip = 0)
+   if (ip != 0) property is required, the caller must set CBcj2Dec::ip after Bcj2Dec_Init()
+*/
+void Bcj2Dec_Init(CBcj2Dec *p);
+
+
+/* Bcj2Dec_Decode():
+   returns:
+     SZ_OK
+     SZ_ERROR_DATA : if data in 5 starting bytes of BCJ2_STREAM_RC stream are not correct
+*/
+SRes Bcj2Dec_Decode(CBcj2Dec *p);
+
+/* To check that decoding was finished you can compare
+   sizes of processed streams with sizes known from another sources.
+   You must do at least one mandatory check from the two following options:
+      - the check for size of processed output (ORIG) stream.
+      - the check for size of processed input  (MAIN) stream.
+   additional optional checks:
+      - the checks for processed sizes of all input streams (MAIN, CALL, JUMP, RC)
+      - the checks Bcj2Dec_IsMaybeFinished*()
+   also before actual decoding you can check that the
+   following condition is met for stream sizes:
+     ( size(ORIG) == size(MAIN) + size(CALL) + size(JUMP) )
+*/
+
+/* (state == BCJ2_STREAM_MAIN) means that decoder is ready for
+      additional input data in BCJ2_STREAM_MAIN stream.
+   Note that (state == BCJ2_STREAM_MAIN) is allowed for non-finished decoding.
+*/
+#define Bcj2Dec_IsMaybeFinished_state_MAIN(_p_) ((_p_)->state == BCJ2_STREAM_MAIN)
+
+/* if the stream decoding was finished correctly, then range decoder
+   part of CBcj2Dec also was finished, and then (CBcj2Dec::code == 0).
+   Note that (CBcj2Dec::code == 0) is allowed for non-finished decoding.
+*/
+#define Bcj2Dec_IsMaybeFinished_code(_p_) ((_p_)->code == 0)
+
+/* use Bcj2Dec_IsMaybeFinished() only as additional check
+    after at least one mandatory check from the two following options:
+      - the check for size of processed output (ORIG) stream.
+      - the check for size of processed input  (MAIN) stream.
+*/
+#define Bcj2Dec_IsMaybeFinished(_p_) ( \
+        Bcj2Dec_IsMaybeFinished_state_MAIN(_p_) && \
+        Bcj2Dec_IsMaybeFinished_code(_p_))
+
+
+
+/* ---------- BCJ2 Encoder ---------- */
+
+typedef enum
+{
+  BCJ2_ENC_FINISH_MODE_CONTINUE,
+  BCJ2_ENC_FINISH_MODE_END_BLOCK,
+  BCJ2_ENC_FINISH_MODE_END_STREAM
+} EBcj2Enc_FinishMode;
+
+/*
+  BCJ2_ENC_FINISH_MODE_CONTINUE:
+     process non finished encoding.
+     It notifies the encoder that additional further calls
+     can provide more input data (src) than provided by current call.
+     In  that case the CBcj2Enc encoder still can move (src) pointer
+     up to (srcLim), but CBcj2Enc encoder can store some of the last
+     processed bytes (up to 4 bytes) from src to internal CBcj2Enc::temp[] buffer.
+   at return:
+       (CBcj2Enc::src will point to position that includes
+       processed data and data copied to (temp[]) buffer)
+       That data from (temp[]) buffer will be used in further calls.
+  
+  BCJ2_ENC_FINISH_MODE_END_BLOCK:
+     finish encoding of current block (ended at srcLim) without RC flushing.
+   at return: if (CBcj2Enc::state == BCJ2_ENC_STATE_ORIG) &&
+                  CBcj2Enc::src == CBcj2Enc::srcLim)
+        :  it shows that block encoding was finished. And the encoder is
+           ready for new (src) data or for stream finish operation.
+     finished block means
+     {
+       CBcj2Enc has completed block encoding up to (srcLim).
+       (1 + 4 bytes) or (2 + 4 bytes) CALL/JUMP cortages will
+       not cross block boundary at (srcLim).
+       temporary CBcj2Enc buffer for (ORIG) src data is empty.
+       3 output uncompressed streams (MAIN, CALL, JUMP) were flushed.
+       RC stream was not flushed. And RC stream will cross block boundary.
+     }
+     Note: some possible implementation of BCJ2 encoder could
+     write branch marker (e8/e8/0f8x) in one call of Bcj2Enc_Encode(),
+     and it could calculate symbol for RC in another call of Bcj2Enc_Encode().
+     BCJ2 encoder uses ip/fileIp/fileSize/relatLimit values to calculate RC symbol.
+     And these CBcj2Enc variables can have different values in different Bcj2Enc_Encode() calls.
+     So caller must finish each block with BCJ2_ENC_FINISH_MODE_END_BLOCK
+     to ensure that RC symbol is calculated and written in proper block.
+    
+  BCJ2_ENC_FINISH_MODE_END_STREAM
+     finish encoding of stream (ended at srcLim) fully including RC flushing.
+   at return: if (CBcj2Enc::state == BCJ2_ENC_STATE_FINISHED)
+        : it shows that stream encoding was finished fully,
+          and all output streams were flushed fully.
+     also Bcj2Enc_IsFinished() can be called.
+*/
+
+
+/*
+  32-bit relative offset in JUMP/CALL commands is
+    - (mod 4 GiB)  for 32-bit x86 code
+    - signed Int32 for 64-bit x86-64 code
+  BCJ2 encoder also does internal relative to absolute address conversions.
+  And there are 2 possible ways to do it:
+    before v23: we used 32-bit variables and (mod 4 GiB) conversion
+    since  v23: we use  64-bit variables and (signed Int32 offset) conversion.
+  The absolute address condition for conversion in v23:
+    ((UInt64)((Int64)ip64 - (Int64)fileIp64 + 5 + (Int32)offset) < (UInt64)fileSize64)
+  note that if (fileSize64 > 2 GiB). there is difference between
+  old (mod 4 GiB) way (v22) and new (signed Int32 offset) way (v23).
+  And new (v23) way is more suitable to encode 64-bit x86-64 code for (fileSize64 > 2 GiB) cases.
+*/
+
+/*
+// for old (v22) way for conversion:
+typedef UInt32 CBcj2Enc_ip_unsigned;
+typedef  Int32 CBcj2Enc_ip_signed;
+#define BCJ2_ENC_FileSize_MAX ((UInt32)1 << 31)
+*/
+typedef UInt64 CBcj2Enc_ip_unsigned;
+typedef  Int64 CBcj2Enc_ip_signed;
+
+/* maximum size of file that can be used for conversion condition */
+#define BCJ2_ENC_FileSize_MAX             ((CBcj2Enc_ip_unsigned)0 - 2)
+
+/* default value of fileSize64_minus1 variable that means
+   that absolute address limitation will not be used */
+#define BCJ2_ENC_FileSizeField_UNLIMITED  ((CBcj2Enc_ip_unsigned)0 - 1)
+
+/* calculate value that later can be set to CBcj2Enc::fileSize64_minus1 */
+#define BCJ2_ENC_GET_FileSizeField_VAL_FROM_FileSize(fileSize) \
+    ((CBcj2Enc_ip_unsigned)(fileSize) - 1)
+
+/* set CBcj2Enc::fileSize64_minus1 variable from size of file */
+#define Bcj2Enc_SET_FileSize(p, fileSize) \
+    (p)->fileSize64_minus1 = BCJ2_ENC_GET_FileSizeField_VAL_FROM_FileSize(fileSize);
+
+
+typedef struct
+{
+  Byte *bufs[BCJ2_NUM_STREAMS];
+  const Byte *lims[BCJ2_NUM_STREAMS];
+  const Byte *src;
+  const Byte *srcLim;
+
+  unsigned state;
+  EBcj2Enc_FinishMode finishMode;
+
+  Byte context;
+  Byte flushRem;
+  Byte isFlushState;
+
+  Byte cache;
+  UInt32 range;
+  UInt64 low;
+  UInt64 cacheSize;
+  
+  // UInt32 context;  // for marker version, it can include marker flag.
+
+  /* (ip64) and (fileIp64) correspond to virtual source stream position
+     that doesn't include data in temp[] */
+  CBcj2Enc_ip_unsigned ip64;         /* current (ip) position */
+  CBcj2Enc_ip_unsigned fileIp64;     /* start (ip) position of current file */
+  CBcj2Enc_ip_unsigned fileSize64_minus1;   /* size of current file (for conversion limitation) */
+  UInt32 relatLimit;  /* (relatLimit <= ((UInt32)1 << 31)) : 0 means disable_conversion */
+  // UInt32 relatExcludeBits;
+
+  UInt32 tempTarget;
+  unsigned tempPos; /* the number of bytes that were copied to temp[] buffer
+                       (tempPos <= 4) outside of Bcj2Enc_Encode() */
+  // Byte temp[4]; // for marker version
+  Byte temp[8];
+  CBcj2Prob probs[2 + 256];
+} CBcj2Enc;
+
+void Bcj2Enc_Init(CBcj2Enc *p);
+
+
+/*
+Bcj2Enc_Encode(): at exit:
+  p->State <  BCJ2_NUM_STREAMS    : we need more buffer space for output stream
+                                    (bufs[p->State] == lims[p->State])
+  p->State == BCJ2_ENC_STATE_ORIG : we need more data in input src stream
+                                    (src == srcLim)
+  p->State == BCJ2_ENC_STATE_FINISHED : after fully encoded stream
+*/
+void Bcj2Enc_Encode(CBcj2Enc *p);
+
+/* Bcj2Enc encoder can look ahead for up 4 bytes of source stream.
+   CBcj2Enc::tempPos : is the number of bytes that were copied from input stream to temp[] buffer.
+   (CBcj2Enc::src) after Bcj2Enc_Encode() is starting position after
+   fully processed data and after data copied to temp buffer.
+   So if the caller needs to get real number of fully processed input
+   bytes (without look ahead data in temp buffer),
+   the caller must subtruct (CBcj2Enc::tempPos) value from processed size
+   value that is calculated based on current (CBcj2Enc::src):
+     cur_processed_pos = Calc_Big_Processed_Pos(enc.src)) -
+        Bcj2Enc_Get_AvailInputSize_in_Temp(&enc);
+*/
+/* get the size of input data that was stored in temp[] buffer: */
+#define Bcj2Enc_Get_AvailInputSize_in_Temp(p) ((p)->tempPos)
+
+#define Bcj2Enc_IsFinished(p) ((p)->flushRem == 0)
+
+/* Note : the decoder supports overlapping of marker (0f 80).
+   But we can eliminate such overlapping cases by setting
+   the limit for relative offset conversion as
+     CBcj2Enc::relatLimit <= (0x0f << 24) == (240 MiB)
+*/
+/* default value for CBcj2Enc::relatLimit */
+#define BCJ2_ENC_RELAT_LIMIT_DEFAULT  ((UInt32)0x0f << 24)
+#define BCJ2_ENC_RELAT_LIMIT_MAX      ((UInt32)1 << 31)
+// #define BCJ2_RELAT_EXCLUDE_NUM_BITS 5
+
+EXTERN_C_END
 
 #endif

C/Bcj2Enc.c

@@ -0,0 +1,506 @@
+/* Bcj2Enc.c -- BCJ2 Encoder converter for x86 code (Branch CALL/JUMP variant2)
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
+
+/* #define SHOW_STAT */
+#ifdef SHOW_STAT
+#include <stdio.h>
+#define PRF2(s) printf("%s ip=%8x  tempPos=%d  src= %8x\n", s, (unsigned)p->ip64, p->tempPos, (unsigned)(p->srcLim - p->src));
+#else
+#define PRF2(s)
+#endif
+
+#include "Bcj2.h"
+#include "CpuArch.h"
+
+#define kTopValue ((UInt32)1 << 24)
+#define kNumBitModelTotalBits 11
+#define kBitModelTotal (1 << kNumBitModelTotalBits)
+#define kNumMoveBits 5
+
+void Bcj2Enc_Init(CBcj2Enc *p)
+{
+  unsigned i;
+  p->state = BCJ2_ENC_STATE_ORIG;
+  p->finishMode = BCJ2_ENC_FINISH_MODE_CONTINUE;
+  p->context = 0;
+  p->flushRem = 5;
+  p->isFlushState = 0;
+  p->cache = 0;
+  p->range = 0xffffffff;
+  p->low = 0;
+  p->cacheSize = 1;
+  p->ip64 = 0;
+  p->fileIp64 = 0;
+  p->fileSize64_minus1 = BCJ2_ENC_FileSizeField_UNLIMITED;
+  p->relatLimit = BCJ2_ENC_RELAT_LIMIT_DEFAULT;
+  // p->relatExcludeBits = 0;
+  p->tempPos = 0;
+  for (i = 0; i < sizeof(p->probs) / sizeof(p->probs[0]); i++)
+    p->probs[i] = kBitModelTotal >> 1;
+}
+
+// Z7_NO_INLINE
+Z7_FORCE_INLINE
+static BoolInt Bcj2_RangeEnc_ShiftLow(CBcj2Enc *p)
+{
+  const UInt32 low = (UInt32)p->low;
+  const unsigned high = (unsigned)
+    #if defined(Z7_MSC_VER_ORIGINAL) \
+        && defined(MY_CPU_X86) \
+        && defined(MY_CPU_LE) \
+        && !defined(MY_CPU_64BIT)
+      // we try to rid of __aullshr() call in MSVS-x86
+      (((const UInt32 *)&p->low)[1]); // [1] : for little-endian only
+    #else
+      (p->low >> 32);
+    #endif
+  if (low < (UInt32)0xff000000 || high != 0)
+  {
+    Byte *buf = p->bufs[BCJ2_STREAM_RC];
+    do
+    {
+      if (buf == p->lims[BCJ2_STREAM_RC])
+      {
+        p->state = BCJ2_STREAM_RC;
+        p->bufs[BCJ2_STREAM_RC] = buf;
+        return True;
+      }
+      *buf++ = (Byte)(p->cache + high);
+      p->cache = 0xff;
+    }
+    while (--p->cacheSize);
+    p->bufs[BCJ2_STREAM_RC] = buf;
+    p->cache = (Byte)(low >> 24);
+  }
+  p->cacheSize++;
+  p->low = low << 8;
+  return False;
+}
+
+
+/*
+We can use 2 alternative versions of code:
+1) non-marker version:
+  Byte CBcj2Enc::context
+  Byte temp[8];
+  Last byte of marker (e8/e9/[0f]8x) can be written to temp[] buffer.
+  Encoder writes last byte of marker (e8/e9/[0f]8x) to dest, only in conjunction
+  with writing branch symbol to range coder in same Bcj2Enc_Encode_2() call.
+
+2) marker version:
+  UInt32 CBcj2Enc::context
+  Byte CBcj2Enc::temp[4];
+  MARKER_FLAG in CBcj2Enc::context shows that CBcj2Enc::context contains finded marker.
+  it's allowed that
+    one call of Bcj2Enc_Encode_2() writes last byte of marker (e8/e9/[0f]8x) to dest,
+    and another call of Bcj2Enc_Encode_2() does offset conversion.
+    So different values of (fileIp) and (fileSize) are possible
+    in these different Bcj2Enc_Encode_2() calls.
+
+Also marker version requires additional if((v & MARKER_FLAG) == 0) check in main loop.
+So we use non-marker version.
+*/
+
+/*
+  Corner cases with overlap in multi-block.
+  before v23: there was one corner case, where converted instruction
+    could start in one sub-stream and finish in next sub-stream.
+  If multi-block (solid) encoding is used,
+    and BCJ2_ENC_FINISH_MODE_END_BLOCK is used for each sub-stream.
+    and (0f) is last byte of previous sub-stream
+    and (8x) is first byte of current sub-stream
+  then (0f 8x) pair is treated as marker by BCJ2 encoder and decoder.
+  BCJ2 encoder can converts 32-bit offset for that (0f 8x) cortage,
+  if that offset meets limit requirements.
+  If encoder allows 32-bit offset conversion for such overlap case,
+  then the data in 3 uncompressed BCJ2 streams for some sub-stream
+  can depend from data of previous sub-stream.
+  That corner case is not big problem, and it's rare case.
+  Since v23.00 we do additional check to prevent conversions in such overlap cases.
+*/
+
+/*
+  Bcj2Enc_Encode_2() output variables at exit:
+  {
+    if (Bcj2Enc_Encode_2() exits with (p->state == BCJ2_ENC_STATE_ORIG))
+    {
+      it means that encoder needs more input data.
+      if (p->srcLim == p->src) at exit, then
+      {
+        (p->finishMode != BCJ2_ENC_FINISH_MODE_END_STREAM)
+        all input data were read and processed, and we are ready for
+        new input data.
+      }
+      else
+      {
+        (p->srcLim != p->src)
+        (p->finishMode == BCJ2_ENC_FINISH_MODE_CONTINUE)
+          The encoder have found e8/e9/0f_8x marker,
+          and p->src points to last byte of that marker,
+          Bcj2Enc_Encode_2() needs more input data to get totally
+          5 bytes (last byte of marker and 32-bit branch offset)
+          as continuous array starting from p->src.
+        (p->srcLim - p->src < 5) requirement is met after exit.
+          So non-processed resedue from p->src to p->srcLim is always less than 5 bytes.
+      }
+    }
+  }
+*/
+
+Z7_NO_INLINE
+static void Bcj2Enc_Encode_2(CBcj2Enc *p)
+{
+  if (!p->isFlushState)
+  {
+    const Byte *src;
+    UInt32 v;
+    {
+      const unsigned state = p->state;
+      if (BCJ2_IS_32BIT_STREAM(state))
+      {
+        Byte *cur = p->bufs[state];
+        if (cur == p->lims[state])
+          return;
+        SetBe32a(cur, p->tempTarget)
+        p->bufs[state] = cur + 4;
+      }
+    }
+    p->state = BCJ2_ENC_STATE_ORIG; // for main reason of exit
+    src = p->src;
+    v = p->context;
+    
+    // #define WRITE_CONTEXT  p->context = v; // for marker version
+    #define WRITE_CONTEXT           p->context = (Byte)v;
+    #define WRITE_CONTEXT_AND_SRC   p->src = src;  WRITE_CONTEXT
+
+    for (;;)
+    {
+      // const Byte *src;
+      // UInt32 v;
+      CBcj2Enc_ip_unsigned ip;
+      if (p->range < kTopValue)
+      {
+        // to reduce register pressure and code size: we save and restore local variables.
+        WRITE_CONTEXT_AND_SRC
+        if (Bcj2_RangeEnc_ShiftLow(p))
+          return;
+        p->range <<= 8;
+        src = p->src;
+        v = p->context;
+      }
+      // src = p->src;
+      // #define MARKER_FLAG  ((UInt32)1 << 17)
+      // if ((v & MARKER_FLAG) == 0) // for marker version
+      {
+        const Byte *srcLim;
+        Byte *dest = p->bufs[BCJ2_STREAM_MAIN];
+        {
+          const SizeT remSrc = (SizeT)(p->srcLim - src);
+          SizeT rem = (SizeT)(p->lims[BCJ2_STREAM_MAIN] - dest);
+          if (rem >= remSrc)
+            rem = remSrc;
+          srcLim = src + rem;
+        }
+        /* p->context contains context of previous byte:
+           bits [0 : 7]  : src[-1], if (src) was changed in this call
+           bits [8 : 31] : are undefined for non-marker version
+        */
+        // v = p->context;
+        #define NUM_SHIFT_BITS  24
+        #define CONV_FLAG  ((UInt32)1 << 16)
+        #define ONE_ITER { \
+          b = src[0]; \
+          *dest++ = (Byte)b; \
+          v = (v << NUM_SHIFT_BITS) | b; \
+          if (((b + (0x100 - 0xe8)) & 0xfe) == 0) break; \
+          if (((v - (((UInt32)0x0f << (NUM_SHIFT_BITS)) + 0x80)) & \
+              ((((UInt32)1 << (4 + NUM_SHIFT_BITS)) - 0x1) << 4)) == 0) break; \
+          src++; if (src == srcLim) { break; } }
+
+        if (src != srcLim)
+        for (;;)
+        {
+          /* clang can generate ineffective code with setne instead of two jcc instructions.
+             we can use 2 iterations and external (unsigned b) to avoid that ineffective code genaration. */
+          unsigned b;
+          ONE_ITER
+          ONE_ITER
+        }
+        
+        ip = p->ip64 + (CBcj2Enc_ip_unsigned)(SizeT)(dest - p->bufs[BCJ2_STREAM_MAIN]);
+        p->bufs[BCJ2_STREAM_MAIN] = dest;
+        p->ip64 = ip;
+
+        if (src == srcLim)
+        {
+          WRITE_CONTEXT_AND_SRC
+          if (src != p->srcLim)
+          {
+            p->state = BCJ2_STREAM_MAIN;
+            return;
+          }
+          /* (p->src == p->srcLim)
+          (p->state == BCJ2_ENC_STATE_ORIG) */
+          if (p->finishMode != BCJ2_ENC_FINISH_MODE_END_STREAM)
+            return;
+          /* (p->finishMode == BCJ2_ENC_FINISH_MODE_END_STREAM */
+          // (p->flushRem == 5);
+          p->isFlushState = 1;
+          break;
+        }
+        src++;
+        // p->src = src;
+      }
+      // ip = p->ip; // for marker version
+      /* marker was found */
+      /* (v) contains marker that was found:
+           bits [NUM_SHIFT_BITS : NUM_SHIFT_BITS + 7]
+                         : value of src[-2] : xx/xx/0f
+           bits [0 : 7]  : value of src[-1] : e8/e9/8x
+      */
+      {
+        {
+        #if NUM_SHIFT_BITS != 24
+          v &= ~(UInt32)CONV_FLAG;
+        #endif
+          // UInt32 relat = 0;
+          if ((SizeT)(p->srcLim - src) >= 4)
+          {
+            /*
+            if (relat != 0 || (Byte)v != 0xe8)
+            BoolInt isBigOffset = True;
+            */
+            const UInt32 relat = GetUi32(src);
+            /*
+            #define EXCLUDE_FLAG  ((UInt32)1 << 4)
+            #define NEED_CONVERT(rel) ((((rel) + EXCLUDE_FLAG) & (0 - EXCLUDE_FLAG * 2)) != 0)
+            if (p->relatExcludeBits != 0)
+            {
+              const UInt32 flag = (UInt32)1 << (p->relatExcludeBits - 1);
+              isBigOffset = (((relat + flag) & (0 - flag * 2)) != 0);
+            }
+            // isBigOffset = False; // for debug
+            */
+            ip -= p->fileIp64;
+            // Use the following if check, if (ip) is 64-bit:
+            if (ip > (((v + 0x20) >> 5) & 1))  // 23.00 : we eliminate milti-block overlap for (Of 80) and (e8/e9)
+            if ((CBcj2Enc_ip_unsigned)((CBcj2Enc_ip_signed)ip + 4 + (Int32)relat) <= p->fileSize64_minus1)
+            if (((UInt32)(relat + p->relatLimit) >> 1) < p->relatLimit)
+              v |= CONV_FLAG;
+          }
+          else if (p->finishMode == BCJ2_ENC_FINISH_MODE_CONTINUE)
+          {
+            // (p->srcLim - src < 4)
+            // /*
+            // for non-marker version
+            p->ip64--; // p->ip = ip - 1;
+            p->bufs[BCJ2_STREAM_MAIN]--;
+            src--;
+            v >>= NUM_SHIFT_BITS;
+            // (0 < p->srcLim - p->src <= 4)
+            // */
+            // v |= MARKER_FLAG; // for marker version
+            /* (p->state == BCJ2_ENC_STATE_ORIG) */
+            WRITE_CONTEXT_AND_SRC
+            return;
+          }
+          {
+            const unsigned c = ((v + 0x17) >> 6) & 1;
+            CBcj2Prob *prob = p->probs + (unsigned)
+                (((0 - c) & (Byte)(v >> NUM_SHIFT_BITS)) + c + ((v >> 5) & 1));
+            /*
+                ((Byte)v == 0xe8 ? 2 + ((Byte)(v >> 8)) :
+                ((Byte)v < 0xe8 ? 0 : 1));  // ((v >> 5) & 1));
+            */
+            const unsigned ttt = *prob;
+            const UInt32 bound = (p->range >> kNumBitModelTotalBits) * ttt;
+            if ((v & CONV_FLAG) == 0)
+            {
+              // static int yyy = 0; yyy++; printf("\n!needConvert = %d\n", yyy);
+              // v = (Byte)v; // for marker version
+              p->range = bound;
+              *prob = (CBcj2Prob)(ttt + ((kBitModelTotal - ttt) >> kNumMoveBits));
+              // WRITE_CONTEXT_AND_SRC
+              continue;
+            }
+            p->low += bound;
+            p->range -= bound;
+            *prob = (CBcj2Prob)(ttt - (ttt >> kNumMoveBits));
+          }
+          // p->context = src[3];
+          {
+            // const unsigned cj = ((Byte)v == 0xe8 ? BCJ2_STREAM_CALL : BCJ2_STREAM_JUMP);
+            const unsigned cj = (((v + 0x57) >> 6) & 1) + BCJ2_STREAM_CALL;
+            ip = p->ip64;
+            v = GetUi32(src); // relat
+            ip += 4;
+            p->ip64 = ip;
+            src += 4;
+            // p->src = src;
+            {
+              const UInt32 absol = (UInt32)ip + v;
+              Byte *cur = p->bufs[cj];
+              v >>= 24;
+              // WRITE_CONTEXT
+              if (cur == p->lims[cj])
+              {
+                p->state = cj;
+                p->tempTarget = absol;
+                WRITE_CONTEXT_AND_SRC
+                return;
+              }
+              SetBe32a(cur, absol)
+              p->bufs[cj] = cur + 4;
+            }
+          }
+        }
+      }
+    } // end of loop
+  }
+
+  for (; p->flushRem != 0; p->flushRem--)
+    if (Bcj2_RangeEnc_ShiftLow(p))
+      return;
+  p->state = BCJ2_ENC_STATE_FINISHED;
+}
+
+
+/*
+BCJ2 encoder needs look ahead for up to 4 bytes in (src) buffer.
+So base function Bcj2Enc_Encode_2()
+  in BCJ2_ENC_FINISH_MODE_CONTINUE mode can return with
+  (p->state == BCJ2_ENC_STATE_ORIG && p->src < p->srcLim)
+Bcj2Enc_Encode() solves that look ahead problem by using p->temp[] buffer.
+  so if (p->state == BCJ2_ENC_STATE_ORIG) after Bcj2Enc_Encode(),
+    then (p->src == p->srcLim).
+  And the caller's code is simpler with Bcj2Enc_Encode().
+*/
+
+Z7_NO_INLINE
+void Bcj2Enc_Encode(CBcj2Enc *p)
+{
+  PRF2("\n----")
+  if (p->tempPos != 0)
+  {
+    /* extra: number of bytes that were copied from (src) to (temp) buffer in this call */
+    unsigned extra = 0;
+    /* We will touch only minimal required number of bytes in input (src) stream.
+       So we will add input bytes from (src) stream to temp[] with step of 1 byte.
+       We don't add new bytes to temp[] before Bcj2Enc_Encode_2() call
+         in first loop iteration because
+         - previous call of Bcj2Enc_Encode() could use another (finishMode),
+         - previous call could finish with (p->state != BCJ2_ENC_STATE_ORIG).
+       the case with full temp[] buffer (p->tempPos == 4) is possible here.
+    */
+    for (;;)
+    {
+      // (0 < p->tempPos <= 5) // in non-marker version
+      /* p->src : the current src data position including extra bytes
+                  that were copied to temp[] buffer in this call */
+      const Byte *src = p->src;
+      const Byte *srcLim = p->srcLim;
+      const EBcj2Enc_FinishMode finishMode = p->finishMode;
+      if (src != srcLim)
+      {
+        /* if there are some src data after the data copied to temp[],
+           then we use MODE_CONTINUE for temp data */
+        p->finishMode = BCJ2_ENC_FINISH_MODE_CONTINUE;
+      }
+      p->src = p->temp;
+      p->srcLim = p->temp + p->tempPos;
+      PRF2("    ")
+      Bcj2Enc_Encode_2(p);
+      {
+        const unsigned num = (unsigned)(p->src - p->temp);
+        const unsigned tempPos = p->tempPos - num;
+        unsigned i;
+        p->tempPos = tempPos;
+        for (i = 0; i < tempPos; i++)
+          p->temp[i] = p->temp[(SizeT)i + num];
+        // tempPos : number of bytes in temp buffer
+        p->src = src;
+        p->srcLim = srcLim;
+        p->finishMode = finishMode;
+        if (p->state != BCJ2_ENC_STATE_ORIG)
+        {
+          // (p->tempPos <= 4) // in non-marker version
+          /* if (the reason of exit from Bcj2Enc_Encode_2()
+                 is not BCJ2_ENC_STATE_ORIG),
+             then we exit from Bcj2Enc_Encode() with same reason */
+          // optional code begin : we rollback (src) and tempPos, if it's possible:
+          if (extra >= tempPos)
+            extra = tempPos;
+          p->src = src - extra;
+          p->tempPos = tempPos - extra;
+          // optional code end : rollback of (src) and tempPos
+          return;
+        }
+        /* (p->tempPos <= 4)
+           (p->state == BCJ2_ENC_STATE_ORIG)
+             so encoder needs more data than in temp[] */
+        if (src == srcLim)
+          return; // src buffer has no more input data.
+        /* (src != srcLim)
+           so we can provide more input data from src for Bcj2Enc_Encode_2() */
+        if (extra >= tempPos)
+        {
+          /* (extra >= tempPos) means that temp buffer contains
+             only data from src buffer of this call.
+             So now we can encode without temp buffer */
+          p->src = src - tempPos; // rollback (src)
+          p->tempPos = 0;
+          break;
+        }
+        // we append one additional extra byte from (src) to temp[] buffer:
+        p->temp[tempPos] = *src;
+        p->tempPos = tempPos + 1;
+        // (0 < p->tempPos <= 5) // in non-marker version
+        p->src = src + 1;
+        extra++;
+      }
+    }
+  }
+
+  PRF2("++++")
+  // (p->tempPos == 0)
+  Bcj2Enc_Encode_2(p);
+  PRF2("====")
+  
+  if (p->state == BCJ2_ENC_STATE_ORIG)
+  {
+    const Byte *src = p->src;
+    const Byte *srcLim = p->srcLim;
+    const unsigned rem = (unsigned)(srcLim - src);
+    /* (rem <= 4) here.
+       if (p->src != p->srcLim), then
+         - we copy non-processed bytes from (p->src) to temp[] buffer,
+         - we set p->src equal to p->srcLim.
+    */
+    if (rem)
+    {
+      unsigned i = 0;
+      p->src = srcLim;
+      p->tempPos = rem;
+      // (0 < p->tempPos <= 4)
+      do
+        p->temp[i] = src[i];
+      while (++i != rem);
+    }
+    // (p->tempPos <= 4)
+    // (p->src == p->srcLim)
+  }
+}
+
+#undef PRF2
+#undef CONV_FLAG
+#undef MARKER_FLAG
+#undef WRITE_CONTEXT
+#undef WRITE_CONTEXT_AND_SRC
+#undef ONE_ITER
+#undef NUM_SHIFT_BITS
+#undef kTopValue
+#undef kNumBitModelTotalBits
+#undef kBitModelTotal
+#undef kNumMoveBits

C/Blake2.h

@@ -0,0 +1,48 @@
+/* Blake2.h -- BLAKE2 Hash
+2023-03-04 : Igor Pavlov : Public domain
+2015 : Samuel Neves : Public domain */
+
+#ifndef ZIP7_INC_BLAKE2_H
+#define ZIP7_INC_BLAKE2_H
+
+#include "7zTypes.h"
+
+EXTERN_C_BEGIN
+
+#define BLAKE2S_BLOCK_SIZE 64
+#define BLAKE2S_DIGEST_SIZE 32
+#define BLAKE2SP_PARALLEL_DEGREE 8
+
+typedef struct
+{
+  UInt32 h[8];
+  UInt32 t[2];
+  UInt32 f[2];
+  Byte buf[BLAKE2S_BLOCK_SIZE];
+  UInt32 bufPos;
+  UInt32 lastNode_f1;
+  UInt32 dummy[2]; /* for sizeof(CBlake2s) alignment */
+} CBlake2s;
+
+/* You need to xor CBlake2s::h[i] with input parameter block after Blake2s_Init0() */
+/*
+void Blake2s_Init0(CBlake2s *p);
+void Blake2s_Update(CBlake2s *p, const Byte *data, size_t size);
+void Blake2s_Final(CBlake2s *p, Byte *digest);
+*/
+
+
+typedef struct
+{
+  CBlake2s S[BLAKE2SP_PARALLEL_DEGREE];
+  unsigned bufPos;
+} CBlake2sp;
+
+
+void Blake2sp_Init(CBlake2sp *p);
+void Blake2sp_Update(CBlake2sp *p, const Byte *data, size_t size);
+void Blake2sp_Final(CBlake2sp *p, Byte *digest);
+
+EXTERN_C_END
+
+#endif

C/Blake2s.c

@@ -0,0 +1,250 @@
+/* Blake2s.c -- BLAKE2s and BLAKE2sp Hash
+2023-03-04 : Igor Pavlov : Public domain
+2015 : Samuel Neves : Public domain */
+
+#include "Precomp.h"
+
+#include <string.h>
+
+#include "Blake2.h"
+#include "CpuArch.h"
+#include "RotateDefs.h"
+
+#define rotr32 rotrFixed
+
+#define BLAKE2S_NUM_ROUNDS 10
+#define BLAKE2S_FINAL_FLAG (~(UInt32)0)
+
+static const UInt32 k_Blake2s_IV[8] =
+{
+  0x6A09E667UL, 0xBB67AE85UL, 0x3C6EF372UL, 0xA54FF53AUL,
+  0x510E527FUL, 0x9B05688CUL, 0x1F83D9ABUL, 0x5BE0CD19UL
+};
+
+static const Byte k_Blake2s_Sigma[BLAKE2S_NUM_ROUNDS][16] =
+{
+  {  0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15 } ,
+  { 14, 10,  4,  8,  9, 15, 13,  6,  1, 12,  0,  2, 11,  7,  5,  3 } ,
+  { 11,  8, 12,  0,  5,  2, 15, 13, 10, 14,  3,  6,  7,  1,  9,  4 } ,
+  {  7,  9,  3,  1, 13, 12, 11, 14,  2,  6,  5, 10,  4,  0, 15,  8 } ,
+  {  9,  0,  5,  7,  2,  4, 10, 15, 14,  1, 11, 12,  6,  8,  3, 13 } ,
+  {  2, 12,  6, 10,  0, 11,  8,  3,  4, 13,  7,  5, 15, 14,  1,  9 } ,
+  { 12,  5,  1, 15, 14, 13,  4, 10,  0,  7,  6,  3,  9,  2,  8, 11 } ,
+  { 13, 11,  7, 14, 12,  1,  3,  9,  5,  0, 15,  4,  8,  6,  2, 10 } ,
+  {  6, 15, 14,  9, 11,  3,  0,  8, 12,  2, 13,  7,  1,  4, 10,  5 } ,
+  { 10,  2,  8,  4,  7,  6,  1,  5, 15, 11,  9, 14,  3, 12, 13 , 0 } ,
+};
+
+
+static void Blake2s_Init0(CBlake2s *p)
+{
+  unsigned i;
+  for (i = 0; i < 8; i++)
+    p->h[i] = k_Blake2s_IV[i];
+  p->t[0] = 0;
+  p->t[1] = 0;
+  p->f[0] = 0;
+  p->f[1] = 0;
+  p->bufPos = 0;
+  p->lastNode_f1 = 0;
+}
+
+
+static void Blake2s_Compress(CBlake2s *p)
+{
+  UInt32 m[16];
+  UInt32 v[16];
+  
+  {
+    unsigned i;
+    
+    for (i = 0; i < 16; i++)
+      m[i] = GetUi32(p->buf + i * sizeof(m[i]));
+    
+    for (i = 0; i < 8; i++)
+      v[i] = p->h[i];
+  }
+
+  v[ 8] = k_Blake2s_IV[0];
+  v[ 9] = k_Blake2s_IV[1];
+  v[10] = k_Blake2s_IV[2];
+  v[11] = k_Blake2s_IV[3];
+  
+  v[12] = p->t[0] ^ k_Blake2s_IV[4];
+  v[13] = p->t[1] ^ k_Blake2s_IV[5];
+  v[14] = p->f[0] ^ k_Blake2s_IV[6];
+  v[15] = p->f[1] ^ k_Blake2s_IV[7];
+
+  #define G(r,i,a,b,c,d) \
+    a += b + m[sigma[2*i+0]];  d ^= a; d = rotr32(d, 16);  c += d;  b ^= c; b = rotr32(b, 12); \
+    a += b + m[sigma[2*i+1]];  d ^= a; d = rotr32(d,  8);  c += d;  b ^= c; b = rotr32(b,  7); \
+
+  #define R(r) \
+    G(r,0,v[ 0],v[ 4],v[ 8],v[12]) \
+    G(r,1,v[ 1],v[ 5],v[ 9],v[13]) \
+    G(r,2,v[ 2],v[ 6],v[10],v[14]) \
+    G(r,3,v[ 3],v[ 7],v[11],v[15]) \
+    G(r,4,v[ 0],v[ 5],v[10],v[15]) \
+    G(r,5,v[ 1],v[ 6],v[11],v[12]) \
+    G(r,6,v[ 2],v[ 7],v[ 8],v[13]) \
+    G(r,7,v[ 3],v[ 4],v[ 9],v[14]) \
+
+  {
+    unsigned r;
+    for (r = 0; r < BLAKE2S_NUM_ROUNDS; r++)
+    {
+      const Byte *sigma = k_Blake2s_Sigma[r];
+      R(r)
+    }
+    /* R(0); R(1); R(2); R(3); R(4); R(5); R(6); R(7); R(8); R(9); */
+  }
+
+  #undef G
+  #undef R
+
+  {
+    unsigned i;
+    for (i = 0; i < 8; i++)
+      p->h[i] ^= v[i] ^ v[i + 8];
+  }
+}
+
+
+#define Blake2s_Increment_Counter(S, inc) \
+  { p->t[0] += (inc); p->t[1] += (p->t[0] < (inc)); }
+
+#define Blake2s_Set_LastBlock(p) \
+  { p->f[0] = BLAKE2S_FINAL_FLAG; p->f[1] = p->lastNode_f1; }
+
+
+static void Blake2s_Update(CBlake2s *p, const Byte *data, size_t size)
+{
+  while (size != 0)
+  {
+    unsigned pos = (unsigned)p->bufPos;
+    unsigned rem = BLAKE2S_BLOCK_SIZE - pos;
+
+    if (size <= rem)
+    {
+      memcpy(p->buf + pos, data, size);
+      p->bufPos += (UInt32)size;
+      return;
+    }
+
+    memcpy(p->buf + pos, data, rem);
+    Blake2s_Increment_Counter(S, BLAKE2S_BLOCK_SIZE)
+    Blake2s_Compress(p);
+    p->bufPos = 0;
+    data += rem;
+    size -= rem;
+  }
+}
+
+
+static void Blake2s_Final(CBlake2s *p, Byte *digest)
+{
+  unsigned i;
+
+  Blake2s_Increment_Counter(S, (UInt32)p->bufPos)
+  Blake2s_Set_LastBlock(p)
+  memset(p->buf + p->bufPos, 0, BLAKE2S_BLOCK_SIZE - p->bufPos);
+  Blake2s_Compress(p);
+
+  for (i = 0; i < 8; i++)
+  {
+    SetUi32(digest + sizeof(p->h[i]) * i, p->h[i])
+  }
+}
+
+
+/* ---------- BLAKE2s ---------- */
+
+/* we need to xor CBlake2s::h[i] with input parameter block after Blake2s_Init0() */
+/*
+typedef struct
+{
+  Byte  digest_length;
+  Byte  key_length;
+  Byte  fanout;
+  Byte  depth;
+  UInt32 leaf_length;
+  Byte  node_offset[6];
+  Byte  node_depth;
+  Byte  inner_length;
+  Byte  salt[BLAKE2S_SALTBYTES];
+  Byte  personal[BLAKE2S_PERSONALBYTES];
+} CBlake2sParam;
+*/
+
+
+static void Blake2sp_Init_Spec(CBlake2s *p, unsigned node_offset, unsigned node_depth)
+{
+  Blake2s_Init0(p);
+  
+  p->h[0] ^= (BLAKE2S_DIGEST_SIZE | ((UInt32)BLAKE2SP_PARALLEL_DEGREE << 16) | ((UInt32)2 << 24));
+  p->h[2] ^= ((UInt32)node_offset);
+  p->h[3] ^= ((UInt32)node_depth << 16) | ((UInt32)BLAKE2S_DIGEST_SIZE << 24);
+  /*
+  P->digest_length = BLAKE2S_DIGEST_SIZE;
+  P->key_length = 0;
+  P->fanout = BLAKE2SP_PARALLEL_DEGREE;
+  P->depth = 2;
+  P->leaf_length = 0;
+  store48(P->node_offset, node_offset);
+  P->node_depth = node_depth;
+  P->inner_length = BLAKE2S_DIGEST_SIZE;
+  */
+}
+
+
+void Blake2sp_Init(CBlake2sp *p)
+{
+  unsigned i;
+  
+  p->bufPos = 0;
+
+  for (i = 0; i < BLAKE2SP_PARALLEL_DEGREE; i++)
+    Blake2sp_Init_Spec(&p->S[i], i, 0);
+
+  p->S[BLAKE2SP_PARALLEL_DEGREE - 1].lastNode_f1 = BLAKE2S_FINAL_FLAG;
+}
+
+
+void Blake2sp_Update(CBlake2sp *p, const Byte *data, size_t size)
+{
+  unsigned pos = p->bufPos;
+  while (size != 0)
+  {
+    unsigned index = pos / BLAKE2S_BLOCK_SIZE;
+    unsigned rem = BLAKE2S_BLOCK_SIZE - (pos & (BLAKE2S_BLOCK_SIZE - 1));
+    if (rem > size)
+      rem = (unsigned)size;
+    Blake2s_Update(&p->S[index], data, rem);
+    size -= rem;
+    data += rem;
+    pos += rem;
+    pos &= (BLAKE2S_BLOCK_SIZE * BLAKE2SP_PARALLEL_DEGREE - 1);
+  }
+  p->bufPos = pos;
+}
+
+
+void Blake2sp_Final(CBlake2sp *p, Byte *digest)
+{
+  CBlake2s R;
+  unsigned i;
+
+  Blake2sp_Init_Spec(&R, 0, 1);
+  R.lastNode_f1 = BLAKE2S_FINAL_FLAG;
+
+  for (i = 0; i < BLAKE2SP_PARALLEL_DEGREE; i++)
+  {
+    Byte hash[BLAKE2S_DIGEST_SIZE];
+    Blake2s_Final(&p->S[i], hash);
+    Blake2s_Update(&R, hash, BLAKE2S_DIGEST_SIZE);
+  }
+
+  Blake2s_Final(&R, digest);
+}
+
+#undef rotr32

C/Bra.c

@@ -1,133 +1,420 @@
-/* Bra.c -- Converters for RISC code
-2008-10-04 : Igor Pavlov : Public domain */
+/* Bra.c -- Branch converters for RISC code
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "Bra.h"
+#include "CpuArch.h"
+#include "RotateDefs.h"
 
-SizeT ARM_Convert(Byte *data, SizeT size, UInt32 ip, int encoding)
+#if defined(MY_CPU_SIZEOF_POINTER) \
+    && ( MY_CPU_SIZEOF_POINTER == 4 \
+      || MY_CPU_SIZEOF_POINTER == 8)
+  #define BR_CONV_USE_OPT_PC_PTR
+#endif
+
+#ifdef BR_CONV_USE_OPT_PC_PTR
+#define BR_PC_INIT  pc -= (UInt32)(SizeT)p;
+#define BR_PC_GET   (pc + (UInt32)(SizeT)p)
+#else
+#define BR_PC_INIT  pc += (UInt32)size;
+#define BR_PC_GET   (pc - (UInt32)(SizeT)(lim - p))
+// #define BR_PC_INIT
+// #define BR_PC_GET   (pc + (UInt32)(SizeT)(p - data))
+#endif
+
+#define BR_CONVERT_VAL(v, c) if (encoding) v += c; else v -= c;
+// #define BR_CONVERT_VAL(v, c) if (!encoding) c = (UInt32)0 - c; v += c;
+
+#define Z7_BRANCH_CONV(name) z7_BranchConv_ ## name
+
+#define Z7_BRANCH_FUNC_MAIN(name) \
+static \
+Z7_FORCE_INLINE \
+Z7_ATTRIB_NO_VECTOR \
+Byte *Z7_BRANCH_CONV(name)(Byte *p, SizeT size, UInt32 pc, int encoding)
+
+#define Z7_BRANCH_FUNC_IMP(name, m, encoding) \
+Z7_NO_INLINE \
+Z7_ATTRIB_NO_VECTOR \
+Byte *m(name)(Byte *data, SizeT size, UInt32 pc) \
+  { return Z7_BRANCH_CONV(name)(data, size, pc, encoding); } \
+
+#ifdef Z7_EXTRACT_ONLY
+#define Z7_BRANCH_FUNCS_IMP(name) \
+  Z7_BRANCH_FUNC_IMP(name, Z7_BRANCH_CONV_DEC, 0)
+#else
+#define Z7_BRANCH_FUNCS_IMP(name) \
+  Z7_BRANCH_FUNC_IMP(name, Z7_BRANCH_CONV_DEC, 0) \
+  Z7_BRANCH_FUNC_IMP(name, Z7_BRANCH_CONV_ENC, 1)
+#endif
+
+#if defined(__clang__)
+#define BR_EXTERNAL_FOR
+#define BR_NEXT_ITERATION  continue;
+#else
+#define BR_EXTERNAL_FOR    for (;;)
+#define BR_NEXT_ITERATION  break;
+#endif
+
+#if defined(__clang__) && (__clang_major__ >= 8) \
+  || defined(__GNUC__) && (__GNUC__ >= 1000) \
+  // GCC is not good for __builtin_expect() here
+  /* || defined(_MSC_VER) && (_MSC_VER >= 1920) */
+  // #define Z7_unlikely [[unlikely]]
+  // #define Z7_LIKELY(x)   (__builtin_expect((x), 1))
+  #define Z7_UNLIKELY(x) (__builtin_expect((x), 0))
+  // #define Z7_likely [[likely]]
+#else
+  // #define Z7_LIKELY(x)   (x)
+  #define Z7_UNLIKELY(x) (x)
+  // #define Z7_likely
+#endif
+
+
+Z7_BRANCH_FUNC_MAIN(ARM64)
 {
-  SizeT i;
-  if (size < 4)
-    return 0;
-  size -= 4;
-  ip += 8;
-  for (i = 0; i <= size; i += 4)
+  // Byte *p = data;
+  const Byte *lim;
+  const UInt32 flag = (UInt32)1 << (24 - 4);
+  const UInt32 mask = ((UInt32)1 << 24) - (flag << 1);
+  size &= ~(SizeT)3;
+  // if (size == 0) return p;
+  lim = p + size;
+  BR_PC_INIT
+  pc -= 4;  // because (p) will point to next instruction
+  
+  BR_EXTERNAL_FOR
   {
-    if (data[i + 3] == 0xEB)
+    // Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE
+    for (;;)
     {
-      UInt32 dest;
-      UInt32 src = ((UInt32)data[i + 2] << 16) | ((UInt32)data[i + 1] << 8) | (data[i + 0]);
-      src <<= 2;
-      if (encoding)
-        dest = ip + (UInt32)i + src;
-      else
-        dest = src - (ip + (UInt32)i);
-      dest >>= 2;
-      data[i + 2] = (Byte)(dest >> 16);
-      data[i + 1] = (Byte)(dest >> 8);
-      data[i + 0] = (Byte)dest;
+      UInt32 v;
+      if Z7_UNLIKELY(p == lim)
+        return p;
+      v = GetUi32a(p);
+      p += 4;
+      if Z7_UNLIKELY(((v - 0x94000000) & 0xfc000000) == 0)
+      {
+        UInt32 c = BR_PC_GET >> 2;
+        BR_CONVERT_VAL(v, c)
+        v &= 0x03ffffff;
+        v |= 0x94000000;
+        SetUi32a(p - 4, v)
+        BR_NEXT_ITERATION
+      }
+      // v = rotlFixed(v, 8);  v += (flag << 8) - 0x90;  if Z7_UNLIKELY((v & ((mask << 8) + 0x9f)) == 0)
+      v -= 0x90000000;  if Z7_UNLIKELY((v & 0x9f000000) == 0)
+      {
+        UInt32 z, c;
+        // v = rotrFixed(v, 8);
+        v += flag; if Z7_UNLIKELY(v & mask) continue;
+        z = (v & 0xffffffe0) | (v >> 26);
+        c = (BR_PC_GET >> (12 - 3)) & ~(UInt32)7;
+        BR_CONVERT_VAL(z, c)
+        v &= 0x1f;
+        v |= 0x90000000;
+        v |= z << 26;
+        v |= 0x00ffffe0 & ((z & (((flag << 1) - 1))) - flag);
+        SetUi32a(p - 4, v)
+      }
     }
   }
-  return i;
 }
+Z7_BRANCH_FUNCS_IMP(ARM64)
 
-SizeT ARMT_Convert(Byte *data, SizeT size, UInt32 ip, int encoding)
+
+Z7_BRANCH_FUNC_MAIN(ARM)
 {
-  SizeT i;
-  if (size < 4)
-    return 0;
-  size -= 4;
-  ip += 4;
-  for (i = 0; i <= size; i += 2)
+  // Byte *p = data;
+  const Byte *lim;
+  size &= ~(SizeT)3;
+  lim = p + size;
+  BR_PC_INIT
+  /* in ARM: branch offset is relative to the +2 instructions from current instruction.
+     (p) will point to next instruction */
+  pc += 8 - 4;
+  
+  for (;;)
   {
-    if ((data[i + 1] & 0xF8) == 0xF0 &&
-        (data[i + 3] & 0xF8) == 0xF8)
+    for (;;)
     {
-      UInt32 dest;
-      UInt32 src =
-        (((UInt32)data[i + 1] & 0x7) << 19) |
-        ((UInt32)data[i + 0] << 11) |
-        (((UInt32)data[i + 3] & 0x7) << 8) |
-        (data[i + 2]);
-      
-      src <<= 1;
-      if (encoding)
-        dest = ip + (UInt32)i + src;
-      else
-        dest = src - (ip + (UInt32)i);
-      dest >>= 1;
-      
-      data[i + 1] = (Byte)(0xF0 | ((dest >> 19) & 0x7));
-      data[i + 0] = (Byte)(dest >> 11);
-      data[i + 3] = (Byte)(0xF8 | ((dest >> 8) & 0x7));
-      data[i + 2] = (Byte)dest;
-      i += 2;
+      if Z7_UNLIKELY(p >= lim) { return p; }  p += 4;  if Z7_UNLIKELY(p[-1] == 0xeb) break;
+      if Z7_UNLIKELY(p >= lim) { return p; }  p += 4;  if Z7_UNLIKELY(p[-1] == 0xeb) break;
+    }
+    {
+      UInt32 v = GetUi32a(p - 4);
+      UInt32 c = BR_PC_GET >> 2;
+      BR_CONVERT_VAL(v, c)
+      v &= 0x00ffffff;
+      v |= 0xeb000000;
+      SetUi32a(p - 4, v)
     }
   }
-  return i;
 }
+Z7_BRANCH_FUNCS_IMP(ARM)
 
-SizeT PPC_Convert(Byte *data, SizeT size, UInt32 ip, int encoding)
+
+Z7_BRANCH_FUNC_MAIN(PPC)
 {
-  SizeT i;
-  if (size < 4)
-    return 0;
-  size -= 4;
-  for (i = 0; i <= size; i += 4)
+  // Byte *p = data;
+  const Byte *lim;
+  size &= ~(SizeT)3;
+  lim = p + size;
+  BR_PC_INIT
+  pc -= 4;  // because (p) will point to next instruction
+  
+  for (;;)
   {
-    if ((data[i] >> 2) == 0x12 && (data[i + 3] & 3) == 1)
+    UInt32 v;
+    for (;;)
     {
-      UInt32 src = ((UInt32)(data[i + 0] & 3) << 24) |
-        ((UInt32)data[i + 1] << 16) |
-        ((UInt32)data[i + 2] << 8) |
-        ((UInt32)data[i + 3] & (~3));
-      
-      UInt32 dest;
-      if (encoding)
-        dest = ip + (UInt32)i + src;
-      else
-        dest = src - (ip + (UInt32)i);
-      data[i + 0] = (Byte)(0x48 | ((dest >> 24) &  0x3));
-      data[i + 1] = (Byte)(dest >> 16);
-      data[i + 2] = (Byte)(dest >> 8);
-      data[i + 3] &= 0x3;
-      data[i + 3] |= dest;
+      if Z7_UNLIKELY(p == lim)
+        return p;
+      // v = GetBe32a(p);
+      v = *(UInt32 *)(void *)p;
+      p += 4;
+      // if ((v & 0xfc000003) == 0x48000001) break;
+      // if ((p[-4] & 0xFC) == 0x48 && (p[-1] & 3) == 1) break;
+      if Z7_UNLIKELY(
+          ((v - Z7_CONV_BE_TO_NATIVE_CONST32(0x48000001))
+              & Z7_CONV_BE_TO_NATIVE_CONST32(0xfc000003)) == 0) break;
+    }
+    {
+      v = Z7_CONV_NATIVE_TO_BE_32(v);
+      {
+        UInt32 c = BR_PC_GET;
+        BR_CONVERT_VAL(v, c)
+      }
+      v &= 0x03ffffff;
+      v |= 0x48000000;
+      SetBe32a(p - 4, v)
     }
   }
-  return i;
 }
+Z7_BRANCH_FUNCS_IMP(PPC)
 
-SizeT SPARC_Convert(Byte *data, SizeT size, UInt32 ip, int encoding)
+
+#ifdef Z7_CPU_FAST_ROTATE_SUPPORTED
+#define BR_SPARC_USE_ROTATE
+#endif
+
+Z7_BRANCH_FUNC_MAIN(SPARC)
 {
-  UInt32 i;
-  if (size < 4)
-    return 0;
-  size -= 4;
-  for (i = 0; i <= size; i += 4)
+  // Byte *p = data;
+  const Byte *lim;
+  const UInt32 flag = (UInt32)1 << 22;
+  size &= ~(SizeT)3;
+  lim = p + size;
+  BR_PC_INIT
+  pc -= 4;  // because (p) will point to next instruction
+  for (;;)
   {
-    if (data[i] == 0x40 && (data[i + 1] & 0xC0) == 0x00 ||
-        data[i] == 0x7F && (data[i + 1] & 0xC0) == 0xC0)
+    UInt32 v;
+    for (;;)
     {
-      UInt32 src =
-        ((UInt32)data[i + 0] << 24) |
-        ((UInt32)data[i + 1] << 16) |
-        ((UInt32)data[i + 2] << 8) |
-        ((UInt32)data[i + 3]);
-      UInt32 dest;
-      
-      src <<= 2;
-      if (encoding)
-        dest = ip + i + src;
-      else
-        dest = src - (ip + i);
-      dest >>= 2;
-      
-      dest = (((0 - ((dest >> 22) & 1)) << 22) & 0x3FFFFFFF) | (dest & 0x3FFFFF) | 0x40000000;
-
-      data[i + 0] = (Byte)(dest >> 24);
-      data[i + 1] = (Byte)(dest >> 16);
-      data[i + 2] = (Byte)(dest >> 8);
-      data[i + 3] = (Byte)dest;
+      if Z7_UNLIKELY(p == lim)
+        return p;
+      /* // the code without GetBe32a():
+      { const UInt32 v = GetUi16a(p) & 0xc0ff; p += 4; if (v == 0x40 || v == 0xc07f) break; }
+      */
+      v = GetBe32a(p);
+      p += 4;
+    #ifdef BR_SPARC_USE_ROTATE
+      v = rotlFixed(v, 2);
+      v += (flag << 2) - 1;
+      if Z7_UNLIKELY((v & (3 - (flag << 3))) == 0)
+    #else
+      v += (UInt32)5 << 29;
+      v ^= (UInt32)7 << 29;
+      v += flag;
+      if Z7_UNLIKELY((v & (0 - (flag << 1))) == 0)
+    #endif
+        break;
+    }
+    {
+      // UInt32 v = GetBe32a(p - 4);
+    #ifndef BR_SPARC_USE_ROTATE
+      v <<= 2;
+    #endif
+      {
+        UInt32 c = BR_PC_GET;
+        BR_CONVERT_VAL(v, c)
+      }
+      v &= (flag << 3) - 1;
+    #ifdef BR_SPARC_USE_ROTATE
+      v -= (flag << 2) - 1;
+      v = rotrFixed(v, 2);
+    #else
+      v -= (flag << 2);
+      v >>= 2;
+      v |= (UInt32)1 << 30;
+    #endif
+      SetBe32a(p - 4, v)
     }
   }
-  return i;
 }
+Z7_BRANCH_FUNCS_IMP(SPARC)
+
+
+Z7_BRANCH_FUNC_MAIN(ARMT)
+{
+  // Byte *p = data;
+  Byte *lim;
+  size &= ~(SizeT)1;
+  // if (size == 0) return p;
+  if (size <= 2) return p;
+  size -= 2;
+  lim = p + size;
+  BR_PC_INIT
+  /* in ARM: branch offset is relative to the +2 instructions from current instruction.
+     (p) will point to the +2 instructions from current instruction */
+  // pc += 4 - 4;
+  // if (encoding) pc -= 0xf800 << 1; else pc += 0xf800 << 1;
+  // #define ARMT_TAIL_PROC { goto armt_tail; }
+  #define ARMT_TAIL_PROC { return p; }
+  
+  do
+  {
+    /* in MSVC 32-bit x86 compilers:
+       UInt32 version : it loads value from memory with movzx
+       Byte   version : it loads value to 8-bit register (AL/CL)
+       movzx version is slightly faster in some cpus
+    */
+    unsigned b1;
+    // Byte / unsigned
+    b1 = p[1];
+    // optimized version to reduce one (p >= lim) check:
+    // unsigned a1 = p[1];  b1 = p[3];  p += 2;  if Z7_LIKELY((b1 & (a1 ^ 8)) < 0xf8)
+    for (;;)
+    {
+      unsigned b3; // Byte / UInt32
+      /* (Byte)(b3) normalization can use low byte computations in MSVC.
+         It gives smaller code, and no loss of speed in some compilers/cpus.
+         But new MSVC 32-bit x86 compilers use more slow load
+         from memory to low byte register in that case.
+         So we try to use full 32-bit computations for faster code.
+      */
+      // if (p >= lim) { ARMT_TAIL_PROC }  b3 = b1 + 8;  b1 = p[3];  p += 2;  if ((b3 & b1) >= 0xf8) break;
+      if Z7_UNLIKELY(p >= lim) { ARMT_TAIL_PROC }  b3 = p[3];  p += 2;  if Z7_UNLIKELY((b3 & (b1 ^ 8)) >= 0xf8) break;
+      if Z7_UNLIKELY(p >= lim) { ARMT_TAIL_PROC }  b1 = p[3];  p += 2;  if Z7_UNLIKELY((b1 & (b3 ^ 8)) >= 0xf8) break;
+    }
+    {
+      /* we can adjust pc for (0xf800) to rid of (& 0x7FF) operation.
+         But gcc/clang for arm64 can use bfi instruction for full code here */
+      UInt32 v =
+          ((UInt32)GetUi16a(p - 2) << 11) |
+          ((UInt32)GetUi16a(p) & 0x7FF);
+      /*
+      UInt32 v =
+            ((UInt32)p[1 - 2] << 19)
+          + (((UInt32)p[1] & 0x7) << 8)
+          + (((UInt32)p[-2] << 11))
+          + (p[0]);
+      */
+      p += 2;
+      {
+        UInt32 c = BR_PC_GET >> 1;
+        BR_CONVERT_VAL(v, c)
+      }
+      SetUi16a(p - 4, (UInt16)(((v >> 11) & 0x7ff) | 0xf000))
+      SetUi16a(p - 2, (UInt16)(v | 0xf800))
+      /*
+      p[-4] = (Byte)(v >> 11);
+      p[-3] = (Byte)(0xf0 | ((v >> 19) & 0x7));
+      p[-2] = (Byte)v;
+      p[-1] = (Byte)(0xf8 | (v >> 8));
+      */
+    }
+  }
+  while (p < lim);
+  return p;
+  // armt_tail:
+  // if ((Byte)((lim[1] & 0xf8)) != 0xf0) { lim += 2; }  return lim;
+  // return (Byte *)(lim + ((Byte)((lim[1] ^ 0xf0) & 0xf8) == 0 ? 0 : 2));
+  // return (Byte *)(lim + (((lim[1] ^ ~0xfu) & ~7u) == 0 ? 0 : 2));
+  // return (Byte *)(lim + 2 - (((((unsigned)lim[1] ^ 8) + 8) >> 7) & 2));
+}
+Z7_BRANCH_FUNCS_IMP(ARMT)
+
+
+// #define BR_IA64_NO_INLINE
+
+Z7_BRANCH_FUNC_MAIN(IA64)
+{
+  // Byte *p = data;
+  const Byte *lim;
+  size &= ~(SizeT)15;
+  lim = p + size;
+  pc -= 1 << 4;
+  pc >>= 4 - 1;
+  // pc -= 1 << 1;
+  
+  for (;;)
+  {
+    unsigned m;
+    for (;;)
+    {
+      if Z7_UNLIKELY(p == lim)
+        return p;
+      m = (unsigned)((UInt32)0x334b0000 >> (*p & 0x1e));
+      p += 16;
+      pc += 1 << 1;
+      if (m &= 3)
+        break;
+    }
+    {
+      p += (ptrdiff_t)m * 5 - 20; // negative value is expected here.
+      do
+      {
+        const UInt32 t =
+          #if defined(MY_CPU_X86_OR_AMD64)
+            // we use 32-bit load here to reduce code size on x86:
+            GetUi32(p);
+          #else
+            GetUi16(p);
+          #endif
+        UInt32 z = GetUi32(p + 1) >> m;
+        p += 5;
+        if (((t >> m) & (0x70 << 1)) == 0
+            && ((z - (0x5000000 << 1)) & (0xf000000 << 1)) == 0)
+        {
+          UInt32 v = (UInt32)((0x8fffff << 1) | 1) & z;
+          z ^= v;
+        #ifdef BR_IA64_NO_INLINE
+          v |= (v & ((UInt32)1 << (23 + 1))) >> 3;
+          {
+            UInt32 c = pc;
+            BR_CONVERT_VAL(v, c)
+          }
+          v &= (0x1fffff << 1) | 1;
+        #else
+          {
+            if (encoding)
+            {
+              // pc &= ~(0xc00000 << 1); // we just need to clear at least 2 bits
+              pc &= (0x1fffff << 1) | 1;
+              v += pc;
+            }
+            else
+            {
+              // pc |= 0xc00000 << 1; // we need to set at least 2 bits
+              pc |= ~(UInt32)((0x1fffff << 1) | 1);
+              v -= pc;
+            }
+          }
+          v &= ~(UInt32)(0x600000 << 1);
+        #endif
+          v += (0x700000 << 1);
+          v &= (0x8fffff << 1) | 1;
+          z |= v;
+          z <<= m;
+          SetUi32(p + 1 - 5, z)
+        }
+        m++;
+      }
+      while (m &= 3); // while (m < 4);
+    }
+  }
+}
+Z7_BRANCH_FUNCS_IMP(IA64)

C/Bra.h

@@ -1,68 +1,99 @@
 /* Bra.h -- Branch converters for executables
-2009-02-07 : Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
 
-#ifndef __BRA_H
-#define __BRA_H
+#ifndef ZIP7_INC_BRA_H
+#define ZIP7_INC_BRA_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
+
+#define Z7_BRANCH_CONV_DEC(name)    z7_BranchConv_   ## name ## _Dec
+#define Z7_BRANCH_CONV_ENC(name)    z7_BranchConv_   ## name ## _Enc
+#define Z7_BRANCH_CONV_ST_DEC(name) z7_BranchConvSt_ ## name ## _Dec
+#define Z7_BRANCH_CONV_ST_ENC(name) z7_BranchConvSt_ ## name ## _Enc
+
+#define Z7_BRANCH_CONV_DECL(name)    Byte * name(Byte *data, SizeT size, UInt32 pc)
+#define Z7_BRANCH_CONV_ST_DECL(name) Byte * name(Byte *data, SizeT size, UInt32 pc, UInt32 *state)
+
+typedef Z7_BRANCH_CONV_DECL(   (*z7_Func_BranchConv));
+typedef Z7_BRANCH_CONV_ST_DECL((*z7_Func_BranchConvSt));
+
+#define Z7_BRANCH_CONV_ST_X86_STATE_INIT_VAL 0
+Z7_BRANCH_CONV_ST_DECL(Z7_BRANCH_CONV_ST_DEC(X86));
+Z7_BRANCH_CONV_ST_DECL(Z7_BRANCH_CONV_ST_ENC(X86));
+
+#define Z7_BRANCH_FUNCS_DECL(name) \
+Z7_BRANCH_CONV_DECL(Z7_BRANCH_CONV_DEC(name)); \
+Z7_BRANCH_CONV_DECL(Z7_BRANCH_CONV_ENC(name));
+
+Z7_BRANCH_FUNCS_DECL(ARM64)
+Z7_BRANCH_FUNCS_DECL(ARM)
+Z7_BRANCH_FUNCS_DECL(ARMT)
+Z7_BRANCH_FUNCS_DECL(PPC)
+Z7_BRANCH_FUNCS_DECL(SPARC)
+Z7_BRANCH_FUNCS_DECL(IA64)
 
 /*
-These functions convert relative addresses to absolute addresses
-in CALL instructions to increase the compression ratio.
-  
-  In:
-    data     - data buffer
-    size     - size of data
-    ip       - current virtual Instruction Pinter (IP) value
-    state    - state variable for x86 converter
-    encoding - 0 (for decoding), 1 (for encoding)
-  
-  Out:
-    state    - state variable for x86 converter
+These functions convert data that contain CPU instructions.
+Each such function converts relative addresses to absolute addresses in some
+branch instructions: CALL (in all converters) and JUMP (X86 converter only).
+Such conversion allows to increase compression ratio, if we compress that data.
 
-  Returns:
-    The number of processed bytes. If you call these functions with multiple calls,
-    you must start next call with first byte after block of processed bytes.
+There are 2 types of converters:
+  Byte * Conv_RISC (Byte *data, SizeT size, UInt32 pc);
+  Byte * ConvSt_X86(Byte *data, SizeT size, UInt32 pc, UInt32 *state);
+Each Converter supports 2 versions: one for encoding
+and one for decoding (_Enc/_Dec postfixes in function name).
+
+In params:
+  data  : data buffer
+  size  : size of data
+  pc    : current virtual Program Counter (Instruction Pinter) value
+In/Out param:
+  state : pointer to state variable (for X86 converter only)
+
+Return:
+  The pointer to position in (data) buffer after last byte that was processed.
+  If the caller calls converter again, it must call it starting with that position.
+  But the caller is allowed to move data in buffer. so pointer to
+  current processed position also will be changed for next call.
+  Also the caller must increase internal (pc) value for next call.
   
+Each converter has some characteristics: Endian, Alignment, LookAhead.
   Type   Endian  Alignment  LookAhead
   
-  x86    little      1          4
+  X86    little      1          4
   ARMT   little      2          2
   ARM    little      4          0
+  ARM64  little      4          0
   PPC     big        4          0
   SPARC   big        4          0
   IA64   little     16          0
 
-  size must be >= Alignment + LookAhead, if it's not last block.
-  If (size < Alignment + LookAhead), converter returns 0.
+  (data) must be aligned for (Alignment).
+  processed size can be calculated as:
+    SizeT processed = Conv(data, size, pc) - data;
+  if (processed == 0)
+    it means that converter needs more data for processing.
+  If (size < Alignment + LookAhead)
+    then (processed == 0) is allowed.
 
-  Example:
-
-    UInt32 ip = 0;
-    for ()
-    {
-      ; size must be >= Alignment + LookAhead, if it's not last block
-      SizeT processed = Convert(data, size, ip, 1);
-      data += processed;
-      size -= processed;
-      ip += processed;
-    }
+Example code for conversion in loop:
+  UInt32 pc = 0;
+  size = 0;
+  for (;;)
+  {
+    size += Load_more_input_data(data + size);
+    SizeT processed = Conv(data, size, pc) - data;
+    if (processed == 0 && no_more_input_data_after_size)
+      break; // we stop convert loop
+    data += processed;
+    size -= processed;
+    pc += processed;
+  }
 */
 
-#define x86_Convert_Init(state) { state = 0; }
-SizeT x86_Convert(Byte *data, SizeT size, UInt32 ip, UInt32 *state, int encoding);
-SizeT ARM_Convert(Byte *data, SizeT size, UInt32 ip, int encoding);
-SizeT ARMT_Convert(Byte *data, SizeT size, UInt32 ip, int encoding);
-SizeT PPC_Convert(Byte *data, SizeT size, UInt32 ip, int encoding);
-SizeT SPARC_Convert(Byte *data, SizeT size, UInt32 ip, int encoding);
-SizeT IA64_Convert(Byte *data, SizeT size, UInt32 ip, int encoding);
-
-#ifdef __cplusplus
-}
-#endif
+EXTERN_C_END
 
 #endif

C/Bra86.c

@@ -1,85 +1,187 @@
-/* Bra86.c -- Converter for x86 code (BCJ)
-2008-10-04 : Igor Pavlov : Public domain */
+/* Bra86.c -- Branch converter for X86 code (BCJ)
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "Bra.h"
+#include "CpuArch.h"
 
-#define Test86MSByte(b) ((b) == 0 || (b) == 0xFF)
 
-const Byte kMaskToAllowedStatus[8] = {1, 1, 1, 0, 1, 0, 0, 0};
-const Byte kMaskToBitNumber[8] = {0, 1, 2, 2, 3, 3, 3, 3};
+#if defined(MY_CPU_SIZEOF_POINTER) \
+    && ( MY_CPU_SIZEOF_POINTER == 4 \
+      || MY_CPU_SIZEOF_POINTER == 8)
+  #define BR_CONV_USE_OPT_PC_PTR
+#endif
 
-SizeT x86_Convert(Byte *data, SizeT size, UInt32 ip, UInt32 *state, int encoding)
+#ifdef BR_CONV_USE_OPT_PC_PTR
+#define BR_PC_INIT  pc -= (UInt32)(SizeT)p; // (MY_uintptr_t)
+#define BR_PC_GET   (pc + (UInt32)(SizeT)p)
+#else
+#define BR_PC_INIT  pc += (UInt32)size;
+#define BR_PC_GET   (pc - (UInt32)(SizeT)(lim - p))
+// #define BR_PC_INIT
+// #define BR_PC_GET   (pc + (UInt32)(SizeT)(p - data))
+#endif
+
+#define BR_CONVERT_VAL(v, c) if (encoding) v += c; else v -= c;
+// #define BR_CONVERT_VAL(v, c) if (!encoding) c = (UInt32)0 - c; v += c;
+
+#define Z7_BRANCH_CONV_ST(name) z7_BranchConvSt_ ## name
+
+#define BR86_NEED_CONV_FOR_MS_BYTE(b) ((((b) + 1) & 0xfe) == 0)
+
+#ifdef MY_CPU_LE_UNALIGN
+  #define BR86_PREPARE_BCJ_SCAN  const UInt32 v = GetUi32(p) ^ 0xe8e8e8e8;
+  #define BR86_IS_BCJ_BYTE(n)    ((v & ((UInt32)0xfe << (n) * 8)) == 0)
+#else
+  #define BR86_PREPARE_BCJ_SCAN
+  // bad for MSVC X86 (partial write to byte reg):
+  #define BR86_IS_BCJ_BYTE(n)    ((p[n - 4] & 0xfe) == 0xe8)
+  // bad for old MSVC (partial write to byte reg):
+  // #define BR86_IS_BCJ_BYTE(n)    (((*p ^ 0xe8) & 0xfe) == 0)
+#endif
+ 
+static
+Z7_FORCE_INLINE
+Z7_ATTRIB_NO_VECTOR
+Byte *Z7_BRANCH_CONV_ST(X86)(Byte *p, SizeT size, UInt32 pc, UInt32 *state, int encoding)
 {
-  SizeT bufferPos = 0, prevPosT;
-  UInt32 prevMask = *state & 0x7;
   if (size < 5)
-    return 0;
-  ip += 5;
-  prevPosT = (SizeT)0 - 1;
+    return p;
+ {
+  // Byte *p = data;
+  const Byte *lim = p + size - 4;
+  unsigned mask = (unsigned)*state;  // & 7;
+#ifdef BR_CONV_USE_OPT_PC_PTR
+  /* if BR_CONV_USE_OPT_PC_PTR is defined: we need to adjust (pc) for (+4),
+        because call/jump offset is relative to the next instruction.
+     if BR_CONV_USE_OPT_PC_PTR is not defined : we don't need to adjust (pc) for (+4),
+         because  BR_PC_GET uses (pc - (lim - p)), and lim was adjusted for (-4) before.
+  */
+  pc += 4;
+#endif
+  BR_PC_INIT
+  goto start;
 
-  for (;;)
+  for (;; mask |= 4)
   {
-    Byte *p = data + bufferPos;
-    Byte *limit = data + size - 4;
-    for (; p < limit; p++)
-      if ((*p & 0xFE) == 0xE8)
-        break;
-    bufferPos = (SizeT)(p - data);
-    if (p >= limit)
-      break;
-    prevPosT = bufferPos - prevPosT;
-    if (prevPosT > 3)
-      prevMask = 0;
-    else
+    // cont: mask |= 4;
+  start:
+    if (p >= lim)
+      goto fin;
     {
-      prevMask = (prevMask << ((int)prevPosT - 1)) & 0x7;
-      if (prevMask != 0)
-      {
-        Byte b = p[4 - kMaskToBitNumber[prevMask]];
-        if (!kMaskToAllowedStatus[prevMask] || Test86MSByte(b))
-        {
-          prevPosT = bufferPos;
-          prevMask = ((prevMask << 1) & 0x7) | 1;
-          bufferPos++;
-          continue;
-        }
-      }
+      BR86_PREPARE_BCJ_SCAN
+      p += 4;
+      if (BR86_IS_BCJ_BYTE(0))  { goto m0; }  mask >>= 1;
+      if (BR86_IS_BCJ_BYTE(1))  { goto m1; }  mask >>= 1;
+      if (BR86_IS_BCJ_BYTE(2))  { goto m2; }  mask = 0;
+      if (BR86_IS_BCJ_BYTE(3))  { goto a3; }
     }
-    prevPosT = bufferPos;
+    goto main_loop;
 
-    if (Test86MSByte(p[4]))
+  m0: p--;
+  m1: p--;
+  m2: p--;
+    if (mask == 0)
+      goto a3;
+    if (p > lim)
+      goto fin_p;
+   
+    // if (((0x17u >> mask) & 1) == 0)
+    if (mask > 4 || mask == 3)
     {
-      UInt32 src = ((UInt32)p[4] << 24) | ((UInt32)p[3] << 16) | ((UInt32)p[2] << 8) | ((UInt32)p[1]);
-      UInt32 dest;
-      for (;;)
-      {
-        Byte b;
-        int index;
-        if (encoding)
-          dest = (ip + (UInt32)bufferPos) + src;
-        else
-          dest = src - (ip + (UInt32)bufferPos);
-        if (prevMask == 0)
-          break;
-        index = kMaskToBitNumber[prevMask] * 8;
-        b = (Byte)(dest >> (24 - index));
-        if (!Test86MSByte(b))
-          break;
-        src = dest ^ ((1 << (32 - index)) - 1);
-      }
-      p[4] = (Byte)(~(((dest >> 24) & 1) - 1));
-      p[3] = (Byte)(dest >> 16);
-      p[2] = (Byte)(dest >> 8);
-      p[1] = (Byte)dest;
-      bufferPos += 5;
+      mask >>= 1;
+      continue; // goto cont;
     }
-    else
+    mask >>= 1;
+    if (BR86_NEED_CONV_FOR_MS_BYTE(p[mask]))
+      continue; // goto cont;
+    // if (!BR86_NEED_CONV_FOR_MS_BYTE(p[3])) continue; // goto cont;
     {
-      prevMask = ((prevMask << 1) & 0x7) | 1;
-      bufferPos++;
+      UInt32 v = GetUi32(p);
+      UInt32 c;
+      v += (1 << 24);  if (v & 0xfe000000) continue; // goto cont;
+      c = BR_PC_GET;
+      BR_CONVERT_VAL(v, c)
+      {
+        mask <<= 3;
+        if (BR86_NEED_CONV_FOR_MS_BYTE(v >> mask))
+        {
+          v ^= (((UInt32)0x100 << mask) - 1);
+          #ifdef MY_CPU_X86
+          // for X86 : we can recalculate (c) to reduce register pressure
+            c = BR_PC_GET;
+          #endif
+          BR_CONVERT_VAL(v, c)
+        }
+        mask = 0;
+      }
+      // v = (v & ((1 << 24) - 1)) - (v & (1 << 24));
+      v &= (1 << 25) - 1;  v -= (1 << 24);
+      SetUi32(p, v)
+      p += 4;
+      goto main_loop;
+    }
+
+  main_loop:
+    if (p >= lim)
+      goto fin;
+    for (;;)
+    {
+      BR86_PREPARE_BCJ_SCAN
+      p += 4;
+      if (BR86_IS_BCJ_BYTE(0))  { goto a0; }
+      if (BR86_IS_BCJ_BYTE(1))  { goto a1; }
+      if (BR86_IS_BCJ_BYTE(2))  { goto a2; }
+      if (BR86_IS_BCJ_BYTE(3))  { goto a3; }
+      if (p >= lim)
+        goto fin;
+    }
+  
+  a0: p--;
+  a1: p--;
+  a2: p--;
+  a3:
+    if (p > lim)
+      goto fin_p;
+    // if (!BR86_NEED_CONV_FOR_MS_BYTE(p[3])) continue; // goto cont;
+    {
+      UInt32 v = GetUi32(p);
+      UInt32 c;
+      v += (1 << 24);  if (v & 0xfe000000) continue; // goto cont;
+      c = BR_PC_GET;
+      BR_CONVERT_VAL(v, c)
+      // v = (v & ((1 << 24) - 1)) - (v & (1 << 24));
+      v &= (1 << 25) - 1;  v -= (1 << 24);
+      SetUi32(p, v)
+      p += 4;
+      goto main_loop;
     }
   }
-  prevPosT = bufferPos - prevPosT;
-  *state = ((prevPosT > 3) ? 0 : ((prevMask << ((int)prevPosT - 1)) & 0x7));
-  return bufferPos;
+
+fin_p:
+  p--;
+fin:
+  // the following processing for tail is optional and can be commented
+  /*
+  lim += 4;
+  for (; p < lim; p++, mask >>= 1)
+    if ((*p & 0xfe) == 0xe8)
+      break;
+  */
+  *state = (UInt32)mask;
+  return p;
+ }
 }
+
+
+#define Z7_BRANCH_CONV_ST_FUNC_IMP(name, m, encoding) \
+Z7_NO_INLINE \
+Z7_ATTRIB_NO_VECTOR \
+Byte *m(name)(Byte *data, SizeT size, UInt32 pc, UInt32 *state) \
+  { return Z7_BRANCH_CONV_ST(name)(data, size, pc, state, encoding); }
+
+Z7_BRANCH_CONV_ST_FUNC_IMP(X86, Z7_BRANCH_CONV_ST_DEC, 0)
+#ifndef Z7_EXTRACT_ONLY
+Z7_BRANCH_CONV_ST_FUNC_IMP(X86, Z7_BRANCH_CONV_ST_ENC, 1)
+#endif

C/BraIA64.c

@@ -1,67 +1,14 @@
 /* BraIA64.c -- Converter for IA-64 code
-2008-10-04 : Igor Pavlov : Public domain */
+2023-02-20 : Igor Pavlov : Public domain */
 
-#include "Bra.h"
+#include "Precomp.h"
 
-static const Byte kBranchTable[32] =
-{
-  0, 0, 0, 0, 0, 0, 0, 0,
-  0, 0, 0, 0, 0, 0, 0, 0,
-  4, 4, 6, 6, 0, 0, 7, 7,
-  4, 4, 0, 0, 4, 4, 0, 0
-};
+// the code was moved to Bra.c
 
-SizeT IA64_Convert(Byte *data, SizeT size, UInt32 ip, int encoding)
-{
-  SizeT i;
-  if (size < 16)
-    return 0;
-  size -= 16;
-  for (i = 0; i <= size; i += 16)
-  {
-    UInt32 instrTemplate = data[i] & 0x1F;
-    UInt32 mask = kBranchTable[instrTemplate];
-    UInt32 bitPos = 5;
-    int slot;
-    for (slot = 0; slot < 3; slot++, bitPos += 41)
-    {
-      UInt32 bytePos, bitRes;
-      UInt64 instruction, instNorm;
-      int j;
-      if (((mask >> slot) & 1) == 0)
-        continue;
-      bytePos = (bitPos >> 3);
-      bitRes = bitPos & 0x7;
-      instruction = 0;
-      for (j = 0; j < 6; j++)
-        instruction += (UInt64)data[i + j + bytePos] << (8 * j);
+#ifdef _MSC_VER
+#pragma warning(disable : 4206) // nonstandard extension used : translation unit is empty
+#endif
 
-      instNorm = instruction >> bitRes;
-      if (((instNorm >> 37) & 0xF) == 0x5 && ((instNorm >> 9) & 0x7) == 0)
-      {
-        UInt32 src = (UInt32)((instNorm >> 13) & 0xFFFFF);
-        UInt32 dest;
-        src |= ((UInt32)(instNorm >> 36) & 1) << 20;
-        
-        src <<= 4;
-        
-        if (encoding)
-          dest = ip + (UInt32)i + src;
-        else
-          dest = src - (ip + (UInt32)i);
-        
-        dest >>= 4;
-        
-        instNorm &= ~((UInt64)(0x8FFFFF) << 13);
-        instNorm |= ((UInt64)(dest & 0xFFFFF) << 13);
-        instNorm |= ((UInt64)(dest & 0x100000) << (36 - 20));
-        
-        instruction &= (1 << bitRes) - 1;
-        instruction |= (instNorm << bitRes);
-        for (j = 0; j < 6; j++)
-          data[i + j + bytePos] = (Byte)(instruction >> (8 * j));
-      }
-    }
-  }
-  return i;
-}
+#if defined(__clang__)
+#pragma GCC diagnostic ignored "-Wempty-translation-unit"
+#endif

C/BwtSort.c

@@ -1,15 +1,13 @@
 /* BwtSort.c -- BWT block sorting
-2008-08-17
-Igor Pavlov
-Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "BwtSort.h"
 #include "Sort.h"
 
 /* #define BLOCK_SORT_USE_HEAP_SORT */
 
-#define NO_INLINE MY_FAST_CALL
-
 /* Don't change it !!! */
 #define kNumHashBytes 2
 #define kNumHashValues (1 << (kNumHashBytes * 8))
@@ -60,7 +58,10 @@ SortGroup - is recursive Range-Sort function with HeapSort optimization for smal
 returns: 1 - if there are groups, 0 - no more groups
 */
 
-UInt32 NO_INLINE SortGroup(UInt32 BlockSize, UInt32 NumSortedBytes, UInt32 groupOffset, UInt32 groupSize, int NumRefBits, UInt32 *Indices
+static
+UInt32
+Z7_FASTCALL
+SortGroup(UInt32 BlockSize, UInt32 NumSortedBytes, UInt32 groupOffset, UInt32 groupSize, int NumRefBits, UInt32 *Indices
   #ifndef BLOCK_SORT_USE_HEAP_SORT
   , UInt32 left, UInt32 range
   #endif
@@ -72,7 +73,7 @@ UInt32 NO_INLINE SortGroup(UInt32 BlockSize, UInt32 NumSortedBytes, UInt32 group
   {
     /*
     #ifndef BLOCK_SORT_EXTERNAL_FLAGS
-    SetFinishedGroupSize(ind2, 1);
+    SetFinishedGroupSize(ind2, 1)
     #endif
     */
     return 0;
@@ -116,7 +117,7 @@ UInt32 NO_INLINE SortGroup(UInt32 BlockSize, UInt32 NumSortedBytes, UInt32 group
     }
     
     HeapSort(temp, groupSize);
-    mask = ((1 << NumRefBits) - 1);
+    mask = (((UInt32)1 << NumRefBits) - 1);
     thereAreGroups = 0;
     
     group = groupOffset;
@@ -314,7 +315,7 @@ UInt32 NO_INLINE SortGroup(UInt32 BlockSize, UInt32 NumSortedBytes, UInt32 group
       #ifndef BLOCK_SORT_EXTERNAL_FLAGS
       UInt32 subGroupSize = ((ind2[j] & ~0xC0000000) >> kNumBitsMax);
       if ((ind2[j] & 0x40000000) != 0)
-        subGroupSize += ((ind2[j + 1] >> kNumBitsMax) << kNumExtra0Bits);
+        subGroupSize += ((ind2[(size_t)j + 1] >> kNumBitsMax) << kNumExtra0Bits);
       subGroupSize++;
       for (;;)
       {
@@ -362,7 +363,7 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
   for (i = 0; i < kNumHashValues; i++)
     counters[i] = 0;
   for (i = 0; i < blockSize - 1; i++)
-    counters[((UInt32)data[i] << 8) | data[i + 1]]++;
+    counters[((UInt32)data[i] << 8) | data[(size_t)i + 1]]++;
   counters[((UInt32)data[i] << 8) | data[0]]++;
 
   Groups = counters + BS_TEMP_SIZE;
@@ -392,11 +393,11 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
     }
 
     for (i = 0; i < blockSize - 1; i++)
-      Groups[i] = counters[((UInt32)data[i] << 8) | data[i + 1]];
+      Groups[i] = counters[((UInt32)data[i] << 8) | data[(size_t)i + 1]];
     Groups[i] = counters[((UInt32)data[i] << 8) | data[0]];
 
     for (i = 0; i < blockSize - 1; i++)
-      Indices[counters[((UInt32)data[i] << 8) | data[i + 1]]++] = i;
+      Indices[counters[((UInt32)data[i] << 8) | data[(size_t)i + 1]]++] = i;
     Indices[counters[((UInt32)data[i] << 8) | data[0]]++] = i;
     
     #ifndef BLOCK_SORT_EXTERNAL_FLAGS
@@ -448,11 +449,11 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
 
       groupSize = ((Indices[i] & ~0xC0000000) >> kNumBitsMax);
       {
-      Bool finishedGroup = ((Indices[i] & 0x80000000) == 0);
+      BoolInt finishedGroup = ((Indices[i] & 0x80000000) == 0);
       if ((Indices[i] & 0x40000000) != 0)
       {
-        groupSize += ((Indices[i + 1] >> kNumBitsMax) << kNumExtra0Bits);
-        Indices[i + 1] &= kIndexMask;
+        groupSize += ((Indices[(size_t)i + 1] >> kNumBitsMax) << kNumExtra0Bits);
+        Indices[(size_t)i + 1] &= kIndexMask;
       }
       Indices[i] &= kIndexMask;
       groupSize++;
@@ -460,10 +461,10 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
       {
         Indices[i - finishedGroupSize] &= kIndexMask;
         if (finishedGroupSize > 1)
-          Indices[i - finishedGroupSize + 1] &= kIndexMask;
+          Indices[(size_t)(i - finishedGroupSize) + 1] &= kIndexMask;
         {
         UInt32 newGroupSize = groupSize + finishedGroupSize;
-        SetFinishedGroupSize(Indices + i - finishedGroupSize, newGroupSize);
+        SetFinishedGroupSize(Indices + i - finishedGroupSize, newGroupSize)
         finishedGroupSize = newGroupSize;
         }
         i += groupSize;
@@ -503,8 +504,8 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
     UInt32 groupSize = ((Indices[i] & ~0xC0000000) >> kNumBitsMax);
     if ((Indices[i] & 0x40000000) != 0)
     {
-      groupSize += ((Indices[i + 1] >> kNumBitsMax) << kNumExtra0Bits);
-      Indices[i + 1] &= kIndexMask;
+      groupSize += ((Indices[(size_t)i + 1] >> kNumBitsMax) << kNumExtra0Bits);
+      Indices[(size_t)i + 1] &= kIndexMask;
     }
     Indices[i] &= kIndexMask;
     groupSize++;
@@ -513,4 +514,3 @@ UInt32 BlockSort(UInt32 *Indices, const Byte *data, UInt32 blockSize)
   #endif
   return Groups[0];
 }
-

C/BwtSort.h

@@ -1,14 +1,12 @@
 /* BwtSort.h -- BWT block sorting
-2009-02-07 : Igor Pavlov : Public domain */
+2023-03-03 : Igor Pavlov : Public domain */
 
-#ifndef __BWT_SORT_H
-#define __BWT_SORT_H
+#ifndef ZIP7_INC_BWT_SORT_H
+#define ZIP7_INC_BWT_SORT_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
 /* use BLOCK_SORT_EXTERNAL_FLAGS if blockSize can be > 1M */
 /* #define BLOCK_SORT_EXTERNAL_FLAGS */
@@ -23,8 +21,6 @@ extern "C" {
 
 UInt32 BlockSort(UInt32 *indices, const Byte *data, UInt32 blockSize);
 
-#ifdef __cplusplus
-}
-#endif
+EXTERN_C_END
 
 #endif

C/Compiler.h

@@ -0,0 +1,161 @@
+/* Compiler.h : Compiler specific defines and pragmas
+2023-04-02 : Igor Pavlov : Public domain */
+
+#ifndef ZIP7_INC_COMPILER_H
+#define ZIP7_INC_COMPILER_H
+
+#if defined(__clang__)
+# define Z7_CLANG_VERSION  (__clang_major__ * 10000 + __clang_minor__ * 100 + __clang_patchlevel__)
+#endif
+#if defined(__clang__) && defined(__apple_build_version__)
+# define Z7_APPLE_CLANG_VERSION   Z7_CLANG_VERSION
+#elif defined(__clang__)
+# define Z7_LLVM_CLANG_VERSION    Z7_CLANG_VERSION
+#elif defined(__GNUC__)
+# define Z7_GCC_VERSION (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__)
+#endif
+
+#ifdef _MSC_VER
+#if !defined(__clang__) && !defined(__GNUC__)
+#define Z7_MSC_VER_ORIGINAL _MSC_VER
+#endif
+#endif
+
+#if defined(__MINGW32__) || defined(__MINGW64__)
+#define Z7_MINGW
+#endif
+
+// #pragma GCC diagnostic ignored "-Wunknown-pragmas"
+
+#ifdef __clang__
+// padding size of '' with 4 bytes to alignment boundary
+#pragma GCC diagnostic ignored "-Wpadded"
+#endif
+
+
+#ifdef _MSC_VER
+
+  #ifdef UNDER_CE
+    #define RPC_NO_WINDOWS_H
+    /* #pragma warning(disable : 4115) // '_RPC_ASYNC_STATE' : named type definition in parentheses */
+    #pragma warning(disable : 4201) // nonstandard extension used : nameless struct/union
+    #pragma warning(disable : 4214) // nonstandard extension used : bit field types other than int
+  #endif
+
+#if defined(_MSC_VER) && _MSC_VER >= 1800
+#pragma warning(disable : 4464) // relative include path contains '..'
+#endif
+
+// == 1200 : -O1 : for __forceinline
+// >= 1900 : -O1 : for printf
+#pragma warning(disable : 4710) // function not inlined
+
+#if _MSC_VER < 1900
+// winnt.h: 'Int64ShllMod32'
+#pragma warning(disable : 4514) // unreferenced inline function has been removed
+#endif
+    
+#if _MSC_VER < 1300
+// #pragma warning(disable : 4702) // unreachable code
+// Bra.c : -O1:
+#pragma warning(disable : 4714) // function marked as __forceinline not inlined
+#endif
+
+/*
+#if _MSC_VER > 1400 && _MSC_VER <= 1900
+// strcat: This function or variable may be unsafe
+// sysinfoapi.h: kit10: GetVersion was declared deprecated
+#pragma warning(disable : 4996)
+#endif
+*/
+
+#pragma warning(disable : 4255)
+
+#if _MSC_VER > 1200
+// -Wall warnings
+
+#pragma warning(disable : 4711) // function selected for automatic inline expansion
+#pragma warning(disable : 4820) // '2' bytes padding added after data member
+
+#if _MSC_VER >= 1400 && _MSC_VER < 1920
+// 1400: string.h: _DBG_MEMCPY_INLINE_
+// 1600 - 191x : smmintrin.h __cplusplus'
+// is not defined as a preprocessor macro, replacing with '0' for '#if/#elif'
+#pragma warning(disable : 4668)
+
+// 1400 - 1600 : WinDef.h : 'FARPROC' :
+// 1900 - 191x : immintrin.h: _readfsbase_u32
+// no function prototype given : converting '()' to '(void)'
+#pragma warning(disable : 4255)
+#endif
+
+#if _MSC_VER >= 1914
+// Compiler will insert Spectre mitigation for memory load if /Qspectre switch specified
+#pragma warning(disable : 5045)
+#endif
+
+#endif // _MSC_VER > 1200
+#endif // _MSC_VER
+
+
+#if defined(__clang__) && (__clang_major__ >= 4)
+  #define Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE \
+    _Pragma("clang loop unroll(disable)") \
+    _Pragma("clang loop vectorize(disable)")
+  #define Z7_ATTRIB_NO_VECTORIZE
+#elif defined(__GNUC__) && (__GNUC__ >= 5)
+  #define Z7_ATTRIB_NO_VECTORIZE __attribute__((optimize("no-tree-vectorize")))
+  // __attribute__((optimize("no-unroll-loops")));
+  #define Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE
+#elif defined(_MSC_VER) && (_MSC_VER >= 1920)
+  #define Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE \
+    _Pragma("loop( no_vector )")
+  #define Z7_ATTRIB_NO_VECTORIZE
+#else
+  #define Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE
+  #define Z7_ATTRIB_NO_VECTORIZE
+#endif
+
+#if defined(MY_CPU_X86_OR_AMD64) && ( \
+       defined(__clang__) && (__clang_major__ >= 4) \
+    || defined(__GNUC__) && (__GNUC__ >= 5))
+  #define Z7_ATTRIB_NO_SSE  __attribute__((__target__("no-sse")))
+#else
+  #define Z7_ATTRIB_NO_SSE
+#endif
+
+#define Z7_ATTRIB_NO_VECTOR \
+  Z7_ATTRIB_NO_VECTORIZE \
+  Z7_ATTRIB_NO_SSE
+
+
+#if defined(__clang__) && (__clang_major__ >= 8) \
+  || defined(__GNUC__) && (__GNUC__ >= 1000) \
+  /* || defined(_MSC_VER) && (_MSC_VER >= 1920) */
+  // GCC is not good for __builtin_expect()
+  #define Z7_LIKELY(x)   (__builtin_expect((x), 1))
+  #define Z7_UNLIKELY(x) (__builtin_expect((x), 0))
+  // #define Z7_unlikely [[unlikely]]
+  // #define Z7_likely [[likely]]
+#else
+  #define Z7_LIKELY(x)   (x)
+  #define Z7_UNLIKELY(x) (x)
+  // #define Z7_likely
+#endif
+
+
+#if (defined(Z7_CLANG_VERSION) && (Z7_CLANG_VERSION >= 36000))
+#define Z7_DIAGNOSCTIC_IGNORE_BEGIN_RESERVED_MACRO_IDENTIFIER \
+  _Pragma("GCC diagnostic push") \
+  _Pragma("GCC diagnostic ignored \"-Wreserved-macro-identifier\"")
+#define Z7_DIAGNOSCTIC_IGNORE_END_RESERVED_MACRO_IDENTIFIER \
+  _Pragma("GCC diagnostic pop")
+#else
+#define Z7_DIAGNOSCTIC_IGNORE_BEGIN_RESERVED_MACRO_IDENTIFIER
+#define Z7_DIAGNOSCTIC_IGNORE_END_RESERVED_MACRO_IDENTIFIER
+#endif
+
+#define UNUSED_VAR(x) (void)x;
+/* #define UNUSED_VAR(x) x=x; */
+
+#endif

C/CpuArch.c

@@ -1,168 +1,823 @@
 /* CpuArch.c -- CPU specific code
-2009-12-12: Igor Pavlov : Public domain */
+2023-05-18 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
+
+// #include <stdio.h>
 
 #include "CpuArch.h"
 
 #ifdef MY_CPU_X86_OR_AMD64
 
-#if (defined(_MSC_VER) && !defined(MY_CPU_AMD64)) || defined(__GNUC__)
-#define USE_ASM
+#undef NEED_CHECK_FOR_CPUID
+#if !defined(MY_CPU_AMD64)
+#define NEED_CHECK_FOR_CPUID
 #endif
 
-#if defined(USE_ASM) && !defined(MY_CPU_AMD64)
-static UInt32 CheckFlag(UInt32 flag)
+/*
+  cpuid instruction supports (subFunction) parameter in ECX,
+  that is used only with some specific (function) parameter values.
+  But we always use only (subFunction==0).
+*/
+/*
+  __cpuid(): MSVC and GCC/CLANG use same function/macro name
+             but parameters are different.
+   We use MSVC __cpuid() parameters style for our z7_x86_cpuid() function.
+*/
+
+#if defined(__GNUC__) /* && (__GNUC__ >= 10) */ \
+    || defined(__clang__) /* && (__clang_major__ >= 10) */
+
+/* there was some CLANG/GCC compilers that have issues with
+   rbx(ebx) handling in asm blocks in -fPIC mode (__PIC__ is defined).
+   compiler's <cpuid.h> contains the macro __cpuid() that is similar to our code.
+   The history of __cpuid() changes in CLANG/GCC:
+   GCC:
+     2007: it preserved ebx for (__PIC__ && __i386__)
+     2013: it preserved rbx and ebx for __PIC__
+     2014: it doesn't preserves rbx and ebx anymore
+     we suppose that (__GNUC__ >= 5) fixed that __PIC__ ebx/rbx problem.
+   CLANG:
+     2014+: it preserves rbx, but only for 64-bit code. No __PIC__ check.
+   Why CLANG cares about 64-bit mode only, and doesn't care about ebx (in 32-bit)?
+   Do we need __PIC__ test for CLANG or we must care about rbx even if
+   __PIC__ is not defined?
+*/
+
+#define ASM_LN "\n"
+   
+#if defined(MY_CPU_AMD64) && defined(__PIC__) \
+    && ((defined (__GNUC__) && (__GNUC__ < 5)) || defined(__clang__))
+
+#define x86_cpuid_MACRO(p, func) { \
+  __asm__ __volatile__ ( \
+    ASM_LN   "mov     %%rbx, %q1"  \
+    ASM_LN   "cpuid"               \
+    ASM_LN   "xchg    %%rbx, %q1"  \
+    : "=a" ((p)[0]), "=&r" ((p)[1]), "=c" ((p)[2]), "=d" ((p)[3]) : "0" (func), "2"(0)); }
+
+  /* "=&r" selects free register. It can select even rbx, if that register is free.
+     "=&D" for (RDI) also works, but the code can be larger with "=&D"
+     "2"(0) means (subFunction = 0),
+     2 is (zero-based) index in the output constraint list "=c" (ECX). */
+
+#elif defined(MY_CPU_X86) && defined(__PIC__) \
+    && ((defined (__GNUC__) && (__GNUC__ < 5)) || defined(__clang__))
+
+#define x86_cpuid_MACRO(p, func) { \
+  __asm__ __volatile__ ( \
+    ASM_LN   "mov     %%ebx, %k1"  \
+    ASM_LN   "cpuid"               \
+    ASM_LN   "xchg    %%ebx, %k1"  \
+    : "=a" ((p)[0]), "=&r" ((p)[1]), "=c" ((p)[2]), "=d" ((p)[3]) : "0" (func), "2"(0)); }
+
+#else
+
+#define x86_cpuid_MACRO(p, func) { \
+  __asm__ __volatile__ ( \
+    ASM_LN   "cpuid"               \
+    : "=a" ((p)[0]), "=b" ((p)[1]), "=c" ((p)[2]), "=d" ((p)[3]) : "0" (func), "2"(0)); }
+
+#endif
+
+
+void Z7_FASTCALL z7_x86_cpuid(UInt32 p[4], UInt32 func)
 {
-  #ifdef _MSC_VER
-  __asm pushfd;
-  __asm pop EAX;
-  __asm mov EDX, EAX;
-  __asm xor EAX, flag;
-  __asm push EAX;
-  __asm popfd;
-  __asm pushfd;
-  __asm pop EAX;
-  __asm xor EAX, EDX;
-  __asm push EDX;
-  __asm popfd;
-  __asm and flag, EAX;
-  #else
-  __asm__ __volatile__ (
-    "pushf\n\t"
-    "pop  %%EAX\n\t"
-    "movl %%EAX,%%EDX\n\t"
-    "xorl %0,%%EAX\n\t"
-    "push %%EAX\n\t"
-    "popf\n\t"
-    "pushf\n\t"
-    "pop  %%EAX\n\t"
-    "xorl %%EDX,%%EAX\n\t"
-    "push %%EDX\n\t"
-    "popf\n\t"
-    "andl %%EAX, %0\n\t":
-    "=c" (flag) : "c" (flag));
-  #endif
-  return flag;
+  x86_cpuid_MACRO(p, func)
 }
-#define CHECK_CPUID_IS_SUPPORTED if (CheckFlag(1 << 18) == 0 || CheckFlag(1 << 21) == 0) return False;
+
+
+Z7_NO_INLINE
+UInt32 Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void)
+{
+ #if defined(NEED_CHECK_FOR_CPUID)
+  #define EFALGS_CPUID_BIT 21
+  UInt32 a;
+  __asm__ __volatile__ (
+    ASM_LN   "pushf"
+    ASM_LN   "pushf"
+    ASM_LN   "pop     %0"
+    // ASM_LN   "movl    %0, %1"
+    // ASM_LN   "xorl    $0x200000, %0"
+    ASM_LN   "btc     %1, %0"
+    ASM_LN   "push    %0"
+    ASM_LN   "popf"
+    ASM_LN   "pushf"
+    ASM_LN   "pop     %0"
+    ASM_LN   "xorl    (%%esp), %0"
+
+    ASM_LN   "popf"
+    ASM_LN
+    : "=&r" (a) // "=a"
+    : "i" (EFALGS_CPUID_BIT)
+    );
+  if ((a & (1 << EFALGS_CPUID_BIT)) == 0)
+    return 0;
+ #endif
+  {
+    UInt32 p[4];
+    x86_cpuid_MACRO(p, 0)
+    return p[0];
+  }
+}
+
+#undef ASM_LN
+
+#elif !defined(_MSC_VER)
+
+/*
+// for gcc/clang and other: we can try to use __cpuid macro:
+#include <cpuid.h>
+void Z7_FASTCALL z7_x86_cpuid(UInt32 p[4], UInt32 func)
+{
+  __cpuid(func, p[0], p[1], p[2], p[3]);
+}
+UInt32 Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void)
+{
+  return (UInt32)__get_cpuid_max(0, NULL);
+}
+*/
+// for unsupported cpuid:
+void Z7_FASTCALL z7_x86_cpuid(UInt32 p[4], UInt32 func)
+{
+  UNUSED_VAR(func)
+  p[0] = p[1] = p[2] = p[3] = 0;
+}
+UInt32 Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void)
+{
+  return 0;
+}
+
+#else // _MSC_VER
+
+#if !defined(MY_CPU_AMD64)
+
+UInt32 __declspec(naked) Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void)
+{
+  #if defined(NEED_CHECK_FOR_CPUID)
+  #define EFALGS_CPUID_BIT 21
+  __asm   pushfd
+  __asm   pushfd
+  /*
+  __asm   pop     eax
+  // __asm   mov     edx, eax
+  __asm   btc     eax, EFALGS_CPUID_BIT
+  __asm   push    eax
+  */
+  __asm   btc     dword ptr [esp], EFALGS_CPUID_BIT
+  __asm   popfd
+  __asm   pushfd
+  __asm   pop     eax
+  // __asm   xor     eax, edx
+  __asm   xor     eax, [esp]
+  // __asm   push    edx
+  __asm   popfd
+  __asm   and     eax, (1 shl EFALGS_CPUID_BIT)
+  __asm   jz end_func
+  #endif
+  __asm   push    ebx
+  __asm   xor     eax, eax    // func
+  __asm   xor     ecx, ecx    // subFunction (optional) for (func == 0)
+  __asm   cpuid
+  __asm   pop     ebx
+  #if defined(NEED_CHECK_FOR_CPUID)
+  end_func:
+  #endif
+  __asm   ret 0
+}
+
+void __declspec(naked) Z7_FASTCALL z7_x86_cpuid(UInt32 p[4], UInt32 func)
+{
+  UNUSED_VAR(p)
+  UNUSED_VAR(func)
+  __asm   push    ebx
+  __asm   push    edi
+  __asm   mov     edi, ecx    // p
+  __asm   mov     eax, edx    // func
+  __asm   xor     ecx, ecx    // subfunction (optional) for (func == 0)
+  __asm   cpuid
+  __asm   mov     [edi     ], eax
+  __asm   mov     [edi +  4], ebx
+  __asm   mov     [edi +  8], ecx
+  __asm   mov     [edi + 12], edx
+  __asm   pop     edi
+  __asm   pop     ebx
+  __asm   ret     0
+}
+
+#else // MY_CPU_AMD64
+
+    #if _MSC_VER >= 1600
+      #include <intrin.h>
+      #define MY_cpuidex  __cpuidex
+    #else
+/*
+ __cpuid (func == (0 or 7)) requires subfunction number in ECX.
+  MSDN: The __cpuid intrinsic clears the ECX register before calling the cpuid instruction.
+   __cpuid() in new MSVC clears ECX.
+   __cpuid() in old MSVC (14.00) x64 doesn't clear ECX
+ We still can use __cpuid for low (func) values that don't require ECX,
+ but __cpuid() in old MSVC will be incorrect for some func values: (func == 7).
+ So here we use the hack for old MSVC to send (subFunction) in ECX register to cpuid instruction,
+ where ECX value is first parameter for FASTCALL / NO_INLINE func,
+ So the caller of MY_cpuidex_HACK() sets ECX as subFunction, and
+ old MSVC for __cpuid() doesn't change ECX and cpuid instruction gets (subFunction) value.
+ 
+DON'T remove Z7_NO_INLINE and Z7_FASTCALL for MY_cpuidex_HACK(): !!!
+*/
+static
+Z7_NO_INLINE void Z7_FASTCALL MY_cpuidex_HACK(UInt32 subFunction, UInt32 func, int *CPUInfo)
+{
+  UNUSED_VAR(subFunction)
+  __cpuid(CPUInfo, func);
+}
+      #define MY_cpuidex(info, func, func2)  MY_cpuidex_HACK(func2, func, info)
+      #pragma message("======== MY_cpuidex_HACK WAS USED ========")
+    #endif // _MSC_VER >= 1600
+
+#if !defined(MY_CPU_AMD64)
+/* inlining for __cpuid() in MSVC x86 (32-bit) produces big ineffective code,
+   so we disable inlining here */
+Z7_NO_INLINE
+#endif
+void Z7_FASTCALL z7_x86_cpuid(UInt32 p[4], UInt32 func)
+{
+  MY_cpuidex((int *)p, (int)func, 0);
+}
+
+Z7_NO_INLINE
+UInt32 Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void)
+{
+  int a[4];
+  MY_cpuidex(a, 0, 0);
+  return a[0];
+}
+
+#endif // MY_CPU_AMD64
+#endif // _MSC_VER
+
+#if defined(NEED_CHECK_FOR_CPUID)
+#define CHECK_CPUID_IS_SUPPORTED { if (z7_x86_cpuid_GetMaxFunc() == 0) return 0; }
 #else
 #define CHECK_CPUID_IS_SUPPORTED
 #endif
+#undef NEED_CHECK_FOR_CPUID
 
-static void MyCPUID(UInt32 function, UInt32 *a, UInt32 *b, UInt32 *c, UInt32 *d)
-{
-  #ifdef USE_ASM
 
-  #ifdef _MSC_VER
-
-  UInt32 a2, b2, c2, d2;
-  __asm xor EBX, EBX;
-  __asm xor ECX, ECX;
-  __asm xor EDX, EDX;
-  __asm mov EAX, function;
-  __asm cpuid;
-  __asm mov a2, EAX;
-  __asm mov b2, EBX;
-  __asm mov c2, ECX;
-  __asm mov d2, EDX;
-
-  *a = a2;
-  *b = b2;
-  *c = c2;
-  *d = d2;
-
-  #else
-
- 	__asm__ __volatile__ (
- 		"cpuid"
- 		: "=a" (*a) ,
- 		  "=b" (*b) ,
- 		  "=c" (*c) ,
- 		  "=d" (*d)
-		: "0" (function)) ;
-
-  #endif
-  
-  #else
-
-  int CPUInfo[4];
-  __cpuid(CPUInfo, function);
-  *a = CPUInfo[0];
-  *b = CPUInfo[1];
-  *c = CPUInfo[2];
-  *d = CPUInfo[3];
-
-  #endif
-}
-
-Bool x86cpuid_CheckAndRead(Cx86cpuid *p)
+static
+BoolInt x86cpuid_Func_1(UInt32 *p)
 {
   CHECK_CPUID_IS_SUPPORTED
-  MyCPUID(0, &p->maxFunc, &p->vendor[0], &p->vendor[2], &p->vendor[1]);
-  MyCPUID(1, &p->ver, &p->b, &p->c, &p->d);
+  z7_x86_cpuid(p, 1);
   return True;
 }
 
-static UInt32 kVendors[][3] =
+/*
+static const UInt32 kVendors[][1] =
 {
-  { 0x756E6547, 0x49656E69, 0x6C65746E},
-  { 0x68747541, 0x69746E65, 0x444D4163},
-  { 0x746E6543, 0x48727561, 0x736C7561}
+  { 0x756E6547 }, // , 0x49656E69, 0x6C65746E },
+  { 0x68747541 }, // , 0x69746E65, 0x444D4163 },
+  { 0x746E6543 }  // , 0x48727561, 0x736C7561 }
 };
+*/
+
+/*
+typedef struct
+{
+  UInt32 maxFunc;
+  UInt32 vendor[3];
+  UInt32 ver;
+  UInt32 b;
+  UInt32 c;
+  UInt32 d;
+} Cx86cpuid;
+
+enum
+{
+  CPU_FIRM_INTEL,
+  CPU_FIRM_AMD,
+  CPU_FIRM_VIA
+};
+int x86cpuid_GetFirm(const Cx86cpuid *p);
+#define x86cpuid_ver_GetFamily(ver) (((ver >> 16) & 0xff0) | ((ver >> 8) & 0xf))
+#define x86cpuid_ver_GetModel(ver)  (((ver >> 12) &  0xf0) | ((ver >> 4) & 0xf))
+#define x86cpuid_ver_GetStepping(ver) (ver & 0xf)
 
 int x86cpuid_GetFirm(const Cx86cpuid *p)
 {
   unsigned i;
-  for (i = 0; i < sizeof(kVendors) / sizeof(kVendors[i]); i++)
+  for (i = 0; i < sizeof(kVendors) / sizeof(kVendors[0]); i++)
   {
     const UInt32 *v = kVendors[i];
-    if (v[0] == p->vendor[0] &&
-        v[1] == p->vendor[1] &&
-        v[2] == p->vendor[2])
+    if (v[0] == p->vendor[0]
+        // && v[1] == p->vendor[1]
+        // && v[2] == p->vendor[2]
+        )
       return (int)i;
   }
   return -1;
 }
 
-Bool CPU_Is_InOrder()
+BoolInt CPU_Is_InOrder()
 {
   Cx86cpuid p;
-  int firm;
   UInt32 family, model;
   if (!x86cpuid_CheckAndRead(&p))
     return True;
-  family = x86cpuid_GetFamily(&p);
-  model = x86cpuid_GetModel(&p);
-  firm = x86cpuid_GetFirm(&p);
-  switch (firm)
+
+  family = x86cpuid_ver_GetFamily(p.ver);
+  model = x86cpuid_ver_GetModel(p.ver);
+
+  switch (x86cpuid_GetFirm(&p))
   {
-    case CPU_FIRM_INTEL: return (family < 6 || (family == 6 && model == 0x100C));
+    case CPU_FIRM_INTEL: return (family < 6 || (family == 6 && (
+        // In-Order Atom CPU
+           model == 0x1C  // 45 nm, N4xx, D4xx, N5xx, D5xx, 230, 330
+        || model == 0x26  // 45 nm, Z6xx
+        || model == 0x27  // 32 nm, Z2460
+        || model == 0x35  // 32 nm, Z2760
+        || model == 0x36  // 32 nm, N2xxx, D2xxx
+        )));
     case CPU_FIRM_AMD: return (family < 5 || (family == 5 && (model < 6 || model == 0xA)));
     case CPU_FIRM_VIA: return (family < 6 || (family == 6 && model < 0xF));
   }
-  return True;
+  return False; // v23 : unknown processors are not In-Order
 }
+*/
+
+#ifdef _WIN32
+#include "7zWindows.h"
+#endif
 
 #if !defined(MY_CPU_AMD64) && defined(_WIN32)
-static Bool CPU_Sys_Is_SSE_Supported()
+
+/* for legacy SSE ia32: there is no user-space cpu instruction to check
+   that OS supports SSE register storing/restoring on context switches.
+   So we need some OS-specific function to check that it's safe to use SSE registers.
+*/
+
+Z7_FORCE_INLINE
+static BoolInt CPU_Sys_Is_SSE_Supported(void)
 {
-  OSVERSIONINFO vi;
-  vi.dwOSVersionInfoSize = sizeof(vi);
-  if (!GetVersionEx(&vi))
-    return False;
-  return (vi.dwMajorVersion >= 5);
+#ifdef _MSC_VER
+  #pragma warning(push)
+  #pragma warning(disable : 4996) // `GetVersion': was declared deprecated
+#endif
+  /* low byte is major version of Windows
+     We suppose that any Windows version since
+     Windows2000 (major == 5) supports SSE registers */
+  return (Byte)GetVersion() >= 5;
+#if defined(_MSC_VER)
+  #pragma warning(pop)
+#endif
 }
 #define CHECK_SYS_SSE_SUPPORT if (!CPU_Sys_Is_SSE_Supported()) return False;
 #else
 #define CHECK_SYS_SSE_SUPPORT
 #endif
 
-Bool CPU_Is_Aes_Supported()
+
+#if !defined(MY_CPU_AMD64)
+
+BoolInt CPU_IsSupported_CMOV(void)
 {
-  Cx86cpuid p;
+  UInt32 a[4];
+  if (!x86cpuid_Func_1(&a[0]))
+    return 0;
+  return (a[3] >> 15) & 1;
+}
+
+BoolInt CPU_IsSupported_SSE(void)
+{
+  UInt32 a[4];
   CHECK_SYS_SSE_SUPPORT
-  if (!x86cpuid_CheckAndRead(&p))
-    return False;
-  return (p.c >> 25) & 1;
+  if (!x86cpuid_Func_1(&a[0]))
+    return 0;
+  return (a[3] >> 25) & 1;
+}
+
+BoolInt CPU_IsSupported_SSE2(void)
+{
+  UInt32 a[4];
+  CHECK_SYS_SSE_SUPPORT
+  if (!x86cpuid_Func_1(&a[0]))
+    return 0;
+  return (a[3] >> 26) & 1;
+}
+
+#endif
+
+
+static UInt32 x86cpuid_Func_1_ECX(void)
+{
+  UInt32 a[4];
+  CHECK_SYS_SSE_SUPPORT
+  if (!x86cpuid_Func_1(&a[0]))
+    return 0;
+  return a[2];
+}
+
+BoolInt CPU_IsSupported_AES(void)
+{
+  return (x86cpuid_Func_1_ECX() >> 25) & 1;
+}
+
+BoolInt CPU_IsSupported_SSSE3(void)
+{
+  return (x86cpuid_Func_1_ECX() >> 9) & 1;
+}
+
+BoolInt CPU_IsSupported_SSE41(void)
+{
+  return (x86cpuid_Func_1_ECX() >> 19) & 1;
+}
+
+BoolInt CPU_IsSupported_SHA(void)
+{
+  CHECK_SYS_SSE_SUPPORT
+
+  if (z7_x86_cpuid_GetMaxFunc() < 7)
+    return False;
+  {
+    UInt32 d[4];
+    z7_x86_cpuid(d, 7);
+    return (d[1] >> 29) & 1;
+  }
+}
+
+/*
+MSVC: _xgetbv() intrinsic is available since VS2010SP1.
+   MSVC also defines (_XCR_XFEATURE_ENABLED_MASK) macro in
+   <immintrin.h> that we can use or check.
+   For any 32-bit x86 we can use asm code in MSVC,
+   but MSVC asm code is huge after compilation.
+   So _xgetbv() is better
+
+ICC: _xgetbv() intrinsic is available (in what version of ICC?)
+   ICC defines (__GNUC___) and it supports gnu assembler
+   also ICC supports MASM style code with -use-msasm switch.
+   but ICC doesn't support __attribute__((__target__))
+
+GCC/CLANG 9:
+  _xgetbv() is macro that works via __builtin_ia32_xgetbv()
+  and we need __attribute__((__target__("xsave")).
+  But with __target__("xsave") the function will be not
+  inlined to function that has no __target__("xsave") attribute.
+  If we want _xgetbv() call inlining, then we should use asm version
+  instead of calling _xgetbv().
+  Note:intrinsic is broke before GCC 8.2:
+    https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85684
+*/
+
+#if    defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 1100) \
+    || defined(_MSC_VER) && (_MSC_VER >= 1600) && (_MSC_FULL_VER >= 160040219)  \
+    || defined(__GNUC__) && (__GNUC__ >= 9) \
+    || defined(__clang__) && (__clang_major__ >= 9)
+// we define ATTRIB_XGETBV, if we want to use predefined _xgetbv() from compiler
+#if defined(__INTEL_COMPILER)
+#define ATTRIB_XGETBV
+#elif defined(__GNUC__) || defined(__clang__)
+// we don't define ATTRIB_XGETBV here, because asm version is better for inlining.
+// #define ATTRIB_XGETBV __attribute__((__target__("xsave")))
+#else
+#define ATTRIB_XGETBV
+#endif
+#endif
+
+#if defined(ATTRIB_XGETBV)
+#include <immintrin.h>
+#endif
+
+
+// XFEATURE_ENABLED_MASK/XCR0
+#define MY_XCR_XFEATURE_ENABLED_MASK 0
+
+#if defined(ATTRIB_XGETBV)
+ATTRIB_XGETBV
+#endif
+static UInt64 x86_xgetbv_0(UInt32 num)
+{
+#if defined(ATTRIB_XGETBV)
+  {
+    return
+      #if (defined(_MSC_VER))
+        _xgetbv(num);
+      #else
+        __builtin_ia32_xgetbv(
+          #if !defined(__clang__)
+            (int)
+          #endif
+            num);
+      #endif
+  }
+
+#elif defined(__GNUC__) || defined(__clang__) || defined(__SUNPRO_CC)
+
+  UInt32 a, d;
+ #if defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 4))
+  __asm__
+  (
+    "xgetbv"
+    : "=a"(a), "=d"(d) : "c"(num) : "cc"
+  );
+ #else // is old gcc
+  __asm__
+  (
+    ".byte 0x0f, 0x01, 0xd0" "\n\t"
+    : "=a"(a), "=d"(d) : "c"(num) : "cc"
+  );
+ #endif
+  return ((UInt64)d << 32) | a;
+  // return a;
+
+#elif defined(_MSC_VER) && !defined(MY_CPU_AMD64)
+  
+  UInt32 a, d;
+  __asm {
+    push eax
+    push edx
+    push ecx
+    mov ecx, num;
+    // xor ecx, ecx // = MY_XCR_XFEATURE_ENABLED_MASK
+    _emit 0x0f
+    _emit 0x01
+    _emit 0xd0
+    mov a, eax
+    mov d, edx
+    pop ecx
+    pop edx
+    pop eax
+  }
+  return ((UInt64)d << 32) | a;
+  // return a;
+
+#else // it's unknown compiler
+  // #error "Need xgetbv function"
+  UNUSED_VAR(num)
+  // for MSVC-X64 we could call external function from external file.
+  /* Actually we had checked OSXSAVE/AVX in cpuid before.
+     So it's expected that OS supports at least AVX and below. */
+  // if (num != MY_XCR_XFEATURE_ENABLED_MASK) return 0; // if not XCR0
+  return
+      // (1 << 0) |  // x87
+        (1 << 1)   // SSE
+      | (1 << 2);  // AVX
+  
+#endif
+}
+
+#ifdef _WIN32
+/*
+  Windows versions do not know about new ISA extensions that
+  can be introduced. But we still can use new extensions,
+  even if Windows doesn't report about supporting them,
+  But we can use new extensions, only if Windows knows about new ISA extension
+  that changes the number or size of registers: SSE, AVX/XSAVE, AVX512
+  So it's enough to check
+    MY_PF_AVX_INSTRUCTIONS_AVAILABLE
+      instead of
+    MY_PF_AVX2_INSTRUCTIONS_AVAILABLE
+*/
+#define MY_PF_XSAVE_ENABLED                            17
+// #define MY_PF_SSSE3_INSTRUCTIONS_AVAILABLE             36
+// #define MY_PF_SSE4_1_INSTRUCTIONS_AVAILABLE            37
+// #define MY_PF_SSE4_2_INSTRUCTIONS_AVAILABLE            38
+// #define MY_PF_AVX_INSTRUCTIONS_AVAILABLE               39
+// #define MY_PF_AVX2_INSTRUCTIONS_AVAILABLE              40
+// #define MY_PF_AVX512F_INSTRUCTIONS_AVAILABLE           41
+#endif
+
+BoolInt CPU_IsSupported_AVX(void)
+{
+  #ifdef _WIN32
+  if (!IsProcessorFeaturePresent(MY_PF_XSAVE_ENABLED))
+    return False;
+  /* PF_AVX_INSTRUCTIONS_AVAILABLE probably is supported starting from
+     some latest Win10 revisions. But we need AVX in older Windows also.
+     So we don't use the following check: */
+  /*
+  if (!IsProcessorFeaturePresent(MY_PF_AVX_INSTRUCTIONS_AVAILABLE))
+    return False;
+  */
+  #endif
+
+  /*
+    OS must use new special XSAVE/XRSTOR instructions to save
+    AVX registers when it required for context switching.
+    At OS statring:
+      OS sets CR4.OSXSAVE flag to signal the processor that OS supports the XSAVE extensions.
+      Also OS sets bitmask in XCR0 register that defines what
+      registers will be processed by XSAVE instruction:
+        XCR0.SSE[bit 0] - x87 registers and state
+        XCR0.SSE[bit 1] - SSE registers and state
+        XCR0.AVX[bit 2] - AVX registers and state
+    CR4.OSXSAVE is reflected to CPUID.1:ECX.OSXSAVE[bit 27].
+       So we can read that bit in user-space.
+    XCR0 is available for reading in user-space by new XGETBV instruction.
+  */
+  {
+    const UInt32 c = x86cpuid_Func_1_ECX();
+    if (0 == (1
+        & (c >> 28)   // AVX instructions are supported by hardware
+        & (c >> 27))) // OSXSAVE bit: XSAVE and related instructions are enabled by OS.
+      return False;
+  }
+
+  /* also we can check
+     CPUID.1:ECX.XSAVE [bit 26] : that shows that
+        XSAVE, XRESTOR, XSETBV, XGETBV instructions are supported by hardware.
+     But that check is redundant, because if OSXSAVE bit is set, then XSAVE is also set */
+
+  /* If OS have enabled XSAVE extension instructions (OSXSAVE == 1),
+     in most cases we expect that OS also will support storing/restoring
+     for AVX and SSE states at least.
+     But to be ensure for that we call user-space instruction
+     XGETBV(0) to get XCR0 value that contains bitmask that defines
+     what exact states(registers) OS have enabled for storing/restoring.
+  */
+
+  {
+    const UInt32 bm = (UInt32)x86_xgetbv_0(MY_XCR_XFEATURE_ENABLED_MASK);
+    // printf("\n=== XGetBV=%d\n", bm);
+    return 1
+        & (bm >> 1)  // SSE state is supported (set by OS) for storing/restoring
+        & (bm >> 2); // AVX state is supported (set by OS) for storing/restoring
+  }
+  // since Win7SP1: we can use GetEnabledXStateFeatures();
+}
+
+
+BoolInt CPU_IsSupported_AVX2(void)
+{
+  if (!CPU_IsSupported_AVX())
+    return False;
+  if (z7_x86_cpuid_GetMaxFunc() < 7)
+    return False;
+  {
+    UInt32 d[4];
+    z7_x86_cpuid(d, 7);
+    // printf("\ncpuid(7): ebx=%8x ecx=%8x\n", d[1], d[2]);
+    return 1
+      & (d[1] >> 5); // avx2
+  }
+}
+
+BoolInt CPU_IsSupported_VAES_AVX2(void)
+{
+  if (!CPU_IsSupported_AVX())
+    return False;
+  if (z7_x86_cpuid_GetMaxFunc() < 7)
+    return False;
+  {
+    UInt32 d[4];
+    z7_x86_cpuid(d, 7);
+    // printf("\ncpuid(7): ebx=%8x ecx=%8x\n", d[1], d[2]);
+    return 1
+      & (d[1] >> 5) // avx2
+      // & (d[1] >> 31) // avx512vl
+      & (d[2] >> 9); // vaes // VEX-256/EVEX
+  }
+}
+
+BoolInt CPU_IsSupported_PageGB(void)
+{
+  CHECK_CPUID_IS_SUPPORTED
+  {
+    UInt32 d[4];
+    z7_x86_cpuid(d, 0x80000000);
+    if (d[0] < 0x80000001)
+      return False;
+    z7_x86_cpuid(d, 0x80000001);
+    return (d[3] >> 26) & 1;
+  }
+}
+
+
+#elif defined(MY_CPU_ARM_OR_ARM64)
+
+#ifdef _WIN32
+
+#include "7zWindows.h"
+
+BoolInt CPU_IsSupported_CRC32(void)  { return IsProcessorFeaturePresent(PF_ARM_V8_CRC32_INSTRUCTIONS_AVAILABLE) ? 1 : 0; }
+BoolInt CPU_IsSupported_CRYPTO(void) { return IsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE) ? 1 : 0; }
+BoolInt CPU_IsSupported_NEON(void)   { return IsProcessorFeaturePresent(PF_ARM_NEON_INSTRUCTIONS_AVAILABLE) ? 1 : 0; }
+
+#else
+
+#if defined(__APPLE__)
+
+/*
+#include <stdio.h>
+#include <string.h>
+static void Print_sysctlbyname(const char *name)
+{
+  size_t bufSize = 256;
+  char buf[256];
+  int res = sysctlbyname(name, &buf, &bufSize, NULL, 0);
+  {
+    int i;
+    printf("\nres = %d : %s : '%s' : bufSize = %d, numeric", res, name, buf, (unsigned)bufSize);
+    for (i = 0; i < 20; i++)
+      printf(" %2x", (unsigned)(Byte)buf[i]);
+
+  }
+}
+*/
+/*
+  Print_sysctlbyname("hw.pagesize");
+  Print_sysctlbyname("machdep.cpu.brand_string");
+*/
+
+static BoolInt z7_sysctlbyname_Get_BoolInt(const char *name)
+{
+  UInt32 val = 0;
+  if (z7_sysctlbyname_Get_UInt32(name, &val) == 0 && val == 1)
+    return 1;
+  return 0;
+}
+
+BoolInt CPU_IsSupported_CRC32(void)
+{
+  return z7_sysctlbyname_Get_BoolInt("hw.optional.armv8_crc32");
+}
+
+BoolInt CPU_IsSupported_NEON(void)
+{
+  return z7_sysctlbyname_Get_BoolInt("hw.optional.neon");
+}
+
+#ifdef MY_CPU_ARM64
+#define APPLE_CRYPTO_SUPPORT_VAL 1
+#else
+#define APPLE_CRYPTO_SUPPORT_VAL 0
+#endif
+
+BoolInt CPU_IsSupported_SHA1(void) { return APPLE_CRYPTO_SUPPORT_VAL; }
+BoolInt CPU_IsSupported_SHA2(void) { return APPLE_CRYPTO_SUPPORT_VAL; }
+BoolInt CPU_IsSupported_AES (void) { return APPLE_CRYPTO_SUPPORT_VAL; }
+
+
+#else // __APPLE__
+
+#include <sys/auxv.h>
+
+#define USE_HWCAP
+
+#ifdef USE_HWCAP
+
+#include <asm/hwcap.h>
+
+  #define MY_HWCAP_CHECK_FUNC_2(name1, name2) \
+  BoolInt CPU_IsSupported_ ## name1() { return (getauxval(AT_HWCAP)  & (HWCAP_  ## name2)) ? 1 : 0; }
+
+#ifdef MY_CPU_ARM64
+  #define MY_HWCAP_CHECK_FUNC(name) \
+  MY_HWCAP_CHECK_FUNC_2(name, name)
+  MY_HWCAP_CHECK_FUNC_2(NEON, ASIMD)
+// MY_HWCAP_CHECK_FUNC (ASIMD)
+#elif defined(MY_CPU_ARM)
+  #define MY_HWCAP_CHECK_FUNC(name) \
+  BoolInt CPU_IsSupported_ ## name() { return (getauxval(AT_HWCAP2) & (HWCAP2_ ## name)) ? 1 : 0; }
+  MY_HWCAP_CHECK_FUNC_2(NEON, NEON)
+#endif
+
+#else // USE_HWCAP
+
+  #define MY_HWCAP_CHECK_FUNC(name) \
+  BoolInt CPU_IsSupported_ ## name() { return 0; }
+  MY_HWCAP_CHECK_FUNC(NEON)
+
+#endif // USE_HWCAP
+
+MY_HWCAP_CHECK_FUNC (CRC32)
+MY_HWCAP_CHECK_FUNC (SHA1)
+MY_HWCAP_CHECK_FUNC (SHA2)
+MY_HWCAP_CHECK_FUNC (AES)
+
+#endif // __APPLE__
+#endif // _WIN32
+
+#endif // MY_CPU_ARM_OR_ARM64
+
+
+
+#ifdef __APPLE__
+
+#include <sys/sysctl.h>
+
+int z7_sysctlbyname_Get(const char *name, void *buf, size_t *bufSize)
+{
+  return sysctlbyname(name, buf, bufSize, NULL, 0);
+}
+
+int z7_sysctlbyname_Get_UInt32(const char *name, UInt32 *val)
+{
+  size_t bufSize = sizeof(*val);
+  const int res = z7_sysctlbyname_Get(name, val, &bufSize);
+  if (res == 0 && bufSize != sizeof(*val))
+    return EFAULT;
+  return res;
 }
 
 #endif

C/CpuArch.h

@@ -1,64 +1,362 @@
 /* CpuArch.h -- CPU specific code
-2010-03-11: Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
 
-#ifndef __CPU_ARCH_H
-#define __CPU_ARCH_H
+#ifndef ZIP7_INC_CPU_ARCH_H
+#define ZIP7_INC_CPU_ARCH_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
 EXTERN_C_BEGIN
 
 /*
 MY_CPU_LE means that CPU is LITTLE ENDIAN.
-If MY_CPU_LE is not defined, we don't know about that property of platform (it can be LITTLE ENDIAN).
+MY_CPU_BE means that CPU is BIG ENDIAN.
+If MY_CPU_LE and MY_CPU_BE are not defined, we don't know about ENDIANNESS of platform.
 
 MY_CPU_LE_UNALIGN means that CPU is LITTLE ENDIAN and CPU supports unaligned memory accesses.
-If MY_CPU_LE_UNALIGN is not defined, we don't know about these properties of platform.
+
+MY_CPU_64BIT means that processor can work with 64-bit registers.
+  MY_CPU_64BIT can be used to select fast code branch
+  MY_CPU_64BIT doesn't mean that (sizeof(void *) == 8)
 */
 
-#if defined(_M_X64) || defined(_M_AMD64) || defined(__x86_64__)
-#define MY_CPU_AMD64
+#if  defined(_M_X64) \
+  || defined(_M_AMD64) \
+  || defined(__x86_64__) \
+  || defined(__AMD64__) \
+  || defined(__amd64__)
+  #define MY_CPU_AMD64
+  #ifdef __ILP32__
+    #define MY_CPU_NAME "x32"
+    #define MY_CPU_SIZEOF_POINTER 4
+  #else
+    #define MY_CPU_NAME "x64"
+    #define MY_CPU_SIZEOF_POINTER 8
+  #endif
+  #define MY_CPU_64BIT
 #endif
 
-#if defined(MY_CPU_AMD64) || defined(_M_IA64)
-#define MY_CPU_64BIT
+
+#if  defined(_M_IX86) \
+  || defined(__i386__)
+  #define MY_CPU_X86
+  #define MY_CPU_NAME "x86"
+  /* #define MY_CPU_32BIT */
+  #define MY_CPU_SIZEOF_POINTER 4
 #endif
 
-#if defined(_M_IX86) || defined(__i386__)
-#define MY_CPU_X86
+
+#if  defined(_M_ARM64) \
+  || defined(__AARCH64EL__) \
+  || defined(__AARCH64EB__) \
+  || defined(__aarch64__)
+  #define MY_CPU_ARM64
+  #ifdef __ILP32__
+    #define MY_CPU_NAME "arm64-32"
+    #define MY_CPU_SIZEOF_POINTER 4
+  #else
+    #define MY_CPU_NAME "arm64"
+    #define MY_CPU_SIZEOF_POINTER 8
+  #endif
+  #define MY_CPU_64BIT
 #endif
 
+
+#if  defined(_M_ARM) \
+  || defined(_M_ARM_NT) \
+  || defined(_M_ARMT) \
+  || defined(__arm__) \
+  || defined(__thumb__) \
+  || defined(__ARMEL__) \
+  || defined(__ARMEB__) \
+  || defined(__THUMBEL__) \
+  || defined(__THUMBEB__)
+  #define MY_CPU_ARM
+
+  #if defined(__thumb__) || defined(__THUMBEL__) || defined(_M_ARMT)
+    #define MY_CPU_ARMT
+    #define MY_CPU_NAME "armt"
+  #else
+    #define MY_CPU_ARM32
+    #define MY_CPU_NAME "arm"
+  #endif
+  /* #define MY_CPU_32BIT */
+  #define MY_CPU_SIZEOF_POINTER 4
+#endif
+
+
+#if  defined(_M_IA64) \
+  || defined(__ia64__)
+  #define MY_CPU_IA64
+  #define MY_CPU_NAME "ia64"
+  #define MY_CPU_64BIT
+#endif
+
+
+#if  defined(__mips64) \
+  || defined(__mips64__) \
+  || (defined(__mips) && (__mips == 64 || __mips == 4 || __mips == 3))
+  #define MY_CPU_NAME "mips64"
+  #define MY_CPU_64BIT
+#elif defined(__mips__)
+  #define MY_CPU_NAME "mips"
+  /* #define MY_CPU_32BIT */
+#endif
+
+
+#if  defined(__ppc64__) \
+  || defined(__powerpc64__) \
+  || defined(__ppc__) \
+  || defined(__powerpc__) \
+  || defined(__PPC__) \
+  || defined(_POWER)
+
+#define MY_CPU_PPC_OR_PPC64
+
+#if  defined(__ppc64__) \
+  || defined(__powerpc64__) \
+  || defined(_LP64) \
+  || defined(__64BIT__)
+  #ifdef __ILP32__
+    #define MY_CPU_NAME "ppc64-32"
+    #define MY_CPU_SIZEOF_POINTER 4
+  #else
+    #define MY_CPU_NAME "ppc64"
+    #define MY_CPU_SIZEOF_POINTER 8
+  #endif
+  #define MY_CPU_64BIT
+#else
+  #define MY_CPU_NAME "ppc"
+  #define MY_CPU_SIZEOF_POINTER 4
+  /* #define MY_CPU_32BIT */
+#endif
+#endif
+
+
+#if  defined(__riscv) \
+  || defined(__riscv__)
+  #if __riscv_xlen == 32
+    #define MY_CPU_NAME "riscv32"
+  #elif __riscv_xlen == 64
+    #define MY_CPU_NAME "riscv64"
+  #else
+    #define MY_CPU_NAME "riscv"
+  #endif
+#endif
+
+
 #if defined(MY_CPU_X86) || defined(MY_CPU_AMD64)
 #define MY_CPU_X86_OR_AMD64
 #endif
 
-#if defined(MY_CPU_X86) || defined(_M_ARM)
-#define MY_CPU_32BIT
+#if defined(MY_CPU_ARM) || defined(MY_CPU_ARM64)
+#define MY_CPU_ARM_OR_ARM64
 #endif
 
-#if defined(_WIN32) && defined(_M_ARM)
-#define MY_CPU_ARM_LE
+
+#ifdef _WIN32
+
+  #ifdef MY_CPU_ARM
+  #define MY_CPU_ARM_LE
+  #endif
+
+  #ifdef MY_CPU_ARM64
+  #define MY_CPU_ARM64_LE
+  #endif
+
+  #ifdef _M_IA64
+  #define MY_CPU_IA64_LE
+  #endif
+
 #endif
 
-#if defined(MY_CPU_X86_OR_AMD64)
-#define MY_CPU_LE_UNALIGN
+
+#if defined(MY_CPU_X86_OR_AMD64) \
+    || defined(MY_CPU_ARM_LE) \
+    || defined(MY_CPU_ARM64_LE) \
+    || defined(MY_CPU_IA64_LE) \
+    || defined(__LITTLE_ENDIAN__) \
+    || defined(__ARMEL__) \
+    || defined(__THUMBEL__) \
+    || defined(__AARCH64EL__) \
+    || defined(__MIPSEL__) \
+    || defined(__MIPSEL) \
+    || defined(_MIPSEL) \
+    || defined(__BFIN__) \
+    || (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__))
+  #define MY_CPU_LE
 #endif
 
-#if defined(MY_CPU_X86_OR_AMD64) || defined(MY_CPU_ARM_LE)
-#define MY_CPU_LE
+#if defined(__BIG_ENDIAN__) \
+    || defined(__ARMEB__) \
+    || defined(__THUMBEB__) \
+    || defined(__AARCH64EB__) \
+    || defined(__MIPSEB__) \
+    || defined(__MIPSEB) \
+    || defined(_MIPSEB) \
+    || defined(__m68k__) \
+    || defined(__s390__) \
+    || defined(__s390x__) \
+    || defined(__zarch__) \
+    || (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__))
+  #define MY_CPU_BE
 #endif
 
-#ifdef MY_CPU_LE_UNALIGN
 
-#define GetUi16(p) (*(const UInt16 *)(p))
-#define GetUi32(p) (*(const UInt32 *)(p))
-#define GetUi64(p) (*(const UInt64 *)(p))
-#define SetUi16(p, d) *(UInt16 *)(p) = (d);
-#define SetUi32(p, d) *(UInt32 *)(p) = (d);
+#if defined(MY_CPU_LE) && defined(MY_CPU_BE)
+  #error Stop_Compiling_Bad_Endian
+#endif
+
+#if !defined(MY_CPU_LE) && !defined(MY_CPU_BE)
+  #error Stop_Compiling_CPU_ENDIAN_must_be_detected_at_compile_time
+#endif
+
+#if defined(MY_CPU_32BIT) && defined(MY_CPU_64BIT)
+  #error Stop_Compiling_Bad_32_64_BIT
+#endif
+
+#ifdef __SIZEOF_POINTER__
+  #ifdef MY_CPU_SIZEOF_POINTER
+    #if MY_CPU_SIZEOF_POINTER != __SIZEOF_POINTER__
+      #error Stop_Compiling_Bad_MY_CPU_PTR_SIZE
+    #endif
+  #else
+    #define MY_CPU_SIZEOF_POINTER  __SIZEOF_POINTER__
+  #endif
+#endif
+
+#if defined(MY_CPU_SIZEOF_POINTER) && (MY_CPU_SIZEOF_POINTER == 4)
+#if defined (_LP64)
+      #error Stop_Compiling_Bad_MY_CPU_PTR_SIZE
+#endif
+#endif
+
+#ifdef _MSC_VER
+  #if _MSC_VER >= 1300
+    #define MY_CPU_pragma_pack_push_1   __pragma(pack(push, 1))
+    #define MY_CPU_pragma_pop           __pragma(pack(pop))
+  #else
+    #define MY_CPU_pragma_pack_push_1
+    #define MY_CPU_pragma_pop
+  #endif
+#else
+  #ifdef __xlC__
+    #define MY_CPU_pragma_pack_push_1   _Pragma("pack(1)")
+    #define MY_CPU_pragma_pop           _Pragma("pack()")
+  #else
+    #define MY_CPU_pragma_pack_push_1   _Pragma("pack(push, 1)")
+    #define MY_CPU_pragma_pop           _Pragma("pack(pop)")
+  #endif
+#endif
+
+
+#ifndef MY_CPU_NAME
+  #ifdef MY_CPU_LE
+    #define MY_CPU_NAME "LE"
+  #elif defined(MY_CPU_BE)
+    #define MY_CPU_NAME "BE"
+  #else
+    /*
+    #define MY_CPU_NAME ""
+    */
+  #endif
+#endif
+
+
+
+
+
+#ifdef __has_builtin
+  #define Z7_has_builtin(x)  __has_builtin(x)
+#else
+  #define Z7_has_builtin(x)  0
+#endif
+
+
+#define Z7_BSWAP32_CONST(v) \
+       ( (((UInt32)(v) << 24)                   ) \
+       | (((UInt32)(v) <<  8) & (UInt32)0xff0000) \
+       | (((UInt32)(v) >>  8) & (UInt32)0xff00  ) \
+       | (((UInt32)(v) >> 24)                   ))
+
+
+#if defined(_MSC_VER) && (_MSC_VER >= 1300)
+
+#include <stdlib.h>
+
+/* Note: these macros will use bswap instruction (486), that is unsupported in 386 cpu */
+
+#pragma intrinsic(_byteswap_ushort)
+#pragma intrinsic(_byteswap_ulong)
+#pragma intrinsic(_byteswap_uint64)
+
+#define Z7_BSWAP16(v)  _byteswap_ushort(v)
+#define Z7_BSWAP32(v)  _byteswap_ulong (v)
+#define Z7_BSWAP64(v)  _byteswap_uint64(v)
+#define Z7_CPU_FAST_BSWAP_SUPPORTED
+
+#elif  (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3))) \
+    || (defined(__clang__) && Z7_has_builtin(__builtin_bswap16))
+ 
+#define Z7_BSWAP16(v)  __builtin_bswap16(v)
+#define Z7_BSWAP32(v)  __builtin_bswap32(v)
+#define Z7_BSWAP64(v)  __builtin_bswap64(v)
+#define Z7_CPU_FAST_BSWAP_SUPPORTED
 
 #else
 
-#define GetUi16(p) (((const Byte *)(p))[0] | ((UInt16)((const Byte *)(p))[1] << 8))
+#define Z7_BSWAP16(v) ((UInt16) \
+       ( ((UInt32)(v) << 8) \
+       | ((UInt32)(v) >> 8) \
+       ))
+
+#define Z7_BSWAP32(v) Z7_BSWAP32_CONST(v)
+
+#define Z7_BSWAP64(v) \
+       ( ( ( (UInt64)(v)                           ) << 8 * 7 ) \
+       | ( ( (UInt64)(v) & ((UInt32)0xff << 8 * 1) ) << 8 * 5 ) \
+       | ( ( (UInt64)(v) & ((UInt32)0xff << 8 * 2) ) << 8 * 3 ) \
+       | ( ( (UInt64)(v) & ((UInt32)0xff << 8 * 3) ) << 8 * 1 ) \
+       | ( ( (UInt64)(v) >> 8 * 1 ) & ((UInt32)0xff << 8 * 3) ) \
+       | ( ( (UInt64)(v) >> 8 * 3 ) & ((UInt32)0xff << 8 * 2) ) \
+       | ( ( (UInt64)(v) >> 8 * 5 ) & ((UInt32)0xff << 8 * 1) ) \
+       | ( ( (UInt64)(v) >> 8 * 7 )                           ) \
+       )
+
+#endif
+
+
+
+#ifdef MY_CPU_LE
+  #if defined(MY_CPU_X86_OR_AMD64) \
+      || defined(MY_CPU_ARM64)
+    #define MY_CPU_LE_UNALIGN
+    #define MY_CPU_LE_UNALIGN_64
+  #elif defined(__ARM_FEATURE_UNALIGNED)
+    /* gcc9 for 32-bit arm can use LDRD instruction that requires 32-bit alignment.
+       So we can't use unaligned 64-bit operations. */
+    #define MY_CPU_LE_UNALIGN
+  #endif
+#endif
+
+
+#ifdef MY_CPU_LE_UNALIGN
+
+#define GetUi16(p) (*(const UInt16 *)(const void *)(p))
+#define GetUi32(p) (*(const UInt32 *)(const void *)(p))
+#ifdef MY_CPU_LE_UNALIGN_64
+#define GetUi64(p) (*(const UInt64 *)(const void *)(p))
+#define SetUi64(p, v) { *(UInt64 *)(void *)(p) = (v); }
+#endif
+
+#define SetUi16(p, v) { *(UInt16 *)(void *)(p) = (v); }
+#define SetUi32(p, v) { *(UInt32 *)(void *)(p) = (v); }
+
+#else
+
+#define GetUi16(p) ( (UInt16) ( \
+             ((const Byte *)(p))[0] | \
+    ((UInt16)((const Byte *)(p))[1] << 8) ))
 
 #define GetUi32(p) ( \
              ((const Byte *)(p))[0]        | \
@@ -66,26 +364,38 @@ If MY_CPU_LE_UNALIGN is not defined, we don't know about these properties of pla
     ((UInt32)((const Byte *)(p))[2] << 16) | \
     ((UInt32)((const Byte *)(p))[3] << 24))
 
-#define GetUi64(p) (GetUi32(p) | ((UInt64)GetUi32(((const Byte *)(p)) + 4) << 32))
+#define SetUi16(p, v) { Byte *_ppp_ = (Byte *)(p); UInt32 _vvv_ = (v); \
+    _ppp_[0] = (Byte)_vvv_; \
+    _ppp_[1] = (Byte)(_vvv_ >> 8); }
 
-#define SetUi16(p, d) { UInt32 _x_ = (d); \
-    ((Byte *)(p))[0] = (Byte)_x_; \
-    ((Byte *)(p))[1] = (Byte)(_x_ >> 8); }
-
-#define SetUi32(p, d) { UInt32 _x_ = (d); \
-    ((Byte *)(p))[0] = (Byte)_x_; \
-    ((Byte *)(p))[1] = (Byte)(_x_ >> 8); \
-    ((Byte *)(p))[2] = (Byte)(_x_ >> 16); \
-    ((Byte *)(p))[3] = (Byte)(_x_ >> 24); }
+#define SetUi32(p, v) { Byte *_ppp_ = (Byte *)(p); UInt32 _vvv_ = (v); \
+    _ppp_[0] = (Byte)_vvv_; \
+    _ppp_[1] = (Byte)(_vvv_ >> 8); \
+    _ppp_[2] = (Byte)(_vvv_ >> 16); \
+    _ppp_[3] = (Byte)(_vvv_ >> 24); }
 
 #endif
 
-#if defined(MY_CPU_LE_UNALIGN) && defined(_WIN64) && (_MSC_VER >= 1300)
 
-#pragma intrinsic(_byteswap_ulong)
-#pragma intrinsic(_byteswap_uint64)
-#define GetBe32(p) _byteswap_ulong(*(const UInt32 *)(const Byte *)(p))
-#define GetBe64(p) _byteswap_uint64(*(const UInt64 *)(const Byte *)(p))
+#ifndef GetUi64
+#define GetUi64(p) (GetUi32(p) | ((UInt64)GetUi32(((const Byte *)(p)) + 4) << 32))
+#endif
+
+#ifndef SetUi64
+#define SetUi64(p, v) { Byte *_ppp2_ = (Byte *)(p); UInt64 _vvv2_ = (v); \
+    SetUi32(_ppp2_    , (UInt32)_vvv2_) \
+    SetUi32(_ppp2_ + 4, (UInt32)(_vvv2_ >> 32)) }
+#endif
+
+
+#if defined(MY_CPU_LE_UNALIGN) && defined(Z7_CPU_FAST_BSWAP_SUPPORTED)
+
+#define GetBe32(p)  Z7_BSWAP32 (*(const UInt32 *)(const void *)(p))
+#define SetBe32(p, v) { (*(UInt32 *)(void *)(p)) = Z7_BSWAP32(v); }
+
+#if defined(MY_CPU_LE_UNALIGN_64)
+#define GetBe64(p)  Z7_BSWAP64 (*(const UInt64 *)(const void *)(p))
+#endif
 
 #else
 
@@ -95,44 +405,119 @@ If MY_CPU_LE_UNALIGN is not defined, we don't know about these properties of pla
     ((UInt32)((const Byte *)(p))[2] <<  8) | \
              ((const Byte *)(p))[3] )
 
-#define GetBe64(p) (((UInt64)GetBe32(p) << 32) | GetBe32(((const Byte *)(p)) + 4))
+#define SetBe32(p, v) { Byte *_ppp_ = (Byte *)(p); UInt32 _vvv_ = (v); \
+    _ppp_[0] = (Byte)(_vvv_ >> 24); \
+    _ppp_[1] = (Byte)(_vvv_ >> 16); \
+    _ppp_[2] = (Byte)(_vvv_ >> 8); \
+    _ppp_[3] = (Byte)_vvv_; }
 
 #endif
 
-#define GetBe16(p) (((UInt16)((const Byte *)(p))[0] << 8) | ((const Byte *)(p))[1])
+#ifndef GetBe64
+#define GetBe64(p) (((UInt64)GetBe32(p) << 32) | GetBe32(((const Byte *)(p)) + 4))
+#endif
+
+#ifndef GetBe16
+#define GetBe16(p) ( (UInt16) ( \
+    ((UInt16)((const Byte *)(p))[0] << 8) | \
+             ((const Byte *)(p))[1] ))
+#endif
+
+
+#if defined(MY_CPU_BE)
+#define Z7_CONV_BE_TO_NATIVE_CONST32(v)  (v)
+#define Z7_CONV_LE_TO_NATIVE_CONST32(v)  Z7_BSWAP32_CONST(v)
+#define Z7_CONV_NATIVE_TO_BE_32(v)       (v)
+#elif defined(MY_CPU_LE)
+#define Z7_CONV_BE_TO_NATIVE_CONST32(v)  Z7_BSWAP32_CONST(v)
+#define Z7_CONV_LE_TO_NATIVE_CONST32(v)  (v)
+#define Z7_CONV_NATIVE_TO_BE_32(v)       Z7_BSWAP32(v)
+#else
+#error Stop_Compiling_Unknown_Endian_CONV
+#endif
+
+
+#if defined(MY_CPU_BE)
+
+#define GetBe32a(p)      (*(const UInt32 *)(const void *)(p))
+#define GetBe16a(p)      (*(const UInt16 *)(const void *)(p))
+#define SetBe32a(p, v)   { *(UInt32 *)(void *)(p) = (v); }
+#define SetBe16a(p, v)   { *(UInt16 *)(void *)(p) = (v); }
+
+#define GetUi32a(p)      GetUi32(p)
+#define GetUi16a(p)      GetUi16(p)
+#define SetUi32a(p, v)   SetUi32(p, v)
+#define SetUi16a(p, v)   SetUi16(p, v)
+
+#elif defined(MY_CPU_LE)
+
+#define GetUi32a(p)      (*(const UInt32 *)(const void *)(p))
+#define GetUi16a(p)      (*(const UInt16 *)(const void *)(p))
+#define SetUi32a(p, v)   { *(UInt32 *)(void *)(p) = (v); }
+#define SetUi16a(p, v)   { *(UInt16 *)(void *)(p) = (v); }
+
+#define GetBe32a(p)      GetBe32(p)
+#define GetBe16a(p)      GetBe16(p)
+#define SetBe32a(p, v)   SetBe32(p, v)
+#define SetBe16a(p, v)   SetBe16(p, v)
+
+#else
+#error Stop_Compiling_Unknown_Endian_CPU_a
+#endif
+
+
+#if defined(MY_CPU_X86_OR_AMD64) \
+  || defined(MY_CPU_ARM_OR_ARM64) \
+  || defined(MY_CPU_PPC_OR_PPC64)
+  #define Z7_CPU_FAST_ROTATE_SUPPORTED
+#endif
 
 
 #ifdef MY_CPU_X86_OR_AMD64
 
-typedef struct
-{
-  UInt32 maxFunc;
-  UInt32 vendor[3];
-  UInt32 ver;
-  UInt32 b;
-  UInt32 c;
-  UInt32 d;
-} Cx86cpuid;
+void Z7_FASTCALL z7_x86_cpuid(UInt32 a[4], UInt32 function);
+UInt32 Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void);
+#if defined(MY_CPU_AMD64)
+#define Z7_IF_X86_CPUID_SUPPORTED
+#else
+#define Z7_IF_X86_CPUID_SUPPORTED if (z7_x86_cpuid_GetMaxFunc())
+#endif
 
-enum
-{
-  CPU_FIRM_INTEL,
-  CPU_FIRM_AMD,
-  CPU_FIRM_VIA
-};
+BoolInt CPU_IsSupported_AES(void);
+BoolInt CPU_IsSupported_AVX(void);
+BoolInt CPU_IsSupported_AVX2(void);
+BoolInt CPU_IsSupported_VAES_AVX2(void);
+BoolInt CPU_IsSupported_CMOV(void);
+BoolInt CPU_IsSupported_SSE(void);
+BoolInt CPU_IsSupported_SSE2(void);
+BoolInt CPU_IsSupported_SSSE3(void);
+BoolInt CPU_IsSupported_SSE41(void);
+BoolInt CPU_IsSupported_SHA(void);
+BoolInt CPU_IsSupported_PageGB(void);
 
-Bool x86cpuid_CheckAndRead(Cx86cpuid *p);
-int x86cpuid_GetFirm(const Cx86cpuid *p);
+#elif defined(MY_CPU_ARM_OR_ARM64)
 
-#define x86cpuid_GetFamily(p) (((p)->ver >> 8) & 0xFF00F)
-#define x86cpuid_GetModel(p) (((p)->ver >> 4) & 0xF00F)
-#define x86cpuid_GetStepping(p) ((p)->ver & 0xF)
+BoolInt CPU_IsSupported_CRC32(void);
+BoolInt CPU_IsSupported_NEON(void);
 
-Bool CPU_Is_InOrder();
-Bool CPU_Is_Aes_Supported();
+#if defined(_WIN32)
+BoolInt CPU_IsSupported_CRYPTO(void);
+#define CPU_IsSupported_SHA1  CPU_IsSupported_CRYPTO
+#define CPU_IsSupported_SHA2  CPU_IsSupported_CRYPTO
+#define CPU_IsSupported_AES   CPU_IsSupported_CRYPTO
+#else
+BoolInt CPU_IsSupported_SHA1(void);
+BoolInt CPU_IsSupported_SHA2(void);
+BoolInt CPU_IsSupported_AES(void);
+#endif
 
 #endif
 
+#if defined(__APPLE__)
+int z7_sysctlbyname_Get(const char *name, void *buf, size_t *bufSize);
+int z7_sysctlbyname_Get_UInt32(const char *name, UInt32 *val);
+#endif
+
 EXTERN_C_END
 
 #endif

C/Delta.c

@@ -1,5 +1,7 @@
 /* Delta.c -- Delta converter
-2009-05-26 : Igor Pavlov : Public domain */
+2021-02-09 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "Delta.h"
 
@@ -10,53 +12,158 @@ void Delta_Init(Byte *state)
     state[i] = 0;
 }
 
-static void MyMemCpy(Byte *dest, const Byte *src, unsigned size)
-{
-  unsigned i;
-  for (i = 0; i < size; i++)
-    dest[i] = src[i];
-}
 
 void Delta_Encode(Byte *state, unsigned delta, Byte *data, SizeT size)
 {
-  Byte buf[DELTA_STATE_SIZE];
-  unsigned j = 0;
-  MyMemCpy(buf, state, delta);
+  Byte temp[DELTA_STATE_SIZE];
+
+  if (size == 0)
+    return;
+
   {
-    SizeT i;
-    for (i = 0; i < size;)
+    unsigned i = 0;
+    do
+      temp[i] = state[i];
+    while (++i != delta);
+  }
+
+  if (size <= delta)
+  {
+    unsigned i = 0, k;
+    do
     {
-      for (j = 0; j < delta && i < size; i++, j++)
+      Byte b = *data;
+      *data++ = (Byte)(b - temp[i]);
+      temp[i] = b;
+    }
+    while (++i != size);
+    
+    k = 0;
+    
+    do
+    {
+      if (i == delta)
+        i = 0;
+      state[k] = temp[i++];
+    }
+    while (++k != delta);
+    
+    return;
+  }
+    
+  {
+    Byte *p = data + size - delta;
+    {
+      unsigned i = 0;
+      do
+        state[i] = *p++;
+      while (++i != delta);
+    }
+    {
+      const Byte *lim = data + delta;
+      ptrdiff_t dif = -(ptrdiff_t)delta;
+      
+      if (((ptrdiff_t)size + dif) & 1)
       {
-        Byte b = data[i];
-        data[i] = (Byte)(b - buf[j]);
-        buf[j] = b;
+        --p;  *p = (Byte)(*p - p[dif]);
       }
+
+      while (p != lim)
+      {
+        --p;  *p = (Byte)(*p - p[dif]);
+        --p;  *p = (Byte)(*p - p[dif]);
+      }
+      
+      dif = -dif;
+      
+      do
+      {
+        --p;  *p = (Byte)(*p - temp[--dif]);
+      }
+      while (dif != 0);
     }
   }
-  if (j == delta)
-    j = 0;
-  MyMemCpy(state, buf + j, delta - j);
-  MyMemCpy(state + delta - j, buf, j);
 }
 
+
 void Delta_Decode(Byte *state, unsigned delta, Byte *data, SizeT size)
 {
-  Byte buf[DELTA_STATE_SIZE];
-  unsigned j = 0;
-  MyMemCpy(buf, state, delta);
+  unsigned i;
+  const Byte *lim;
+
+  if (size == 0)
+    return;
+  
+  i = 0;
+  lim = data + size;
+  
+  if (size <= delta)
   {
-    SizeT i;
-    for (i = 0; i < size;)
+    do
+      *data = (Byte)(*data + state[i++]);
+    while (++data != lim);
+
+    for (; delta != i; state++, delta--)
+      *state = state[i];
+    data -= i;
+  }
+  else
+  {
+    /*
+    #define B(n) b ## n
+    #define I(n) Byte B(n) = state[n];
+    #define U(n) { B(n) = (Byte)((B(n)) + *data++); data[-1] = (B(n)); }
+    #define F(n) if (data != lim) { U(n) }
+
+    if (delta == 1)
     {
-      for (j = 0; j < delta && i < size; i++, j++)
+      I(0)
+      if ((lim - data) & 1) { U(0) }
+      while (data != lim) { U(0) U(0) }
+      data -= 1;
+    }
+    else if (delta == 2)
+    {
+      I(0) I(1)
+      lim -= 1; while (data < lim) { U(0) U(1) }
+      lim += 1; F(0)
+      data -= 2;
+    }
+    else if (delta == 3)
+    {
+      I(0) I(1) I(2)
+      lim -= 2; while (data < lim) { U(0) U(1) U(2) }
+      lim += 2; F(0) F(1)
+      data -= 3;
+    }
+    else if (delta == 4)
+    {
+      I(0) I(1) I(2) I(3)
+      lim -= 3; while (data < lim) { U(0) U(1) U(2) U(3) }
+      lim += 3; F(0) F(1) F(2)
+      data -= 4;
+    }
+    else
+    */
+    {
+      do
       {
-        buf[j] = data[i] = (Byte)(buf[j] + data[i]);
+        *data = (Byte)(*data + state[i++]);
+        data++;
+      }
+      while (i != delta);
+  
+      {
+        ptrdiff_t dif = -(ptrdiff_t)delta;
+        do
+          *data = (Byte)(*data + data[dif]);
+        while (++data != lim);
+        data += dif;
       }
     }
   }
-  if (j == delta)
-    j = 0;
-  MyMemCpy(state, buf + j, delta - j);
-  MyMemCpy(state + delta - j, buf, j);
+
+  do
+    *state++ = *data;
+  while (++data != lim);
 }

C/Delta.h

@@ -1,14 +1,12 @@
 /* Delta.h -- Delta converter
-2009-04-15 : Igor Pavlov : Public domain */
+2023-03-03 : Igor Pavlov : Public domain */
 
-#ifndef __DELTA_H
-#define __DELTA_H
+#ifndef ZIP7_INC_DELTA_H
+#define ZIP7_INC_DELTA_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
 #define DELTA_STATE_SIZE 256
 
@@ -16,8 +14,6 @@ void Delta_Init(Byte *state);
 void Delta_Encode(Byte *state, unsigned delta, Byte *data, SizeT size);
 void Delta_Decode(Byte *state, unsigned delta, Byte *data, SizeT size);
 
-#ifdef __cplusplus
-}
-#endif
+EXTERN_C_END
 
 #endif

C/DllSecur.c

@@ -0,0 +1,111 @@
+/* DllSecur.c -- DLL loading security
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
+
+#ifdef _WIN32
+
+#include "7zWindows.h"
+
+#include "DllSecur.h"
+
+#ifndef UNDER_CE
+
+#if (defined(__GNUC__) && (__GNUC__ >= 8)) || defined(__clang__)
+  // #pragma GCC diagnostic ignored "-Wcast-function-type"
+#endif
+
+#if defined(__clang__) || defined(__GNUC__)
+typedef void (*Z7_voidFunction)(void);
+#define MY_CAST_FUNC (Z7_voidFunction)
+#elif defined(_MSC_VER) && _MSC_VER > 1920
+#define MY_CAST_FUNC  (void *)
+// #pragma warning(disable : 4191) // 'type cast': unsafe conversion from 'FARPROC' to 'void (__cdecl *)()'
+#else
+#define MY_CAST_FUNC
+#endif
+
+typedef BOOL (WINAPI *Func_SetDefaultDllDirectories)(DWORD DirectoryFlags);
+
+#define MY_LOAD_LIBRARY_SEARCH_USER_DIRS 0x400
+#define MY_LOAD_LIBRARY_SEARCH_SYSTEM32  0x800
+
+#define DELIM "\0"
+
+static const char * const g_Dlls =
+         "userenv"
+  DELIM  "setupapi"
+  DELIM  "apphelp"
+  DELIM  "propsys"
+  DELIM  "dwmapi"
+  DELIM  "cryptbase"
+  DELIM  "oleacc"
+  DELIM  "clbcatq"
+  DELIM  "version"
+  #ifndef _CONSOLE
+  DELIM  "uxtheme"
+  #endif
+  DELIM;
+
+#endif
+
+#ifdef __clang__
+  #pragma GCC diagnostic ignored "-Wdeprecated-declarations"
+#endif
+#if defined (_MSC_VER) && _MSC_VER >= 1900
+// sysinfoapi.h: kit10: GetVersion was declared deprecated
+#pragma warning(disable : 4996)
+#endif
+
+#define IF_NON_VISTA_SET_DLL_DIRS_AND_RETURN \
+    if ((UInt16)GetVersion() != 6) { \
+      const \
+       Func_SetDefaultDllDirectories setDllDirs = \
+      (Func_SetDefaultDllDirectories) MY_CAST_FUNC GetProcAddress(GetModuleHandle(TEXT("kernel32.dll")), \
+           "SetDefaultDllDirectories"); \
+      if (setDllDirs) if (setDllDirs(MY_LOAD_LIBRARY_SEARCH_SYSTEM32 | MY_LOAD_LIBRARY_SEARCH_USER_DIRS)) return; }
+
+void My_SetDefaultDllDirectories(void)
+{
+  #ifndef UNDER_CE
+  IF_NON_VISTA_SET_DLL_DIRS_AND_RETURN
+  #endif
+}
+
+
+void LoadSecurityDlls(void)
+{
+  #ifndef UNDER_CE
+  // at Vista (ver 6.0) : CoCreateInstance(CLSID_ShellLink, ...) doesn't work after SetDefaultDllDirectories() : Check it ???
+  IF_NON_VISTA_SET_DLL_DIRS_AND_RETURN
+  {
+    wchar_t buf[MAX_PATH + 100];
+    const char *dll;
+    unsigned pos = GetSystemDirectoryW(buf, MAX_PATH + 2);
+    if (pos == 0 || pos > MAX_PATH)
+      return;
+    if (buf[pos - 1] != '\\')
+      buf[pos++] = '\\';
+    for (dll = g_Dlls; *dll != 0;)
+    {
+      wchar_t *dest = &buf[pos];
+      for (;;)
+      {
+        const char c = *dll++;
+        if (c == 0)
+          break;
+        *dest++ = (Byte)c;
+      }
+      dest[0] = '.';
+      dest[1] = 'd';
+      dest[2] = 'l';
+      dest[3] = 'l';
+      dest[4] = 0;
+      // lstrcatW(buf, L".dll");
+      LoadLibraryExW(buf, NULL, LOAD_WITH_ALTERED_SEARCH_PATH);
+    }
+  }
+  #endif
+}
+
+#endif // _WIN32

C/DllSecur.h

@@ -0,0 +1,20 @@
+/* DllSecur.h -- DLL loading for security
+2023-03-03 : Igor Pavlov : Public domain */
+
+#ifndef ZIP7_INC_DLL_SECUR_H
+#define ZIP7_INC_DLL_SECUR_H
+
+#include "7zTypes.h"
+
+EXTERN_C_BEGIN
+
+#ifdef _WIN32
+
+void My_SetDefaultDllDirectories(void);
+void LoadSecurityDlls(void);
+
+#endif
+
+EXTERN_C_END
+
+#endif

C/HuffEnc.c

@@ -1,12 +1,14 @@
 /* HuffEnc.c -- functions for Huffman encoding
-2009-09-02 : Igor Pavlov : Public domain */
+2023-03-04 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "HuffEnc.h"
 #include "Sort.h"
 
 #define kMaxLen 16
 #define NUM_BITS 10
-#define MASK ((1 << NUM_BITS) - 1)
+#define MASK (((unsigned)1 << NUM_BITS) - 1)
 
 #define NUM_COUNTERS 64
 
@@ -104,14 +106,14 @@ void Huffman_Generate(const UInt32 *freqs, UInt32 *p, Byte *lens, UInt32 numSymb
       
       p[--e] &= MASK;
       lenCounters[1] = 2;
-      while (e > 0)
+      while (e != 0)
       {
         UInt32 len = (p[p[--e] >> NUM_BITS] >> NUM_BITS) + 1;
         p[e] = (p[e] & MASK) | (len << NUM_BITS);
         if (len >= maxLen)
           for (len = maxLen - 1; lenCounters[len] == 0; len--);
         lenCounters[len]--;
-        lenCounters[len + 1] += 2;
+        lenCounters[(size_t)len + 1] += 2;
       }
       
       {
@@ -119,8 +121,8 @@ void Huffman_Generate(const UInt32 *freqs, UInt32 *p, Byte *lens, UInt32 numSymb
         i = 0;
         for (len = maxLen; len != 0; len--)
         {
-          UInt32 num;
-          for (num = lenCounters[len]; num != 0; num--)
+          UInt32 k;
+          for (k = lenCounters[len]; k != 0; k--)
             lens[p[i++] & MASK] = (Byte)len;
         }
       }
@@ -131,16 +133,22 @@ void Huffman_Generate(const UInt32 *freqs, UInt32 *p, Byte *lens, UInt32 numSymb
           UInt32 code = 0;
           UInt32 len;
           for (len = 1; len <= kMaxLen; len++)
-            nextCodes[len] = code = (code + lenCounters[len - 1]) << 1;
+            nextCodes[len] = code = (code + lenCounters[(size_t)len - 1]) << 1;
         }
         /* if (code + lenCounters[kMaxLen] - 1 != (1 << kMaxLen) - 1) throw 1; */
 
         {
-          UInt32 i;
-          for (i = 0; i < numSymbols; i++)
-            p[i] = nextCodes[lens[i]]++;
+          UInt32 k;
+          for (k = 0; k < numSymbols; k++)
+            p[k] = nextCodes[lens[k]]++;
         }
       }
     }
   }
 }
+
+#undef kMaxLen
+#undef NUM_BITS
+#undef MASK
+#undef NUM_COUNTERS
+#undef HUFFMAN_SPEED_OPT

C/HuffEnc.h

@@ -1,14 +1,12 @@
 /* HuffEnc.h -- Huffman encoding
-2009-02-07 : Igor Pavlov : Public domain */
+2023-03-05 : Igor Pavlov : Public domain */
 
-#ifndef __HUFF_ENC_H
-#define __HUFF_ENC_H
+#ifndef ZIP7_INC_HUFF_ENC_H
+#define ZIP7_INC_HUFF_ENC_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
 /*
 Conditions:
@@ -20,8 +18,6 @@ Conditions:
  
 void Huffman_Generate(const UInt32 *freqs, UInt32 *p, Byte *lens, UInt32 num, UInt32 maxLen);
 
-#ifdef __cplusplus
-}
-#endif
+EXTERN_C_END
 
 #endif

C/LzFind.h

@@ -1,80 +1,121 @@
 /* LzFind.h -- Match finder for LZ algorithms
-2009-04-22 : Igor Pavlov : Public domain */
+2023-03-04 : Igor Pavlov : Public domain */
 
-#ifndef __LZ_FIND_H
-#define __LZ_FIND_H
+#ifndef ZIP7_INC_LZ_FIND_H
+#define ZIP7_INC_LZ_FIND_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
 typedef UInt32 CLzRef;
 
-typedef struct _CMatchFinder
+typedef struct
 {
-  Byte *buffer;
+  const Byte *buffer;
   UInt32 pos;
   UInt32 posLimit;
-  UInt32 streamPos;
+  UInt32 streamPos;  /* wrap over Zero is allowed (streamPos < pos). Use (UInt32)(streamPos - pos) */
   UInt32 lenLimit;
 
   UInt32 cyclicBufferPos;
   UInt32 cyclicBufferSize; /* it must be = (historySize + 1) */
 
+  Byte streamEndWasReached;
+  Byte btMode;
+  Byte bigHash;
+  Byte directInput;
+
   UInt32 matchMaxLen;
   CLzRef *hash;
   CLzRef *son;
   UInt32 hashMask;
   UInt32 cutValue;
 
-  Byte *bufferBase;
-  ISeqInStream *stream;
-  int streamEndWasReached;
-
+  Byte *bufBase;
+  ISeqInStreamPtr stream;
+  
   UInt32 blockSize;
   UInt32 keepSizeBefore;
   UInt32 keepSizeAfter;
 
   UInt32 numHashBytes;
-  int directInput;
   size_t directInputRem;
-  int btMode;
-  int bigHash;
   UInt32 historySize;
   UInt32 fixedHashSize;
-  UInt32 hashSizeSum;
-  UInt32 numSons;
+  Byte numHashBytes_Min;
+  Byte numHashOutBits;
+  Byte _pad2_[2];
   SRes result;
   UInt32 crc[256];
+  size_t numRefs;
+
+  UInt64 expectedDataSize;
 } CMatchFinder;
 
-#define Inline_MatchFinder_GetPointerToCurrentPos(p) ((p)->buffer)
-#define Inline_MatchFinder_GetIndexByte(p, index) ((p)->buffer[(Int32)(index)])
+#define Inline_MatchFinder_GetPointerToCurrentPos(p) ((const Byte *)(p)->buffer)
 
-#define Inline_MatchFinder_GetNumAvailableBytes(p) ((p)->streamPos - (p)->pos)
+#define Inline_MatchFinder_GetNumAvailableBytes(p) ((UInt32)((p)->streamPos - (p)->pos))
 
+/*
+#define Inline_MatchFinder_IsFinishedOK(p) \
+    ((p)->streamEndWasReached \
+        && (p)->streamPos == (p)->pos \
+        && (!(p)->directInput || (p)->directInputRem == 0))
+*/
+      
 int MatchFinder_NeedMove(CMatchFinder *p);
-Byte *MatchFinder_GetPointerToCurrentPos(CMatchFinder *p);
+/* Byte *MatchFinder_GetPointerToCurrentPos(CMatchFinder *p); */
 void MatchFinder_MoveBlock(CMatchFinder *p);
 void MatchFinder_ReadIfRequired(CMatchFinder *p);
 
 void MatchFinder_Construct(CMatchFinder *p);
 
-/* Conditions:
-     historySize <= 3 GB
-     keepAddBufferBefore + matchMaxLen + keepAddBufferAfter < 511MB
+/* (directInput = 0) is default value.
+   It's required to provide correct (directInput) value
+   before calling MatchFinder_Create().
+   You can set (directInput) by any of the following calls:
+     - MatchFinder_SET_DIRECT_INPUT_BUF()
+     - MatchFinder_SET_STREAM()
+     - MatchFinder_SET_STREAM_MODE()
 */
+
+#define MatchFinder_SET_DIRECT_INPUT_BUF(p, _src_, _srcLen_) { \
+  (p)->stream = NULL; \
+  (p)->directInput = 1; \
+  (p)->buffer = (_src_); \
+  (p)->directInputRem = (_srcLen_); }
+
+/*
+#define MatchFinder_SET_STREAM_MODE(p) { \
+  (p)->directInput = 0; }
+*/
+
+#define MatchFinder_SET_STREAM(p, _stream_) { \
+  (p)->stream = _stream_; \
+  (p)->directInput = 0; }
+  
+
 int MatchFinder_Create(CMatchFinder *p, UInt32 historySize,
     UInt32 keepAddBufferBefore, UInt32 matchMaxLen, UInt32 keepAddBufferAfter,
-    ISzAlloc *alloc);
-void MatchFinder_Free(CMatchFinder *p, ISzAlloc *alloc);
-void MatchFinder_Normalize3(UInt32 subValue, CLzRef *items, UInt32 numItems);
-void MatchFinder_ReduceOffsets(CMatchFinder *p, UInt32 subValue);
+    ISzAllocPtr alloc);
+void MatchFinder_Free(CMatchFinder *p, ISzAllocPtr alloc);
+void MatchFinder_Normalize3(UInt32 subValue, CLzRef *items, size_t numItems);
+
+/*
+#define MatchFinder_INIT_POS(p, val) \
+    (p)->pos = (val); \
+    (p)->streamPos = (val);
+*/
+
+// void MatchFinder_ReduceOffsets(CMatchFinder *p, UInt32 subValue);
+#define MatchFinder_REDUCE_OFFSETS(p, subValue) \
+    (p)->pos -= (subValue); \
+    (p)->streamPos -= (subValue);
+
 
 UInt32 * GetMatchesSpec1(UInt32 lenLimit, UInt32 curMatch, UInt32 pos, const Byte *buffer, CLzRef *son,
-    UInt32 _cyclicBufferPos, UInt32 _cyclicBufferSize, UInt32 _cutValue,
+    size_t _cyclicBufferPos, UInt32 _cyclicBufferSize, UInt32 _cutValue,
     UInt32 *distances, UInt32 maxLen);
 
 /*
@@ -84,32 +125,35 @@ Conditions:
 */
 
 typedef void (*Mf_Init_Func)(void *object);
-typedef Byte (*Mf_GetIndexByte_Func)(void *object, Int32 index);
 typedef UInt32 (*Mf_GetNumAvailableBytes_Func)(void *object);
 typedef const Byte * (*Mf_GetPointerToCurrentPos_Func)(void *object);
-typedef UInt32 (*Mf_GetMatches_Func)(void *object, UInt32 *distances);
+typedef UInt32 * (*Mf_GetMatches_Func)(void *object, UInt32 *distances);
 typedef void (*Mf_Skip_Func)(void *object, UInt32);
 
-typedef struct _IMatchFinder
+typedef struct
 {
   Mf_Init_Func Init;
-  Mf_GetIndexByte_Func GetIndexByte;
   Mf_GetNumAvailableBytes_Func GetNumAvailableBytes;
   Mf_GetPointerToCurrentPos_Func GetPointerToCurrentPos;
   Mf_GetMatches_Func GetMatches;
   Mf_Skip_Func Skip;
-} IMatchFinder;
+} IMatchFinder2;
 
-void MatchFinder_CreateVTable(CMatchFinder *p, IMatchFinder *vTable);
+void MatchFinder_CreateVTable(CMatchFinder *p, IMatchFinder2 *vTable);
 
+void MatchFinder_Init_LowHash(CMatchFinder *p);
+void MatchFinder_Init_HighHash(CMatchFinder *p);
+void MatchFinder_Init_4(CMatchFinder *p);
 void MatchFinder_Init(CMatchFinder *p);
-UInt32 Bt3Zip_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances);
-UInt32 Hc3Zip_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances);
+
+UInt32* Bt3Zip_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances);
+UInt32* Hc3Zip_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances);
+
 void Bt3Zip_MatchFinder_Skip(CMatchFinder *p, UInt32 num);
 void Hc3Zip_MatchFinder_Skip(CMatchFinder *p, UInt32 num);
 
-#ifdef __cplusplus
-}
-#endif
+void LzFindPrepare(void);
+
+EXTERN_C_END
 
 #endif

C/LzFindMt.h

@@ -1,41 +1,34 @@
 /* LzFindMt.h -- multithreaded Match finder for LZ algorithms
-2009-02-07 : Igor Pavlov : Public domain */
+2023-03-05 : Igor Pavlov : Public domain */
 
-#ifndef __LZ_FIND_MT_H
-#define __LZ_FIND_MT_H
+#ifndef ZIP7_INC_LZ_FIND_MT_H
+#define ZIP7_INC_LZ_FIND_MT_H
 
 #include "LzFind.h"
 #include "Threads.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
-#define kMtHashBlockSize (1 << 13)
-#define kMtHashNumBlocks (1 << 3)
-#define kMtHashNumBlocksMask (kMtHashNumBlocks - 1)
-
-#define kMtBtBlockSize (1 << 14)
-#define kMtBtNumBlocks (1 << 6)
-#define kMtBtNumBlocksMask (kMtBtNumBlocks - 1)
-
-typedef struct _CMtSync
+typedef struct
 {
-  Bool wasCreated;
-  Bool needStart;
-  Bool exit;
-  Bool stopWriting;
-
+  UInt32 numProcessedBlocks;
   CThread thread;
+  UInt64 affinity;
+
+  BoolInt wasCreated;
+  BoolInt needStart;
+  BoolInt csWasInitialized;
+  BoolInt csWasEntered;
+
+  BoolInt exit;
+  BoolInt stopWriting;
+
   CAutoResetEvent canStart;
-  CAutoResetEvent wasStarted;
   CAutoResetEvent wasStopped;
   CSemaphore freeSemaphore;
   CSemaphore filledSemaphore;
-  Bool csWasInitialized;
-  Bool csWasEntered;
   CCriticalSection cs;
-  UInt32 numProcessedBlocks;
+  // UInt32 numBlocks_Sent;
 } CMtSync;
 
 typedef UInt32 * (*Mf_Mix_Matches)(void *p, UInt32 matchMinPos, UInt32 *distances);
@@ -46,23 +39,28 @@ typedef UInt32 * (*Mf_Mix_Matches)(void *p, UInt32 matchMinPos, UInt32 *distance
 typedef void (*Mf_GetHeads)(const Byte *buffer, UInt32 pos,
   UInt32 *hash, UInt32 hashMask, UInt32 *heads, UInt32 numHeads, const UInt32 *crc);
 
-typedef struct _CMatchFinderMt
+typedef struct
 {
   /* LZ */
   const Byte *pointerToCurPos;
   UInt32 *btBuf;
-  UInt32 btBufPos;
-  UInt32 btBufPosLimit;
+  const UInt32 *btBufPos;
+  const UInt32 *btBufPosLimit;
   UInt32 lzPos;
   UInt32 btNumAvailBytes;
 
   UInt32 *hash;
   UInt32 fixedHashSize;
+  // UInt32 hash4Mask;
   UInt32 historySize;
   const UInt32 *crc;
 
   Mf_Mix_Matches MixMatchesFunc;
-  
+  UInt32 failure_LZ_BT; // failure in BT transfered to LZ
+  // UInt32 failure_LZ_LZ; // failure in LZ tables
+  UInt32 failureBuf[1];
+  // UInt32 crc[256];
+
   /* LZ + BT */
   CMtSync btSync;
   Byte btDummy[kMtCacheLineDummy];
@@ -72,14 +70,16 @@ typedef struct _CMatchFinderMt
   UInt32 hashBufPos;
   UInt32 hashBufPosLimit;
   UInt32 hashNumAvail;
+  UInt32 failure_BT;
+
 
   CLzRef *son;
   UInt32 matchMaxLen;
   UInt32 numHashBytes;
   UInt32 pos;
-  Byte *buffer;
+  const Byte *buffer;
   UInt32 cyclicBufferPos;
-  UInt32 cyclicBufferSize; /* it must be historySize + 1 */
+  UInt32 cyclicBufferSize; /* it must be = (historySize + 1) */
   UInt32 cutValue;
 
   /* BT + Hash */
@@ -89,17 +89,21 @@ typedef struct _CMatchFinderMt
   /* Hash */
   Mf_GetHeads GetHeadsFunc;
   CMatchFinder *MatchFinder;
+  // CMatchFinder MatchFinder;
 } CMatchFinderMt;
 
+// only for Mt part
 void MatchFinderMt_Construct(CMatchFinderMt *p);
-void MatchFinderMt_Destruct(CMatchFinderMt *p, ISzAlloc *alloc);
+void MatchFinderMt_Destruct(CMatchFinderMt *p, ISzAllocPtr alloc);
+
 SRes MatchFinderMt_Create(CMatchFinderMt *p, UInt32 historySize, UInt32 keepAddBufferBefore,
-    UInt32 matchMaxLen, UInt32 keepAddBufferAfter, ISzAlloc *alloc);
-void MatchFinderMt_CreateVTable(CMatchFinderMt *p, IMatchFinder *vTable);
+    UInt32 matchMaxLen, UInt32 keepAddBufferAfter, ISzAllocPtr alloc);
+void MatchFinderMt_CreateVTable(CMatchFinderMt *p, IMatchFinder2 *vTable);
+
+/* call MatchFinderMt_InitMt() before IMatchFinder::Init() */
+SRes MatchFinderMt_InitMt(CMatchFinderMt *p);
 void MatchFinderMt_ReleaseStream(CMatchFinderMt *p);
 
-#ifdef __cplusplus
-}
-#endif
+EXTERN_C_END
 
 #endif

C/LzFindOpt.c

@@ -0,0 +1,578 @@
+/* LzFindOpt.c -- multithreaded Match finder for LZ algorithms
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
+
+#include "CpuArch.h"
+#include "LzFind.h"
+
+// #include "LzFindMt.h"
+
+// #define LOG_ITERS
+
+// #define LOG_THREAD
+
+#ifdef LOG_THREAD
+#include <stdio.h>
+#define PRF(x) x
+#else
+// #define PRF(x)
+#endif
+
+#ifdef LOG_ITERS
+#include <stdio.h>
+UInt64 g_NumIters_Tree;
+UInt64 g_NumIters_Loop;
+UInt64 g_NumIters_Bytes;
+#define LOG_ITER(x) x
+#else
+#define LOG_ITER(x)
+#endif
+
+// ---------- BT THREAD ----------
+
+#define USE_SON_PREFETCH
+#define USE_LONG_MATCH_OPT
+
+#define kEmptyHashValue 0
+
+// #define CYC_TO_POS_OFFSET 0
+
+// #define CYC_TO_POS_OFFSET 1 // for debug
+
+/*
+Z7_NO_INLINE
+UInt32 * Z7_FASTCALL GetMatchesSpecN_1(const Byte *lenLimit, size_t pos, const Byte *cur, CLzRef *son,
+    UInt32 _cutValue, UInt32 *d, size_t _maxLen, const UInt32 *hash, const UInt32 *limit, const UInt32 *size, UInt32 *posRes)
+{
+  do
+  {
+    UInt32 delta;
+    if (hash == size)
+      break;
+    delta = *hash++;
+
+    if (delta == 0 || delta > (UInt32)pos)
+      return NULL;
+
+    lenLimit++;
+
+    if (delta == (UInt32)pos)
+    {
+      CLzRef *ptr1 = son + ((size_t)pos << 1) - CYC_TO_POS_OFFSET * 2;
+      *d++ = 0;
+      ptr1[0] = kEmptyHashValue;
+      ptr1[1] = kEmptyHashValue;
+    }
+else
+{
+  UInt32 *_distances = ++d;
+
+  CLzRef *ptr0 = son + ((size_t)(pos) << 1) - CYC_TO_POS_OFFSET * 2 + 1;
+  CLzRef *ptr1 = son + ((size_t)(pos) << 1) - CYC_TO_POS_OFFSET * 2;
+
+  const Byte *len0 = cur, *len1 = cur;
+  UInt32 cutValue = _cutValue;
+  const Byte *maxLen = cur + _maxLen;
+
+  for (LOG_ITER(g_NumIters_Tree++);;)
+  {
+    LOG_ITER(g_NumIters_Loop++);
+    {
+      const ptrdiff_t diff = (ptrdiff_t)0 - (ptrdiff_t)delta;
+      CLzRef *pair = son + ((size_t)(((ptrdiff_t)pos - CYC_TO_POS_OFFSET) + diff) << 1);
+      const Byte *len = (len0 < len1 ? len0 : len1);
+
+    #ifdef USE_SON_PREFETCH
+      const UInt32 pair0 = *pair;
+    #endif
+
+      if (len[diff] == len[0])
+      {
+        if (++len != lenLimit && len[diff] == len[0])
+          while (++len != lenLimit)
+          {
+            LOG_ITER(g_NumIters_Bytes++);
+            if (len[diff] != len[0])
+              break;
+          }
+        if (maxLen < len)
+        {
+          maxLen = len;
+          *d++ = (UInt32)(len - cur);
+          *d++ = delta - 1;
+          
+          if (len == lenLimit)
+          {
+            const UInt32 pair1 = pair[1];
+            *ptr1 =
+              #ifdef USE_SON_PREFETCH
+                pair0;
+              #else
+                pair[0];
+              #endif
+            *ptr0 = pair1;
+
+            _distances[-1] = (UInt32)(d - _distances);
+
+            #ifdef USE_LONG_MATCH_OPT
+
+                if (hash == size || *hash != delta || lenLimit[diff] != lenLimit[0] || d >= limit)
+                  break;
+
+            {
+              for (;;)
+              {
+                hash++;
+                pos++;
+                cur++;
+                lenLimit++;
+                {
+                  CLzRef *ptr = son + ((size_t)(pos) << 1) - CYC_TO_POS_OFFSET * 2;
+                  #if 0
+                  *(UInt64 *)(void *)ptr = ((const UInt64 *)(const void *)ptr)[diff];
+                  #else
+                  const UInt32 p0 = ptr[0 + (diff * 2)];
+                  const UInt32 p1 = ptr[1 + (diff * 2)];
+                  ptr[0] = p0;
+                  ptr[1] = p1;
+                  // ptr[0] = ptr[0 + (diff * 2)];
+                  // ptr[1] = ptr[1 + (diff * 2)];
+                  #endif
+                }
+                // PrintSon(son + 2, pos - 1);
+                // printf("\npos = %x delta = %x\n", pos, delta);
+                len++;
+                *d++ = 2;
+                *d++ = (UInt32)(len - cur);
+                *d++ = delta - 1;
+                if (hash == size || *hash != delta || lenLimit[diff] != lenLimit[0] || d >= limit)
+                  break;
+              }
+            }
+            #endif
+
+            break;
+          }
+        }
+      }
+
+      {
+        const UInt32 curMatch = (UInt32)pos - delta; // (UInt32)(pos + diff);
+        if (len[diff] < len[0])
+        {
+          delta = pair[1];
+          if (delta >= curMatch)
+            return NULL;
+          *ptr1 = curMatch;
+          ptr1 = pair + 1;
+          len1 = len;
+        }
+        else
+        {
+          delta = *pair;
+          if (delta >= curMatch)
+            return NULL;
+          *ptr0 = curMatch;
+          ptr0 = pair;
+          len0 = len;
+        }
+
+        delta = (UInt32)pos - delta;
+ 
+        if (--cutValue == 0 || delta >= pos)
+        {
+          *ptr0 = *ptr1 = kEmptyHashValue;
+          _distances[-1] = (UInt32)(d - _distances);
+          break;
+        }
+      }
+    }
+  } // for (tree iterations)
+}
+    pos++;
+    cur++;
+  }
+  while (d < limit);
+  *posRes = (UInt32)pos;
+  return d;
+}
+*/
+
+/* define cbs if you use 2 functions.
+       GetMatchesSpecN_1() :  (pos <  _cyclicBufferSize)
+       GetMatchesSpecN_2() :  (pos >= _cyclicBufferSize)
+
+  do not define cbs if you use 1 function:
+       GetMatchesSpecN_2()
+*/
+
+// #define cbs _cyclicBufferSize
+
+/*
+  we use size_t for (pos) and (_cyclicBufferPos_ instead of UInt32
+  to eliminate "movsx" BUG in old MSVC x64 compiler.
+*/
+
+UInt32 * Z7_FASTCALL GetMatchesSpecN_2(const Byte *lenLimit, size_t pos, const Byte *cur, CLzRef *son,
+    UInt32 _cutValue, UInt32 *d, size_t _maxLen, const UInt32 *hash, const UInt32 *limit, const UInt32 *size,
+    size_t _cyclicBufferPos, UInt32 _cyclicBufferSize,
+    UInt32 *posRes);
+
+Z7_NO_INLINE
+UInt32 * Z7_FASTCALL GetMatchesSpecN_2(const Byte *lenLimit, size_t pos, const Byte *cur, CLzRef *son,
+    UInt32 _cutValue, UInt32 *d, size_t _maxLen, const UInt32 *hash, const UInt32 *limit, const UInt32 *size,
+    size_t _cyclicBufferPos, UInt32 _cyclicBufferSize,
+    UInt32 *posRes)
+{
+  do // while (hash != size)
+  {
+    UInt32 delta;
+    
+  #ifndef cbs
+    UInt32 cbs;
+  #endif
+
+    if (hash == size)
+      break;
+
+    delta = *hash++;
+
+    if (delta == 0)
+      return NULL;
+
+    lenLimit++;
+
+  #ifndef cbs
+    cbs = _cyclicBufferSize;
+    if ((UInt32)pos < cbs)
+    {
+      if (delta > (UInt32)pos)
+        return NULL;
+      cbs = (UInt32)pos;
+    }
+  #endif
+
+    if (delta >= cbs)
+    {
+      CLzRef *ptr1 = son + ((size_t)_cyclicBufferPos << 1);
+      *d++ = 0;
+      ptr1[0] = kEmptyHashValue;
+      ptr1[1] = kEmptyHashValue;
+    }
+else
+{
+  UInt32 *_distances = ++d;
+
+  CLzRef *ptr0 = son + ((size_t)_cyclicBufferPos << 1) + 1;
+  CLzRef *ptr1 = son + ((size_t)_cyclicBufferPos << 1);
+
+  UInt32 cutValue = _cutValue;
+  const Byte *len0 = cur, *len1 = cur;
+  const Byte *maxLen = cur + _maxLen;
+
+  // if (cutValue == 0) { *ptr0 = *ptr1 = kEmptyHashValue; } else
+  for (LOG_ITER(g_NumIters_Tree++);;)
+  {
+    LOG_ITER(g_NumIters_Loop++);
+    {
+      // SPEC code
+      CLzRef *pair = son + ((size_t)((ptrdiff_t)_cyclicBufferPos - (ptrdiff_t)delta
+          + (ptrdiff_t)(UInt32)(_cyclicBufferPos < delta ? cbs : 0)
+          ) << 1);
+
+      const ptrdiff_t diff = (ptrdiff_t)0 - (ptrdiff_t)delta;
+      const Byte *len = (len0 < len1 ? len0 : len1);
+
+    #ifdef USE_SON_PREFETCH
+      const UInt32 pair0 = *pair;
+    #endif
+
+      if (len[diff] == len[0])
+      {
+        if (++len != lenLimit && len[diff] == len[0])
+          while (++len != lenLimit)
+          {
+            LOG_ITER(g_NumIters_Bytes++);
+            if (len[diff] != len[0])
+              break;
+          }
+        if (maxLen < len)
+        {
+          maxLen = len;
+          *d++ = (UInt32)(len - cur);
+          *d++ = delta - 1;
+          
+          if (len == lenLimit)
+          {
+            const UInt32 pair1 = pair[1];
+            *ptr1 =
+              #ifdef USE_SON_PREFETCH
+                pair0;
+              #else
+                pair[0];
+              #endif
+            *ptr0 = pair1;
+
+            _distances[-1] = (UInt32)(d - _distances);
+
+            #ifdef USE_LONG_MATCH_OPT
+
+                if (hash == size || *hash != delta || lenLimit[diff] != lenLimit[0] || d >= limit)
+                  break;
+
+            {
+              for (;;)
+              {
+                *d++ = 2;
+                *d++ = (UInt32)(lenLimit - cur);
+                *d++ = delta - 1;
+                cur++;
+                lenLimit++;
+                // SPEC
+                _cyclicBufferPos++;
+                {
+                  // SPEC code
+                  CLzRef *dest = son + ((size_t)(_cyclicBufferPos) << 1);
+                  const CLzRef *src = dest + ((diff
+                      + (ptrdiff_t)(UInt32)((_cyclicBufferPos < delta) ? cbs : 0)) << 1);
+                  // CLzRef *ptr = son + ((size_t)(pos) << 1) - CYC_TO_POS_OFFSET * 2;
+                  #if 0
+                  *(UInt64 *)(void *)dest = *((const UInt64 *)(const void *)src);
+                  #else
+                  const UInt32 p0 = src[0];
+                  const UInt32 p1 = src[1];
+                  dest[0] = p0;
+                  dest[1] = p1;
+                  #endif
+                }
+                pos++;
+                hash++;
+                if (hash == size || *hash != delta || lenLimit[diff] != lenLimit[0] || d >= limit)
+                  break;
+              } // for() end for long matches
+            }
+            #endif
+
+            break; // break from TREE iterations
+          }
+        }
+      }
+      {
+        const UInt32 curMatch = (UInt32)pos - delta; // (UInt32)(pos + diff);
+        if (len[diff] < len[0])
+        {
+          delta = pair[1];
+          *ptr1 = curMatch;
+          ptr1 = pair + 1;
+          len1 = len;
+          if (delta >= curMatch)
+            return NULL;
+        }
+        else
+        {
+          delta = *pair;
+          *ptr0 = curMatch;
+          ptr0 = pair;
+          len0 = len;
+          if (delta >= curMatch)
+            return NULL;
+        }
+        delta = (UInt32)pos - delta;
+ 
+        if (--cutValue == 0 || delta >= cbs)
+        {
+          *ptr0 = *ptr1 = kEmptyHashValue;
+          _distances[-1] = (UInt32)(d - _distances);
+          break;
+        }
+      }
+    }
+  } // for (tree iterations)
+}
+    pos++;
+    _cyclicBufferPos++;
+    cur++;
+  }
+  while (d < limit);
+  *posRes = (UInt32)pos;
+  return d;
+}
+
+
+
+/*
+typedef UInt32 uint32plus; // size_t
+
+UInt32 * Z7_FASTCALL GetMatchesSpecN_3(uint32plus lenLimit, size_t pos, const Byte *cur, CLzRef *son,
+    UInt32 _cutValue, UInt32 *d, uint32plus _maxLen, const UInt32 *hash, const UInt32 *limit, const UInt32 *size,
+    size_t _cyclicBufferPos, UInt32 _cyclicBufferSize,
+    UInt32 *posRes)
+{
+  do // while (hash != size)
+  {
+    UInt32 delta;
+
+  #ifndef cbs
+    UInt32 cbs;
+  #endif
+
+    if (hash == size)
+      break;
+
+    delta = *hash++;
+
+    if (delta == 0)
+      return NULL;
+
+  #ifndef cbs
+    cbs = _cyclicBufferSize;
+    if ((UInt32)pos < cbs)
+    {
+      if (delta > (UInt32)pos)
+        return NULL;
+      cbs = (UInt32)pos;
+    }
+  #endif
+    
+    if (delta >= cbs)
+    {
+      CLzRef *ptr1 = son + ((size_t)_cyclicBufferPos << 1);
+      *d++ = 0;
+      ptr1[0] = kEmptyHashValue;
+      ptr1[1] = kEmptyHashValue;
+    }
+else
+{
+  CLzRef *ptr0 = son + ((size_t)_cyclicBufferPos << 1) + 1;
+  CLzRef *ptr1 = son + ((size_t)_cyclicBufferPos << 1);
+  UInt32 *_distances = ++d;
+  uint32plus len0 = 0, len1 = 0;
+  UInt32 cutValue = _cutValue;
+  uint32plus maxLen = _maxLen;
+  // lenLimit++; // const Byte *lenLimit = cur + _lenLimit;
+
+  for (LOG_ITER(g_NumIters_Tree++);;)
+  {
+    LOG_ITER(g_NumIters_Loop++);
+    {
+      // const ptrdiff_t diff = (ptrdiff_t)0 - (ptrdiff_t)delta;
+      CLzRef *pair = son + ((size_t)((ptrdiff_t)_cyclicBufferPos - delta
+          + (ptrdiff_t)(UInt32)(_cyclicBufferPos < delta ? cbs : 0)
+          ) << 1);
+      const Byte *pb = cur - delta;
+      uint32plus len = (len0 < len1 ? len0 : len1);
+
+    #ifdef USE_SON_PREFETCH
+      const UInt32 pair0 = *pair;
+    #endif
+
+      if (pb[len] == cur[len])
+      {
+        if (++len != lenLimit && pb[len] == cur[len])
+          while (++len != lenLimit)
+            if (pb[len] != cur[len])
+              break;
+        if (maxLen < len)
+        {
+          maxLen = len;
+          *d++ = (UInt32)len;
+          *d++ = delta - 1;
+          if (len == lenLimit)
+          {
+            {
+              const UInt32 pair1 = pair[1];
+              *ptr0 = pair1;
+              *ptr1 =
+              #ifdef USE_SON_PREFETCH
+                pair0;
+              #else
+                pair[0];
+              #endif
+            }
+
+            _distances[-1] = (UInt32)(d - _distances);
+
+            #ifdef USE_LONG_MATCH_OPT
+
+                if (hash == size || *hash != delta || pb[lenLimit] != cur[lenLimit] || d >= limit)
+                  break;
+
+            {
+              const ptrdiff_t diff = (ptrdiff_t)0 - (ptrdiff_t)delta;
+              for (;;)
+              {
+                *d++ = 2;
+                *d++ = (UInt32)lenLimit;
+                *d++ = delta - 1;
+                _cyclicBufferPos++;
+                {
+                  CLzRef *dest = son + ((size_t)_cyclicBufferPos << 1);
+                  const CLzRef *src = dest + ((diff +
+                      (ptrdiff_t)(UInt32)(_cyclicBufferPos < delta ? cbs : 0)) << 1);
+                #if 0
+                  *(UInt64 *)(void *)dest = *((const UInt64 *)(const void *)src);
+                #else
+                  const UInt32 p0 = src[0];
+                  const UInt32 p1 = src[1];
+                  dest[0] = p0;
+                  dest[1] = p1;
+                #endif
+                }
+                hash++;
+                pos++;
+                cur++;
+                pb++;
+                if (hash == size || *hash != delta || pb[lenLimit] != cur[lenLimit] || d >= limit)
+                  break;
+              }
+            }
+            #endif
+
+            break;
+          }
+        }
+      }
+      {
+        const UInt32 curMatch = (UInt32)pos - delta;
+        if (pb[len] < cur[len])
+        {
+          delta = pair[1];
+          *ptr1 = curMatch;
+          ptr1 = pair + 1;
+          len1 = len;
+        }
+        else
+        {
+          delta = *pair;
+          *ptr0 = curMatch;
+          ptr0 = pair;
+          len0 = len;
+        }
+
+        {
+          if (delta >= curMatch)
+            return NULL;
+          delta = (UInt32)pos - delta;
+          if (delta >= cbs
+              // delta >= _cyclicBufferSize || delta >= pos
+              || --cutValue == 0)
+          {
+            *ptr0 = *ptr1 = kEmptyHashValue;
+            _distances[-1] = (UInt32)(d - _distances);
+            break;
+          }
+        }
+      }
+    }
+  } // for (tree iterations)
+}
+    pos++;
+    _cyclicBufferPos++;
+    cur++;
+  }
+  while (d < limit);
+  *posRes = (UInt32)pos;
+  return d;
+}
+*/

C/LzHash.h

@@ -1,54 +1,34 @@
-/* LzHash.h -- HASH functions for LZ algorithms
-2009-02-07 : Igor Pavlov : Public domain */
+/* LzHash.h -- HASH constants for LZ algorithms
+2023-03-05 : Igor Pavlov : Public domain */
 
-#ifndef __LZ_HASH_H
-#define __LZ_HASH_H
+#ifndef ZIP7_INC_LZ_HASH_H
+#define ZIP7_INC_LZ_HASH_H
+
+/*
+  (kHash2Size >= (1 <<  8)) : Required
+  (kHash3Size >= (1 << 16)) : Required
+*/
 
 #define kHash2Size (1 << 10)
 #define kHash3Size (1 << 16)
-#define kHash4Size (1 << 20)
+// #define kHash4Size (1 << 20)
 
 #define kFix3HashSize (kHash2Size)
 #define kFix4HashSize (kHash2Size + kHash3Size)
-#define kFix5HashSize (kHash2Size + kHash3Size + kHash4Size)
+// #define kFix5HashSize (kHash2Size + kHash3Size + kHash4Size)
 
-#define HASH2_CALC hashValue = cur[0] | ((UInt32)cur[1] << 8);
+/*
+  We use up to 3 crc values for hash:
+    crc0
+    crc1 << Shift_1
+    crc2 << Shift_2
+  (Shift_1 = 5) and (Shift_2 = 10) is good tradeoff.
+  Small values for Shift are not good for collision rate.
+  Big value for Shift_2 increases the minimum size
+  of hash table, that will be slow for small files.
+*/
 
-#define HASH3_CALC { \
-  UInt32 temp = p->crc[cur[0]] ^ cur[1]; \
-  hash2Value = temp & (kHash2Size - 1); \
-  hashValue = (temp ^ ((UInt32)cur[2] << 8)) & p->hashMask; }
-
-#define HASH4_CALC { \
-  UInt32 temp = p->crc[cur[0]] ^ cur[1]; \
-  hash2Value = temp & (kHash2Size - 1); \
-  hash3Value = (temp ^ ((UInt32)cur[2] << 8)) & (kHash3Size - 1); \
-  hashValue = (temp ^ ((UInt32)cur[2] << 8) ^ (p->crc[cur[3]] << 5)) & p->hashMask; }
-
-#define HASH5_CALC { \
-  UInt32 temp = p->crc[cur[0]] ^ cur[1]; \
-  hash2Value = temp & (kHash2Size - 1); \
-  hash3Value = (temp ^ ((UInt32)cur[2] << 8)) & (kHash3Size - 1); \
-  hash4Value = (temp ^ ((UInt32)cur[2] << 8) ^ (p->crc[cur[3]] << 5)); \
-  hashValue = (hash4Value ^ (p->crc[cur[4]] << 3)) & p->hashMask; \
-  hash4Value &= (kHash4Size - 1); }
-
-/* #define HASH_ZIP_CALC hashValue = ((cur[0] | ((UInt32)cur[1] << 8)) ^ p->crc[cur[2]]) & 0xFFFF; */
-#define HASH_ZIP_CALC hashValue = ((cur[2] | ((UInt32)cur[0] << 8)) ^ p->crc[cur[1]]) & 0xFFFF;
-
-
-#define MT_HASH2_CALC \
-  hash2Value = (p->crc[cur[0]] ^ cur[1]) & (kHash2Size - 1);
-
-#define MT_HASH3_CALC { \
-  UInt32 temp = p->crc[cur[0]] ^ cur[1]; \
-  hash2Value = temp & (kHash2Size - 1); \
-  hash3Value = (temp ^ ((UInt32)cur[2] << 8)) & (kHash3Size - 1); }
-
-#define MT_HASH4_CALC { \
-  UInt32 temp = p->crc[cur[0]] ^ cur[1]; \
-  hash2Value = temp & (kHash2Size - 1); \
-  hash3Value = (temp ^ ((UInt32)cur[2] << 8)) & (kHash3Size - 1); \
-  hash4Value = (temp ^ ((UInt32)cur[2] << 8) ^ (p->crc[cur[3]] << 5)) & (kHash4Size - 1); }
+#define kLzHash_CrcShift_1 5
+#define kLzHash_CrcShift_2 10
 
 #endif

C/Lzma2Dec.c

@@ -1,8 +1,10 @@
 /* Lzma2Dec.c -- LZMA2 Decoder
-2009-05-03 : Igor Pavlov : Public domain */
+2023-03-03 : Igor Pavlov : Public domain */
 
 /* #define SHOW_DEBUG_INFO */
 
+#include "Precomp.h"
+
 #ifdef SHOW_DEBUG_INFO
 #include <stdio.h>
 #endif
@@ -12,28 +14,22 @@
 #include "Lzma2Dec.h"
 
 /*
-00000000  -  EOS
-00000001 U U  -  Uncompressed Reset Dic
-00000010 U U  -  Uncompressed No Reset
-100uuuuu U U P P  -  LZMA no reset
-101uuuuu U U P P  -  LZMA reset state
-110uuuuu U U P P S  -  LZMA reset state + new prop
-111uuuuu U U P P S  -  LZMA reset state + new prop + reset dic
+00000000  -  End of data
+00000001 U U  -  Uncompressed, reset dic, need reset state and set new prop
+00000010 U U  -  Uncompressed, no reset
+100uuuuu U U P P  -  LZMA, no reset
+101uuuuu U U P P  -  LZMA, reset state
+110uuuuu U U P P S  -  LZMA, reset state + set new prop
+111uuuuu U U P P S  -  LZMA, reset state + set new prop, reset dic
 
   u, U - Unpack Size
   P - Pack Size
   S - Props
 */
 
-#define LZMA2_CONTROL_LZMA (1 << 7)
-#define LZMA2_CONTROL_COPY_NO_RESET 2
 #define LZMA2_CONTROL_COPY_RESET_DIC 1
-#define LZMA2_CONTROL_EOF 0
 
-#define LZMA2_IS_UNCOMPRESSED_STATE(p) (((p)->control & LZMA2_CONTROL_LZMA) == 0)
-
-#define LZMA2_GET_LZMA_MODE(p) (((p)->control >> 5) & 3)
-#define LZMA2_IS_THERE_PROP(mode) ((mode) >= 2)
+#define LZMA2_IS_UNCOMPRESSED_STATE(p) (((p)->control & (1 << 7)) == 0)
 
 #define LZMA2_LCLP_MAX 4
 #define LZMA2_DIC_SIZE_FROM_PROP(p) (((UInt32)2 | ((p) & 1)) << ((p) / 2 + 11))
@@ -72,47 +68,57 @@ static SRes Lzma2Dec_GetOldProps(Byte prop, Byte *props)
   return SZ_OK;
 }
 
-SRes Lzma2Dec_AllocateProbs(CLzma2Dec *p, Byte prop, ISzAlloc *alloc)
+SRes Lzma2Dec_AllocateProbs(CLzma2Dec *p, Byte prop, ISzAllocPtr alloc)
 {
   Byte props[LZMA_PROPS_SIZE];
-  RINOK(Lzma2Dec_GetOldProps(prop, props));
+  RINOK(Lzma2Dec_GetOldProps(prop, props))
   return LzmaDec_AllocateProbs(&p->decoder, props, LZMA_PROPS_SIZE, alloc);
 }
 
-SRes Lzma2Dec_Allocate(CLzma2Dec *p, Byte prop, ISzAlloc *alloc)
+SRes Lzma2Dec_Allocate(CLzma2Dec *p, Byte prop, ISzAllocPtr alloc)
 {
   Byte props[LZMA_PROPS_SIZE];
-  RINOK(Lzma2Dec_GetOldProps(prop, props));
+  RINOK(Lzma2Dec_GetOldProps(prop, props))
   return LzmaDec_Allocate(&p->decoder, props, LZMA_PROPS_SIZE, alloc);
 }
 
 void Lzma2Dec_Init(CLzma2Dec *p)
 {
   p->state = LZMA2_STATE_CONTROL;
-  p->needInitDic = True;
-  p->needInitState = True;
-  p->needInitProp = True;
+  p->needInitLevel = 0xE0;
+  p->isExtraMode = False;
+  p->unpackSize = 0;
+  
+  // p->decoder.dicPos = 0; // we can use it instead of full init
   LzmaDec_Init(&p->decoder);
 }
 
-static ELzma2State Lzma2Dec_UpdateState(CLzma2Dec *p, Byte b)
+// ELzma2State
+static unsigned Lzma2Dec_UpdateState(CLzma2Dec *p, Byte b)
 {
-  switch(p->state)
+  switch (p->state)
   {
     case LZMA2_STATE_CONTROL:
+      p->isExtraMode = False;
       p->control = b;
-      PRF(printf("\n %4X ", p->decoder.dicPos));
-      PRF(printf(" %2X", b));
-      if (p->control == 0)
+      PRF(printf("\n %8X", (unsigned)p->decoder.dicPos));
+      PRF(printf(" %02X", (unsigned)b));
+      if (b == 0)
         return LZMA2_STATE_FINISHED;
       if (LZMA2_IS_UNCOMPRESSED_STATE(p))
       {
-        if ((p->control & 0x7F) > 2)
+        if (b == LZMA2_CONTROL_COPY_RESET_DIC)
+          p->needInitLevel = 0xC0;
+        else if (b > 2 || p->needInitLevel == 0xE0)
           return LZMA2_STATE_ERROR;
-        p->unpackSize = 0;
       }
       else
-        p->unpackSize = (UInt32)(p->control & 0x1F) << 16;
+      {
+        if (b < p->needInitLevel)
+          return LZMA2_STATE_ERROR;
+        p->needInitLevel = 0;
+        p->unpackSize = (UInt32)(b & 0x1F) << 16;
+      }
       return LZMA2_STATE_UNPACK0;
     
     case LZMA2_STATE_UNPACK0:
@@ -122,8 +128,8 @@ static ELzma2State Lzma2Dec_UpdateState(CLzma2Dec *p, Byte b)
     case LZMA2_STATE_UNPACK1:
       p->unpackSize |= (UInt32)b;
       p->unpackSize++;
-      PRF(printf(" %8d", p->unpackSize));
-      return (LZMA2_IS_UNCOMPRESSED_STATE(p)) ? LZMA2_STATE_DATA : LZMA2_STATE_PACK0;
+      PRF(printf(" %7u", (unsigned)p->unpackSize));
+      return LZMA2_IS_UNCOMPRESSED_STATE(p) ? LZMA2_STATE_DATA : LZMA2_STATE_PACK0;
     
     case LZMA2_STATE_PACK0:
       p->packSize = (UInt32)b << 8;
@@ -132,24 +138,23 @@ static ELzma2State Lzma2Dec_UpdateState(CLzma2Dec *p, Byte b)
     case LZMA2_STATE_PACK1:
       p->packSize |= (UInt32)b;
       p->packSize++;
-      PRF(printf(" %8d", p->packSize));
-      return LZMA2_IS_THERE_PROP(LZMA2_GET_LZMA_MODE(p)) ? LZMA2_STATE_PROP:
-        (p->needInitProp ? LZMA2_STATE_ERROR : LZMA2_STATE_DATA);
+      // if (p->packSize < 5) return LZMA2_STATE_ERROR;
+      PRF(printf(" %5u", (unsigned)p->packSize));
+      return (p->control & 0x40) ? LZMA2_STATE_PROP : LZMA2_STATE_DATA;
 
     case LZMA2_STATE_PROP:
     {
-      int lc, lp;
+      unsigned lc, lp;
       if (b >= (9 * 5 * 5))
         return LZMA2_STATE_ERROR;
       lc = b % 9;
       b /= 9;
-      p->decoder.prop.pb = b / 5;
+      p->decoder.prop.pb = (Byte)(b / 5);
       lp = b % 5;
       if (lc + lp > LZMA2_LCLP_MAX)
         return LZMA2_STATE_ERROR;
-      p->decoder.prop.lc = lc;
-      p->decoder.prop.lp = lp;
-      p->needInitProp = False;
+      p->decoder.prop.lc = (Byte)lc;
+      p->decoder.prop.lp = (Byte)lp;
       return LZMA2_STATE_DATA;
     }
   }
@@ -165,7 +170,8 @@ static void LzmaDec_UpdateWithUncompressed(CLzmaDec *p, const Byte *src, SizeT s
   p->processedPos += (UInt32)size;
 }
 
-void LzmaDec_InitDicAndState(CLzmaDec *p, Bool initDic, Bool initState);
+void LzmaDec_InitDicAndState(CLzmaDec *p, BoolInt initDic, BoolInt initState);
+
 
 SRes Lzma2Dec_DecodeToDic(CLzma2Dec *p, SizeT dicLimit,
     const Byte *src, SizeT *srcLen, ELzmaFinishMode finishMode, ELzmaStatus *status)
@@ -174,16 +180,24 @@ SRes Lzma2Dec_DecodeToDic(CLzma2Dec *p, SizeT dicLimit,
   *srcLen = 0;
   *status = LZMA_STATUS_NOT_SPECIFIED;
 
-  while (p->state != LZMA2_STATE_FINISHED)
+  while (p->state != LZMA2_STATE_ERROR)
   {
-    SizeT dicPos = p->decoder.dicPos;
-    if (p->state == LZMA2_STATE_ERROR)
-      return SZ_ERROR_DATA;
+    SizeT dicPos;
+
+    if (p->state == LZMA2_STATE_FINISHED)
+    {
+      *status = LZMA_STATUS_FINISHED_WITH_MARK;
+      return SZ_OK;
+    }
+    
+    dicPos = p->decoder.dicPos;
+    
     if (dicPos == dicLimit && finishMode == LZMA_FINISH_ANY)
     {
       *status = LZMA_STATUS_NOT_FINISHED;
       return SZ_OK;
     }
+
     if (p->state != LZMA2_STATE_DATA && p->state != LZMA2_STATE_DATA_CONT)
     {
       if (*srcLen == inSize)
@@ -193,22 +207,25 @@ SRes Lzma2Dec_DecodeToDic(CLzma2Dec *p, SizeT dicLimit,
       }
       (*srcLen)++;
       p->state = Lzma2Dec_UpdateState(p, *src++);
+      if (dicPos == dicLimit && p->state != LZMA2_STATE_FINISHED)
+        break;
       continue;
     }
+    
     {
-      SizeT destSizeCur = dicLimit - dicPos;
-      SizeT srcSizeCur = inSize - *srcLen;
+      SizeT inCur = inSize - *srcLen;
+      SizeT outCur = dicLimit - dicPos;
       ELzmaFinishMode curFinishMode = LZMA_FINISH_ANY;
       
-      if (p->unpackSize <= destSizeCur)
+      if (outCur >= p->unpackSize)
       {
-        destSizeCur = (SizeT)p->unpackSize;
+        outCur = (SizeT)p->unpackSize;
         curFinishMode = LZMA_FINISH_END;
       }
 
       if (LZMA2_IS_UNCOMPRESSED_STATE(p))
       {
-        if (*srcLen == inSize)
+        if (inCur == 0)
         {
           *status = LZMA_STATUS_NEEDS_MORE_INPUT;
           return SZ_OK;
@@ -216,141 +233,259 @@ SRes Lzma2Dec_DecodeToDic(CLzma2Dec *p, SizeT dicLimit,
 
         if (p->state == LZMA2_STATE_DATA)
         {
-          Bool initDic = (p->control == LZMA2_CONTROL_COPY_RESET_DIC);
-          if (initDic)
-            p->needInitProp = p->needInitState = True;
-          else if (p->needInitDic)
-            return SZ_ERROR_DATA;
-          p->needInitDic = False;
+          BoolInt initDic = (p->control == LZMA2_CONTROL_COPY_RESET_DIC);
           LzmaDec_InitDicAndState(&p->decoder, initDic, False);
         }
 
-        if (srcSizeCur > destSizeCur)
-          srcSizeCur = destSizeCur;
+        if (inCur > outCur)
+          inCur = outCur;
+        if (inCur == 0)
+          break;
 
-        if (srcSizeCur == 0)
-          return SZ_ERROR_DATA;
+        LzmaDec_UpdateWithUncompressed(&p->decoder, src, inCur);
 
-        LzmaDec_UpdateWithUncompressed(&p->decoder, src, srcSizeCur);
-
-        src += srcSizeCur;
-        *srcLen += srcSizeCur;
-        p->unpackSize -= (UInt32)srcSizeCur;
+        src += inCur;
+        *srcLen += inCur;
+        p->unpackSize -= (UInt32)inCur;
         p->state = (p->unpackSize == 0) ? LZMA2_STATE_CONTROL : LZMA2_STATE_DATA_CONT;
       }
       else
       {
-        SizeT outSizeProcessed;
         SRes res;
 
         if (p->state == LZMA2_STATE_DATA)
         {
-          int mode = LZMA2_GET_LZMA_MODE(p);
-          Bool initDic = (mode == 3);
-          Bool initState = (mode > 0);
-          if ((!initDic && p->needInitDic) || (!initState && p->needInitState))
-            return SZ_ERROR_DATA;
-          
+          BoolInt initDic = (p->control >= 0xE0);
+          BoolInt initState = (p->control >= 0xA0);
           LzmaDec_InitDicAndState(&p->decoder, initDic, initState);
-          p->needInitDic = False;
-          p->needInitState = False;
           p->state = LZMA2_STATE_DATA_CONT;
         }
-        if (srcSizeCur > p->packSize)
-          srcSizeCur = (SizeT)p->packSize;
-          
-        res = LzmaDec_DecodeToDic(&p->decoder, dicPos + destSizeCur, src, &srcSizeCur, curFinishMode, status);
+  
+        if (inCur > p->packSize)
+          inCur = (SizeT)p->packSize;
         
-        src += srcSizeCur;
-        *srcLen += srcSizeCur;
-        p->packSize -= (UInt32)srcSizeCur;
+        res = LzmaDec_DecodeToDic(&p->decoder, dicPos + outCur, src, &inCur, curFinishMode, status);
 
-        outSizeProcessed = p->decoder.dicPos - dicPos;
-        p->unpackSize -= (UInt32)outSizeProcessed;
+        src += inCur;
+        *srcLen += inCur;
+        p->packSize -= (UInt32)inCur;
+        outCur = p->decoder.dicPos - dicPos;
+        p->unpackSize -= (UInt32)outCur;
 
-        RINOK(res);
+        if (res != 0)
+          break;
+        
         if (*status == LZMA_STATUS_NEEDS_MORE_INPUT)
-          return res;
-
-        if (srcSizeCur == 0 && outSizeProcessed == 0)
         {
-          if (*status != LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK ||
-              p->unpackSize != 0 || p->packSize != 0)
-            return SZ_ERROR_DATA;
+          if (p->packSize == 0)
+            break;
+          return SZ_OK;
+        }
+
+        if (inCur == 0 && outCur == 0)
+        {
+          if (*status != LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK
+              || p->unpackSize != 0
+              || p->packSize != 0)
+            break;
           p->state = LZMA2_STATE_CONTROL;
         }
-        if (*status == LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK)
-          *status = LZMA_STATUS_NOT_FINISHED;
+        
+        *status = LZMA_STATUS_NOT_SPECIFIED;
       }
     }
   }
-  *status = LZMA_STATUS_FINISHED_WITH_MARK;
-  return SZ_OK;
+  
+  *status = LZMA_STATUS_NOT_SPECIFIED;
+  p->state = LZMA2_STATE_ERROR;
+  return SZ_ERROR_DATA;
 }
 
+
+
+
+ELzma2ParseStatus Lzma2Dec_Parse(CLzma2Dec *p,
+    SizeT outSize,
+    const Byte *src, SizeT *srcLen,
+    int checkFinishBlock)
+{
+  SizeT inSize = *srcLen;
+  *srcLen = 0;
+
+  while (p->state != LZMA2_STATE_ERROR)
+  {
+    if (p->state == LZMA2_STATE_FINISHED)
+      return (ELzma2ParseStatus)LZMA_STATUS_FINISHED_WITH_MARK;
+
+    if (outSize == 0 && !checkFinishBlock)
+      return (ELzma2ParseStatus)LZMA_STATUS_NOT_FINISHED;
+    
+    if (p->state != LZMA2_STATE_DATA && p->state != LZMA2_STATE_DATA_CONT)
+    {
+      if (*srcLen == inSize)
+        return (ELzma2ParseStatus)LZMA_STATUS_NEEDS_MORE_INPUT;
+      (*srcLen)++;
+
+      p->state = Lzma2Dec_UpdateState(p, *src++);
+
+      if (p->state == LZMA2_STATE_UNPACK0)
+      {
+        // if (p->decoder.dicPos != 0)
+        if (p->control == LZMA2_CONTROL_COPY_RESET_DIC || p->control >= 0xE0)
+          return LZMA2_PARSE_STATUS_NEW_BLOCK;
+        // if (outSize == 0) return LZMA_STATUS_NOT_FINISHED;
+      }
+
+      // The following code can be commented.
+      // It's not big problem, if we read additional input bytes.
+      // It will be stopped later in LZMA2_STATE_DATA / LZMA2_STATE_DATA_CONT state.
+
+      if (outSize == 0 && p->state != LZMA2_STATE_FINISHED)
+      {
+        // checkFinishBlock is true. So we expect that block must be finished,
+        // We can return LZMA_STATUS_NOT_SPECIFIED or LZMA_STATUS_NOT_FINISHED here
+        // break;
+        return (ELzma2ParseStatus)LZMA_STATUS_NOT_FINISHED;
+      }
+
+      if (p->state == LZMA2_STATE_DATA)
+        return LZMA2_PARSE_STATUS_NEW_CHUNK;
+
+      continue;
+    }
+
+    if (outSize == 0)
+      return (ELzma2ParseStatus)LZMA_STATUS_NOT_FINISHED;
+
+    {
+      SizeT inCur = inSize - *srcLen;
+
+      if (LZMA2_IS_UNCOMPRESSED_STATE(p))
+      {
+        if (inCur == 0)
+          return (ELzma2ParseStatus)LZMA_STATUS_NEEDS_MORE_INPUT;
+        if (inCur > p->unpackSize)
+          inCur = p->unpackSize;
+        if (inCur > outSize)
+          inCur = outSize;
+        p->decoder.dicPos += inCur;
+        src += inCur;
+        *srcLen += inCur;
+        outSize -= inCur;
+        p->unpackSize -= (UInt32)inCur;
+        p->state = (p->unpackSize == 0) ? LZMA2_STATE_CONTROL : LZMA2_STATE_DATA_CONT;
+      }
+      else
+      {
+        p->isExtraMode = True;
+
+        if (inCur == 0)
+        {
+          if (p->packSize != 0)
+            return (ELzma2ParseStatus)LZMA_STATUS_NEEDS_MORE_INPUT;
+        }
+        else if (p->state == LZMA2_STATE_DATA)
+        {
+          p->state = LZMA2_STATE_DATA_CONT;
+          if (*src != 0)
+          {
+            // first byte of lzma chunk must be Zero
+            *srcLen += 1;
+            p->packSize--;
+            break;
+          }
+        }
+  
+        if (inCur > p->packSize)
+          inCur = (SizeT)p->packSize;
+
+        src += inCur;
+        *srcLen += inCur;
+        p->packSize -= (UInt32)inCur;
+
+        if (p->packSize == 0)
+        {
+          SizeT rem = outSize;
+          if (rem > p->unpackSize)
+            rem = p->unpackSize;
+          p->decoder.dicPos += rem;
+          p->unpackSize -= (UInt32)rem;
+          outSize -= rem;
+          if (p->unpackSize == 0)
+            p->state = LZMA2_STATE_CONTROL;
+        }
+      }
+    }
+  }
+  
+  p->state = LZMA2_STATE_ERROR;
+  return (ELzma2ParseStatus)LZMA_STATUS_NOT_SPECIFIED;
+}
+
+
+
+
 SRes Lzma2Dec_DecodeToBuf(CLzma2Dec *p, Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen, ELzmaFinishMode finishMode, ELzmaStatus *status)
 {
   SizeT outSize = *destLen, inSize = *srcLen;
   *srcLen = *destLen = 0;
+  
   for (;;)
   {
-    SizeT srcSizeCur = inSize, outSizeCur, dicPos;
+    SizeT inCur = inSize, outCur, dicPos;
     ELzmaFinishMode curFinishMode;
     SRes res;
+    
     if (p->decoder.dicPos == p->decoder.dicBufSize)
       p->decoder.dicPos = 0;
     dicPos = p->decoder.dicPos;
-    if (outSize > p->decoder.dicBufSize - dicPos)
+    curFinishMode = LZMA_FINISH_ANY;
+    outCur = p->decoder.dicBufSize - dicPos;
+    
+    if (outCur >= outSize)
     {
-      outSizeCur = p->decoder.dicBufSize;
-      curFinishMode = LZMA_FINISH_ANY;
-    }
-    else
-    {
-      outSizeCur = dicPos + outSize;
+      outCur = outSize;
       curFinishMode = finishMode;
     }
 
-    res = Lzma2Dec_DecodeToDic(p, outSizeCur, src, &srcSizeCur, curFinishMode, status);
-    src += srcSizeCur;
-    inSize -= srcSizeCur;
-    *srcLen += srcSizeCur;
-    outSizeCur = p->decoder.dicPos - dicPos;
-    memcpy(dest, p->decoder.dic + dicPos, outSizeCur);
-    dest += outSizeCur;
-    outSize -= outSizeCur;
-    *destLen += outSizeCur;
+    res = Lzma2Dec_DecodeToDic(p, dicPos + outCur, src, &inCur, curFinishMode, status);
+    
+    src += inCur;
+    inSize -= inCur;
+    *srcLen += inCur;
+    outCur = p->decoder.dicPos - dicPos;
+    memcpy(dest, p->decoder.dic + dicPos, outCur);
+    dest += outCur;
+    outSize -= outCur;
+    *destLen += outCur;
     if (res != 0)
       return res;
-    if (outSizeCur == 0 || outSize == 0)
+    if (outCur == 0 || outSize == 0)
       return SZ_OK;
   }
 }
 
+
 SRes Lzma2Decode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen,
-    Byte prop, ELzmaFinishMode finishMode, ELzmaStatus *status, ISzAlloc *alloc)
+    Byte prop, ELzmaFinishMode finishMode, ELzmaStatus *status, ISzAllocPtr alloc)
 {
-  CLzma2Dec decoder;
+  CLzma2Dec p;
   SRes res;
   SizeT outSize = *destLen, inSize = *srcLen;
-  Byte props[LZMA_PROPS_SIZE];
-
-  Lzma2Dec_Construct(&decoder);
-
   *destLen = *srcLen = 0;
   *status = LZMA_STATUS_NOT_SPECIFIED;
-  decoder.decoder.dic = dest;
-  decoder.decoder.dicBufSize = outSize;
-
-  RINOK(Lzma2Dec_GetOldProps(prop, props));
-  RINOK(LzmaDec_AllocateProbs(&decoder.decoder, props, LZMA_PROPS_SIZE, alloc));
-  
+  Lzma2Dec_CONSTRUCT(&p)
+  RINOK(Lzma2Dec_AllocateProbs(&p, prop, alloc))
+  p.decoder.dic = dest;
+  p.decoder.dicBufSize = outSize;
+  Lzma2Dec_Init(&p);
   *srcLen = inSize;
-  res = Lzma2Dec_DecodeToDic(&decoder, outSize, src, srcLen, finishMode, status);
-  *destLen = decoder.decoder.dicPos;
+  res = Lzma2Dec_DecodeToDic(&p, outSize, src, srcLen, finishMode, status);
+  *destLen = p.decoder.dicPos;
   if (res == SZ_OK && *status == LZMA_STATUS_NEEDS_MORE_INPUT)
     res = SZ_ERROR_INPUT_EOF;
-
-  LzmaDec_FreeProbs(&decoder.decoder, alloc);
+  Lzma2Dec_FreeProbs(&p, alloc);
   return res;
 }
+
+#undef PRF

C/Lzma2Dec.h

@@ -1,38 +1,36 @@
 /* Lzma2Dec.h -- LZMA2 Decoder
-2009-05-03 : Igor Pavlov : Public domain */
+2023-03-03 : Igor Pavlov : Public domain */
 
-#ifndef __LZMA2_DEC_H
-#define __LZMA2_DEC_H
+#ifndef ZIP7_INC_LZMA2_DEC_H
+#define ZIP7_INC_LZMA2_DEC_H
 
 #include "LzmaDec.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
 /* ---------- State Interface ---------- */
 
 typedef struct
 {
-  CLzmaDec decoder;
+  unsigned state;
+  Byte control;
+  Byte needInitLevel;
+  Byte isExtraMode;
+  Byte _pad_;
   UInt32 packSize;
   UInt32 unpackSize;
-  int state;
-  Byte control;
-  Bool needInitDic;
-  Bool needInitState;
-  Bool needInitProp;
+  CLzmaDec decoder;
 } CLzma2Dec;
 
-#define Lzma2Dec_Construct(p) LzmaDec_Construct(&(p)->decoder)
-#define Lzma2Dec_FreeProbs(p, alloc) LzmaDec_FreeProbs(&(p)->decoder, alloc);
-#define Lzma2Dec_Free(p, alloc) LzmaDec_Free(&(p)->decoder, alloc);
+#define Lzma2Dec_CONSTRUCT(p)  LzmaDec_CONSTRUCT(&(p)->decoder)
+#define Lzma2Dec_Construct(p)  Lzma2Dec_CONSTRUCT(p)
+#define Lzma2Dec_FreeProbs(p, alloc)  LzmaDec_FreeProbs(&(p)->decoder, alloc)
+#define Lzma2Dec_Free(p, alloc)  LzmaDec_Free(&(p)->decoder, alloc)
 
-SRes Lzma2Dec_AllocateProbs(CLzma2Dec *p, Byte prop, ISzAlloc *alloc);
-SRes Lzma2Dec_Allocate(CLzma2Dec *p, Byte prop, ISzAlloc *alloc);
+SRes Lzma2Dec_AllocateProbs(CLzma2Dec *p, Byte prop, ISzAllocPtr alloc);
+SRes Lzma2Dec_Allocate(CLzma2Dec *p, Byte prop, ISzAllocPtr alloc);
 void Lzma2Dec_Init(CLzma2Dec *p);
 
-
 /*
 finishMode:
   It has meaning only if the decoding reaches output limit (*destLen or dicLimit).
@@ -55,6 +53,47 @@ SRes Lzma2Dec_DecodeToBuf(CLzma2Dec *p, Byte *dest, SizeT *destLen,
     const Byte *src, SizeT *srcLen, ELzmaFinishMode finishMode, ELzmaStatus *status);
 
 
+/* ---------- LZMA2 block and chunk parsing ---------- */
+
+/*
+Lzma2Dec_Parse() parses compressed data stream up to next independent block or next chunk data.
+It can return LZMA_STATUS_* code or LZMA2_PARSE_STATUS_* code:
+  - LZMA2_PARSE_STATUS_NEW_BLOCK - there is new block, and 1 additional byte (control byte of next block header) was read from input.
+  - LZMA2_PARSE_STATUS_NEW_CHUNK - there is new chunk, and only lzma2 header of new chunk was read.
+                                   CLzma2Dec::unpackSize contains unpack size of that chunk
+*/
+
+typedef enum
+{
+/*
+  LZMA_STATUS_NOT_SPECIFIED                 // data error
+  LZMA_STATUS_FINISHED_WITH_MARK
+  LZMA_STATUS_NOT_FINISHED                  //
+  LZMA_STATUS_NEEDS_MORE_INPUT
+  LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK   // unused
+*/
+  LZMA2_PARSE_STATUS_NEW_BLOCK = LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK + 1,
+  LZMA2_PARSE_STATUS_NEW_CHUNK
+} ELzma2ParseStatus;
+
+ELzma2ParseStatus Lzma2Dec_Parse(CLzma2Dec *p,
+    SizeT outSize,   // output size
+    const Byte *src, SizeT *srcLen,
+    int checkFinishBlock   // set (checkFinishBlock = 1), if it must read full input data, if decoder.dicPos reaches blockMax position.
+    );
+
+/*
+LZMA2 parser doesn't decode LZMA chunks, so we must read
+  full input LZMA chunk to decode some part of LZMA chunk.
+
+Lzma2Dec_GetUnpackExtra() returns the value that shows
+    max possible number of output bytes that can be output by decoder
+    at current input positon.
+*/
+
+#define Lzma2Dec_GetUnpackExtra(p)  ((p)->isExtraMode ? (p)->unpackSize : 0)
+
+
 /* ---------- One Call Interface ---------- */
 
 /*
@@ -75,10 +114,8 @@ Returns:
 */
 
 SRes Lzma2Decode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen,
-    Byte prop, ELzmaFinishMode finishMode, ELzmaStatus *status, ISzAlloc *alloc);
+    Byte prop, ELzmaFinishMode finishMode, ELzmaStatus *status, ISzAllocPtr alloc);
 
-#ifdef __cplusplus
-}
-#endif
+EXTERN_C_END
 
 #endif

C/Lzma2DecMt.h

@@ -0,0 +1,81 @@
+/* Lzma2DecMt.h -- LZMA2 Decoder Multi-thread
+2023-04-13 : Igor Pavlov : Public domain */
+
+#ifndef ZIP7_INC_LZMA2_DEC_MT_H
+#define ZIP7_INC_LZMA2_DEC_MT_H
+
+#include "7zTypes.h"
+
+EXTERN_C_BEGIN
+
+typedef struct
+{
+  size_t inBufSize_ST;
+  size_t outStep_ST;
+  
+  #ifndef Z7_ST
+  unsigned numThreads;
+  size_t inBufSize_MT;
+  size_t outBlockMax;
+  size_t inBlockMax;
+  #endif
+} CLzma2DecMtProps;
+
+/* init to single-thread mode */
+void Lzma2DecMtProps_Init(CLzma2DecMtProps *p);
+
+
+/* ---------- CLzma2DecMtHandle Interface ---------- */
+
+/* Lzma2DecMt_ * functions can return the following exit codes:
+SRes:
+  SZ_OK           - OK
+  SZ_ERROR_MEM    - Memory allocation error
+  SZ_ERROR_PARAM  - Incorrect paramater in props
+  SZ_ERROR_WRITE  - ISeqOutStream write callback error
+  // SZ_ERROR_OUTPUT_EOF - output buffer overflow - version with (Byte *) output
+  SZ_ERROR_PROGRESS - some break from progress callback
+  SZ_ERROR_THREAD - error in multithreading functions (only for Mt version)
+*/
+
+typedef struct CLzma2DecMt CLzma2DecMt;
+typedef CLzma2DecMt * CLzma2DecMtHandle;
+// Z7_DECLARE_HANDLE(CLzma2DecMtHandle)
+
+CLzma2DecMtHandle Lzma2DecMt_Create(ISzAllocPtr alloc, ISzAllocPtr allocMid);
+void Lzma2DecMt_Destroy(CLzma2DecMtHandle p);
+
+SRes Lzma2DecMt_Decode(CLzma2DecMtHandle p,
+    Byte prop,
+    const CLzma2DecMtProps *props,
+    ISeqOutStreamPtr outStream,
+    const UInt64 *outDataSize, // NULL means undefined
+    int finishMode,            // 0 - partial unpacking is allowed, 1 - if lzma2 stream must be finished
+    // Byte *outBuf, size_t *outBufSize,
+    ISeqInStreamPtr inStream,
+    // const Byte *inData, size_t inDataSize,
+    
+    // out variables:
+    UInt64 *inProcessed,
+    int *isMT,  /* out: (*isMT == 0), if single thread decoding was used */
+
+    // UInt64 *outProcessed,
+    ICompressProgressPtr progress);
+
+
+/* ---------- Read from CLzma2DecMtHandle Interface ---------- */
+
+SRes Lzma2DecMt_Init(CLzma2DecMtHandle pp,
+    Byte prop,
+    const CLzma2DecMtProps *props,
+    const UInt64 *outDataSize, int finishMode,
+    ISeqInStreamPtr inStream);
+
+SRes Lzma2DecMt_Read(CLzma2DecMtHandle pp,
+    Byte *data, size_t *outSize,
+    UInt64 *inStreamProcessed);
+
+
+EXTERN_C_END
+
+#endif

C/Lzma2Enc.h

@@ -1,20 +1,22 @@
 /* Lzma2Enc.h -- LZMA2 Encoder
-2009-02-07 : Igor Pavlov : Public domain */
+2023-04-13 : Igor Pavlov : Public domain */
 
-#ifndef __LZMA2_ENC_H
-#define __LZMA2_ENC_H
+#ifndef ZIP7_INC_LZMA2_ENC_H
+#define ZIP7_INC_LZMA2_ENC_H
 
 #include "LzmaEnc.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
+
+#define LZMA2_ENC_PROPS_BLOCK_SIZE_AUTO   0
+#define LZMA2_ENC_PROPS_BLOCK_SIZE_SOLID  ((UInt64)(Int64)-1)
 
 typedef struct
 {
   CLzmaEncProps lzmaProps;
-  size_t blockSize;
-  int numBlockThreads;
+  UInt64 blockSize;
+  int numBlockThreads_Reduced;
+  int numBlockThreads_Max;
   int numTotalThreads;
 } CLzma2EncProps;
 
@@ -24,43 +26,32 @@ void Lzma2EncProps_Normalize(CLzma2EncProps *p);
 /* ---------- CLzmaEnc2Handle Interface ---------- */
 
 /* Lzma2Enc_* functions can return the following exit codes:
-Returns:
+SRes:
   SZ_OK           - OK
   SZ_ERROR_MEM    - Memory allocation error
   SZ_ERROR_PARAM  - Incorrect paramater in props
-  SZ_ERROR_WRITE  - Write callback error
+  SZ_ERROR_WRITE  - ISeqOutStream write callback error
+  SZ_ERROR_OUTPUT_EOF - output buffer overflow - version with (Byte *) output
   SZ_ERROR_PROGRESS - some break from progress callback
-  SZ_ERROR_THREAD - errors in multithreading functions (only for Mt version)
+  SZ_ERROR_THREAD - error in multithreading functions (only for Mt version)
 */
 
-typedef void * CLzma2EncHandle;
+typedef struct CLzma2Enc CLzma2Enc;
+typedef CLzma2Enc * CLzma2EncHandle;
+// Z7_DECLARE_HANDLE(CLzma2EncHandle)
 
-CLzma2EncHandle Lzma2Enc_Create(ISzAlloc *alloc, ISzAlloc *allocBig);
+CLzma2EncHandle Lzma2Enc_Create(ISzAllocPtr alloc, ISzAllocPtr allocBig);
 void Lzma2Enc_Destroy(CLzma2EncHandle p);
 SRes Lzma2Enc_SetProps(CLzma2EncHandle p, const CLzma2EncProps *props);
+void Lzma2Enc_SetDataSize(CLzma2EncHandle p, UInt64 expectedDataSiize);
 Byte Lzma2Enc_WriteProperties(CLzma2EncHandle p);
-SRes Lzma2Enc_Encode(CLzma2EncHandle p,
-    ISeqOutStream *outStream, ISeqInStream *inStream, ICompressProgress *progress);
+SRes Lzma2Enc_Encode2(CLzma2EncHandle p,
+    ISeqOutStreamPtr outStream,
+    Byte *outBuf, size_t *outBufSize,
+    ISeqInStreamPtr inStream,
+    const Byte *inData, size_t inDataSize,
+    ICompressProgressPtr progress);
 
-/* ---------- One Call Interface ---------- */
-
-/* Lzma2Encode
-Return code:
-  SZ_OK               - OK
-  SZ_ERROR_MEM        - Memory allocation error
-  SZ_ERROR_PARAM      - Incorrect paramater
-  SZ_ERROR_OUTPUT_EOF - output buffer overflow
-  SZ_ERROR_THREAD     - errors in multithreading functions (only for Mt version)
-*/
-
-/*
-SRes Lzma2Encode(Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen,
-    const CLzmaEncProps *props, Byte *propsEncoded, int writeEndMark,
-    ICompressProgress *progress, ISzAlloc *alloc, ISzAlloc *allocBig);
-*/
-
-#ifdef __cplusplus
-}
-#endif
+EXTERN_C_END
 
 #endif

C/Lzma86.h

@@ -1,10 +1,10 @@
 /* Lzma86.h -- LZMA + x86 (BCJ) Filter
-2009-08-14 : Igor Pavlov : Public domain */
+2023-03-03 : Igor Pavlov : Public domain */
 
-#ifndef __LZMA86_H
-#define __LZMA86_H
+#ifndef ZIP7_INC_LZMA86_H
+#define ZIP7_INC_LZMA86_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
 EXTERN_C_BEGIN
 

C/Lzma86Dec.c

@@ -1,5 +1,7 @@
 /* Lzma86Dec.c -- LZMA + x86 (BCJ) Filter Decoder
-2009-08-14 : Igor Pavlov : Public domain */
+2023-03-03 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include "Lzma86.h"
 
@@ -7,9 +9,6 @@
 #include "Bra.h"
 #include "LzmaDec.h"
 
-static void *SzAlloc(void *p, size_t size) { p = p; return MyAlloc(size); }
-static void SzFree(void *p, void *address) { p = p; MyFree(address); }
-
 SRes Lzma86_GetUnpackSize(const Byte *src, SizeT srcLen, UInt64 *unpackSize)
 {
   unsigned i;
@@ -23,7 +22,6 @@ SRes Lzma86_GetUnpackSize(const Byte *src, SizeT srcLen, UInt64 *unpackSize)
 
 SRes Lzma86_Decode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen)
 {
-  ISzAlloc g_Alloc = { SzAlloc, SzFree };
   SRes res;
   int useFilter;
   SizeT inSizePure;
@@ -48,9 +46,8 @@ SRes Lzma86_Decode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen)
     return res;
   if (useFilter == 1)
   {
-    UInt32 x86State;
-    x86_Convert_Init(x86State);
-    x86_Convert(dest, *destLen, 0, &x86State, 0);
+    UInt32 x86State = Z7_BRANCH_CONV_ST_X86_STATE_INIT_VAL;
+    z7_BranchConvSt_X86_Dec(dest, *destLen, 0, &x86State);
   }
   return SZ_OK;
 }

C/Lzma86Enc.c

@@ -1,5 +1,7 @@
 /* Lzma86Enc.c -- LZMA + x86 (BCJ) Filter Encoder
-2009-08-14 : Igor Pavlov : Public domain */
+2023-03-03 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
 #include <string.h>
 
@@ -9,18 +11,12 @@
 #include "Bra.h"
 #include "LzmaEnc.h"
 
-#define SZE_OUT_OVERFLOW SZE_DATA_ERROR
-
-static void *SzAlloc(void *p, size_t size) { p = p; return MyAlloc(size); }
-static void SzFree(void *p, void *address) { p = p; MyFree(address); }
-
 int Lzma86_Encode(Byte *dest, size_t *destLen, const Byte *src, size_t srcLen,
     int level, UInt32 dictSize, int filterMode)
 {
-  ISzAlloc g_Alloc = { SzAlloc, SzFree };
   size_t outSize2 = *destLen;
   Byte *filteredStream;
-  Bool useFilter;
+  BoolInt useFilter;
   int mainResult = SZ_ERROR_OUTPUT_EOF;
   CLzmaEncProps props;
   LzmaEncProps_Init(&props);
@@ -50,15 +46,14 @@ int Lzma86_Encode(Byte *dest, size_t *destLen, const Byte *src, size_t srcLen,
       memcpy(filteredStream, src, srcLen);
     }
     {
-      UInt32 x86State;
-      x86_Convert_Init(x86State);
-      x86_Convert(filteredStream, srcLen, 0, &x86State, 1);
+      UInt32 x86State = Z7_BRANCH_CONV_ST_X86_STATE_INIT_VAL;
+      z7_BranchConvSt_X86_Enc(filteredStream, srcLen, 0, &x86State);
     }
   }
 
   {
     size_t minSize = 0;
-    Bool bestIsFiltered = False;
+    BoolInt bestIsFiltered = False;
 
     /* passes for SZ_FILTER_AUTO:
         0 - BCJ + LZMA
@@ -73,7 +68,7 @@ int Lzma86_Encode(Byte *dest, size_t *destLen, const Byte *src, size_t srcLen,
       size_t outSizeProcessed = outSize2 - LZMA86_HEADER_SIZE;
       size_t outPropsSize = 5;
       SRes curRes;
-      Bool curModeIsFiltered = (numPasses > 1 && i == numPasses - 1);
+      BoolInt curModeIsFiltered = (numPasses > 1 && i == numPasses - 1);
       if (curModeIsFiltered && !bestIsFiltered)
         break;
       if (useFilter && i == 0)
@@ -99,7 +94,7 @@ int Lzma86_Encode(Byte *dest, size_t *destLen, const Byte *src, size_t srcLen,
         }
       }
     }
-    dest[0] = (bestIsFiltered ? 1 : 0);
+    dest[0] = (Byte)(bestIsFiltered ? 1 : 0);
     *destLen = LZMA86_HEADER_SIZE + minSize;
   }
   if (useFilter)

C/LzmaDec.h

@@ -1,33 +1,36 @@
 /* LzmaDec.h -- LZMA Decoder
-2009-02-07 : Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
 
-#ifndef __LZMA_DEC_H
-#define __LZMA_DEC_H
+#ifndef ZIP7_INC_LZMA_DEC_H
+#define ZIP7_INC_LZMA_DEC_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
-/* #define _LZMA_PROB32 */
-/* _LZMA_PROB32 can increase the speed on some CPUs,
+/* #define Z7_LZMA_PROB32 */
+/* Z7_LZMA_PROB32 can increase the speed on some CPUs,
    but memory usage for CLzmaDec::probs will be doubled in that case */
 
-#ifdef _LZMA_PROB32
-#define CLzmaProb UInt32
+typedef
+#ifdef Z7_LZMA_PROB32
+  UInt32
 #else
-#define CLzmaProb UInt16
+  UInt16
 #endif
+  CLzmaProb;
 
 
 /* ---------- LZMA Properties ---------- */
 
 #define LZMA_PROPS_SIZE 5
 
-typedef struct _CLzmaProps
+typedef struct
 {
-  unsigned lc, lp, pb;
+  Byte lc;
+  Byte lp;
+  Byte pb;
+  Byte _pad_;
   UInt32 dicSize;
 } CLzmaProps;
 
@@ -49,32 +52,35 @@ SRes LzmaProps_Decode(CLzmaProps *p, const Byte *data, unsigned size);
 
 typedef struct
 {
+  /* Don't change this structure. ASM code can use it. */
   CLzmaProps prop;
   CLzmaProb *probs;
+  CLzmaProb *probs_1664;
   Byte *dic;
-  const Byte *buf;
-  UInt32 range, code;
-  SizeT dicPos;
   SizeT dicBufSize;
+  SizeT dicPos;
+  const Byte *buf;
+  UInt32 range;
+  UInt32 code;
   UInt32 processedPos;
   UInt32 checkDicSize;
-  unsigned state;
   UInt32 reps[4];
-  unsigned remainLen;
-  int needFlush;
-  int needInitState;
+  UInt32 state;
+  UInt32 remainLen;
+
   UInt32 numProbs;
   unsigned tempBufSize;
   Byte tempBuf[LZMA_REQUIRED_INPUT_MAX];
 } CLzmaDec;
 
-#define LzmaDec_Construct(p) { (p)->dic = 0; (p)->probs = 0; }
+#define LzmaDec_CONSTRUCT(p) { (p)->dic = NULL; (p)->probs = NULL; }
+#define LzmaDec_Construct(p) LzmaDec_CONSTRUCT(p)
 
 void LzmaDec_Init(CLzmaDec *p);
 
 /* There are two types of LZMA streams:
-     0) Stream with end mark. That end mark adds about 6 bytes to compressed size.
-     1) Stream without end mark. You must know exact uncompressed size to decompress such stream. */
+     - Stream with end mark. That end mark adds about 6 bytes to compressed size.
+     - Stream without end mark. You must know exact uncompressed size to decompress such stream. */
 
 typedef enum
 {
@@ -131,11 +137,11 @@ LzmaDec_Allocate* can return:
   SZ_ERROR_UNSUPPORTED - Unsupported properties
 */
    
-SRes LzmaDec_AllocateProbs(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAlloc *alloc);
-void LzmaDec_FreeProbs(CLzmaDec *p, ISzAlloc *alloc);
+SRes LzmaDec_AllocateProbs(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAllocPtr alloc);
+void LzmaDec_FreeProbs(CLzmaDec *p, ISzAllocPtr alloc);
 
-SRes LzmaDec_Allocate(CLzmaDec *state, const Byte *prop, unsigned propsSize, ISzAlloc *alloc);
-void LzmaDec_Free(CLzmaDec *state, ISzAlloc *alloc);
+SRes LzmaDec_Allocate(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAllocPtr alloc);
+void LzmaDec_Free(CLzmaDec *p, ISzAllocPtr alloc);
 
 /* ---------- Dictionary Interface ---------- */
 
@@ -144,7 +150,7 @@ void LzmaDec_Free(CLzmaDec *state, ISzAlloc *alloc);
    You must work with CLzmaDec variables directly in this interface.
 
    STEPS:
-     LzmaDec_Constr()
+     LzmaDec_Construct()
      LzmaDec_Allocate()
      for (each new stream)
      {
@@ -176,6 +182,7 @@ Returns:
       LZMA_STATUS_NEEDS_MORE_INPUT
       LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK
   SZ_ERROR_DATA - Data error
+  SZ_ERROR_FAIL - Some unexpected error: internal error of code, memory corruption or hardware failure
 */
 
 SRes LzmaDec_DecodeToDic(CLzmaDec *p, SizeT dicLimit,
@@ -218,14 +225,13 @@ Returns:
   SZ_ERROR_MEM  - Memory allocation error
   SZ_ERROR_UNSUPPORTED - Unsupported properties
   SZ_ERROR_INPUT_EOF - It needs more bytes in input buffer (src).
+  SZ_ERROR_FAIL - Some unexpected error: internal error of code, memory corruption or hardware failure
 */
 
 SRes LzmaDecode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen,
     const Byte *propData, unsigned propSize, ELzmaFinishMode finishMode,
-    ELzmaStatus *status, ISzAlloc *alloc);
+    ELzmaStatus *status, ISzAllocPtr alloc);
 
-#ifdef __cplusplus
-}
-#endif
+EXTERN_C_END
 
 #endif

C/LzmaEnc.h

@@ -1,23 +1,21 @@
 /*  LzmaEnc.h -- LZMA Encoder
-2009-02-07 : Igor Pavlov : Public domain */
+2023-04-13 : Igor Pavlov : Public domain */
 
-#ifndef __LZMA_ENC_H
-#define __LZMA_ENC_H
+#ifndef ZIP7_INC_LZMA_ENC_H
+#define ZIP7_INC_LZMA_ENC_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
 #define LZMA_PROPS_SIZE 5
 
-typedef struct _CLzmaEncProps
+typedef struct
 {
-  int level;       /*  0 <= level <= 9 */
+  int level;       /* 0 <= level <= 9 */
   UInt32 dictSize; /* (1 << 12) <= dictSize <= (1 << 27) for 32-bit version
-                      (1 << 12) <= dictSize <= (1 << 30) for 64-bit version
-                       default = (1 << 24) */
+                      (1 << 12) <= dictSize <= (3 << 29) for 64-bit version
+                      default = (1 << 24) */
   int lc;          /* 0 <= lc <= 8, default = 3 */
   int lp;          /* 0 <= lp <= 4, default = 0 */
   int pb;          /* 0 <= pb <= 4, default = 2 */
@@ -25,9 +23,17 @@ typedef struct _CLzmaEncProps
   int fb;          /* 5 <= fb <= 273, default = 32 */
   int btMode;      /* 0 - hashChain Mode, 1 - binTree mode - normal, default = 1 */
   int numHashBytes; /* 2, 3 or 4, default = 4 */
-  UInt32 mc;        /* 1 <= mc <= (1 << 30), default = 32 */
+  unsigned numHashOutBits;  /* default = ? */
+  UInt32 mc;       /* 1 <= mc <= (1 << 30), default = 32 */
   unsigned writeEndMark;  /* 0 - do not write EOPM, 1 - write EOPM, default = 0 */
   int numThreads;  /* 1 or 2, default = 2 */
+
+  // int _pad;
+
+  UInt64 reduceSize; /* estimated size of data that will be compressed. default = (UInt64)(Int64)-1.
+                        Encoder uses this value to reduce dictionary size */
+
+  UInt64 affinity;
 } CLzmaEncProps;
 
 void LzmaEncProps_Init(CLzmaEncProps *p);
@@ -37,44 +43,41 @@ UInt32 LzmaEncProps_GetDictSize(const CLzmaEncProps *props2);
 
 /* ---------- CLzmaEncHandle Interface ---------- */
 
-/* LzmaEnc_* functions can return the following exit codes:
-Returns:
+/* LzmaEnc* functions can return the following exit codes:
+SRes:
   SZ_OK           - OK
   SZ_ERROR_MEM    - Memory allocation error
   SZ_ERROR_PARAM  - Incorrect paramater in props
-  SZ_ERROR_WRITE  - Write callback error.
+  SZ_ERROR_WRITE  - ISeqOutStream write callback error
+  SZ_ERROR_OUTPUT_EOF - output buffer overflow - version with (Byte *) output
   SZ_ERROR_PROGRESS - some break from progress callback
-  SZ_ERROR_THREAD - errors in multithreading functions (only for Mt version)
+  SZ_ERROR_THREAD - error in multithreading functions (only for Mt version)
 */
 
-typedef void * CLzmaEncHandle;
+typedef struct CLzmaEnc CLzmaEnc;
+typedef CLzmaEnc * CLzmaEncHandle;
+// Z7_DECLARE_HANDLE(CLzmaEncHandle)
+
+CLzmaEncHandle LzmaEnc_Create(ISzAllocPtr alloc);
+void LzmaEnc_Destroy(CLzmaEncHandle p, ISzAllocPtr alloc, ISzAllocPtr allocBig);
 
-CLzmaEncHandle LzmaEnc_Create(ISzAlloc *alloc);
-void LzmaEnc_Destroy(CLzmaEncHandle p, ISzAlloc *alloc, ISzAlloc *allocBig);
 SRes LzmaEnc_SetProps(CLzmaEncHandle p, const CLzmaEncProps *props);
+void LzmaEnc_SetDataSize(CLzmaEncHandle p, UInt64 expectedDataSiize);
 SRes LzmaEnc_WriteProperties(CLzmaEncHandle p, Byte *properties, SizeT *size);
-SRes LzmaEnc_Encode(CLzmaEncHandle p, ISeqOutStream *outStream, ISeqInStream *inStream,
-    ICompressProgress *progress, ISzAlloc *alloc, ISzAlloc *allocBig);
+unsigned LzmaEnc_IsWriteEndMark(CLzmaEncHandle p);
+
+SRes LzmaEnc_Encode(CLzmaEncHandle p, ISeqOutStreamPtr outStream, ISeqInStreamPtr inStream,
+    ICompressProgressPtr progress, ISzAllocPtr alloc, ISzAllocPtr allocBig);
 SRes LzmaEnc_MemEncode(CLzmaEncHandle p, Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen,
-    int writeEndMark, ICompressProgress *progress, ISzAlloc *alloc, ISzAlloc *allocBig);
+    int writeEndMark, ICompressProgressPtr progress, ISzAllocPtr alloc, ISzAllocPtr allocBig);
+
 
 /* ---------- One Call Interface ---------- */
 
-/* LzmaEncode
-Return code:
-  SZ_OK               - OK
-  SZ_ERROR_MEM        - Memory allocation error
-  SZ_ERROR_PARAM      - Incorrect paramater
-  SZ_ERROR_OUTPUT_EOF - output buffer overflow
-  SZ_ERROR_THREAD     - errors in multithreading functions (only for Mt version)
-*/
-
 SRes LzmaEncode(Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen,
     const CLzmaEncProps *props, Byte *propsEncoded, SizeT *propsSize, int writeEndMark,
-    ICompressProgress *progress, ISzAlloc *alloc, ISzAlloc *allocBig);
+    ICompressProgressPtr progress, ISzAllocPtr alloc, ISzAllocPtr allocBig);
 
-#ifdef __cplusplus
-}
-#endif
+EXTERN_C_END
 
 #endif

C/LzmaLib.c

@@ -1,18 +1,14 @@
 /* LzmaLib.c -- LZMA library wrapper
-2008-08-05
-Igor Pavlov
-Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
 
-#include "LzmaEnc.h"
-#include "LzmaDec.h"
 #include "Alloc.h"
+#include "LzmaDec.h"
+#include "LzmaEnc.h"
 #include "LzmaLib.h"
 
-static void *SzAlloc(void *p, size_t size) { p = p; return MyAlloc(size); }
-static void SzFree(void *p, void *address) { p = p; MyFree(address); }
-static ISzAlloc g_Alloc = { SzAlloc, SzFree };
-
-MY_STDAPI LzmaCompress(unsigned char *dest, size_t  *destLen, const unsigned char *src, size_t  srcLen,
+Z7_STDAPI LzmaCompress(unsigned char *dest, size_t *destLen, const unsigned char *src, size_t srcLen,
   unsigned char *outProps, size_t *outPropsSize,
   int level, /* 0 <= level <= 9, default = 5 */
   unsigned dictSize, /* use (1 << N) or (3 << N). 4 KB < dictSize <= 128 MB */
@@ -38,7 +34,7 @@ MY_STDAPI LzmaCompress(unsigned char *dest, size_t  *destLen, const unsigned cha
 }
 
 
-MY_STDAPI LzmaUncompress(unsigned char *dest, size_t  *destLen, const unsigned char *src, size_t  *srcLen,
+Z7_STDAPI LzmaUncompress(unsigned char *dest, size_t *destLen, const unsigned char *src, size_t *srcLen,
   const unsigned char *props, size_t propsSize)
 {
   ELzmaStatus status;

C/LzmaLib.h

@@ -1,16 +1,14 @@
 /* LzmaLib.h -- LZMA library interface
-2009-04-07 : Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
 
-#ifndef __LZMA_LIB_H
-#define __LZMA_LIB_H
+#ifndef ZIP7_INC_LZMA_LIB_H
+#define ZIP7_INC_LZMA_LIB_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
-#define MY_STDAPI int MY_STD_CALL
+#define Z7_STDAPI int Z7_STDCALL
 
 #define LZMA_PROPS_SIZE 5
 
@@ -42,14 +40,16 @@ outPropsSize -
 level - compression level: 0 <= level <= 9;
 
   level dictSize algo  fb
-    0:    16 KB   0    32
-    1:    64 KB   0    32
-    2:   256 KB   0    32
-    3:     1 MB   0    32
-    4:     4 MB   0    32
+    0:    64 KB   0    32
+    1:   256 KB   0    32
+    2:     1 MB   0    32
+    3:     4 MB   0    32
+    4:    16 MB   0    32
     5:    16 MB   1    32
     6:    32 MB   1    32
-    7+:   64 MB   1    64
+    7:    32 MB   1    64
+    8:    64 MB   1    64
+    9:    64 MB   1    64
  
   The default value for "level" is 5.
 
@@ -85,6 +85,11 @@ fb - Word size (the number of fast bytes).
 numThreads - The number of thereads. 1 or 2. The default value is 2.
      Fast mode (algo = 0) can use only 1 thread.
 
+In:
+  dest     - output data buffer
+  destLen  - output data buffer size
+  src      - input data
+  srcLen   - input data size
 Out:
   destLen  - processed output size
 Returns:
@@ -95,7 +100,7 @@ Returns:
   SZ_ERROR_THREAD     - errors in multithreading functions (only for Mt version)
 */
 
-MY_STDAPI LzmaCompress(unsigned char *dest, size_t *destLen, const unsigned char *src, size_t srcLen,
+Z7_STDAPI LzmaCompress(unsigned char *dest, size_t *destLen, const unsigned char *src, size_t srcLen,
   unsigned char *outProps, size_t *outPropsSize, /* *outPropsSize must be = 5 */
   int level,      /* 0 <= level <= 9, default = 5 */
   unsigned dictSize,  /* default = (1 << 24) */
@@ -110,8 +115,8 @@ MY_STDAPI LzmaCompress(unsigned char *dest, size_t *destLen, const unsigned char
 LzmaUncompress
 --------------
 In:
-  dest     - output data
-  destLen  - output data size
+  dest     - output data buffer
+  destLen  - output data buffer size
   src      - input data
   srcLen   - input data size
 Out:
@@ -125,11 +130,9 @@ Returns:
   SZ_ERROR_INPUT_EOF   - it needs more bytes in input buffer (src)
 */
 
-MY_STDAPI LzmaUncompress(unsigned char *dest, size_t *destLen, const unsigned char *src, SizeT *srcLen,
+Z7_STDAPI LzmaUncompress(unsigned char *dest, size_t *destLen, const unsigned char *src, SizeT *srcLen,
   const unsigned char *props, size_t propsSize);
 
-#ifdef __cplusplus
-}
-#endif
+EXTERN_C_END
 
 #endif

C/MtCoder.c

@@ -1,327 +1,571 @@
 /* MtCoder.c -- Multi-thread Coder
-2009-03-26 : Igor Pavlov : Public domain */
+2023-04-13 : Igor Pavlov : Public domain */
 
-#include <stdio.h>
+#include "Precomp.h"
 
 #include "MtCoder.h"
 
-void LoopThread_Construct(CLoopThread *p)
+#ifndef Z7_ST
+
+static SRes MtProgressThunk_Progress(ICompressProgressPtr pp, UInt64 inSize, UInt64 outSize)
 {
-  Thread_Construct(&p->thread);
-  Event_Construct(&p->startEvent);
-  Event_Construct(&p->finishedEvent);
+  Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(CMtProgressThunk)
+  UInt64 inSize2 = 0;
+  UInt64 outSize2 = 0;
+  if (inSize != (UInt64)(Int64)-1)
+  {
+    inSize2 = inSize - p->inSize;
+    p->inSize = inSize;
+  }
+  if (outSize != (UInt64)(Int64)-1)
+  {
+    outSize2 = outSize - p->outSize;
+    p->outSize = outSize;
+  }
+  return MtProgress_ProgressAdd(p->mtProgress, inSize2, outSize2);
 }
 
-void LoopThread_Close(CLoopThread *p)
+
+void MtProgressThunk_CreateVTable(CMtProgressThunk *p)
 {
-  Thread_Close(&p->thread);
-  Event_Close(&p->startEvent);
-  Event_Close(&p->finishedEvent);
+  p->vt.Progress = MtProgressThunk_Progress;
 }
 
-static THREAD_FUNC_RET_TYPE THREAD_FUNC_CALL_TYPE LoopThreadFunc(void *pp)
+
+
+#define RINOK_THREAD(x) { if ((x) != 0) return SZ_ERROR_THREAD; }
+
+
+static THREAD_FUNC_DECL ThreadFunc(void *pp);
+
+
+static SRes MtCoderThread_CreateAndStart(CMtCoderThread *t)
 {
-  CLoopThread *p = (CLoopThread *)pp;
+  WRes wres = AutoResetEvent_OptCreate_And_Reset(&t->startEvent);
+  if (wres == 0)
+  {
+    t->stop = False;
+    if (!Thread_WasCreated(&t->thread))
+      wres = Thread_Create(&t->thread, ThreadFunc, t);
+    if (wres == 0)
+      wres = Event_Set(&t->startEvent);
+  }
+  if (wres == 0)
+    return SZ_OK;
+  return MY_SRes_HRESULT_FROM_WRes(wres);
+}
+
+
+static void MtCoderThread_Destruct(CMtCoderThread *t)
+{
+  if (Thread_WasCreated(&t->thread))
+  {
+    t->stop = 1;
+    Event_Set(&t->startEvent);
+    Thread_Wait_Close(&t->thread);
+  }
+
+  Event_Close(&t->startEvent);
+
+  if (t->inBuf)
+  {
+    ISzAlloc_Free(t->mtCoder->allocBig, t->inBuf);
+    t->inBuf = NULL;
+  }
+}
+
+
+
+
+/*
+  ThreadFunc2() returns:
+  SZ_OK           - in all normal cases (even for stream error or memory allocation error)
+  SZ_ERROR_THREAD - in case of failure in system synch function
+*/
+
+static SRes ThreadFunc2(CMtCoderThread *t)
+{
+  CMtCoder *mtc = t->mtCoder;
+
   for (;;)
   {
-    if (Event_Wait(&p->startEvent) != 0)
-      return SZ_ERROR_THREAD;
-    if (p->stop)
-      return 0;
-    p->res = p->func(p->param);
-    if (Event_Set(&p->finishedEvent) != 0)
-      return SZ_ERROR_THREAD;
-  }
-}
+    unsigned bi;
+    SRes res;
+    SRes res2;
+    BoolInt finished;
+    unsigned bufIndex;
+    size_t size;
+    const Byte *inData;
+    UInt64 readProcessed = 0;
+    
+    RINOK_THREAD(Event_Wait(&mtc->readEvent))
 
-WRes LoopThread_Create(CLoopThread *p)
-{
-  p->stop = 0;
-  RINOK(AutoResetEvent_CreateNotSignaled(&p->startEvent));
-  RINOK(AutoResetEvent_CreateNotSignaled(&p->finishedEvent));
-  return Thread_Create(&p->thread, LoopThreadFunc, p);
-}
+    /* after Event_Wait(&mtc->readEvent) we must call Event_Set(&mtc->readEvent) in any case to unlock another threads */
 
-WRes LoopThread_StopAndWait(CLoopThread *p)
-{
-  p->stop = 1;
-  if (Event_Set(&p->startEvent) != 0)
-    return SZ_ERROR_THREAD;
-  return Thread_Wait(&p->thread);
-}
-
-WRes LoopThread_StartSubThread(CLoopThread *p) { return Event_Set(&p->startEvent); }
-WRes LoopThread_WaitSubThread(CLoopThread *p) { return Event_Wait(&p->finishedEvent); }
-
-static SRes Progress(ICompressProgress *p, UInt64 inSize, UInt64 outSize)
-{
-  return (p && p->Progress(p, inSize, outSize) != SZ_OK) ? SZ_ERROR_PROGRESS : SZ_OK;
-}
-
-static void MtProgress_Init(CMtProgress *p, ICompressProgress *progress)
-{
-  unsigned i;
-  for (i = 0; i < NUM_MT_CODER_THREADS_MAX; i++)
-    p->inSizes[i] = p->outSizes[i] = 0;
-  p->totalInSize = p->totalOutSize = 0;
-  p->progress = progress;
-  p->res = SZ_OK;
-}
-
-static void MtProgress_Reinit(CMtProgress *p, unsigned index)
-{
-  p->inSizes[index] = 0;
-  p->outSizes[index] = 0;
-}
-
-#define UPDATE_PROGRESS(size, prev, total) \
-  if (size != (UInt64)(Int64)-1) { total += size - prev; prev = size; }
-
-SRes MtProgress_Set(CMtProgress *p, unsigned index, UInt64 inSize, UInt64 outSize)
-{
-  SRes res;
-  CriticalSection_Enter(&p->cs);
-  UPDATE_PROGRESS(inSize, p->inSizes[index], p->totalInSize)
-  UPDATE_PROGRESS(outSize, p->outSizes[index], p->totalOutSize)
-  if (p->res == SZ_OK)
-    p->res = Progress(p->progress, p->totalInSize, p->totalOutSize);
-  res = p->res;
-  CriticalSection_Leave(&p->cs);
-  return res;
-}
-
-static void MtProgress_SetError(CMtProgress *p, SRes res)
-{
-  CriticalSection_Enter(&p->cs);
-  if (p->res == SZ_OK)
-    p->res = res;
-  CriticalSection_Leave(&p->cs);
-}
-
-static void MtCoder_SetError(CMtCoder* p, SRes res)
-{
-  CriticalSection_Enter(&p->cs);
-  if (p->res == SZ_OK)
-    p->res = res;
-  CriticalSection_Leave(&p->cs);
-}
-
-/* ---------- MtThread ---------- */
-
-void CMtThread_Construct(CMtThread *p, CMtCoder *mtCoder)
-{
-  p->mtCoder = mtCoder;
-  p->outBuf = 0;
-  p->inBuf = 0;
-  Event_Construct(&p->canRead);
-  Event_Construct(&p->canWrite);
-  LoopThread_Construct(&p->thread);
-}
-
-#define RINOK_THREAD(x) { if((x) != 0) return SZ_ERROR_THREAD; }
-
-static void CMtThread_CloseEvents(CMtThread *p)
-{
-  Event_Close(&p->canRead);
-  Event_Close(&p->canWrite);
-}
-
-static void CMtThread_Destruct(CMtThread *p)
-{
-  CMtThread_CloseEvents(p);
-
-  if (Thread_WasCreated(&p->thread.thread))
-  {
-    LoopThread_StopAndWait(&p->thread);
-    LoopThread_Close(&p->thread);
-  }
-
-  if (p->mtCoder->alloc)
-    IAlloc_Free(p->mtCoder->alloc, p->outBuf);
-  p->outBuf = 0;
-
-  if (p->mtCoder->alloc)
-    IAlloc_Free(p->mtCoder->alloc, p->inBuf);
-  p->inBuf = 0;
-}
-
-#define MY_BUF_ALLOC(buf, size, newSize) \
-  if (buf == 0 || size != newSize) \
-  { IAlloc_Free(p->mtCoder->alloc, buf); \
-    size = newSize; buf = IAlloc_Alloc(p->mtCoder->alloc, size); \
-    if (buf == 0) return SZ_ERROR_MEM; }
-
-static SRes CMtThread_Prepare(CMtThread *p)
-{
-  MY_BUF_ALLOC(p->inBuf, p->inBufSize, p->mtCoder->blockSize)
-  MY_BUF_ALLOC(p->outBuf, p->outBufSize, p->mtCoder->destBlockSize)
-
-  p->stopReading = False;
-  p->stopWriting = False;
-  RINOK_THREAD(AutoResetEvent_CreateNotSignaled(&p->canRead));
-  RINOK_THREAD(AutoResetEvent_CreateNotSignaled(&p->canWrite));
-
-  return SZ_OK;
-}
-
-static SRes FullRead(ISeqInStream *stream, Byte *data, size_t *processedSize)
-{
-  size_t size = *processedSize;
-  *processedSize = 0;
-  while (size != 0)
-  {
-    size_t curSize = size;
-    SRes res = stream->Read(stream, data, &curSize);
-    *processedSize += curSize;
-    data += curSize;
-    size -= curSize;
-    RINOK(res);
-    if (curSize == 0)
-      return SZ_OK;
-  }
-  return SZ_OK;
-}
-
-#define GET_NEXT_THREAD(p) &p->mtCoder->threads[p->index == p->mtCoder->numThreads  - 1 ? 0 : p->index + 1]
-
-static SRes MtThread_Process(CMtThread *p, Bool *stop)
-{
-  CMtThread *next;
-  *stop = True;
-  if (Event_Wait(&p->canRead) != 0)
-    return SZ_ERROR_THREAD;
-  
-  next = GET_NEXT_THREAD(p);
-  
-  if (p->stopReading)
-  {
-    next->stopReading = True;
-    return Event_Set(&next->canRead) == 0 ? SZ_OK : SZ_ERROR_THREAD;
-  }
-
-  {
-    size_t size = p->mtCoder->blockSize;
-    size_t destSize = p->outBufSize;
-
-    RINOK(FullRead(p->mtCoder->inStream, p->inBuf, &size));
-    next->stopReading = *stop = (size != p->mtCoder->blockSize);
-    if (Event_Set(&next->canRead) != 0)
-      return SZ_ERROR_THREAD;
-
-    RINOK(p->mtCoder->mtCallback->Code(p->mtCoder->mtCallback, p->index,
-        p->outBuf, &destSize, p->inBuf, size, *stop));
-
-    MtProgress_Reinit(&p->mtCoder->mtProgress, p->index);
-
-    if (Event_Wait(&p->canWrite) != 0)
-      return SZ_ERROR_THREAD;
-    if (p->stopWriting)
-      return SZ_ERROR_FAIL;
-    if (p->mtCoder->outStream->Write(p->mtCoder->outStream, p->outBuf, destSize) != destSize)
-      return SZ_ERROR_WRITE;
-    return Event_Set(&next->canWrite) == 0 ? SZ_OK : SZ_ERROR_THREAD;
-  }
-}
-
-static THREAD_FUNC_RET_TYPE THREAD_FUNC_CALL_TYPE ThreadFunc(void *pp)
-{
-  CMtThread *p = (CMtThread *)pp;
-  for (;;)
-  {
-    Bool stop;
-    CMtThread *next = GET_NEXT_THREAD(p);
-    SRes res = MtThread_Process(p, &stop);
-    if (res != SZ_OK)
+    if (mtc->stopReading)
     {
-      MtCoder_SetError(p->mtCoder, res);
-      MtProgress_SetError(&p->mtCoder->mtProgress, res);
-      next->stopReading = True;
-      next->stopWriting = True;
-      Event_Set(&next->canRead);
-      Event_Set(&next->canWrite);
-      return res;
+      return Event_Set(&mtc->readEvent) == 0 ? SZ_OK : SZ_ERROR_THREAD;
     }
-    if (stop)
+
+    res = MtProgress_GetError(&mtc->mtProgress);
+    
+    size = 0;
+    inData = NULL;
+    finished = True;
+
+    if (res == SZ_OK)
+    {
+      size = mtc->blockSize;
+      if (mtc->inStream)
+      {
+        if (!t->inBuf)
+        {
+          t->inBuf = (Byte *)ISzAlloc_Alloc(mtc->allocBig, mtc->blockSize);
+          if (!t->inBuf)
+            res = SZ_ERROR_MEM;
+        }
+        if (res == SZ_OK)
+        {
+          res = SeqInStream_ReadMax(mtc->inStream, t->inBuf, &size);
+          readProcessed = mtc->readProcessed + size;
+          mtc->readProcessed = readProcessed;
+        }
+        if (res != SZ_OK)
+        {
+          mtc->readRes = res;
+          /* after reading error - we can stop encoding of previous blocks */
+          MtProgress_SetError(&mtc->mtProgress, res);
+        }
+        else
+          finished = (size != mtc->blockSize);
+      }
+      else
+      {
+        size_t rem;
+        readProcessed = mtc->readProcessed;
+        rem = mtc->inDataSize - (size_t)readProcessed;
+        if (size > rem)
+          size = rem;
+        inData = mtc->inData + (size_t)readProcessed;
+        readProcessed += size;
+        mtc->readProcessed = readProcessed;
+        finished = (mtc->inDataSize == (size_t)readProcessed);
+      }
+    }
+
+    /* we must get some block from blocksSemaphore before Event_Set(&mtc->readEvent) */
+
+    res2 = SZ_OK;
+
+    if (Semaphore_Wait(&mtc->blocksSemaphore) != 0)
+    {
+      res2 = SZ_ERROR_THREAD;
+      if (res == SZ_OK)
+      {
+        res = res2;
+        // MtProgress_SetError(&mtc->mtProgress, res);
+      }
+    }
+
+    bi = mtc->blockIndex;
+
+    if (++mtc->blockIndex >= mtc->numBlocksMax)
+      mtc->blockIndex = 0;
+
+    bufIndex = (unsigned)(int)-1;
+
+    if (res == SZ_OK)
+      res = MtProgress_GetError(&mtc->mtProgress);
+
+    if (res != SZ_OK)
+      finished = True;
+
+    if (!finished)
+    {
+      if (mtc->numStartedThreads < mtc->numStartedThreadsLimit
+          && mtc->expectedDataSize != readProcessed)
+      {
+        res = MtCoderThread_CreateAndStart(&mtc->threads[mtc->numStartedThreads]);
+        if (res == SZ_OK)
+          mtc->numStartedThreads++;
+        else
+        {
+          MtProgress_SetError(&mtc->mtProgress, res);
+          finished = True;
+        }
+      }
+    }
+
+    if (finished)
+      mtc->stopReading = True;
+
+    RINOK_THREAD(Event_Set(&mtc->readEvent))
+
+    if (res2 != SZ_OK)
+      return res2;
+
+    if (res == SZ_OK)
+    {
+      CriticalSection_Enter(&mtc->cs);
+      bufIndex = mtc->freeBlockHead;
+      mtc->freeBlockHead = mtc->freeBlockList[bufIndex];
+      CriticalSection_Leave(&mtc->cs);
+      
+      res = mtc->mtCallback->Code(mtc->mtCallbackObject, t->index, bufIndex,
+          mtc->inStream ? t->inBuf : inData, size, finished);
+      
+      // MtProgress_Reinit(&mtc->mtProgress, t->index);
+
+      if (res != SZ_OK)
+        MtProgress_SetError(&mtc->mtProgress, res);
+    }
+
+    {
+      CMtCoderBlock *block = &mtc->blocks[bi];
+      block->res = res;
+      block->bufIndex = bufIndex;
+      block->finished = finished;
+    }
+    
+    #ifdef MTCODER_USE_WRITE_THREAD
+      RINOK_THREAD(Event_Set(&mtc->writeEvents[bi]))
+    #else
+    {
+      unsigned wi;
+      {
+        CriticalSection_Enter(&mtc->cs);
+        wi = mtc->writeIndex;
+        if (wi == bi)
+          mtc->writeIndex = (unsigned)(int)-1;
+        else
+          mtc->ReadyBlocks[bi] = True;
+        CriticalSection_Leave(&mtc->cs);
+      }
+
+      if (wi != bi)
+      {
+        if (res != SZ_OK || finished)
+          return 0;
+        continue;
+      }
+
+      if (mtc->writeRes != SZ_OK)
+        res = mtc->writeRes;
+
+      for (;;)
+      {
+        if (res == SZ_OK && bufIndex != (unsigned)(int)-1)
+        {
+          res = mtc->mtCallback->Write(mtc->mtCallbackObject, bufIndex);
+          if (res != SZ_OK)
+          {
+            mtc->writeRes = res;
+            MtProgress_SetError(&mtc->mtProgress, res);
+          }
+        }
+
+        if (++wi >= mtc->numBlocksMax)
+          wi = 0;
+        {
+          BoolInt isReady;
+
+          CriticalSection_Enter(&mtc->cs);
+          
+          if (bufIndex != (unsigned)(int)-1)
+          {
+            mtc->freeBlockList[bufIndex] = mtc->freeBlockHead;
+            mtc->freeBlockHead = bufIndex;
+          }
+          
+          isReady = mtc->ReadyBlocks[wi];
+          
+          if (isReady)
+            mtc->ReadyBlocks[wi] = False;
+          else
+            mtc->writeIndex = wi;
+          
+          CriticalSection_Leave(&mtc->cs);
+
+          RINOK_THREAD(Semaphore_Release1(&mtc->blocksSemaphore))
+
+          if (!isReady)
+            break;
+        }
+
+        {
+          CMtCoderBlock *block = &mtc->blocks[wi];
+          if (res == SZ_OK && block->res != SZ_OK)
+            res = block->res;
+          bufIndex = block->bufIndex;
+          finished = block->finished;
+        }
+      }
+    }
+    #endif
+      
+    if (finished || res != SZ_OK)
       return 0;
   }
 }
 
-void MtCoder_Construct(CMtCoder* p)
+
+static THREAD_FUNC_DECL ThreadFunc(void *pp)
+{
+  CMtCoderThread *t = (CMtCoderThread *)pp;
+  for (;;)
+  {
+    if (Event_Wait(&t->startEvent) != 0)
+      return (THREAD_FUNC_RET_TYPE)SZ_ERROR_THREAD;
+    if (t->stop)
+      return 0;
+    {
+      SRes res = ThreadFunc2(t);
+      CMtCoder *mtc = t->mtCoder;
+      if (res != SZ_OK)
+      {
+        MtProgress_SetError(&mtc->mtProgress, res);
+      }
+      
+      #ifndef MTCODER_USE_WRITE_THREAD
+      {
+        unsigned numFinished = (unsigned)InterlockedIncrement(&mtc->numFinishedThreads);
+        if (numFinished == mtc->numStartedThreads)
+          if (Event_Set(&mtc->finishedEvent) != 0)
+            return (THREAD_FUNC_RET_TYPE)SZ_ERROR_THREAD;
+      }
+      #endif
+    }
+  }
+}
+
+
+
+void MtCoder_Construct(CMtCoder *p)
 {
   unsigned i;
-  p->alloc = 0;
-  for (i = 0; i < NUM_MT_CODER_THREADS_MAX; i++)
+  
+  p->blockSize = 0;
+  p->numThreadsMax = 0;
+  p->expectedDataSize = (UInt64)(Int64)-1;
+
+  p->inStream = NULL;
+  p->inData = NULL;
+  p->inDataSize = 0;
+
+  p->progress = NULL;
+  p->allocBig = NULL;
+
+  p->mtCallback = NULL;
+  p->mtCallbackObject = NULL;
+
+  p->allocatedBufsSize = 0;
+
+  Event_Construct(&p->readEvent);
+  Semaphore_Construct(&p->blocksSemaphore);
+
+  for (i = 0; i < MTCODER_THREADS_MAX; i++)
   {
-    CMtThread *t = &p->threads[i];
+    CMtCoderThread *t = &p->threads[i];
+    t->mtCoder = p;
     t->index = i;
-    CMtThread_Construct(t, p);
+    t->inBuf = NULL;
+    t->stop = False;
+    Event_Construct(&t->startEvent);
+    Thread_CONSTRUCT(&t->thread)
   }
+
+  #ifdef MTCODER_USE_WRITE_THREAD
+    for (i = 0; i < MTCODER_BLOCKS_MAX; i++)
+      Event_Construct(&p->writeEvents[i]);
+  #else
+    Event_Construct(&p->finishedEvent);
+  #endif
+
   CriticalSection_Init(&p->cs);
   CriticalSection_Init(&p->mtProgress.cs);
 }
 
-void MtCoder_Destruct(CMtCoder* p)
+
+
+
+static void MtCoder_Free(CMtCoder *p)
 {
   unsigned i;
-  for (i = 0; i < NUM_MT_CODER_THREADS_MAX; i++)
-    CMtThread_Destruct(&p->threads[i]);
+
+  /*
+  p->stopReading = True;
+  if (Event_IsCreated(&p->readEvent))
+    Event_Set(&p->readEvent);
+  */
+
+  for (i = 0; i < MTCODER_THREADS_MAX; i++)
+    MtCoderThread_Destruct(&p->threads[i]);
+
+  Event_Close(&p->readEvent);
+  Semaphore_Close(&p->blocksSemaphore);
+
+  #ifdef MTCODER_USE_WRITE_THREAD
+    for (i = 0; i < MTCODER_BLOCKS_MAX; i++)
+      Event_Close(&p->writeEvents[i]);
+  #else
+    Event_Close(&p->finishedEvent);
+  #endif
+}
+
+
+void MtCoder_Destruct(CMtCoder *p)
+{
+  MtCoder_Free(p);
+
   CriticalSection_Delete(&p->cs);
   CriticalSection_Delete(&p->mtProgress.cs);
 }
 
+
 SRes MtCoder_Code(CMtCoder *p)
 {
-  unsigned i, numThreads = p->numThreads;
+  unsigned numThreads = p->numThreadsMax;
+  unsigned numBlocksMax;
+  unsigned i;
   SRes res = SZ_OK;
-  p->res = SZ_OK;
+
+  if (numThreads > MTCODER_THREADS_MAX)
+    numThreads = MTCODER_THREADS_MAX;
+  numBlocksMax = MTCODER_GET_NUM_BLOCKS_FROM_THREADS(numThreads);
+  
+  if (p->blockSize < ((UInt32)1 << 26)) numBlocksMax++;
+  if (p->blockSize < ((UInt32)1 << 24)) numBlocksMax++;
+  if (p->blockSize < ((UInt32)1 << 22)) numBlocksMax++;
+
+  if (numBlocksMax > MTCODER_BLOCKS_MAX)
+    numBlocksMax = MTCODER_BLOCKS_MAX;
+
+  if (p->blockSize != p->allocatedBufsSize)
+  {
+    for (i = 0; i < MTCODER_THREADS_MAX; i++)
+    {
+      CMtCoderThread *t = &p->threads[i];
+      if (t->inBuf)
+      {
+        ISzAlloc_Free(p->allocBig, t->inBuf);
+        t->inBuf = NULL;
+      }
+    }
+    p->allocatedBufsSize = p->blockSize;
+  }
+
+  p->readRes = SZ_OK;
 
   MtProgress_Init(&p->mtProgress, p->progress);
 
-  for (i = 0; i < numThreads; i++)
+  #ifdef MTCODER_USE_WRITE_THREAD
+    for (i = 0; i < numBlocksMax; i++)
+    {
+      RINOK_THREAD(AutoResetEvent_OptCreate_And_Reset(&p->writeEvents[i]))
+    }
+  #else
+    RINOK_THREAD(AutoResetEvent_OptCreate_And_Reset(&p->finishedEvent))
+  #endif
+
   {
-    RINOK(CMtThread_Prepare(&p->threads[i]));
+    RINOK_THREAD(AutoResetEvent_OptCreate_And_Reset(&p->readEvent))
+    RINOK_THREAD(Semaphore_OptCreateInit(&p->blocksSemaphore, numBlocksMax, numBlocksMax))
   }
 
-  for (i = 0; i < numThreads; i++)
+  for (i = 0; i < MTCODER_BLOCKS_MAX - 1; i++)
+    p->freeBlockList[i] = i + 1;
+  p->freeBlockList[MTCODER_BLOCKS_MAX - 1] = (unsigned)(int)-1;
+  p->freeBlockHead = 0;
+
+  p->readProcessed = 0;
+  p->blockIndex = 0;
+  p->numBlocksMax = numBlocksMax;
+  p->stopReading = False;
+
+  #ifndef MTCODER_USE_WRITE_THREAD
+    p->writeIndex = 0;
+    p->writeRes = SZ_OK;
+    for (i = 0; i < MTCODER_BLOCKS_MAX; i++)
+      p->ReadyBlocks[i] = False;
+    p->numFinishedThreads = 0;
+  #endif
+
+  p->numStartedThreadsLimit = numThreads;
+  p->numStartedThreads = 0;
+
+  // for (i = 0; i < numThreads; i++)
   {
-    CMtThread *t = &p->threads[i];
-    CLoopThread *lt = &t->thread;
+    CMtCoderThread *nextThread = &p->threads[p->numStartedThreads++];
+    RINOK(MtCoderThread_CreateAndStart(nextThread))
+  }
 
-    if (!Thread_WasCreated(&lt->thread))
+  RINOK_THREAD(Event_Set(&p->readEvent))
+
+  #ifdef MTCODER_USE_WRITE_THREAD
+  {
+    unsigned bi = 0;
+
+    for (;; bi++)
     {
-      lt->func = ThreadFunc;
-      lt->param = t;
+      if (bi >= numBlocksMax)
+        bi = 0;
+
+      RINOK_THREAD(Event_Wait(&p->writeEvents[bi]))
 
-      if (LoopThread_Create(lt) != SZ_OK)
       {
-        res = SZ_ERROR_THREAD;
-        break;
+        const CMtCoderBlock *block = &p->blocks[bi];
+        unsigned bufIndex = block->bufIndex;
+        BoolInt finished = block->finished;
+        if (res == SZ_OK && block->res != SZ_OK)
+          res = block->res;
+
+        if (bufIndex != (unsigned)(int)-1)
+        {
+          if (res == SZ_OK)
+          {
+            res = p->mtCallback->Write(p->mtCallbackObject, bufIndex);
+            if (res != SZ_OK)
+              MtProgress_SetError(&p->mtProgress, res);
+          }
+          
+          CriticalSection_Enter(&p->cs);
+          {
+            p->freeBlockList[bufIndex] = p->freeBlockHead;
+            p->freeBlockHead = bufIndex;
+          }
+          CriticalSection_Leave(&p->cs);
+        }
+        
+        RINOK_THREAD(Semaphore_Release1(&p->blocksSemaphore))
+
+        if (finished)
+          break;
       }
     }
   }
+  #else
+  {
+    WRes wres = Event_Wait(&p->finishedEvent);
+    res = MY_SRes_HRESULT_FROM_WRes(wres);
+  }
+  #endif
 
   if (res == SZ_OK)
-  {
-    unsigned j;
-    for (i = 0; i < numThreads; i++)
-    {
-      CMtThread *t = &p->threads[i];
-      if (LoopThread_StartSubThread(&t->thread) != SZ_OK || i == 10)
-      {
-        res = SZ_ERROR_THREAD;
-        p->threads[0].stopReading = True;
-        break;
-      }
-    }
+    res = p->readRes;
 
-    Event_Set(&p->threads[0].canWrite);
-    Event_Set(&p->threads[0].canRead);
+  if (res == SZ_OK)
+    res = p->mtProgress.res;
 
-    for (j = 0; j < i; j++)
-      LoopThread_WaitSubThread(&p->threads[j].thread);
-  }
+  #ifndef MTCODER_USE_WRITE_THREAD
+    if (res == SZ_OK)
+      res = p->writeRes;
+  #endif
 
-  for (i = 0; i < numThreads; i++)
-    CMtThread_CloseEvents(&p->threads[i]);
-  return (res == SZ_OK) ? p->res : res;
+  if (res != SZ_OK)
+    MtCoder_Free(p);
+  return res;
 }
+
+#endif
+
+#undef RINOK_THREAD

C/MtCoder.h

@@ -1,98 +1,141 @@
 /* MtCoder.h -- Multi-thread Coder
-2009-11-19 : Igor Pavlov : Public domain */
+2023-04-13 : Igor Pavlov : Public domain */
 
-#ifndef __MT_CODER_H
-#define __MT_CODER_H
+#ifndef ZIP7_INC_MT_CODER_H
+#define ZIP7_INC_MT_CODER_H
 
-#include "Threads.h"
+#include "MtDec.h"
 
 EXTERN_C_BEGIN
 
-typedef struct
-{
-  CThread thread;
-  CAutoResetEvent startEvent;
-  CAutoResetEvent finishedEvent;
-  int stop;
-  
-  THREAD_FUNC_TYPE func;
-  LPVOID param;
-  THREAD_FUNC_RET_TYPE res;
-} CLoopThread;
+/*
+  if (    defined MTCODER_USE_WRITE_THREAD) : main thread writes all data blocks to output stream
+  if (not defined MTCODER_USE_WRITE_THREAD) : any coder thread can write data blocks to output stream
+*/
+/* #define MTCODER_USE_WRITE_THREAD */
 
-void LoopThread_Construct(CLoopThread *p);
-void LoopThread_Close(CLoopThread *p);
-WRes LoopThread_Create(CLoopThread *p);
-WRes LoopThread_StopAndWait(CLoopThread *p);
-WRes LoopThread_StartSubThread(CLoopThread *p);
-WRes LoopThread_WaitSubThread(CLoopThread *p);
-
-#ifndef _7ZIP_ST
-#define NUM_MT_CODER_THREADS_MAX 32
+#ifndef Z7_ST
+  #define MTCODER_GET_NUM_BLOCKS_FROM_THREADS(numThreads) ((numThreads) + (numThreads) / 8 + 1)
+  #define MTCODER_THREADS_MAX 64
+  #define MTCODER_BLOCKS_MAX (MTCODER_GET_NUM_BLOCKS_FROM_THREADS(MTCODER_THREADS_MAX) + 3)
 #else
-#define NUM_MT_CODER_THREADS_MAX 1
+  #define MTCODER_THREADS_MAX 1
+  #define MTCODER_BLOCKS_MAX 1
 #endif
 
-typedef struct
-{
-  UInt64 totalInSize;
-  UInt64 totalOutSize;
-  ICompressProgress *progress;
-  SRes res;
-  CCriticalSection cs;
-  UInt64 inSizes[NUM_MT_CODER_THREADS_MAX];
-  UInt64 outSizes[NUM_MT_CODER_THREADS_MAX];
-} CMtProgress;
 
-SRes MtProgress_Set(CMtProgress *p, unsigned index, UInt64 inSize, UInt64 outSize);
+#ifndef Z7_ST
 
-struct _CMtCoder;
 
 typedef struct
 {
-  struct _CMtCoder *mtCoder;
-  Byte *outBuf;
-  size_t outBufSize;
-  Byte *inBuf;
-  size_t inBufSize;
+  ICompressProgress vt;
+  CMtProgress *mtProgress;
+  UInt64 inSize;
+  UInt64 outSize;
+} CMtProgressThunk;
+
+void MtProgressThunk_CreateVTable(CMtProgressThunk *p);
+    
+#define MtProgressThunk_INIT(p) { (p)->inSize = 0; (p)->outSize = 0; }
+
+
+struct CMtCoder_;
+
+
+typedef struct
+{
+  struct CMtCoder_ *mtCoder;
   unsigned index;
-  CLoopThread thread;
+  int stop;
+  Byte *inBuf;
+
+  CAutoResetEvent startEvent;
+  CThread thread;
+} CMtCoderThread;
 
-  Bool stopReading;
-  Bool stopWriting;
-  CAutoResetEvent canRead;
-  CAutoResetEvent canWrite;
-} CMtThread;
 
 typedef struct
 {
-  SRes (*Code)(void *p, unsigned index, Byte *dest, size_t *destSize,
+  SRes (*Code)(void *p, unsigned coderIndex, unsigned outBufIndex,
       const Byte *src, size_t srcSize, int finished);
-} IMtCoderCallback;
+  SRes (*Write)(void *p, unsigned outBufIndex);
+} IMtCoderCallback2;
 
-typedef struct _CMtCoder
+
+typedef struct
 {
-  size_t blockSize;
-  size_t destBlockSize;
-  unsigned numThreads;
-  
-  ISeqInStream *inStream;
-  ISeqOutStream *outStream;
-  ICompressProgress *progress;
-  ISzAlloc *alloc;
-
-  IMtCoderCallback *mtCallback;
-  CCriticalSection cs;
   SRes res;
+  unsigned bufIndex;
+  BoolInt finished;
+} CMtCoderBlock;
+
+
+typedef struct CMtCoder_
+{
+  /* input variables */
+  
+  size_t blockSize;        /* size of input block */
+  unsigned numThreadsMax;
+  UInt64 expectedDataSize;
+
+  ISeqInStreamPtr inStream;
+  const Byte *inData;
+  size_t inDataSize;
+
+  ICompressProgressPtr progress;
+  ISzAllocPtr allocBig;
+
+  IMtCoderCallback2 *mtCallback;
+  void *mtCallbackObject;
+
+  
+  /* internal variables */
+  
+  size_t allocatedBufsSize;
+
+  CAutoResetEvent readEvent;
+  CSemaphore blocksSemaphore;
+
+  BoolInt stopReading;
+  SRes readRes;
+
+  #ifdef MTCODER_USE_WRITE_THREAD
+    CAutoResetEvent writeEvents[MTCODER_BLOCKS_MAX];
+  #else
+    CAutoResetEvent finishedEvent;
+    SRes writeRes;
+    unsigned writeIndex;
+    Byte ReadyBlocks[MTCODER_BLOCKS_MAX];
+    LONG numFinishedThreads;
+  #endif
+
+  unsigned numStartedThreadsLimit;
+  unsigned numStartedThreads;
+
+  unsigned numBlocksMax;
+  unsigned blockIndex;
+  UInt64 readProcessed;
+
+  CCriticalSection cs;
+
+  unsigned freeBlockHead;
+  unsigned freeBlockList[MTCODER_BLOCKS_MAX];
 
   CMtProgress mtProgress;
-  CMtThread threads[NUM_MT_CODER_THREADS_MAX];
+  CMtCoderBlock blocks[MTCODER_BLOCKS_MAX];
+  CMtCoderThread threads[MTCODER_THREADS_MAX];
 } CMtCoder;
 
-void MtCoder_Construct(CMtCoder* p);
-void MtCoder_Destruct(CMtCoder* p);
+
+void MtCoder_Construct(CMtCoder *p);
+void MtCoder_Destruct(CMtCoder *p);
 SRes MtCoder_Code(CMtCoder *p);
 
+
+#endif
+
+
 EXTERN_C_END
 
 #endif

C/MtDec.h

@@ -0,0 +1,202 @@
+/* MtDec.h -- Multi-thread Decoder
+2023-04-02 : Igor Pavlov : Public domain */
+
+#ifndef ZIP7_INC_MT_DEC_H
+#define ZIP7_INC_MT_DEC_H
+
+#include "7zTypes.h"
+
+#ifndef Z7_ST
+#include "Threads.h"
+#endif
+
+EXTERN_C_BEGIN
+
+#ifndef Z7_ST
+
+#ifndef Z7_ST
+  #define MTDEC_THREADS_MAX 32
+#else
+  #define MTDEC_THREADS_MAX 1
+#endif
+
+
+typedef struct
+{
+  ICompressProgressPtr progress;
+  SRes res;
+  UInt64 totalInSize;
+  UInt64 totalOutSize;
+  CCriticalSection cs;
+} CMtProgress;
+
+void MtProgress_Init(CMtProgress *p, ICompressProgressPtr progress);
+SRes MtProgress_Progress_ST(CMtProgress *p);
+SRes MtProgress_ProgressAdd(CMtProgress *p, UInt64 inSize, UInt64 outSize);
+SRes MtProgress_GetError(CMtProgress *p);
+void MtProgress_SetError(CMtProgress *p, SRes res);
+
+struct CMtDec;
+
+typedef struct
+{
+  struct CMtDec_ *mtDec;
+  unsigned index;
+  void *inBuf;
+
+  size_t inDataSize_Start; // size of input data in start block
+  UInt64 inDataSize;       // total size of input data in all blocks
+
+  CThread thread;
+  CAutoResetEvent canRead;
+  CAutoResetEvent canWrite;
+  void  *allocaPtr;
+} CMtDecThread;
+
+void MtDecThread_FreeInBufs(CMtDecThread *t);
+
+
+typedef enum
+{
+  MTDEC_PARSE_CONTINUE, // continue this block with more input data
+  MTDEC_PARSE_OVERFLOW, // MT buffers overflow, need switch to single-thread
+  MTDEC_PARSE_NEW,      // new block
+  MTDEC_PARSE_END       // end of block threading. But we still can return to threading after Write(&needContinue)
+} EMtDecParseState;
+
+typedef struct
+{
+  // in
+  int startCall;
+  const Byte *src;
+  size_t srcSize;
+      // in  : (srcSize == 0) is allowed
+      // out : it's allowed to return less that actually was used ?
+  int srcFinished;
+
+  // out
+  EMtDecParseState state;
+  BoolInt canCreateNewThread;
+  UInt64 outPos; // check it (size_t)
+} CMtDecCallbackInfo;
+
+
+typedef struct
+{
+  void (*Parse)(void *p, unsigned coderIndex, CMtDecCallbackInfo *ci);
+  
+  // PreCode() and Code():
+  // (SRes_return_result != SZ_OK) means stop decoding, no need another blocks
+  SRes (*PreCode)(void *p, unsigned coderIndex);
+  SRes (*Code)(void *p, unsigned coderIndex,
+      const Byte *src, size_t srcSize, int srcFinished,
+      UInt64 *inCodePos, UInt64 *outCodePos, int *stop);
+  // stop - means stop another Code calls
+
+
+  /* Write() must be called, if Parse() was called
+      set (needWrite) if
+      {
+         && (was not interrupted by progress)
+         && (was not interrupted in previous block)
+      }
+
+    out:
+      if (*needContinue), decoder still need to continue decoding with new iteration,
+         even after MTDEC_PARSE_END
+      if (*canRecode), we didn't flush current block data, so we still can decode current block later.
+  */
+  SRes (*Write)(void *p, unsigned coderIndex,
+      BoolInt needWriteToStream,
+      const Byte *src, size_t srcSize, BoolInt isCross,
+      // int srcFinished,
+      BoolInt *needContinue,
+      BoolInt *canRecode);
+
+} IMtDecCallback2;
+
+
+
+typedef struct CMtDec_
+{
+  /* input variables */
+  
+  size_t inBufSize;        /* size of input block */
+  unsigned numThreadsMax;
+  // size_t inBlockMax;
+  unsigned numThreadsMax_2;
+
+  ISeqInStreamPtr inStream;
+  // const Byte *inData;
+  // size_t inDataSize;
+
+  ICompressProgressPtr progress;
+  ISzAllocPtr alloc;
+
+  IMtDecCallback2 *mtCallback;
+  void *mtCallbackObject;
+
+  
+  /* internal variables */
+  
+  size_t allocatedBufsSize;
+
+  BoolInt exitThread;
+  WRes exitThreadWRes;
+
+  UInt64 blockIndex;
+  BoolInt isAllocError;
+  BoolInt overflow;
+  SRes threadingErrorSRes;
+
+  BoolInt needContinue;
+
+  // CAutoResetEvent finishedEvent;
+
+  SRes readRes;
+  SRes codeRes;
+
+  BoolInt wasInterrupted;
+
+  unsigned numStartedThreads_Limit;
+  unsigned numStartedThreads;
+
+  Byte *crossBlock;
+  size_t crossStart;
+  size_t crossEnd;
+  UInt64 readProcessed;
+  BoolInt readWasFinished;
+  UInt64 inProcessed;
+
+  unsigned filledThreadStart;
+  unsigned numFilledThreads;
+
+  #ifndef Z7_ST
+  BoolInt needInterrupt;
+  UInt64 interruptIndex;
+  CMtProgress mtProgress;
+  CMtDecThread threads[MTDEC_THREADS_MAX];
+  #endif
+} CMtDec;
+
+
+void MtDec_Construct(CMtDec *p);
+void MtDec_Destruct(CMtDec *p);
+
+/*
+MtDec_Code() returns:
+  SZ_OK - in most cases
+  MY_SRes_HRESULT_FROM_WRes(WRes_error) - in case of unexpected error in threading function
+*/
+  
+SRes MtDec_Code(CMtDec *p);
+Byte *MtDec_GetCrossBuff(CMtDec *p);
+
+int MtDec_PrepareRead(CMtDec *p);
+const Byte *MtDec_Read(CMtDec *p, size_t *inLim);
+
+#endif
+
+EXTERN_C_END
+
+#endif

C/Ppmd.h

@@ -1,16 +1,24 @@
 /* Ppmd.h -- PPMD codec common code
-2010-03-12 : Igor Pavlov : Public domain
+2023-03-05 : Igor Pavlov : Public domain
 This code is based on PPMd var.H (2001): Dmitry Shkarin : Public domain */
 
-#ifndef __PPMD_H
-#define __PPMD_H
+#ifndef ZIP7_INC_PPMD_H
+#define ZIP7_INC_PPMD_H
 
-#include "Types.h"
 #include "CpuArch.h"
 
 EXTERN_C_BEGIN
 
-#ifdef MY_CPU_32BIT
+#if defined(MY_CPU_SIZEOF_POINTER) && (MY_CPU_SIZEOF_POINTER == 4)
+/*
+   PPMD code always uses 32-bit internal fields in PPMD structures to store internal references in main block.
+   if (PPMD_32BIT is     defined), the PPMD code stores internal pointers to 32-bit reference fields.
+   if (PPMD_32BIT is NOT defined), the PPMD code stores internal UInt32 offsets to reference fields.
+   if (pointer size is 64-bit), then (PPMD_32BIT) mode is not allowed,
+   if (pointer size is 32-bit), then (PPMD_32BIT) mode is optional,
+     and it's allowed to disable PPMD_32BIT mode even if pointer is 32-bit.
+   PPMD code works slightly faster in (PPMD_32BIT) mode.
+*/
   #define PPMD_32BIT
 #endif
 
@@ -29,6 +37,9 @@ EXTERN_C_BEGIN
 #define PPMD_N4 ((128 + 3 - 1 * PPMD_N1 - 2 * PPMD_N2 - 3 * PPMD_N3) / 4)
 #define PPMD_NUM_INDEXES (PPMD_N1 + PPMD_N2 + PPMD_N3 + PPMD_N4)
 
+MY_CPU_pragma_pack_push_1
+/* Most compilers works OK here even without #pragma pack(push, 1), but some GCC compilers need it. */
+
 /* SEE-contexts for PPM-contexts with masked symbols */
 typedef struct
 {
@@ -37,44 +48,121 @@ typedef struct
   Byte Count;  /* Count to next change of Shift */
 } CPpmd_See;
 
-#define Ppmd_See_Update(p)  if ((p)->Shift < PPMD_PERIOD_BITS && --(p)->Count == 0) \
-    { (p)->Summ <<= 1; (p)->Count = (Byte)(3 << (p)->Shift++); }
+#define Ppmd_See_UPDATE(p) \
+  { if ((p)->Shift < PPMD_PERIOD_BITS && --(p)->Count == 0) \
+    { (p)->Summ = (UInt16)((p)->Summ << 1); \
+      (p)->Count = (Byte)(3 << (p)->Shift++); }}
+
 
 typedef struct
 {
   Byte Symbol;
   Byte Freq;
-  UInt16 SuccessorLow;
-  UInt16 SuccessorHigh;
+  UInt16 Successor_0;
+  UInt16 Successor_1;
 } CPpmd_State;
 
-typedef
-  #ifdef PPMD_32BIT
-    CPpmd_State *
-  #else
-    UInt32
-  #endif
-  CPpmd_State_Ref;
+typedef struct CPpmd_State2_
+{
+  Byte Symbol;
+  Byte Freq;
+} CPpmd_State2;
 
-typedef
-  #ifdef PPMD_32BIT
-    void *
-  #else
-    UInt32
-  #endif
-  CPpmd_Void_Ref;
+typedef struct CPpmd_State4_
+{
+  UInt16 Successor_0;
+  UInt16 Successor_1;
+} CPpmd_State4;
+
+MY_CPU_pragma_pop
+
+/*
+   PPMD code can write full CPpmd_State structure data to CPpmd*_Context
+      at (byte offset = 2) instead of some fields of original CPpmd*_Context structure.
+   
+   If we use pointers to different types, but that point to shared
+   memory space, we can have aliasing problem (strict aliasing).
+   
+   XLC compiler in -O2 mode can change the order of memory write instructions
+   in relation to read instructions, if we have use pointers to different types.
+   
+   To solve that aliasing problem we use combined CPpmd*_Context structure
+   with unions that contain the fields from both structures:
+   the original CPpmd*_Context and CPpmd_State.
+   So we can access the fields from both structures via one pointer,
+   and the compiler doesn't change the order of write instructions
+   in relation to read instructions.
+
+   If we don't use memory write instructions to shared memory in
+   some local code, and we use only reading instructions (read only),
+   then probably it's safe to use pointers to different types for reading.
+*/
+  
+
+
+#ifdef PPMD_32BIT
+
+  #define Ppmd_Ref_Type(type)   type *
+  #define Ppmd_GetRef(p, ptr)   (ptr)
+  #define Ppmd_GetPtr(p, ptr)   (ptr)
+  #define Ppmd_GetPtr_Type(p, ptr, note_type) (ptr)
+
+#else
+
+  #define Ppmd_Ref_Type(type)   UInt32
+  #define Ppmd_GetRef(p, ptr)   ((UInt32)((Byte *)(ptr) - (p)->Base))
+  #define Ppmd_GetPtr(p, offs)  ((void *)((p)->Base + (offs)))
+  #define Ppmd_GetPtr_Type(p, offs, type) ((type *)Ppmd_GetPtr(p, offs))
+
+#endif // PPMD_32BIT
+
+
+typedef Ppmd_Ref_Type(CPpmd_State) CPpmd_State_Ref;
+typedef Ppmd_Ref_Type(void)        CPpmd_Void_Ref;
+typedef Ppmd_Ref_Type(Byte)        CPpmd_Byte_Ref;
+
+
+/*
+#ifdef MY_CPU_LE_UNALIGN
+// the unaligned 32-bit access latency can be too large, if the data is not in L1 cache.
+#define Ppmd_GET_SUCCESSOR(p) ((CPpmd_Void_Ref)*(const UInt32 *)(const void *)&(p)->Successor_0)
+#define Ppmd_SET_SUCCESSOR(p, v) *(UInt32 *)(void *)(void *)&(p)->Successor_0 = (UInt32)(v)
+
+#else
+*/
+
+/*
+   We can write 16-bit halves to 32-bit (Successor) field in any selected order.
+   But the native order is more consistent way.
+   So we use the native order, if LE/BE order can be detected here at compile time.
+*/
+
+#ifdef MY_CPU_BE
+
+  #define Ppmd_GET_SUCCESSOR(p) \
+    ( (CPpmd_Void_Ref) (((UInt32)(p)->Successor_0 << 16) | (p)->Successor_1) )
+
+  #define Ppmd_SET_SUCCESSOR(p, v) { \
+    (p)->Successor_0 = (UInt16)(((UInt32)(v) >> 16) /* & 0xFFFF */); \
+    (p)->Successor_1 = (UInt16)((UInt32)(v) /* & 0xFFFF */); }
+
+#else
+
+  #define Ppmd_GET_SUCCESSOR(p) \
+    ( (CPpmd_Void_Ref) ((p)->Successor_0 | ((UInt32)(p)->Successor_1 << 16)) )
+
+  #define Ppmd_SET_SUCCESSOR(p, v) { \
+    (p)->Successor_0 = (UInt16)((UInt32)(v) /* & 0xFFFF */); \
+    (p)->Successor_1 = (UInt16)(((UInt32)(v) >> 16) /* & 0xFFFF */); }
+
+#endif
+
+// #endif
 
-typedef
-  #ifdef PPMD_32BIT
-    Byte *
-  #else
-    UInt32
-  #endif
-  CPpmd_Byte_Ref;
 
 #define PPMD_SetAllBitsIn256Bytes(p) \
-  { unsigned i; for (i = 0; i < 256 / sizeof(p[0]); i += 8) { \
-  p[i+7] = p[i+6] = p[i+5] = p[i+4] = p[i+3] = p[i+2] = p[i+1] = p[i+0] = ~(size_t)0; }}
+  { size_t z; for (z = 0; z < 256 / sizeof(p[0]); z += 8) { \
+  p[z+7] = p[z+6] = p[z+5] = p[z+4] = p[z+3] = p[z+2] = p[z+1] = p[z+0] = ~(size_t)0; }}
 
 EXTERN_C_END
  

C/Ppmd7.h

@@ -1,13 +1,11 @@
-/* Ppmd7.h -- PPMdH compression codec
-2010-03-12 : Igor Pavlov : Public domain
-This code is based on PPMd var.H (2001): Dmitry Shkarin : Public domain */
+/* Ppmd7.h -- Ppmd7 (PPMdH) compression codec
+2023-04-02 : Igor Pavlov : Public domain
+This code is based on:
+  PPMd var.H (2001): Dmitry Shkarin : Public domain */
+ 
 
-/* This code supports virtual RangeDecoder and includes the implementation
-of RangeCoder from 7z, instead of RangeCoder from original PPMd var.H.
-If you need the compatibility with original PPMd var.H, you can use external RangeDecoder */
-
-#ifndef __PPMD7_H
-#define __PPMD7_H
+#ifndef ZIP7_INC_PPMD7_H
+#define ZIP7_INC_PPMD7_H
 
 #include "Ppmd.h"
 
@@ -21,23 +19,56 @@ EXTERN_C_BEGIN
 
 struct CPpmd7_Context_;
 
-typedef
-  #ifdef PPMD_32BIT
-    struct CPpmd7_Context_ *
-  #else
-    UInt32
-  #endif
-  CPpmd7_Context_Ref;
+typedef Ppmd_Ref_Type(struct CPpmd7_Context_) CPpmd7_Context_Ref;
+
+// MY_CPU_pragma_pack_push_1
 
 typedef struct CPpmd7_Context_
 {
   UInt16 NumStats;
-  UInt16 SummFreq;
-  CPpmd_State_Ref Stats;
+
+
+  union
+  {
+    UInt16 SummFreq;
+    CPpmd_State2 State2;
+  } Union2;
+
+  union
+  {
+    CPpmd_State_Ref Stats;
+    CPpmd_State4 State4;
+  } Union4;
+
   CPpmd7_Context_Ref Suffix;
 } CPpmd7_Context;
 
-#define Ppmd7Context_OneState(p) ((CPpmd_State *)&(p)->SummFreq)
+// MY_CPU_pragma_pop
+
+#define Ppmd7Context_OneState(p) ((CPpmd_State *)&(p)->Union2)
+
+
+
+
+typedef struct
+{
+  UInt32 Range;
+  UInt32 Code;
+  UInt32 Low;
+  IByteInPtr Stream;
+} CPpmd7_RangeDec;
+
+
+typedef struct
+{
+  UInt32 Range;
+  Byte Cache;
+  // Byte _dummy_[3];
+  UInt64 Low;
+  UInt64 CacheSize;
+  IByteOutPtr Stream;
+} CPpmd7z_RangeEnc;
+
 
 typedef struct
 {
@@ -48,92 +79,102 @@ typedef struct
 
   UInt32 Size;
   UInt32 GlueCount;
-  Byte *Base, *LoUnit, *HiUnit, *Text, *UnitsStart;
   UInt32 AlignOffset;
+  Byte *Base, *LoUnit, *HiUnit, *Text, *UnitsStart;
 
-  Byte Indx2Units[PPMD_NUM_INDEXES];
+
+  
+  
+  union
+  {
+    CPpmd7_RangeDec dec;
+    CPpmd7z_RangeEnc enc;
+  } rc;
+  
+  Byte Indx2Units[PPMD_NUM_INDEXES + 2]; // +2 for alignment
   Byte Units2Indx[128];
   CPpmd_Void_Ref FreeList[PPMD_NUM_INDEXES];
-  Byte NS2Indx[256], NS2BSIndx[256], HB2Flag[256];
+
+  Byte NS2BSIndx[256], NS2Indx[256];
+  Byte ExpEscape[16];
   CPpmd_See DummySee, See[25][16];
   UInt16 BinSumm[128][64];
+  // int LastSymbol;
 } CPpmd7;
 
+
 void Ppmd7_Construct(CPpmd7 *p);
-Bool Ppmd7_Alloc(CPpmd7 *p, UInt32 size, ISzAlloc *alloc);
-void Ppmd7_Free(CPpmd7 *p, ISzAlloc *alloc);
+BoolInt Ppmd7_Alloc(CPpmd7 *p, UInt32 size, ISzAllocPtr alloc);
+void Ppmd7_Free(CPpmd7 *p, ISzAllocPtr alloc);
 void Ppmd7_Init(CPpmd7 *p, unsigned maxOrder);
 #define Ppmd7_WasAllocated(p) ((p)->Base != NULL)
 
 
 /* ---------- Internal Functions ---------- */
 
-extern const Byte PPMD7_kExpEscape[16];
-
-#ifdef PPMD_32BIT
-  #define Ppmd7_GetPtr(p, ptr) (ptr)
-  #define Ppmd7_GetContext(p, ptr) (ptr)
-  #define Ppmd7_GetStats(p, ctx) ((ctx)->Stats)
-#else
-  #define Ppmd7_GetPtr(p, offs) ((void *)((p)->Base + (offs)))
-  #define Ppmd7_GetContext(p, offs) ((CPpmd7_Context *)Ppmd7_GetPtr((p), (offs)))
-  #define Ppmd7_GetStats(p, ctx) ((CPpmd_State *)Ppmd7_GetPtr((p), ((ctx)->Stats)))
-#endif
+#define Ppmd7_GetPtr(p, ptr)     Ppmd_GetPtr(p, ptr)
+#define Ppmd7_GetContext(p, ptr) Ppmd_GetPtr_Type(p, ptr, CPpmd7_Context)
+#define Ppmd7_GetStats(p, ctx)   Ppmd_GetPtr_Type(p, (ctx)->Union4.Stats, CPpmd_State)
 
 void Ppmd7_Update1(CPpmd7 *p);
 void Ppmd7_Update1_0(CPpmd7 *p);
 void Ppmd7_Update2(CPpmd7 *p);
-void Ppmd7_UpdateBin(CPpmd7 *p);
+
+#define PPMD7_HiBitsFlag_3(sym) ((((unsigned)sym + 0xC0) >> (8 - 3)) & (1 << 3))
+#define PPMD7_HiBitsFlag_4(sym) ((((unsigned)sym + 0xC0) >> (8 - 4)) & (1 << 4))
+// #define PPMD7_HiBitsFlag_3(sym) ((sym) < 0x40 ? 0 : (1 << 3))
+// #define PPMD7_HiBitsFlag_4(sym) ((sym) < 0x40 ? 0 : (1 << 4))
 
 #define Ppmd7_GetBinSumm(p) \
-    &p->BinSumm[Ppmd7Context_OneState(p->MinContext)->Freq - 1][p->PrevSuccess + \
-    p->NS2BSIndx[Ppmd7_GetContext(p, p->MinContext->Suffix)->NumStats - 1] + \
-    (p->HiBitsFlag = p->HB2Flag[p->FoundState->Symbol]) + \
-    2 * p->HB2Flag[Ppmd7Context_OneState(p->MinContext)->Symbol] + \
-    ((p->RunLength >> 26) & 0x20)]
+    &p->BinSumm[(size_t)(unsigned)Ppmd7Context_OneState(p->MinContext)->Freq - 1] \
+    [ p->PrevSuccess + ((p->RunLength >> 26) & 0x20) \
+    + p->NS2BSIndx[(size_t)Ppmd7_GetContext(p, p->MinContext->Suffix)->NumStats - 1] \
+    + PPMD7_HiBitsFlag_4(Ppmd7Context_OneState(p->MinContext)->Symbol) \
+    + (p->HiBitsFlag = PPMD7_HiBitsFlag_3(p->FoundState->Symbol)) ]
 
 CPpmd_See *Ppmd7_MakeEscFreq(CPpmd7 *p, unsigned numMasked, UInt32 *scale);
 
 
+/*
+We support two versions of Ppmd7 (PPMdH) methods that use same CPpmd7 structure:
+  1) Ppmd7a_*: original PPMdH
+  2) Ppmd7z_*: modified PPMdH with 7z Range Coder
+Ppmd7_*: the structures and functions that are common for both versions of PPMd7 (PPMdH)
+*/
+
 /* ---------- Decode ---------- */
 
-typedef struct
-{
-  UInt32 (*GetThreshold)(void *p, UInt32 total);
-  void (*Decode)(void *p, UInt32 start, UInt32 size);
-  UInt32 (*DecodeBit)(void *p, UInt32 size0);
-} IPpmd7_RangeDec;
+#define PPMD7_SYM_END    (-1)
+#define PPMD7_SYM_ERROR  (-2)
 
-typedef struct
-{
-  IPpmd7_RangeDec p;
-  UInt32 Range;
-  UInt32 Code;
-  IByteIn *Stream;
-} CPpmd7z_RangeDec;
+/*
+You must set (CPpmd7::rc.dec.Stream) before Ppmd7*_RangeDec_Init()
 
-void Ppmd7z_RangeDec_CreateVTable(CPpmd7z_RangeDec *p);
-Bool Ppmd7z_RangeDec_Init(CPpmd7z_RangeDec *p);
+Ppmd7*_DecodeSymbol()
+out:
+  >= 0 : decoded byte
+    -1 : PPMD7_SYM_END   : End of payload marker
+    -2 : PPMD7_SYM_ERROR : Data error
+*/
+
+/* Ppmd7a_* : original PPMdH */
+BoolInt Ppmd7a_RangeDec_Init(CPpmd7_RangeDec *p);
+#define Ppmd7a_RangeDec_IsFinishedOK(p) ((p)->Code == 0)
+int Ppmd7a_DecodeSymbol(CPpmd7 *p);
+
+/* Ppmd7z_* : modified PPMdH with 7z Range Coder */
+BoolInt Ppmd7z_RangeDec_Init(CPpmd7_RangeDec *p);
 #define Ppmd7z_RangeDec_IsFinishedOK(p) ((p)->Code == 0)
-
-int Ppmd7_DecodeSymbol(CPpmd7 *p, IPpmd7_RangeDec *rc);
+int Ppmd7z_DecodeSymbol(CPpmd7 *p);
+// Byte *Ppmd7z_DecodeSymbols(CPpmd7 *p, Byte *buf, const Byte *lim);
 
 
 /* ---------- Encode ---------- */
 
-typedef struct
-{
-  UInt64 Low;
-  UInt32 Range;
-  Byte Cache;
-  UInt64 CacheSize;
-  IByteOut *Stream;
-} CPpmd7z_RangeEnc;
-
-void Ppmd7z_RangeEnc_Init(CPpmd7z_RangeEnc *p);
-void Ppmd7z_RangeEnc_FlushData(CPpmd7z_RangeEnc *p);
-
-void Ppmd7_EncodeSymbol(CPpmd7 *p, CPpmd7z_RangeEnc *rc, int symbol);
+void Ppmd7z_Init_RangeEnc(CPpmd7 *p);
+void Ppmd7z_Flush_RangeEnc(CPpmd7 *p);
+// void Ppmd7z_EncodeSymbol(CPpmd7 *p, int symbol);
+void Ppmd7z_EncodeSymbols(CPpmd7 *p, const Byte *buf, const Byte *lim);
 
 EXTERN_C_END
  

C/Ppmd7Dec.c

@@ -1,187 +1,312 @@
-/* Ppmd7Dec.c -- PPMdH Decoder
-2010-03-12 : Igor Pavlov : Public domain
-This code is based on PPMd var.H (2001): Dmitry Shkarin : Public domain */
+/* Ppmd7Dec.c -- Ppmd7z (PPMdH with 7z Range Coder) Decoder
+2023-04-02 : Igor Pavlov : Public domain
+This code is based on:
+  PPMd var.H (2001): Dmitry Shkarin : Public domain */
+
+
+#include "Precomp.h"
 
 #include "Ppmd7.h"
 
-#define kTopValue (1 << 24)
+#define kTopValue ((UInt32)1 << 24)
 
-Bool Ppmd7z_RangeDec_Init(CPpmd7z_RangeDec *p)
+
+#define READ_BYTE(p) IByteIn_Read((p)->Stream)
+
+BoolInt Ppmd7z_RangeDec_Init(CPpmd7_RangeDec *p)
 {
   unsigned i;
   p->Code = 0;
   p->Range = 0xFFFFFFFF;
-  if (p->Stream->Read((void *)p->Stream) != 0)
+  if (READ_BYTE(p) != 0)
     return False;
   for (i = 0; i < 4; i++)
-    p->Code = (p->Code << 8) | p->Stream->Read((void *)p->Stream);
+    p->Code = (p->Code << 8) | READ_BYTE(p);
   return (p->Code < 0xFFFFFFFF);
 }
 
-static UInt32 Range_GetThreshold(void *pp, UInt32 total)
+#define RC_NORM_BASE(p) if ((p)->Range < kTopValue) \
+  { (p)->Code = ((p)->Code << 8) | READ_BYTE(p); (p)->Range <<= 8;
+
+#define RC_NORM_1(p)  RC_NORM_BASE(p) }
+#define RC_NORM(p)    RC_NORM_BASE(p) RC_NORM_BASE(p) }}
+
+// we must use only one type of Normalization from two: LOCAL or REMOTE
+#define RC_NORM_LOCAL(p)    // RC_NORM(p)
+#define RC_NORM_REMOTE(p)   RC_NORM(p)
+
+#define R (&p->rc.dec)
+
+Z7_FORCE_INLINE
+// Z7_NO_INLINE
+static void Ppmd7z_RD_Decode(CPpmd7 *p, UInt32 start, UInt32 size)
 {
-  CPpmd7z_RangeDec *p = (CPpmd7z_RangeDec *)pp;
-  return (p->Code) / (p->Range /= total);
+
+  
+  R->Code -= start * R->Range;
+  R->Range *= size;
+  RC_NORM_LOCAL(R)
 }
 
-static void Range_Normalize(CPpmd7z_RangeDec *p)
-{
-  if (p->Range < kTopValue)
-  {
-    p->Code = (p->Code << 8) | p->Stream->Read((void *)p->Stream);
-    p->Range <<= 8;
-    if (p->Range < kTopValue)
-    {
-      p->Code = (p->Code << 8) | p->Stream->Read((void *)p->Stream);
-      p->Range <<= 8;
-    }
-  }
-}
-
-static void Range_Decode(void *pp, UInt32 start, UInt32 size)
-{
-  CPpmd7z_RangeDec *p = (CPpmd7z_RangeDec *)pp;
-  p->Code -= start * p->Range;
-  p->Range *= size;
-  Range_Normalize(p);
-}
-
-static UInt32 Range_DecodeBit(void *pp, UInt32 size0)
-{
-  CPpmd7z_RangeDec *p = (CPpmd7z_RangeDec *)pp;
-  UInt32 newBound = (p->Range >> 14) * size0;
-  UInt32 symbol;
-  if (p->Code < newBound)
-  {
-    symbol = 0;
-    p->Range = newBound;
-  }
-  else
-  {
-    symbol = 1;
-    p->Code -= newBound;
-    p->Range -= newBound;
-  }
-  Range_Normalize(p);
-  return symbol;
-}
-
-void Ppmd7z_RangeDec_CreateVTable(CPpmd7z_RangeDec *p)
-{
-  p->p.GetThreshold = Range_GetThreshold;
-  p->p.Decode = Range_Decode;
-  p->p.DecodeBit = Range_DecodeBit;
-}
+#define RC_Decode(start, size)  Ppmd7z_RD_Decode(p, start, size);
+#define RC_DecodeFinal(start, size)  RC_Decode(start, size)  RC_NORM_REMOTE(R)
+#define RC_GetThreshold(total)  (R->Code / (R->Range /= (total)))
 
 
-#define MASK(sym) ((signed char *)charMask)[sym]
+#define CTX(ref) ((CPpmd7_Context *)Ppmd7_GetContext(p, ref))
+// typedef CPpmd7_Context * CTX_PTR;
+#define SUCCESSOR(p) Ppmd_GET_SUCCESSOR(p)
+void Ppmd7_UpdateModel(CPpmd7 *p);
 
-int Ppmd7_DecodeSymbol(CPpmd7 *p, IPpmd7_RangeDec *rc)
+#define MASK(sym) ((unsigned char *)charMask)[sym]
+// Z7_FORCE_INLINE
+// static
+int Ppmd7z_DecodeSymbol(CPpmd7 *p)
 {
   size_t charMask[256 / sizeof(size_t)];
+
   if (p->MinContext->NumStats != 1)
   {
     CPpmd_State *s = Ppmd7_GetStats(p, p->MinContext);
     unsigned i;
     UInt32 count, hiCnt;
-    if ((count = rc->GetThreshold(rc, p->MinContext->SummFreq)) < (hiCnt = s->Freq))
+    const UInt32 summFreq = p->MinContext->Union2.SummFreq;
+
+    
+    
+    
+    count = RC_GetThreshold(summFreq);
+    hiCnt = count;
+    
+    if ((Int32)(count -= s->Freq) < 0)
     {
-      Byte symbol;
-      rc->Decode(rc, 0, s->Freq);
+      Byte sym;
+      RC_DecodeFinal(0, s->Freq)
       p->FoundState = s;
-      symbol = s->Symbol;
+      sym = s->Symbol;
       Ppmd7_Update1_0(p);
-      return symbol;
+      return sym;
     }
+  
     p->PrevSuccess = 0;
-    i = p->MinContext->NumStats - 1;
+    i = (unsigned)p->MinContext->NumStats - 1;
+    
     do
     {
-      if ((hiCnt += (++s)->Freq) > count)
+      if ((Int32)(count -= (++s)->Freq) < 0)
       {
-        Byte symbol;
-        rc->Decode(rc, hiCnt - s->Freq, s->Freq);
+        Byte sym;
+        RC_DecodeFinal((hiCnt - count) - s->Freq, s->Freq)
         p->FoundState = s;
-        symbol = s->Symbol;
+        sym = s->Symbol;
         Ppmd7_Update1(p);
-        return symbol;
+        return sym;
       }
     }
     while (--i);
-    if (count >= p->MinContext->SummFreq)
-      return -2;
-    p->HiBitsFlag = p->HB2Flag[p->FoundState->Symbol];
-    rc->Decode(rc, hiCnt, p->MinContext->SummFreq - hiCnt);
-    PPMD_SetAllBitsIn256Bytes(charMask);
-    MASK(s->Symbol) = 0;
-    i = p->MinContext->NumStats - 1;
-    do { MASK((--s)->Symbol) = 0; } while (--i);
+    
+    if (hiCnt >= summFreq)
+      return PPMD7_SYM_ERROR;
+    
+    hiCnt -= count;
+    RC_Decode(hiCnt, summFreq - hiCnt)
+
+    p->HiBitsFlag = PPMD7_HiBitsFlag_3(p->FoundState->Symbol);
+    PPMD_SetAllBitsIn256Bytes(charMask)
+    // i = p->MinContext->NumStats - 1;
+    // do { MASK((--s)->Symbol) = 0; } while (--i);
+    {
+      CPpmd_State *s2 = Ppmd7_GetStats(p, p->MinContext);
+      MASK(s->Symbol) = 0;
+      do
+      {
+        unsigned sym0 = s2[0].Symbol;
+        unsigned sym1 = s2[1].Symbol;
+        s2 += 2;
+        MASK(sym0) = 0;
+        MASK(sym1) = 0;
+      }
+      while (s2 < s);
+    }
   }
   else
   {
+    CPpmd_State *s = Ppmd7Context_OneState(p->MinContext);
     UInt16 *prob = Ppmd7_GetBinSumm(p);
-    if (rc->DecodeBit(rc, *prob) == 0)
+    UInt32 pr = *prob;
+    UInt32 size0 = (R->Range >> 14) * pr;
+    pr = PPMD_UPDATE_PROB_1(pr);
+
+    if (R->Code < size0)
     {
-      Byte symbol;
-      *prob = (UInt16)PPMD_UPDATE_PROB_0(*prob);
-      symbol = (p->FoundState = Ppmd7Context_OneState(p->MinContext))->Symbol;
-      Ppmd7_UpdateBin(p);
-      return symbol;
+      Byte sym;
+      *prob = (UInt16)(pr + (1 << PPMD_INT_BITS));
+      
+      // RangeDec_DecodeBit0(size0);
+      R->Range = size0;
+      RC_NORM_1(R)
+      /* we can use single byte normalization here because of
+         (min(BinSumm[][]) = 95) > (1 << (14 - 8)) */
+
+      // sym = (p->FoundState = Ppmd7Context_OneState(p->MinContext))->Symbol;
+      // Ppmd7_UpdateBin(p);
+      {
+        unsigned freq = s->Freq;
+        CPpmd7_Context *c = CTX(SUCCESSOR(s));
+        sym = s->Symbol;
+        p->FoundState = s;
+        p->PrevSuccess = 1;
+        p->RunLength++;
+        s->Freq = (Byte)(freq + (freq < 128));
+        // NextContext(p);
+        if (p->OrderFall == 0 && (const Byte *)c > p->Text)
+          p->MaxContext = p->MinContext = c;
+        else
+          Ppmd7_UpdateModel(p);
+      }
+      return sym;
     }
-    *prob = (UInt16)PPMD_UPDATE_PROB_1(*prob);
-    p->InitEsc = PPMD7_kExpEscape[*prob >> 10];
-    PPMD_SetAllBitsIn256Bytes(charMask);
+
+    *prob = (UInt16)pr;
+    p->InitEsc = p->ExpEscape[pr >> 10];
+
+    // RangeDec_DecodeBit1(size0);
+    
+    R->Code -= size0;
+    R->Range -= size0;
+    RC_NORM_LOCAL(R)
+    
+    PPMD_SetAllBitsIn256Bytes(charMask)
     MASK(Ppmd7Context_OneState(p->MinContext)->Symbol) = 0;
     p->PrevSuccess = 0;
   }
+
   for (;;)
   {
-    CPpmd_State *ps[256], *s;
+    CPpmd_State *s, *s2;
     UInt32 freqSum, count, hiCnt;
+
     CPpmd_See *see;
-    unsigned i, num, numMasked = p->MinContext->NumStats;
+    CPpmd7_Context *mc;
+    unsigned numMasked;
+    RC_NORM_REMOTE(R)
+    mc = p->MinContext;
+    numMasked = mc->NumStats;
+
     do
     {
       p->OrderFall++;
-      if (!p->MinContext->Suffix)
-        return -1;
-      p->MinContext = Ppmd7_GetContext(p, p->MinContext->Suffix);
+      if (!mc->Suffix)
+        return PPMD7_SYM_END;
+      mc = Ppmd7_GetContext(p, mc->Suffix);
     }
-    while (p->MinContext->NumStats == numMasked);
-    hiCnt = 0;
-    s = Ppmd7_GetStats(p, p->MinContext);
-    i = 0;
-    num = p->MinContext->NumStats - numMasked;
-    do
-    {
-      int k = (int)(MASK(s->Symbol));
-      hiCnt += (s->Freq & k);
-      ps[i] = s++;
-      i -= k;
-    }
-    while (i != num);
+    while (mc->NumStats == numMasked);
     
+    s = Ppmd7_GetStats(p, mc);
+
+    {
+      unsigned num = mc->NumStats;
+      unsigned num2 = num / 2;
+      
+      num &= 1;
+      hiCnt = (s->Freq & (unsigned)(MASK(s->Symbol))) & (0 - (UInt32)num);
+      s += num;
+      p->MinContext = mc;
+
+      do
+      {
+        unsigned sym0 = s[0].Symbol;
+        unsigned sym1 = s[1].Symbol;
+        s += 2;
+        hiCnt += (s[-2].Freq & (unsigned)(MASK(sym0)));
+        hiCnt += (s[-1].Freq & (unsigned)(MASK(sym1)));
+      }
+      while (--num2);
+    }
+
     see = Ppmd7_MakeEscFreq(p, numMasked, &freqSum);
     freqSum += hiCnt;
-    count = rc->GetThreshold(rc, freqSum);
+
+
+
+
+    count = RC_GetThreshold(freqSum);
     
     if (count < hiCnt)
     {
-      Byte symbol;
-      CPpmd_State **pps = ps;
-      for (hiCnt = 0; (hiCnt += (*pps)->Freq) <= count; pps++);
-      s = *pps;
-      rc->Decode(rc, hiCnt - s->Freq, s->Freq);
-      Ppmd_See_Update(see);
+      Byte sym;
+
+      s = Ppmd7_GetStats(p, p->MinContext);
+      hiCnt = count;
+      // count -= s->Freq & (unsigned)(MASK(s->Symbol));
+      // if ((Int32)count >= 0)
+      {
+        for (;;)
+        {
+          count -= s->Freq & (unsigned)(MASK((s)->Symbol)); s++; if ((Int32)count < 0) break;
+          // count -= s->Freq & (unsigned)(MASK((s)->Symbol)); s++; if ((Int32)count < 0) break;
+        }
+      }
+      s--;
+      RC_DecodeFinal((hiCnt - count) - s->Freq, s->Freq)
+
+      // new (see->Summ) value can overflow over 16-bits in some rare cases
+      Ppmd_See_UPDATE(see)
       p->FoundState = s;
-      symbol = s->Symbol;
+      sym = s->Symbol;
       Ppmd7_Update2(p);
-      return symbol;
+      return sym;
     }
+
     if (count >= freqSum)
-      return -2;
-    rc->Decode(rc, hiCnt, freqSum - hiCnt);
+      return PPMD7_SYM_ERROR;
+    
+    RC_Decode(hiCnt, freqSum - hiCnt)
+
+    // We increase (see->Summ) for sum of Freqs of all non_Masked symbols.
+    // new (see->Summ) value can overflow over 16-bits in some rare cases
     see->Summ = (UInt16)(see->Summ + freqSum);
-    do { MASK(ps[--i]->Symbol) = 0; } while (i != 0);
+
+    s = Ppmd7_GetStats(p, p->MinContext);
+    s2 = s + p->MinContext->NumStats;
+    do
+    {
+      MASK(s->Symbol) = 0;
+      s++;
+    }
+    while (s != s2);
   }
 }
+
+/*
+Byte *Ppmd7z_DecodeSymbols(CPpmd7 *p, Byte *buf, const Byte *lim)
+{
+  int sym = 0;
+  if (buf != lim)
+  do
+  {
+    sym = Ppmd7z_DecodeSymbol(p);
+    if (sym < 0)
+      break;
+    *buf = (Byte)sym;
+  }
+  while (++buf < lim);
+  p->LastSymbol = sym;
+  return buf;
+}
+*/
+
+#undef kTopValue
+#undef READ_BYTE
+#undef RC_NORM_BASE
+#undef RC_NORM_1
+#undef RC_NORM
+#undef RC_NORM_LOCAL
+#undef RC_NORM_REMOTE
+#undef R
+#undef RC_Decode
+#undef RC_DecodeFinal
+#undef RC_GetThreshold
+#undef CTX
+#undef SUCCESSOR
+#undef MASK

C/Ppmd7Enc.c

@@ -1,102 +1,123 @@
-/* Ppmd7Enc.c -- PPMdH Encoder
-2010-03-12 : Igor Pavlov : Public domain
-This code is based on PPMd var.H (2001): Dmitry Shkarin : Public domain */
+/* Ppmd7Enc.c -- Ppmd7z (PPMdH with 7z Range Coder) Encoder
+2023-04-02 : Igor Pavlov : Public domain
+This code is based on:
+  PPMd var.H (2001): Dmitry Shkarin : Public domain */
+
+
+#include "Precomp.h"
 
 #include "Ppmd7.h"
 
-#define kTopValue (1 << 24)
+#define kTopValue ((UInt32)1 << 24)
 
-void Ppmd7z_RangeEnc_Init(CPpmd7z_RangeEnc *p)
+#define R (&p->rc.enc)
+
+void Ppmd7z_Init_RangeEnc(CPpmd7 *p)
 {
-  p->Low = 0;
-  p->Range = 0xFFFFFFFF;
-  p->Cache = 0;
-  p->CacheSize = 1;
+  R->Low = 0;
+  R->Range = 0xFFFFFFFF;
+  R->Cache = 0;
+  R->CacheSize = 1;
 }
 
-static void RangeEnc_ShiftLow(CPpmd7z_RangeEnc *p)
+Z7_NO_INLINE
+static void Ppmd7z_RangeEnc_ShiftLow(CPpmd7 *p)
 {
-  if ((UInt32)p->Low < (UInt32)0xFF000000 || (unsigned)(p->Low >> 32) != 0)
+  if ((UInt32)R->Low < (UInt32)0xFF000000 || (unsigned)(R->Low >> 32) != 0)
   {
-    Byte temp = p->Cache;
+    Byte temp = R->Cache;
     do
     {
-      p->Stream->Write(p->Stream, (Byte)(temp + (Byte)(p->Low >> 32)));
+      IByteOut_Write(R->Stream, (Byte)(temp + (Byte)(R->Low >> 32)));
       temp = 0xFF;
     }
-    while(--p->CacheSize != 0);
-    p->Cache = (Byte)((UInt32)p->Low >> 24);
+    while (--R->CacheSize != 0);
+    R->Cache = (Byte)((UInt32)R->Low >> 24);
   }
-  p->CacheSize++;
-  p->Low = (UInt32)p->Low << 8;
+  R->CacheSize++;
+  R->Low = (UInt32)((UInt32)R->Low << 8);
 }
 
-static void RangeEnc_Encode(CPpmd7z_RangeEnc *p, UInt32 start, UInt32 size, UInt32 total)
+#define RC_NORM_BASE(p) if (R->Range < kTopValue) { R->Range <<= 8;  Ppmd7z_RangeEnc_ShiftLow(p);
+#define RC_NORM_1(p)    RC_NORM_BASE(p) }
+#define RC_NORM(p)      RC_NORM_BASE(p)  RC_NORM_BASE(p) }}
+
+// we must use only one type of Normalization from two: LOCAL or REMOTE
+#define RC_NORM_LOCAL(p)    // RC_NORM(p)
+#define RC_NORM_REMOTE(p)   RC_NORM(p)
+
+/*
+#define Ppmd7z_RangeEnc_Encode(p, start, _size_) \
+  { UInt32 size = _size_; \
+    R->Low += start * R->Range; \
+    R->Range *= size; \
+    RC_NORM_LOCAL(p); }
+*/
+
+Z7_FORCE_INLINE
+// Z7_NO_INLINE
+static void Ppmd7z_RangeEnc_Encode(CPpmd7 *p, UInt32 start, UInt32 size)
 {
-  p->Low += start * (p->Range /= total);
-  p->Range *= size;
-  while (p->Range < kTopValue)
-  {
-    p->Range <<= 8;
-    RangeEnc_ShiftLow(p);
-  }
+  R->Low += start * R->Range;
+  R->Range *= size;
+  RC_NORM_LOCAL(p)
 }
 
-static void RangeEnc_EncodeBit_0(CPpmd7z_RangeEnc *p, UInt32 size0)
-{
-  p->Range = (p->Range >> 14) * size0;
-  while (p->Range < kTopValue)
-  {
-    p->Range <<= 8;
-    RangeEnc_ShiftLow(p);
-  }
-}
-
-static void RangeEnc_EncodeBit_1(CPpmd7z_RangeEnc *p, UInt32 size0)
-{
-  UInt32 newBound = (p->Range >> 14) * size0;
-  p->Low += newBound;
-  p->Range -= newBound;
-  while (p->Range < kTopValue)
-  {
-    p->Range <<= 8;
-    RangeEnc_ShiftLow(p);
-  }
-}
-
-void Ppmd7z_RangeEnc_FlushData(CPpmd7z_RangeEnc *p)
+void Ppmd7z_Flush_RangeEnc(CPpmd7 *p)
 {
   unsigned i;
   for (i = 0; i < 5; i++)
-    RangeEnc_ShiftLow(p);
+    Ppmd7z_RangeEnc_ShiftLow(p);
 }
 
 
-#define MASK(sym) ((signed char *)charMask)[sym]
 
-void Ppmd7_EncodeSymbol(CPpmd7 *p, CPpmd7z_RangeEnc *rc, int symbol)
+#define RC_Encode(start, size)  Ppmd7z_RangeEnc_Encode(p, start, size);
+#define RC_EncodeFinal(start, size)  RC_Encode(start, size) RC_NORM_REMOTE(p)
+
+#define CTX(ref) ((CPpmd7_Context *)Ppmd7_GetContext(p, ref))
+#define SUFFIX(ctx) CTX((ctx)->Suffix)
+// typedef CPpmd7_Context * CTX_PTR;
+#define SUCCESSOR(p) Ppmd_GET_SUCCESSOR(p)
+
+void Ppmd7_UpdateModel(CPpmd7 *p);
+
+#define MASK(sym) ((unsigned char *)charMask)[sym]
+
+Z7_FORCE_INLINE
+static
+void Ppmd7z_EncodeSymbol(CPpmd7 *p, int symbol)
 {
   size_t charMask[256 / sizeof(size_t)];
+  
   if (p->MinContext->NumStats != 1)
   {
     CPpmd_State *s = Ppmd7_GetStats(p, p->MinContext);
     UInt32 sum;
     unsigned i;
+   
+
+    
+    
+    R->Range /= p->MinContext->Union2.SummFreq;
+    
     if (s->Symbol == symbol)
     {
-      RangeEnc_Encode(rc, 0, s->Freq, p->MinContext->SummFreq);
+      // R->Range /= p->MinContext->Union2.SummFreq;
+      RC_EncodeFinal(0, s->Freq)
       p->FoundState = s;
       Ppmd7_Update1_0(p);
       return;
     }
     p->PrevSuccess = 0;
     sum = s->Freq;
-    i = p->MinContext->NumStats - 1;
+    i = (unsigned)p->MinContext->NumStats - 1;
     do
     {
       if ((++s)->Symbol == symbol)
       {
-        RangeEnc_Encode(rc, sum, s->Freq, p->MinContext->SummFreq);
+        // R->Range /= p->MinContext->Union2.SummFreq;
+        RC_EncodeFinal(sum, s->Freq)
         p->FoundState = s;
         Ppmd7_Update1(p);
         return;
@@ -104,82 +125,214 @@ void Ppmd7_EncodeSymbol(CPpmd7 *p, CPpmd7z_RangeEnc *rc, int symbol)
       sum += s->Freq;
     }
     while (--i);
+
+    // R->Range /= p->MinContext->Union2.SummFreq;
+    RC_Encode(sum, p->MinContext->Union2.SummFreq - sum)
     
-    p->HiBitsFlag = p->HB2Flag[p->FoundState->Symbol];
-    PPMD_SetAllBitsIn256Bytes(charMask);
-    MASK(s->Symbol) = 0;
-    i = p->MinContext->NumStats - 1;
-    do { MASK((--s)->Symbol) = 0; } while (--i);
-    RangeEnc_Encode(rc, sum, p->MinContext->SummFreq - sum, p->MinContext->SummFreq);
+    p->HiBitsFlag = PPMD7_HiBitsFlag_3(p->FoundState->Symbol);
+    PPMD_SetAllBitsIn256Bytes(charMask)
+    // MASK(s->Symbol) = 0;
+    // i = p->MinContext->NumStats - 1;
+    // do { MASK((--s)->Symbol) = 0; } while (--i);
+    {
+      CPpmd_State *s2 = Ppmd7_GetStats(p, p->MinContext);
+      MASK(s->Symbol) = 0;
+      do
+      {
+        unsigned sym0 = s2[0].Symbol;
+        unsigned sym1 = s2[1].Symbol;
+        s2 += 2;
+        MASK(sym0) = 0;
+        MASK(sym1) = 0;
+      }
+      while (s2 < s);
+    }
   }
   else
   {
     UInt16 *prob = Ppmd7_GetBinSumm(p);
     CPpmd_State *s = Ppmd7Context_OneState(p->MinContext);
+    UInt32 pr = *prob;
+    const UInt32 bound = (R->Range >> 14) * pr;
+    pr = PPMD_UPDATE_PROB_1(pr);
     if (s->Symbol == symbol)
     {
-      RangeEnc_EncodeBit_0(rc, *prob);
-      *prob = (UInt16)PPMD_UPDATE_PROB_0(*prob);
-      p->FoundState = s;
-      Ppmd7_UpdateBin(p);
+      *prob = (UInt16)(pr + (1 << PPMD_INT_BITS));
+      // RangeEnc_EncodeBit_0(p, bound);
+      R->Range = bound;
+      RC_NORM_1(p)
+      
+      // p->FoundState = s;
+      // Ppmd7_UpdateBin(p);
+      {
+        const unsigned freq = s->Freq;
+        CPpmd7_Context *c = CTX(SUCCESSOR(s));
+        p->FoundState = s;
+        p->PrevSuccess = 1;
+        p->RunLength++;
+        s->Freq = (Byte)(freq + (freq < 128));
+        // NextContext(p);
+        if (p->OrderFall == 0 && (const Byte *)c > p->Text)
+          p->MaxContext = p->MinContext = c;
+        else
+          Ppmd7_UpdateModel(p);
+      }
       return;
     }
-    else
-    {
-      RangeEnc_EncodeBit_1(rc, *prob);
-      *prob = (UInt16)PPMD_UPDATE_PROB_1(*prob);
-      p->InitEsc = PPMD7_kExpEscape[*prob >> 10];
-      PPMD_SetAllBitsIn256Bytes(charMask);
-      MASK(s->Symbol) = 0;
-      p->PrevSuccess = 0;
-    }
+
+    *prob = (UInt16)pr;
+    p->InitEsc = p->ExpEscape[pr >> 10];
+    // RangeEnc_EncodeBit_1(p, bound);
+    R->Low += bound;
+    R->Range -= bound;
+    RC_NORM_LOCAL(p)
+    
+    PPMD_SetAllBitsIn256Bytes(charMask)
+    MASK(s->Symbol) = 0;
+    p->PrevSuccess = 0;
   }
+
   for (;;)
   {
-    UInt32 escFreq;
     CPpmd_See *see;
     CPpmd_State *s;
-    UInt32 sum;
-    unsigned i, numMasked = p->MinContext->NumStats;
+    UInt32 sum, escFreq;
+    CPpmd7_Context *mc;
+    unsigned i, numMasked;
+    
+    RC_NORM_REMOTE(p)
+
+    mc = p->MinContext;
+    numMasked = mc->NumStats;
+
     do
     {
       p->OrderFall++;
-      if (!p->MinContext->Suffix)
+      if (!mc->Suffix)
         return; /* EndMarker (symbol = -1) */
-      p->MinContext = Ppmd7_GetContext(p, p->MinContext->Suffix);
+      mc = Ppmd7_GetContext(p, mc->Suffix);
+      i = mc->NumStats;
     }
-    while (p->MinContext->NumStats == numMasked);
+    while (i == numMasked);
+
+    p->MinContext = mc;
     
-    see = Ppmd7_MakeEscFreq(p, numMasked, &escFreq);
-    s = Ppmd7_GetStats(p, p->MinContext);
+    // see = Ppmd7_MakeEscFreq(p, numMasked, &escFreq);
+    {
+      if (i != 256)
+      {
+        unsigned nonMasked = i - numMasked;
+        see = p->See[(unsigned)p->NS2Indx[(size_t)nonMasked - 1]]
+            + p->HiBitsFlag
+            + (nonMasked < (unsigned)SUFFIX(mc)->NumStats - i)
+            + 2 * (unsigned)(mc->Union2.SummFreq < 11 * i)
+            + 4 * (unsigned)(numMasked > nonMasked);
+        {
+          // if (see->Summ) field is larger than 16-bit, we need only low 16 bits of Summ
+          unsigned summ = (UInt16)see->Summ; // & 0xFFFF
+          unsigned r = (summ >> see->Shift);
+          see->Summ = (UInt16)(summ - r);
+          escFreq = r + (r == 0);
+        }
+      }
+      else
+      {
+        see = &p->DummySee;
+        escFreq = 1;
+      }
+    }
+
+    s = Ppmd7_GetStats(p, mc);
     sum = 0;
-    i = p->MinContext->NumStats;
+    // i = mc->NumStats;
+
     do
     {
-      int cur = s->Symbol;
-      if (cur == symbol)
+      const unsigned cur = s->Symbol;
+      if ((int)cur == symbol)
       {
-        UInt32 low = sum;
-        CPpmd_State *s1 = s;
-        do
+        const UInt32 low = sum;
+        const UInt32 freq = s->Freq;
+        unsigned num2;
+
+        Ppmd_See_UPDATE(see)
+        p->FoundState = s;
+        sum += escFreq;
+
+        num2 = i / 2;
+        i &= 1;
+        sum += freq & (0 - (UInt32)i);
+        if (num2 != 0)
         {
-          sum += (s->Freq & (int)(MASK(s->Symbol)));
-          s++;
+          s += i;
+          for (;;)
+          {
+            unsigned sym0 = s[0].Symbol;
+            unsigned sym1 = s[1].Symbol;
+            s += 2;
+            sum += (s[-2].Freq & (unsigned)(MASK(sym0)));
+            sum += (s[-1].Freq & (unsigned)(MASK(sym1)));
+            if (--num2 == 0)
+              break;
+          }
         }
-        while (--i);
-        RangeEnc_Encode(rc, low, s1->Freq, sum + escFreq);
-        Ppmd_See_Update(see);
-        p->FoundState = s1;
+
+        
+        R->Range /= sum;
+        RC_EncodeFinal(low, freq)
         Ppmd7_Update2(p);
         return;
       }
-      sum += (s->Freq & (int)(MASK(cur)));
-      MASK(cur) = 0;
+      sum += (s->Freq & (unsigned)(MASK(cur)));
       s++;
     }
     while (--i);
     
-    RangeEnc_Encode(rc, sum, escFreq, sum + escFreq);
-    see->Summ = (UInt16)(see->Summ + sum + escFreq);
+    {
+      const UInt32 total = sum + escFreq;
+      see->Summ = (UInt16)(see->Summ + total);
+
+      R->Range /= total;
+      RC_Encode(sum, escFreq)
+    }
+
+    {
+      const CPpmd_State *s2 = Ppmd7_GetStats(p, p->MinContext);
+      s--;
+      MASK(s->Symbol) = 0;
+      do
+      {
+        const unsigned sym0 = s2[0].Symbol;
+        const unsigned sym1 = s2[1].Symbol;
+        s2 += 2;
+        MASK(sym0) = 0;
+        MASK(sym1) = 0;
+      }
+      while (s2 < s);
+    }
   }
 }
+
+
+void Ppmd7z_EncodeSymbols(CPpmd7 *p, const Byte *buf, const Byte *lim)
+{
+  for (; buf < lim; buf++)
+  {
+    Ppmd7z_EncodeSymbol(p, *buf);
+  }
+}
+
+#undef kTopValue
+#undef WRITE_BYTE
+#undef RC_NORM_BASE
+#undef RC_NORM_1
+#undef RC_NORM
+#undef RC_NORM_LOCAL
+#undef RC_NORM_REMOTE
+#undef R
+#undef RC_Encode
+#undef RC_EncodeFinal
+#undef SUFFIX
+#undef CTX
+#undef SUCCESSOR
+#undef MASK

C/Ppmd7aDec.c

@@ -0,0 +1,295 @@
+/* Ppmd7aDec.c -- PPMd7a (PPMdH) Decoder
+2023-04-02 : Igor Pavlov : Public domain
+This code is based on:
+  PPMd var.H (2001): Dmitry Shkarin : Public domain
+  Carryless rangecoder (1999): Dmitry Subbotin : Public domain */
+
+#include "Precomp.h"
+
+#include "Ppmd7.h"
+
+#define kTop ((UInt32)1 << 24)
+#define kBot ((UInt32)1 << 15)
+
+#define READ_BYTE(p) IByteIn_Read((p)->Stream)
+
+BoolInt Ppmd7a_RangeDec_Init(CPpmd7_RangeDec *p)
+{
+  unsigned i;
+  p->Code = 0;
+  p->Range = 0xFFFFFFFF;
+  p->Low = 0;
+  
+  for (i = 0; i < 4; i++)
+    p->Code = (p->Code << 8) | READ_BYTE(p);
+  return (p->Code < 0xFFFFFFFF);
+}
+
+#define RC_NORM(p) \
+  while ((p->Low ^ (p->Low + p->Range)) < kTop \
+    || (p->Range < kBot && ((p->Range = (0 - p->Low) & (kBot - 1)), 1))) { \
+      p->Code = (p->Code << 8) | READ_BYTE(p); \
+      p->Range <<= 8; p->Low <<= 8; }
+
+// we must use only one type of Normalization from two: LOCAL or REMOTE
+#define RC_NORM_LOCAL(p)    // RC_NORM(p)
+#define RC_NORM_REMOTE(p)   RC_NORM(p)
+
+#define R (&p->rc.dec)
+
+Z7_FORCE_INLINE
+// Z7_NO_INLINE
+static void Ppmd7a_RD_Decode(CPpmd7 *p, UInt32 start, UInt32 size)
+{
+  start *= R->Range;
+  R->Low += start;
+  R->Code -= start;
+  R->Range *= size;
+  RC_NORM_LOCAL(R)
+}
+
+#define RC_Decode(start, size)  Ppmd7a_RD_Decode(p, start, size);
+#define RC_DecodeFinal(start, size)  RC_Decode(start, size)  RC_NORM_REMOTE(R)
+#define RC_GetThreshold(total)  (R->Code / (R->Range /= (total)))
+
+
+#define CTX(ref) ((CPpmd7_Context *)Ppmd7_GetContext(p, ref))
+typedef CPpmd7_Context * CTX_PTR;
+#define SUCCESSOR(p) Ppmd_GET_SUCCESSOR(p)
+void Ppmd7_UpdateModel(CPpmd7 *p);
+
+#define MASK(sym) ((unsigned char *)charMask)[sym]
+
+
+int Ppmd7a_DecodeSymbol(CPpmd7 *p)
+{
+  size_t charMask[256 / sizeof(size_t)];
+
+  if (p->MinContext->NumStats != 1)
+  {
+    CPpmd_State *s = Ppmd7_GetStats(p, p->MinContext);
+    unsigned i;
+    UInt32 count, hiCnt;
+    const UInt32 summFreq = p->MinContext->Union2.SummFreq;
+
+    if (summFreq > R->Range)
+      return PPMD7_SYM_ERROR;
+
+    count = RC_GetThreshold(summFreq);
+    hiCnt = count;
+    
+    if ((Int32)(count -= s->Freq) < 0)
+    {
+      Byte sym;
+      RC_DecodeFinal(0, s->Freq)
+      p->FoundState = s;
+      sym = s->Symbol;
+      Ppmd7_Update1_0(p);
+      return sym;
+    }
+  
+    p->PrevSuccess = 0;
+    i = (unsigned)p->MinContext->NumStats - 1;
+    
+    do
+    {
+      if ((Int32)(count -= (++s)->Freq) < 0)
+      {
+        Byte sym;
+        RC_DecodeFinal((hiCnt - count) - s->Freq, s->Freq)
+        p->FoundState = s;
+        sym = s->Symbol;
+        Ppmd7_Update1(p);
+        return sym;
+      }
+    }
+    while (--i);
+    
+    if (hiCnt >= summFreq)
+      return PPMD7_SYM_ERROR;
+    
+    hiCnt -= count;
+    RC_Decode(hiCnt, summFreq - hiCnt)
+
+    p->HiBitsFlag = PPMD7_HiBitsFlag_3(p->FoundState->Symbol);
+    PPMD_SetAllBitsIn256Bytes(charMask)
+    // i = p->MinContext->NumStats - 1;
+    // do { MASK((--s)->Symbol) = 0; } while (--i);
+    {
+      CPpmd_State *s2 = Ppmd7_GetStats(p, p->MinContext);
+      MASK(s->Symbol) = 0;
+      do
+      {
+        unsigned sym0 = s2[0].Symbol;
+        unsigned sym1 = s2[1].Symbol;
+        s2 += 2;
+        MASK(sym0) = 0;
+        MASK(sym1) = 0;
+      }
+      while (s2 < s);
+    }
+  }
+  else
+  {
+    CPpmd_State *s = Ppmd7Context_OneState(p->MinContext);
+    UInt16 *prob = Ppmd7_GetBinSumm(p);
+    UInt32 pr = *prob;
+    UInt32 size0 = (R->Range >> 14) * pr;
+    pr = PPMD_UPDATE_PROB_1(pr);
+
+    if (R->Code < size0)
+    {
+      Byte sym;
+      *prob = (UInt16)(pr + (1 << PPMD_INT_BITS));
+      
+      // RangeDec_DecodeBit0(size0);
+      R->Range = size0;
+      RC_NORM(R)
+      
+        
+        
+      // sym = (p->FoundState = Ppmd7Context_OneState(p->MinContext))->Symbol;
+      // Ppmd7_UpdateBin(p);
+      {
+        unsigned freq = s->Freq;
+        CTX_PTR c = CTX(SUCCESSOR(s));
+        sym = s->Symbol;
+        p->FoundState = s;
+        p->PrevSuccess = 1;
+        p->RunLength++;
+        s->Freq = (Byte)(freq + (freq < 128));
+        // NextContext(p);
+        if (p->OrderFall == 0 && (const Byte *)c > p->Text)
+          p->MaxContext = p->MinContext = c;
+        else
+          Ppmd7_UpdateModel(p);
+      }
+      return sym;
+    }
+
+    *prob = (UInt16)pr;
+    p->InitEsc = p->ExpEscape[pr >> 10];
+
+    // RangeDec_DecodeBit1(size0);
+    R->Low += size0;
+    R->Code -= size0;
+    R->Range = (R->Range & ~((UInt32)PPMD_BIN_SCALE - 1)) - size0;
+    RC_NORM_LOCAL(R)
+    
+    PPMD_SetAllBitsIn256Bytes(charMask)
+    MASK(Ppmd7Context_OneState(p->MinContext)->Symbol) = 0;
+    p->PrevSuccess = 0;
+  }
+
+  for (;;)
+  {
+    CPpmd_State *s, *s2;
+    UInt32 freqSum, count, hiCnt;
+
+    CPpmd_See *see;
+    CPpmd7_Context *mc;
+    unsigned numMasked;
+    RC_NORM_REMOTE(R)
+    mc = p->MinContext;
+    numMasked = mc->NumStats;
+
+    do
+    {
+      p->OrderFall++;
+      if (!mc->Suffix)
+        return PPMD7_SYM_END;
+      mc = Ppmd7_GetContext(p, mc->Suffix);
+    }
+    while (mc->NumStats == numMasked);
+    
+    s = Ppmd7_GetStats(p, mc);
+
+    {
+      unsigned num = mc->NumStats;
+      unsigned num2 = num / 2;
+      
+      num &= 1;
+      hiCnt = (s->Freq & (unsigned)(MASK(s->Symbol))) & (0 - (UInt32)num);
+      s += num;
+      p->MinContext = mc;
+
+      do
+      {
+        unsigned sym0 = s[0].Symbol;
+        unsigned sym1 = s[1].Symbol;
+        s += 2;
+        hiCnt += (s[-2].Freq & (unsigned)(MASK(sym0)));
+        hiCnt += (s[-1].Freq & (unsigned)(MASK(sym1)));
+      }
+      while (--num2);
+    }
+
+    see = Ppmd7_MakeEscFreq(p, numMasked, &freqSum);
+    freqSum += hiCnt;
+
+    if (freqSum > R->Range)
+      return PPMD7_SYM_ERROR;
+
+    count = RC_GetThreshold(freqSum);
+    
+    if (count < hiCnt)
+    {
+      Byte sym;
+
+      s = Ppmd7_GetStats(p, p->MinContext);
+      hiCnt = count;
+      // count -= s->Freq & (unsigned)(MASK(s->Symbol));
+      // if ((Int32)count >= 0)
+      {
+        for (;;)
+        {
+          count -= s->Freq & (unsigned)(MASK((s)->Symbol)); s++; if ((Int32)count < 0) break;
+          // count -= s->Freq & (unsigned)(MASK((s)->Symbol)); s++; if ((Int32)count < 0) break;
+        }
+      }
+      s--;
+      RC_DecodeFinal((hiCnt - count) - s->Freq, s->Freq)
+
+      // new (see->Summ) value can overflow over 16-bits in some rare cases
+      Ppmd_See_UPDATE(see)
+      p->FoundState = s;
+      sym = s->Symbol;
+      Ppmd7_Update2(p);
+      return sym;
+    }
+
+    if (count >= freqSum)
+      return PPMD7_SYM_ERROR;
+    
+    RC_Decode(hiCnt, freqSum - hiCnt)
+
+    // We increase (see->Summ) for sum of Freqs of all non_Masked symbols.
+    // new (see->Summ) value can overflow over 16-bits in some rare cases
+    see->Summ = (UInt16)(see->Summ + freqSum);
+
+    s = Ppmd7_GetStats(p, p->MinContext);
+    s2 = s + p->MinContext->NumStats;
+    do
+    {
+      MASK(s->Symbol) = 0;
+      s++;
+    }
+    while (s != s2);
+  }
+}
+
+#undef kTop
+#undef kBot
+#undef READ_BYTE
+#undef RC_NORM_BASE
+#undef RC_NORM_1
+#undef RC_NORM
+#undef RC_NORM_LOCAL
+#undef RC_NORM_REMOTE
+#undef R
+#undef RC_Decode
+#undef RC_DecodeFinal
+#undef RC_GetThreshold
+#undef CTX
+#undef SUCCESSOR
+#undef MASK

C/Ppmd8.h

@@ -1,11 +1,11 @@
-/* Ppmd8.h -- PPMdI codec
-2010-03-12 : Igor Pavlov : Public domain
+/* Ppmd8.h -- Ppmd8 (PPMdI) compression codec
+2023-04-02 : Igor Pavlov : Public domain
 This code is based on:
   PPMd var.I (2002): Dmitry Shkarin : Public domain
   Carryless rangecoder (1999): Dmitry Subbotin : Public domain */
 
-#ifndef __PPMD8_H
-#define __PPMD8_H
+#ifndef ZIP7_INC_PPMD8_H
+#define ZIP7_INC_PPMD8_H
 
 #include "Ppmd.h"
 
@@ -14,29 +14,44 @@ EXTERN_C_BEGIN
 #define PPMD8_MIN_ORDER 2
 #define PPMD8_MAX_ORDER 16
 
+
+
+
 struct CPpmd8_Context_;
 
-typedef
-  #ifdef PPMD_32BIT
-    struct CPpmd8_Context_ *
-  #else
-    UInt32
-  #endif
-  CPpmd8_Context_Ref;
+typedef Ppmd_Ref_Type(struct CPpmd8_Context_) CPpmd8_Context_Ref;
+
+// MY_CPU_pragma_pack_push_1
 
 typedef struct CPpmd8_Context_
 {
   Byte NumStats;
   Byte Flags;
-  UInt16 SummFreq;
-  CPpmd_State_Ref Stats;
+  
+  union
+  {
+    UInt16 SummFreq;
+    CPpmd_State2 State2;
+  } Union2;
+  
+  union
+  {
+    CPpmd_State_Ref Stats;
+    CPpmd_State4 State4;
+  } Union4;
+
   CPpmd8_Context_Ref Suffix;
 } CPpmd8_Context;
 
-#define Ppmd8Context_OneState(p) ((CPpmd_State *)&(p)->SummFreq)
+// MY_CPU_pragma_pop
 
-/* There is some bug in FREEZE mode (including original code,
-so we disable FREEZE mode support */
+#define Ppmd8Context_OneState(p) ((CPpmd_State *)&(p)->Union2)
+
+/* PPMdI code rev.2 contains the fix over PPMdI code rev.1.
+   But the code PPMdI.2 is not compatible with PPMdI.1 for some files compressed
+   in FREEZE mode. So we disable FREEZE mode support. */
+
+// #define PPMD8_FREEZE_SUPPORT
 
 enum
 {
@@ -45,87 +60,121 @@ enum
   #ifdef PPMD8_FREEZE_SUPPORT
   , PPMD8_RESTORE_METHOD_FREEZE
   #endif
+  , PPMD8_RESTORE_METHOD_UNSUPPPORTED
 };
 
+
+
+
+
+
+
+
 typedef struct
 {
   CPpmd8_Context *MinContext, *MaxContext;
   CPpmd_State *FoundState;
-  unsigned OrderFall, InitEsc, PrevSuccess, MaxOrder;
+  unsigned OrderFall, InitEsc, PrevSuccess, MaxOrder, RestoreMethod;
   Int32 RunLength, InitRL; /* must be 32-bit at least */
 
   UInt32 Size;
   UInt32 GlueCount;
-  Byte *Base, *LoUnit, *HiUnit, *Text, *UnitsStart;
   UInt32 AlignOffset;
-  unsigned RestoreMethod;
+  Byte *Base, *LoUnit, *HiUnit, *Text, *UnitsStart;
 
-  /* Range Coder */
   UInt32 Range;
   UInt32 Code;
   UInt32 Low;
   union
   {
-    IByteIn *In;
-    IByteOut *Out;
+    IByteInPtr In;
+    IByteOutPtr Out;
   } Stream;
 
-  Byte Indx2Units[PPMD_NUM_INDEXES];
+  Byte Indx2Units[PPMD_NUM_INDEXES + 2]; // +2 for alignment
   Byte Units2Indx[128];
   CPpmd_Void_Ref FreeList[PPMD_NUM_INDEXES];
   UInt32 Stamps[PPMD_NUM_INDEXES];
-
   Byte NS2BSIndx[256], NS2Indx[260];
+  Byte ExpEscape[16];
   CPpmd_See DummySee, See[24][32];
   UInt16 BinSumm[25][64];
+
 } CPpmd8;
 
+
 void Ppmd8_Construct(CPpmd8 *p);
-Bool Ppmd8_Alloc(CPpmd8 *p, UInt32 size, ISzAlloc *alloc);
-void Ppmd8_Free(CPpmd8 *p, ISzAlloc *alloc);
+BoolInt Ppmd8_Alloc(CPpmd8 *p, UInt32 size, ISzAllocPtr alloc);
+void Ppmd8_Free(CPpmd8 *p, ISzAllocPtr alloc);
 void Ppmd8_Init(CPpmd8 *p, unsigned maxOrder, unsigned restoreMethod);
 #define Ppmd8_WasAllocated(p) ((p)->Base != NULL)
 
 
 /* ---------- Internal Functions ---------- */
 
-extern const Byte PPMD8_kExpEscape[16];
-
-#ifdef PPMD_32BIT
-  #define Ppmd8_GetPtr(p, ptr) (ptr)
-  #define Ppmd8_GetContext(p, ptr) (ptr)
-  #define Ppmd8_GetStats(p, ctx) ((ctx)->Stats)
-#else
-  #define Ppmd8_GetPtr(p, offs) ((void *)((p)->Base + (offs)))
-  #define Ppmd8_GetContext(p, offs) ((CPpmd8_Context *)Ppmd8_GetPtr((p), (offs)))
-  #define Ppmd8_GetStats(p, ctx) ((CPpmd_State *)Ppmd8_GetPtr((p), ((ctx)->Stats)))
-#endif
+#define Ppmd8_GetPtr(p, ptr)     Ppmd_GetPtr(p, ptr)
+#define Ppmd8_GetContext(p, ptr) Ppmd_GetPtr_Type(p, ptr, CPpmd8_Context)
+#define Ppmd8_GetStats(p, ctx)   Ppmd_GetPtr_Type(p, (ctx)->Union4.Stats, CPpmd_State)
 
 void Ppmd8_Update1(CPpmd8 *p);
 void Ppmd8_Update1_0(CPpmd8 *p);
 void Ppmd8_Update2(CPpmd8 *p);
-void Ppmd8_UpdateBin(CPpmd8 *p);
+
+
+
+
+
 
 #define Ppmd8_GetBinSumm(p) \
-    &p->BinSumm[p->NS2Indx[Ppmd8Context_OneState(p->MinContext)->Freq - 1]][ \
-    p->NS2BSIndx[Ppmd8_GetContext(p, p->MinContext->Suffix)->NumStats] + \
-    p->PrevSuccess + p->MinContext->Flags + ((p->RunLength >> 26) & 0x20)]
+    &p->BinSumm[p->NS2Indx[(size_t)Ppmd8Context_OneState(p->MinContext)->Freq - 1]] \
+    [ p->PrevSuccess + ((p->RunLength >> 26) & 0x20) \
+    + p->NS2BSIndx[Ppmd8_GetContext(p, p->MinContext->Suffix)->NumStats] + \
+    + p->MinContext->Flags ]
+
 
 CPpmd_See *Ppmd8_MakeEscFreq(CPpmd8 *p, unsigned numMasked, UInt32 *scale);
 
 
+/* 20.01: the original PPMdI encoder and decoder probably could work incorrectly in some rare cases,
+   where the original PPMdI code can give "Divide by Zero" operation.
+   We use the following fix to allow correct working of encoder and decoder in any cases.
+   We correct (Escape_Freq) and (_sum_), if (_sum_) is larger than p->Range) */
+#define PPMD8_CORRECT_SUM_RANGE(p, _sum_) if (_sum_ > p->Range /* /1 */) _sum_ = p->Range;
+
+
 /* ---------- Decode ---------- */
 
-Bool Ppmd8_RangeDec_Init(CPpmd8 *p);
+#define PPMD8_SYM_END    (-1)
+#define PPMD8_SYM_ERROR  (-2)
+
+/*
+You must set (CPpmd8::Stream.In) before Ppmd8_RangeDec_Init()
+
+Ppmd8_DecodeSymbol()
+out:
+  >= 0 : decoded byte
+    -1 : PPMD8_SYM_END   : End of payload marker
+    -2 : PPMD8_SYM_ERROR : Data error
+*/
+
+
+BoolInt Ppmd8_Init_RangeDec(CPpmd8 *p);
 #define Ppmd8_RangeDec_IsFinishedOK(p) ((p)->Code == 0)
-int Ppmd8_DecodeSymbol(CPpmd8 *p); /* returns: -1 as EndMarker, -2 as DataError */
+int Ppmd8_DecodeSymbol(CPpmd8 *p);
+
+
+
+
+
+
 
 
 /* ---------- Encode ---------- */
 
-#define Ppmd8_RangeEnc_Init(p) { (p)->Low = 0; (p)->Range = 0xFFFFFFFF; }
-void Ppmd8_RangeEnc_FlushData(CPpmd8 *p);
-void Ppmd8_EncodeSymbol(CPpmd8 *p, int symbol); /* symbol = -1 means EndMarker */
+#define Ppmd8_Init_RangeEnc(p) { (p)->Low = 0; (p)->Range = 0xFFFFFFFF; }
+void Ppmd8_Flush_RangeEnc(CPpmd8 *p);
+void Ppmd8_EncodeSymbol(CPpmd8 *p, int symbol);
+
 
 EXTERN_C_END
  

C/Ppmd8Dec.c

@@ -1,155 +1,295 @@
-/* Ppmd8Dec.c -- PPMdI Decoder
-2010-03-12 : Igor Pavlov : Public domain
+/* Ppmd8Dec.c -- Ppmd8 (PPMdI) Decoder
+2023-04-02 : Igor Pavlov : Public domain
 This code is based on:
   PPMd var.I (2002): Dmitry Shkarin : Public domain
   Carryless rangecoder (1999): Dmitry Subbotin : Public domain */
 
+#include "Precomp.h"
+
 #include "Ppmd8.h"
 
-#define kTop (1 << 24)
-#define kBot (1 << 15)
+#define kTop ((UInt32)1 << 24)
+#define kBot ((UInt32)1 << 15)
 
-Bool Ppmd8_RangeDec_Init(CPpmd8 *p)
+#define READ_BYTE(p) IByteIn_Read((p)->Stream.In)
+
+BoolInt Ppmd8_Init_RangeDec(CPpmd8 *p)
 {
   unsigned i;
-  p->Low = 0;
-  p->Range = 0xFFFFFFFF;
   p->Code = 0;
+  p->Range = 0xFFFFFFFF;
+  p->Low = 0;
+  
   for (i = 0; i < 4; i++)
-    p->Code = (p->Code << 8) | p->Stream.In->Read(p->Stream.In);
+    p->Code = (p->Code << 8) | READ_BYTE(p);
   return (p->Code < 0xFFFFFFFF);
 }
 
-static UInt32 RangeDec_GetThreshold(CPpmd8 *p, UInt32 total)
+#define RC_NORM(p) \
+  while ((p->Low ^ (p->Low + p->Range)) < kTop \
+    || (p->Range < kBot && ((p->Range = (0 - p->Low) & (kBot - 1)), 1))) { \
+      p->Code = (p->Code << 8) | READ_BYTE(p); \
+      p->Range <<= 8; p->Low <<= 8; }
+
+// we must use only one type of Normalization from two: LOCAL or REMOTE
+#define RC_NORM_LOCAL(p)    // RC_NORM(p)
+#define RC_NORM_REMOTE(p)   RC_NORM(p)
+
+#define R p
+
+Z7_FORCE_INLINE
+// Z7_NO_INLINE
+static void Ppmd8_RD_Decode(CPpmd8 *p, UInt32 start, UInt32 size)
 {
-  return p->Code / (p->Range /= total);
+  start *= R->Range;
+  R->Low += start;
+  R->Code -= start;
+  R->Range *= size;
+  RC_NORM_LOCAL(R)
 }
 
-static void RangeDec_Decode(CPpmd8 *p, UInt32 start, UInt32 size)
-{
-  start *= p->Range;
-  p->Low += start;
-  p->Code -= start;
-  p->Range *= size;
+#define RC_Decode(start, size)  Ppmd8_RD_Decode(p, start, size);
+#define RC_DecodeFinal(start, size)  RC_Decode(start, size)  RC_NORM_REMOTE(R)
+#define RC_GetThreshold(total)  (R->Code / (R->Range /= (total)))
 
-  while ((p->Low ^ (p->Low + p->Range)) < kTop ||
-      p->Range < kBot && ((p->Range = (0 - p->Low) & (kBot - 1)), 1))
-  {
-    p->Code = (p->Code << 8) | p->Stream.In->Read(p->Stream.In);
-    p->Range <<= 8;
-    p->Low <<= 8;
-  }
-}
 
-#define MASK(sym) ((signed char *)charMask)[sym]
+#define CTX(ref) ((CPpmd8_Context *)Ppmd8_GetContext(p, ref))
+// typedef CPpmd8_Context * CTX_PTR;
+#define SUCCESSOR(p) Ppmd_GET_SUCCESSOR(p)
+void Ppmd8_UpdateModel(CPpmd8 *p);
+
+#define MASK(sym) ((unsigned char *)charMask)[sym]
+
 
 int Ppmd8_DecodeSymbol(CPpmd8 *p)
 {
   size_t charMask[256 / sizeof(size_t)];
+
   if (p->MinContext->NumStats != 0)
   {
     CPpmd_State *s = Ppmd8_GetStats(p, p->MinContext);
     unsigned i;
     UInt32 count, hiCnt;
-    if ((count = RangeDec_GetThreshold(p, p->MinContext->SummFreq)) < (hiCnt = s->Freq))
+    UInt32 summFreq = p->MinContext->Union2.SummFreq;
+
+    PPMD8_CORRECT_SUM_RANGE(p, summFreq)
+
+
+    count = RC_GetThreshold(summFreq);
+    hiCnt = count;
+    
+    if ((Int32)(count -= s->Freq) < 0)
     {
-      Byte symbol;
-      RangeDec_Decode(p, 0, s->Freq);
+      Byte sym;
+      RC_DecodeFinal(0, s->Freq)
       p->FoundState = s;
-      symbol = s->Symbol;
+      sym = s->Symbol;
       Ppmd8_Update1_0(p);
-      return symbol;
+      return sym;
     }
+    
     p->PrevSuccess = 0;
     i = p->MinContext->NumStats;
+    
     do
     {
-      if ((hiCnt += (++s)->Freq) > count)
+      if ((Int32)(count -= (++s)->Freq) < 0)
       {
-        Byte symbol;
-        RangeDec_Decode(p, hiCnt - s->Freq, s->Freq);
+        Byte sym;
+        RC_DecodeFinal((hiCnt - count) - s->Freq, s->Freq)
         p->FoundState = s;
-        symbol = s->Symbol;
+        sym = s->Symbol;
         Ppmd8_Update1(p);
-        return symbol;
+        return sym;
       }
     }
     while (--i);
-    if (count >= p->MinContext->SummFreq)
-      return -2;
-    RangeDec_Decode(p, hiCnt, p->MinContext->SummFreq - hiCnt);
-    PPMD_SetAllBitsIn256Bytes(charMask);
-    MASK(s->Symbol) = 0;
-    i = p->MinContext->NumStats;
-    do { MASK((--s)->Symbol) = 0; } while (--i);
+    
+    if (hiCnt >= summFreq)
+      return PPMD8_SYM_ERROR;
+
+    hiCnt -= count;
+    RC_Decode(hiCnt, summFreq - hiCnt)
+    
+    
+    PPMD_SetAllBitsIn256Bytes(charMask)
+    // i = p->MinContext->NumStats - 1;
+    // do { MASK((--s)->Symbol) = 0; } while (--i);
+    {
+      CPpmd_State *s2 = Ppmd8_GetStats(p, p->MinContext);
+      MASK(s->Symbol) = 0;
+      do
+      {
+        unsigned sym0 = s2[0].Symbol;
+        unsigned sym1 = s2[1].Symbol;
+        s2 += 2;
+        MASK(sym0) = 0;
+        MASK(sym1) = 0;
+      }
+      while (s2 < s);
+    }
   }
   else
   {
+    CPpmd_State *s = Ppmd8Context_OneState(p->MinContext);
     UInt16 *prob = Ppmd8_GetBinSumm(p);
-    if (((p->Code / (p->Range >>= 14)) < *prob))
+    UInt32 pr = *prob;
+    UInt32 size0 = (R->Range >> 14) * pr;
+    pr = PPMD_UPDATE_PROB_1(pr);
+    
+    if (R->Code < size0)
     {
-      Byte symbol;
-      RangeDec_Decode(p, 0, *prob);
-      *prob = (UInt16)PPMD_UPDATE_PROB_0(*prob);
-      symbol = (p->FoundState = Ppmd8Context_OneState(p->MinContext))->Symbol;
-      Ppmd8_UpdateBin(p);
-      return symbol;
+      Byte sym;
+      *prob = (UInt16)(pr + (1 << PPMD_INT_BITS));
+      
+      // RangeDec_DecodeBit0(size0);
+      R->Range = size0;
+      RC_NORM(R)
+      
+      
+        
+      // sym = (p->FoundState = Ppmd8Context_OneState(p->MinContext))->Symbol;
+      // Ppmd8_UpdateBin(p);
+      {
+        unsigned freq = s->Freq;
+        CPpmd8_Context *c = CTX(SUCCESSOR(s));
+        sym = s->Symbol;
+        p->FoundState = s;
+        p->PrevSuccess = 1;
+        p->RunLength++;
+        s->Freq = (Byte)(freq + (freq < 196));
+        // NextContext(p);
+        if (p->OrderFall == 0 && (const Byte *)c >= p->UnitsStart)
+          p->MaxContext = p->MinContext = c;
+        else
+          Ppmd8_UpdateModel(p);
+      }
+      return sym;
     }
-    RangeDec_Decode(p, *prob, (1 << 14) - *prob);
-    *prob = (UInt16)PPMD_UPDATE_PROB_1(*prob);
-    p->InitEsc = PPMD8_kExpEscape[*prob >> 10];
-    PPMD_SetAllBitsIn256Bytes(charMask);
+    
+    *prob = (UInt16)pr;
+    p->InitEsc = p->ExpEscape[pr >> 10];
+    
+    // RangeDec_DecodeBit1(rc2, size0);
+    R->Low += size0;
+    R->Code -= size0;
+    R->Range = (R->Range & ~((UInt32)PPMD_BIN_SCALE - 1)) - size0;
+    RC_NORM_LOCAL(R)
+    
+    PPMD_SetAllBitsIn256Bytes(charMask)
     MASK(Ppmd8Context_OneState(p->MinContext)->Symbol) = 0;
     p->PrevSuccess = 0;
   }
+  
   for (;;)
   {
-    CPpmd_State *ps[256], *s;
+    CPpmd_State *s, *s2;
     UInt32 freqSum, count, hiCnt;
+    UInt32 freqSum2;
     CPpmd_See *see;
-    unsigned i, num, numMasked = p->MinContext->NumStats;
+    CPpmd8_Context *mc;
+    unsigned numMasked;
+    RC_NORM_REMOTE(R)
+    mc = p->MinContext;
+    numMasked = mc->NumStats;
+    
     do
     {
       p->OrderFall++;
-      if (!p->MinContext->Suffix)
-        return -1;
-      p->MinContext = Ppmd8_GetContext(p, p->MinContext->Suffix);
+      if (!mc->Suffix)
+        return PPMD8_SYM_END;
+      mc = Ppmd8_GetContext(p, mc->Suffix);
     }
-    while (p->MinContext->NumStats == numMasked);
-    hiCnt = 0;
-    s = Ppmd8_GetStats(p, p->MinContext);
-    i = 0;
-    num = p->MinContext->NumStats - numMasked;
-    do
+    while (mc->NumStats == numMasked);
+    
+    s = Ppmd8_GetStats(p, mc);
+
     {
-      int k = (int)(MASK(s->Symbol));
-      hiCnt += (s->Freq & k);
-      ps[i] = s++;
-      i -= k;
+      unsigned num = (unsigned)mc->NumStats + 1;
+      unsigned num2 = num / 2;
+
+      num &= 1;
+      hiCnt = (s->Freq & (unsigned)(MASK(s->Symbol))) & (0 - (UInt32)num);
+      s += num;
+      p->MinContext = mc;
+
+      do
+      {
+        unsigned sym0 = s[0].Symbol;
+        unsigned sym1 = s[1].Symbol;
+        s += 2;
+        hiCnt += (s[-2].Freq & (unsigned)(MASK(sym0)));
+        hiCnt += (s[-1].Freq & (unsigned)(MASK(sym1)));
+      }
+      while (--num2);
     }
-    while (i != num);
     
     see = Ppmd8_MakeEscFreq(p, numMasked, &freqSum);
     freqSum += hiCnt;
-    count = RangeDec_GetThreshold(p, freqSum);
+    freqSum2 = freqSum;
+    PPMD8_CORRECT_SUM_RANGE(R, freqSum2)
+
+
+    count = RC_GetThreshold(freqSum2);
     
     if (count < hiCnt)
     {
-      Byte symbol;
-      CPpmd_State **pps = ps;
-      for (hiCnt = 0; (hiCnt += (*pps)->Freq) <= count; pps++);
-      s = *pps;
-      RangeDec_Decode(p, hiCnt - s->Freq, s->Freq);
-      Ppmd_See_Update(see);
+      Byte sym;
+      // Ppmd_See_UPDATE(see) // new (see->Summ) value can overflow over 16-bits in some rare cases
+      s = Ppmd8_GetStats(p, p->MinContext);
+      hiCnt = count;
+
+      
+      {
+        for (;;)
+        {
+          count -= s->Freq & (unsigned)(MASK((s)->Symbol)); s++; if ((Int32)count < 0) break;
+          // count -= s->Freq & (unsigned)(MASK((s)->Symbol)); s++; if ((Int32)count < 0) break;
+        }
+      }
+      s--;
+      RC_DecodeFinal((hiCnt - count) - s->Freq, s->Freq)
+
+      // new (see->Summ) value can overflow over 16-bits in some rare cases
+      Ppmd_See_UPDATE(see)
       p->FoundState = s;
-      symbol = s->Symbol;
+      sym = s->Symbol;
       Ppmd8_Update2(p);
-      return symbol;
+      return sym;
     }
-    if (count >= freqSum)
-      return -2;
-    RangeDec_Decode(p, hiCnt, freqSum - hiCnt);
+
+    if (count >= freqSum2)
+      return PPMD8_SYM_ERROR;
+    
+    RC_Decode(hiCnt, freqSum2 - hiCnt)
+    
+    // We increase (see->Summ) for sum of Freqs of all non_Masked symbols.
+    // new (see->Summ) value can overflow over 16-bits in some rare cases
     see->Summ = (UInt16)(see->Summ + freqSum);
-    do { MASK(ps[--i]->Symbol) = 0; } while (i != 0);
+    
+    s = Ppmd8_GetStats(p, p->MinContext);
+    s2 = s + p->MinContext->NumStats + 1;
+    do
+    {
+      MASK(s->Symbol) = 0;
+      s++;
+    }
+    while (s != s2);
   }
 }
+
+#undef kTop
+#undef kBot
+#undef READ_BYTE
+#undef RC_NORM_BASE
+#undef RC_NORM_1
+#undef RC_NORM
+#undef RC_NORM_LOCAL
+#undef RC_NORM_REMOTE
+#undef R
+#undef RC_Decode
+#undef RC_DecodeFinal
+#undef RC_GetThreshold
+#undef CTX
+#undef SUCCESSOR
+#undef MASK

C/Ppmd8Enc.c

@@ -1,67 +1,110 @@
-/* Ppmd8Enc.c -- PPMdI Encoder
-2010-03-12 : Igor Pavlov : Public domain
+/* Ppmd8Enc.c -- Ppmd8 (PPMdI) Encoder
+2023-04-02 : Igor Pavlov : Public domain
 This code is based on:
   PPMd var.I (2002): Dmitry Shkarin : Public domain
   Carryless rangecoder (1999): Dmitry Subbotin : Public domain */
 
+#include "Precomp.h"
+
 #include "Ppmd8.h"
 
-#define kTop (1 << 24)
-#define kBot (1 << 15)
+#define kTop ((UInt32)1 << 24)
+#define kBot ((UInt32)1 << 15)
 
-void Ppmd8_RangeEnc_FlushData(CPpmd8 *p)
+#define WRITE_BYTE(p) IByteOut_Write(p->Stream.Out, (Byte)(p->Low >> 24))
+
+void Ppmd8_Flush_RangeEnc(CPpmd8 *p)
 {
   unsigned i;
   for (i = 0; i < 4; i++, p->Low <<= 8 )
-    p->Stream.Out->Write(p->Stream.Out, (Byte)(p->Low >> 24));
+    WRITE_BYTE(p);
 }
 
-static void RangeEnc_Normalize(CPpmd8 *p)
+
+
+
+
+
+#define RC_NORM(p) \
+  while ((p->Low ^ (p->Low + p->Range)) < kTop \
+    || (p->Range < kBot && ((p->Range = (0 - p->Low) & (kBot - 1)), 1))) \
+    { WRITE_BYTE(p); p->Range <<= 8; p->Low <<= 8; }
+
+
+
+
+
+
+
+
+
+
+
+
+
+// we must use only one type of Normalization from two: LOCAL or REMOTE
+#define RC_NORM_LOCAL(p)    // RC_NORM(p)
+#define RC_NORM_REMOTE(p)   RC_NORM(p)
+
+// #define RC_PRE(total) p->Range /= total;
+// #define RC_PRE(total)
+
+#define R p
+
+
+
+
+Z7_FORCE_INLINE
+// Z7_NO_INLINE
+static void Ppmd8_RangeEnc_Encode(CPpmd8 *p, UInt32 start, UInt32 size, UInt32 total)
 {
-  while ((p->Low ^ (p->Low + p->Range)) < kTop ||
-      p->Range < kBot && ((p->Range = (0 - p->Low) & (kBot - 1)), 1))
-  {
-    p->Stream.Out->Write(p->Stream.Out, (Byte)(p->Low >> 24));
-    p->Range <<= 8;
-    p->Low <<= 8;
-  }
-}
-
-static void RangeEnc_Encode(CPpmd8 *p, UInt32 start, UInt32 size, UInt32 total)
-{
-  p->Low += start * (p->Range /= total);
-  p->Range *= size;
-  RangeEnc_Normalize(p);
-}
-
-static void RangeEnc_EncodeBit_0(CPpmd8 *p, UInt32 size0)
-{
-  p->Range >>= 14;
-  p->Range *= size0;
-  RangeEnc_Normalize(p);
-}
-
-static void RangeEnc_EncodeBit_1(CPpmd8 *p, UInt32 size0)
-{
-  p->Low += size0 * (p->Range >>= 14);
-  p->Range *= ((1 << 14) - size0);
-  RangeEnc_Normalize(p);
+  R->Low += start * (R->Range /= total);
+  R->Range *= size;
+  RC_NORM_LOCAL(R)
 }
 
 
-#define MASK(sym) ((signed char *)charMask)[sym]
 
+
+
+
+
+
+
+
+#define RC_Encode(start, size, total)  Ppmd8_RangeEnc_Encode(p, start, size, total);
+#define RC_EncodeFinal(start, size, total)  RC_Encode(start, size, total)  RC_NORM_REMOTE(p)
+
+#define CTX(ref) ((CPpmd8_Context *)Ppmd8_GetContext(p, ref))
+
+// typedef CPpmd8_Context * CTX_PTR;
+#define SUCCESSOR(p) Ppmd_GET_SUCCESSOR(p)
+
+void Ppmd8_UpdateModel(CPpmd8 *p);
+
+#define MASK(sym) ((unsigned char *)charMask)[sym]
+
+// Z7_FORCE_INLINE
+// static
 void Ppmd8_EncodeSymbol(CPpmd8 *p, int symbol)
 {
   size_t charMask[256 / sizeof(size_t)];
+  
   if (p->MinContext->NumStats != 0)
   {
     CPpmd_State *s = Ppmd8_GetStats(p, p->MinContext);
     UInt32 sum;
     unsigned i;
+    UInt32 summFreq = p->MinContext->Union2.SummFreq;
+    
+    PPMD8_CORRECT_SUM_RANGE(p, summFreq)
+
+    // RC_PRE(summFreq);
+
     if (s->Symbol == symbol)
     {
-      RangeEnc_Encode(p, 0, s->Freq, p->MinContext->SummFreq);
+
+      RC_EncodeFinal(0, s->Freq, summFreq)
       p->FoundState = s;
       Ppmd8_Update1_0(p);
       return;
@@ -73,7 +116,8 @@ void Ppmd8_EncodeSymbol(CPpmd8 *p, int symbol)
     {
       if ((++s)->Symbol == symbol)
       {
-        RangeEnc_Encode(p, sum, s->Freq, p->MinContext->SummFreq);
+
+        RC_EncodeFinal(sum, s->Freq, summFreq)
         p->FoundState = s;
         Ppmd8_Update1(p);
         return;
@@ -82,80 +126,213 @@ void Ppmd8_EncodeSymbol(CPpmd8 *p, int symbol)
     }
     while (--i);
     
-    PPMD_SetAllBitsIn256Bytes(charMask);
-    MASK(s->Symbol) = 0;
-    i = p->MinContext->NumStats;
-    do { MASK((--s)->Symbol) = 0; } while (--i);
-    RangeEnc_Encode(p, sum, p->MinContext->SummFreq - sum, p->MinContext->SummFreq);
+    
+    RC_Encode(sum, summFreq - sum, summFreq)
+    
+    
+    PPMD_SetAllBitsIn256Bytes(charMask)
+    // MASK(s->Symbol) = 0;
+    // i = p->MinContext->NumStats;
+    // do { MASK((--s)->Symbol) = 0; } while (--i);
+    {
+      CPpmd_State *s2 = Ppmd8_GetStats(p, p->MinContext);
+      MASK(s->Symbol) = 0;
+      do
+      {
+        unsigned sym0 = s2[0].Symbol;
+        unsigned sym1 = s2[1].Symbol;
+        s2 += 2;
+        MASK(sym0) = 0;
+        MASK(sym1) = 0;
+      }
+      while (s2 < s);
+    }
   }
   else
   {
     UInt16 *prob = Ppmd8_GetBinSumm(p);
     CPpmd_State *s = Ppmd8Context_OneState(p->MinContext);
+    UInt32 pr = *prob;
+    const UInt32 bound = (R->Range >> 14) * pr;
+    pr = PPMD_UPDATE_PROB_1(pr);
     if (s->Symbol == symbol)
     {
-      RangeEnc_EncodeBit_0(p, *prob);
-      *prob = (UInt16)PPMD_UPDATE_PROB_0(*prob);
-      p->FoundState = s;
-      Ppmd8_UpdateBin(p);
+      *prob = (UInt16)(pr + (1 << PPMD_INT_BITS));
+      // RangeEnc_EncodeBit_0(p, bound);
+      R->Range = bound;
+      RC_NORM(R)
+
+      // p->FoundState = s;
+      // Ppmd8_UpdateBin(p);
+      {
+        const unsigned freq = s->Freq;
+        CPpmd8_Context *c = CTX(SUCCESSOR(s));
+        p->FoundState = s;
+        p->PrevSuccess = 1;
+        p->RunLength++;
+        s->Freq = (Byte)(freq + (freq < 196)); // Ppmd8 (196)
+        // NextContext(p);
+        if (p->OrderFall == 0 && (const Byte *)c >= p->UnitsStart)
+          p->MaxContext = p->MinContext = c;
+        else
+          Ppmd8_UpdateModel(p);
+      }
       return;
     }
-    else
-    {
-      RangeEnc_EncodeBit_1(p, *prob);
-      *prob = (UInt16)PPMD_UPDATE_PROB_1(*prob);
-      p->InitEsc = PPMD8_kExpEscape[*prob >> 10];
-      PPMD_SetAllBitsIn256Bytes(charMask);
-      MASK(s->Symbol) = 0;
-      p->PrevSuccess = 0;
-    }
+
+    *prob = (UInt16)pr;
+    p->InitEsc = p->ExpEscape[pr >> 10];
+    // RangeEnc_EncodeBit_1(p, bound);
+    R->Low += bound;
+    R->Range = (R->Range & ~((UInt32)PPMD_BIN_SCALE - 1)) - bound;
+    RC_NORM_LOCAL(R)
+
+    PPMD_SetAllBitsIn256Bytes(charMask)
+    MASK(s->Symbol) = 0;
+    p->PrevSuccess = 0;
   }
+
   for (;;)
   {
-    UInt32 escFreq;
     CPpmd_See *see;
     CPpmd_State *s;
-    UInt32 sum;
-    unsigned i, numMasked = p->MinContext->NumStats;
+    UInt32 sum, escFreq;
+    CPpmd8_Context *mc;
+    unsigned i, numMasked;
+
+    RC_NORM_REMOTE(p)
+
+    mc = p->MinContext;
+    numMasked = mc->NumStats;
+
     do
     {
       p->OrderFall++;
-      if (!p->MinContext->Suffix)
+      if (!mc->Suffix)
         return; /* EndMarker (symbol = -1) */
-      p->MinContext = Ppmd8_GetContext(p, p->MinContext->Suffix);
+      mc = Ppmd8_GetContext(p, mc->Suffix);
+
     }
-    while (p->MinContext->NumStats == numMasked);
+    while (mc->NumStats == numMasked);
     
+    p->MinContext = mc;
+
     see = Ppmd8_MakeEscFreq(p, numMasked, &escFreq);
+    
+    
+    
+    
+    
+    
+    
+    
+    
+
+    
+    
+    
+    
+    
+    
+
+    
+    
+    
+    
+    
+    
+    
     s = Ppmd8_GetStats(p, p->MinContext);
     sum = 0;
-    i = p->MinContext->NumStats + 1;
+    i = (unsigned)p->MinContext->NumStats + 1;
+
     do
     {
-      int cur = s->Symbol;
-      if (cur == symbol)
+      const unsigned cur = s->Symbol;
+      if ((int)cur == symbol)
       {
-        UInt32 low = sum;
-        CPpmd_State *s1 = s;
-        do
+        const UInt32 low = sum;
+        const UInt32 freq = s->Freq;
+        unsigned num2;
+
+        Ppmd_See_UPDATE(see)
+        p->FoundState = s;
+        sum += escFreq;
+
+        num2 = i / 2;
+        i &= 1;
+        sum += freq & (0 - (UInt32)i);
+        if (num2 != 0)
         {
-          sum += (s->Freq & (int)(MASK(s->Symbol)));
-          s++;
+          s += i;
+          for (;;)
+          {
+            unsigned sym0 = s[0].Symbol;
+            unsigned sym1 = s[1].Symbol;
+            s += 2;
+            sum += (s[-2].Freq & (unsigned)(MASK(sym0)));
+            sum += (s[-1].Freq & (unsigned)(MASK(sym1)));
+            if (--num2 == 0)
+              break;
+          }
         }
-        while (--i);
-        RangeEnc_Encode(p, low, s1->Freq, sum + escFreq);
-        Ppmd_See_Update(see);
-        p->FoundState = s1;
+
+        PPMD8_CORRECT_SUM_RANGE(p, sum)
+
+        RC_EncodeFinal(low, freq, sum)
         Ppmd8_Update2(p);
         return;
       }
-      sum += (s->Freq & (int)(MASK(cur)));
-      MASK(cur) = 0;
+      sum += (s->Freq & (unsigned)(MASK(cur)));
       s++;
     }
     while (--i);
     
-    RangeEnc_Encode(p, sum, escFreq, sum + escFreq);
-    see->Summ = (UInt16)(see->Summ + sum + escFreq);
+    {
+      UInt32 total = sum + escFreq;
+      see->Summ = (UInt16)(see->Summ + total);
+      PPMD8_CORRECT_SUM_RANGE(p, total)
+
+      RC_Encode(sum, total - sum, total)
+    }
+
+    {
+      const CPpmd_State *s2 = Ppmd8_GetStats(p, p->MinContext);
+      s--;
+      MASK(s->Symbol) = 0;
+      do
+      {
+        const unsigned sym0 = s2[0].Symbol;
+        const unsigned sym1 = s2[1].Symbol;
+        s2 += 2;
+        MASK(sym0) = 0;
+        MASK(sym1) = 0;
+      }
+      while (s2 < s);
+    }
   }
 }
+
+
+
+
+
+
+
+
+
+#undef kTop
+#undef kBot
+#undef WRITE_BYTE
+#undef RC_NORM_BASE
+#undef RC_NORM_1
+#undef RC_NORM
+#undef RC_NORM_LOCAL
+#undef RC_NORM_REMOTE
+#undef R
+#undef RC_Encode
+#undef RC_EncodeFinal
+
+#undef CTX
+#undef SUCCESSOR
+#undef MASK

C/Precomp.h

@@ -0,0 +1,10 @@
+/* Precomp.h -- StdAfx
+2023-04-02 : Igor Pavlov : Public domain */
+
+#ifndef ZIP7_INC_PRECOMP_H
+#define ZIP7_INC_PRECOMP_H
+
+#include "Compiler.h"
+/* #include "7zTypes.h" */
+
+#endif

C/RotateDefs.h

@@ -1,19 +1,49 @@
 /* RotateDefs.h -- Rotate functions
-2009-02-07 : Igor Pavlov : Public domain */
+2023-06-18 : Igor Pavlov : Public domain */
 
-#ifndef __ROTATE_DEFS_H
-#define __ROTATE_DEFS_H
+#ifndef ZIP7_INC_ROTATE_DEFS_H
+#define ZIP7_INC_ROTATE_DEFS_H
 
 #ifdef _MSC_VER
 
 #include <stdlib.h>
+
+/* don't use _rotl with old MINGW. It can insert slow call to function. */
+ 
+/* #if (_MSC_VER >= 1200) */
+#pragma intrinsic(_rotl)
+#pragma intrinsic(_rotr)
+/* #endif */
+
 #define rotlFixed(x, n) _rotl((x), (n))
 #define rotrFixed(x, n) _rotr((x), (n))
 
+#if (_MSC_VER >= 1300)
+#define Z7_ROTL64(x, n) _rotl64((x), (n))
+#define Z7_ROTR64(x, n) _rotr64((x), (n))
+#else
+#define Z7_ROTL64(x, n) (((x) << (n)) | ((x) >> (64 - (n))))
+#define Z7_ROTR64(x, n) (((x) >> (n)) | ((x) << (64 - (n))))
+#endif
+
 #else
 
+/* new compilers can translate these macros to fast commands. */
+
+#if defined(__clang__) && (__clang_major__ >= 4) \
+  || defined(__GNUC__) && (__GNUC__ >= 5)
+/* GCC 4.9.0 and clang 3.5 can recognize more correct version: */
+#define rotlFixed(x, n) (((x) << (n)) | ((x) >> (-(n) & 31)))
+#define rotrFixed(x, n) (((x) >> (n)) | ((x) << (-(n) & 31)))
+#define Z7_ROTL64(x, n) (((x) << (n)) | ((x) >> (-(n) & 63)))
+#define Z7_ROTR64(x, n) (((x) >> (n)) | ((x) << (-(n) & 63)))
+#else
+/* for old GCC / clang: */
 #define rotlFixed(x, n) (((x) << (n)) | ((x) >> (32 - (n))))
 #define rotrFixed(x, n) (((x) >> (n)) | ((x) << (32 - (n))))
+#define Z7_ROTL64(x, n) (((x) << (n)) | ((x) >> (64 - (n))))
+#define Z7_ROTR64(x, n) (((x) >> (n)) | ((x) << (64 - (n))))
+#endif
 
 #endif
 

C/Sha1.c

@@ -0,0 +1,498 @@
+/* Sha1.c -- SHA-1 Hash
+2023-04-02 : Igor Pavlov : Public domain
+This code is based on public domain code of Steve Reid from Wei Dai's Crypto++ library. */
+
+#include "Precomp.h"
+
+#include <string.h>
+
+#include "CpuArch.h"
+#include "RotateDefs.h"
+#include "Sha1.h"
+
+#if defined(_MSC_VER) && (_MSC_VER < 1900)
+// #define USE_MY_MM
+#endif
+
+#ifdef MY_CPU_X86_OR_AMD64
+  #ifdef _MSC_VER
+    #if _MSC_VER >= 1200
+      #define Z7_COMPILER_SHA1_SUPPORTED
+    #endif
+  #elif defined(__clang__)
+    #if (__clang_major__ >= 8) // fix that check
+      #define Z7_COMPILER_SHA1_SUPPORTED
+    #endif
+  #elif defined(__GNUC__)
+    #if (__GNUC__ >= 8) // fix that check
+      #define Z7_COMPILER_SHA1_SUPPORTED
+    #endif
+  #elif defined(__INTEL_COMPILER)
+    #if (__INTEL_COMPILER >= 1800) // fix that check
+      #define Z7_COMPILER_SHA1_SUPPORTED
+    #endif
+  #endif
+#elif defined(MY_CPU_ARM_OR_ARM64)
+  #ifdef _MSC_VER
+    #if _MSC_VER >= 1910 && _MSC_VER >= 1929 && _MSC_FULL_VER >= 192930037
+      #define Z7_COMPILER_SHA1_SUPPORTED
+    #endif
+  #elif defined(__clang__)
+    #if (__clang_major__ >= 8) // fix that check
+      #define Z7_COMPILER_SHA1_SUPPORTED
+    #endif
+  #elif defined(__GNUC__)
+    #if (__GNUC__ >= 6) // fix that check
+      #define Z7_COMPILER_SHA1_SUPPORTED
+    #endif
+  #endif
+#endif
+
+void Z7_FASTCALL Sha1_UpdateBlocks(UInt32 state[5], const Byte *data, size_t numBlocks);
+
+#ifdef Z7_COMPILER_SHA1_SUPPORTED
+  void Z7_FASTCALL Sha1_UpdateBlocks_HW(UInt32 state[5], const Byte *data, size_t numBlocks);
+
+  static SHA1_FUNC_UPDATE_BLOCKS g_SHA1_FUNC_UPDATE_BLOCKS = Sha1_UpdateBlocks;
+  static SHA1_FUNC_UPDATE_BLOCKS g_SHA1_FUNC_UPDATE_BLOCKS_HW;
+
+  #define SHA1_UPDATE_BLOCKS(p) p->func_UpdateBlocks
+#else
+  #define SHA1_UPDATE_BLOCKS(p) Sha1_UpdateBlocks
+#endif
+
+
+BoolInt Sha1_SetFunction(CSha1 *p, unsigned algo)
+{
+  SHA1_FUNC_UPDATE_BLOCKS func = Sha1_UpdateBlocks;
+  
+  #ifdef Z7_COMPILER_SHA1_SUPPORTED
+    if (algo != SHA1_ALGO_SW)
+    {
+      if (algo == SHA1_ALGO_DEFAULT)
+        func = g_SHA1_FUNC_UPDATE_BLOCKS;
+      else
+      {
+        if (algo != SHA1_ALGO_HW)
+          return False;
+        func = g_SHA1_FUNC_UPDATE_BLOCKS_HW;
+        if (!func)
+          return False;
+      }
+    }
+  #else
+    if (algo > 1)
+      return False;
+  #endif
+
+  p->func_UpdateBlocks = func;
+  return True;
+}
+
+
+/* define it for speed optimization */
+// #define Z7_SHA1_UNROLL
+
+// allowed unroll steps: (1, 2, 4, 5, 20)
+
+#undef Z7_SHA1_BIG_W
+#ifdef Z7_SHA1_UNROLL
+  #define STEP_PRE  20
+  #define STEP_MAIN 20
+#else
+  #define Z7_SHA1_BIG_W
+  #define STEP_PRE  5
+  #define STEP_MAIN 5
+#endif
+
+
+#ifdef Z7_SHA1_BIG_W
+  #define kNumW 80
+  #define w(i) W[i]
+#else
+  #define kNumW 16
+  #define w(i) W[(i)&15]
+#endif
+
+#define w0(i) (W[i] = GetBe32(data + (size_t)(i) * 4))
+#define w1(i) (w(i) = rotlFixed(w((size_t)(i)-3) ^ w((size_t)(i)-8) ^ w((size_t)(i)-14) ^ w((size_t)(i)-16), 1))
+
+#define f0(x,y,z)  ( 0x5a827999 + (z^(x&(y^z))) )
+#define f1(x,y,z)  ( 0x6ed9eba1 + (x^y^z) )
+#define f2(x,y,z)  ( 0x8f1bbcdc + ((x&y)|(z&(x|y))) )
+#define f3(x,y,z)  ( 0xca62c1d6 + (x^y^z) )
+
+/*
+#define T1(fx, ww) \
+    tmp = e + fx(b,c,d) + ww + rotlFixed(a, 5); \
+    e = d; \
+    d = c; \
+    c = rotlFixed(b, 30); \
+    b = a; \
+    a = tmp; \
+*/
+
+#define T5(a,b,c,d,e, fx, ww) \
+    e += fx(b,c,d) + ww + rotlFixed(a, 5); \
+    b = rotlFixed(b, 30); \
+
+
+/*
+#define R1(i, fx, wx) \
+    T1 ( fx, wx(i)); \
+
+#define R2(i, fx, wx) \
+    R1 ( (i)    , fx, wx); \
+    R1 ( (i) + 1, fx, wx); \
+
+#define R4(i, fx, wx) \
+    R2 ( (i)    , fx, wx); \
+    R2 ( (i) + 2, fx, wx); \
+*/
+
+#define M5(i, fx, wx0, wx1) \
+    T5 ( a,b,c,d,e, fx, wx0((i)  ) ) \
+    T5 ( e,a,b,c,d, fx, wx1((i)+1) ) \
+    T5 ( d,e,a,b,c, fx, wx1((i)+2) ) \
+    T5 ( c,d,e,a,b, fx, wx1((i)+3) ) \
+    T5 ( b,c,d,e,a, fx, wx1((i)+4) ) \
+
+#define R5(i, fx, wx) \
+    M5 ( i, fx, wx, wx) \
+
+
+#if STEP_PRE > 5
+
+  #define R20_START \
+    R5 (  0, f0, w0) \
+    R5 (  5, f0, w0) \
+    R5 ( 10, f0, w0) \
+    M5 ( 15, f0, w0, w1) \
+  
+  #elif STEP_PRE == 5
+  
+  #define R20_START \
+    { size_t i; for (i = 0; i < 15; i += STEP_PRE) \
+      { R5(i, f0, w0) } } \
+    M5 ( 15, f0, w0, w1) \
+
+#else
+
+  #if STEP_PRE == 1
+    #define R_PRE R1
+  #elif STEP_PRE == 2
+    #define R_PRE R2
+  #elif STEP_PRE == 4
+    #define R_PRE R4
+  #endif
+
+  #define R20_START \
+    { size_t i; for (i = 0; i < 16; i += STEP_PRE) \
+      { R_PRE(i, f0, w0) } } \
+    R4 ( 16, f0, w1) \
+
+#endif
+
+
+
+#if STEP_MAIN > 5
+
+  #define R20(ii, fx) \
+    R5 ( (ii)     , fx, w1) \
+    R5 ( (ii) + 5 , fx, w1) \
+    R5 ( (ii) + 10, fx, w1) \
+    R5 ( (ii) + 15, fx, w1) \
+
+#else
+
+  #if STEP_MAIN == 1
+    #define R_MAIN R1
+  #elif STEP_MAIN == 2
+    #define R_MAIN R2
+  #elif STEP_MAIN == 4
+    #define R_MAIN R4
+  #elif STEP_MAIN == 5
+    #define R_MAIN R5
+  #endif
+
+  #define R20(ii, fx)  \
+    { size_t i; for (i = (ii); i < (ii) + 20; i += STEP_MAIN) \
+      { R_MAIN(i, fx, w1) } } \
+
+#endif
+
+
+
+void Sha1_InitState(CSha1 *p)
+{
+  p->count = 0;
+  p->state[0] = 0x67452301;
+  p->state[1] = 0xEFCDAB89;
+  p->state[2] = 0x98BADCFE;
+  p->state[3] = 0x10325476;
+  p->state[4] = 0xC3D2E1F0;
+}
+
+void Sha1_Init(CSha1 *p)
+{
+  p->func_UpdateBlocks =
+  #ifdef Z7_COMPILER_SHA1_SUPPORTED
+      g_SHA1_FUNC_UPDATE_BLOCKS;
+  #else
+      NULL;
+  #endif
+  Sha1_InitState(p);
+}
+
+
+Z7_NO_INLINE
+void Z7_FASTCALL Sha1_UpdateBlocks(UInt32 state[5], const Byte *data, size_t numBlocks)
+{
+  UInt32 a, b, c, d, e;
+  UInt32 W[kNumW];
+  // if (numBlocks != 0x1264378347) return;
+  if (numBlocks == 0)
+    return;
+
+  a = state[0];
+  b = state[1];
+  c = state[2];
+  d = state[3];
+  e = state[4];
+
+  do
+  {
+  #if STEP_PRE < 5 || STEP_MAIN < 5
+  UInt32 tmp;
+  #endif
+
+  R20_START
+  R20(20, f1)
+  R20(40, f2)
+  R20(60, f3)
+
+  a += state[0];
+  b += state[1];
+  c += state[2];
+  d += state[3];
+  e += state[4];
+
+  state[0] = a;
+  state[1] = b;
+  state[2] = c;
+  state[3] = d;
+  state[4] = e;
+
+  data += 64;
+  }
+  while (--numBlocks);
+}
+
+
+#define Sha1_UpdateBlock(p) SHA1_UPDATE_BLOCKS(p)(p->state, p->buffer, 1)
+
+void Sha1_Update(CSha1 *p, const Byte *data, size_t size)
+{
+  if (size == 0)
+    return;
+
+  {
+    unsigned pos = (unsigned)p->count & 0x3F;
+    unsigned num;
+    
+    p->count += size;
+    
+    num = 64 - pos;
+    if (num > size)
+    {
+      memcpy(p->buffer + pos, data, size);
+      return;
+    }
+    
+    if (pos != 0)
+    {
+      size -= num;
+      memcpy(p->buffer + pos, data, num);
+      data += num;
+      Sha1_UpdateBlock(p);
+    }
+  }
+  {
+    size_t numBlocks = size >> 6;
+    SHA1_UPDATE_BLOCKS(p)(p->state, data, numBlocks);
+    size &= 0x3F;
+    if (size == 0)
+      return;
+    data += (numBlocks << 6);
+    memcpy(p->buffer, data, size);
+  }
+}
+
+
+void Sha1_Final(CSha1 *p, Byte *digest)
+{
+  unsigned pos = (unsigned)p->count & 0x3F;
+  
+
+  p->buffer[pos++] = 0x80;
+  
+  if (pos > (64 - 8))
+  {
+    while (pos != 64) { p->buffer[pos++] = 0; }
+    // memset(&p->buf.buffer[pos], 0, 64 - pos);
+    Sha1_UpdateBlock(p);
+    pos = 0;
+  }
+
+  /*
+  if (pos & 3)
+  {
+    p->buffer[pos] = 0;
+    p->buffer[pos + 1] = 0;
+    p->buffer[pos + 2] = 0;
+    pos += 3;
+    pos &= ~3;
+  }
+  {
+    for (; pos < 64 - 8; pos += 4)
+      *(UInt32 *)(&p->buffer[pos]) = 0;
+  }
+  */
+
+  memset(&p->buffer[pos], 0, (64 - 8) - pos);
+
+  {
+    const UInt64 numBits = (p->count << 3);
+    SetBe32(p->buffer + 64 - 8, (UInt32)(numBits >> 32))
+    SetBe32(p->buffer + 64 - 4, (UInt32)(numBits))
+  }
+  
+  Sha1_UpdateBlock(p);
+
+  SetBe32(digest,      p->state[0])
+  SetBe32(digest + 4,  p->state[1])
+  SetBe32(digest + 8,  p->state[2])
+  SetBe32(digest + 12, p->state[3])
+  SetBe32(digest + 16, p->state[4])
+  
+
+
+
+  Sha1_InitState(p);
+}
+
+
+void Sha1_PrepareBlock(const CSha1 *p, Byte *block, unsigned size)
+{
+  const UInt64 numBits = (p->count + size) << 3;
+  SetBe32(&((UInt32 *)(void *)block)[SHA1_NUM_BLOCK_WORDS - 2], (UInt32)(numBits >> 32))
+  SetBe32(&((UInt32 *)(void *)block)[SHA1_NUM_BLOCK_WORDS - 1], (UInt32)(numBits))
+  // SetBe32((UInt32 *)(block + size), 0x80000000);
+  SetUi32((UInt32 *)(void *)(block + size), 0x80)
+  size += 4;
+  while (size != (SHA1_NUM_BLOCK_WORDS - 2) * 4)
+  {
+    *((UInt32 *)(void *)(block + size)) = 0;
+    size += 4;
+  }
+}
+
+void Sha1_GetBlockDigest(const CSha1 *p, const Byte *data, Byte *destDigest)
+{
+  MY_ALIGN (16)
+  UInt32 st[SHA1_NUM_DIGEST_WORDS];
+
+  st[0] = p->state[0];
+  st[1] = p->state[1];
+  st[2] = p->state[2];
+  st[3] = p->state[3];
+  st[4] = p->state[4];
+  
+  SHA1_UPDATE_BLOCKS(p)(st, data, 1);
+
+  SetBe32(destDigest + 0    , st[0])
+  SetBe32(destDigest + 1 * 4, st[1])
+  SetBe32(destDigest + 2 * 4, st[2])
+  SetBe32(destDigest + 3 * 4, st[3])
+  SetBe32(destDigest + 4 * 4, st[4])
+}
+
+
+void Sha1Prepare(void)
+{
+  #ifdef Z7_COMPILER_SHA1_SUPPORTED
+  SHA1_FUNC_UPDATE_BLOCKS f, f_hw;
+  f = Sha1_UpdateBlocks;
+  f_hw = NULL;
+  #ifdef MY_CPU_X86_OR_AMD64
+  #ifndef USE_MY_MM
+  if (CPU_IsSupported_SHA()
+      && CPU_IsSupported_SSSE3()
+      // && CPU_IsSupported_SSE41()
+      )
+  #endif
+  #else
+  if (CPU_IsSupported_SHA1())
+  #endif
+  {
+    // printf("\n========== HW SHA1 ======== \n");
+    #if defined(MY_CPU_ARM_OR_ARM64) && defined(_MSC_VER)
+    /* there was bug in MSVC compiler for ARM64 -O2 before version VS2019 16.10 (19.29.30037).
+       It generated incorrect SHA-1 code.
+       21.03 : we test sha1-hardware code at runtime initialization */
+
+      #pragma message("== SHA1 code: MSC compiler : failure-check code was inserted")
+
+      UInt32 state[5] = { 0, 1, 2, 3, 4 } ;
+      Byte data[64];
+      unsigned i;
+      for (i = 0; i < sizeof(data); i += 2)
+      {
+        data[i    ] = (Byte)(i);
+        data[i + 1] = (Byte)(i + 1);
+      }
+
+      Sha1_UpdateBlocks_HW(state, data, sizeof(data) / 64);
+    
+      if (   state[0] != 0x9acd7297
+          || state[1] != 0x4624d898
+          || state[2] != 0x0bf079f0
+          || state[3] != 0x031e61b3
+          || state[4] != 0x8323fe20)
+      {
+        // printf("\n========== SHA-1 hardware version failure ======== \n");
+      }
+      else
+    #endif
+      {
+        f = f_hw = Sha1_UpdateBlocks_HW;
+      }
+  }
+  g_SHA1_FUNC_UPDATE_BLOCKS    = f;
+  g_SHA1_FUNC_UPDATE_BLOCKS_HW = f_hw;
+  #endif
+}
+
+#undef kNumW
+#undef w
+#undef w0
+#undef w1
+#undef f0
+#undef f1
+#undef f2
+#undef f3
+#undef T1
+#undef T5
+#undef M5
+#undef R1
+#undef R2
+#undef R4
+#undef R5
+#undef R20_START
+#undef R_PRE
+#undef R_MAIN
+#undef STEP_PRE
+#undef STEP_MAIN
+#undef Z7_SHA1_BIG_W
+#undef Z7_SHA1_UNROLL
+#undef Z7_COMPILER_SHA1_SUPPORTED

C/Sha1.h

@@ -0,0 +1,76 @@
+/* Sha1.h -- SHA-1 Hash
+2023-04-02 : Igor Pavlov : Public domain */
+
+#ifndef ZIP7_INC_SHA1_H
+#define ZIP7_INC_SHA1_H
+
+#include "7zTypes.h"
+
+EXTERN_C_BEGIN
+
+#define SHA1_NUM_BLOCK_WORDS  16
+#define SHA1_NUM_DIGEST_WORDS  5
+
+#define SHA1_BLOCK_SIZE   (SHA1_NUM_BLOCK_WORDS * 4)
+#define SHA1_DIGEST_SIZE  (SHA1_NUM_DIGEST_WORDS * 4)
+
+typedef void (Z7_FASTCALL *SHA1_FUNC_UPDATE_BLOCKS)(UInt32 state[5], const Byte *data, size_t numBlocks);
+
+/*
+  if (the system supports different SHA1 code implementations)
+  {
+    (CSha1::func_UpdateBlocks) will be used
+    (CSha1::func_UpdateBlocks) can be set by
+       Sha1_Init()        - to default (fastest)
+       Sha1_SetFunction() - to any algo
+  }
+  else
+  {
+    (CSha1::func_UpdateBlocks) is ignored.
+  }
+*/
+
+typedef struct
+{
+  SHA1_FUNC_UPDATE_BLOCKS func_UpdateBlocks;
+  UInt64 count;
+  UInt64 _pad_2[2];
+  UInt32 state[SHA1_NUM_DIGEST_WORDS];
+  UInt32 _pad_3[3];
+  Byte buffer[SHA1_BLOCK_SIZE];
+} CSha1;
+
+
+#define SHA1_ALGO_DEFAULT 0
+#define SHA1_ALGO_SW      1
+#define SHA1_ALGO_HW      2
+
+/*
+Sha1_SetFunction()
+return:
+  0 - (algo) value is not supported, and func_UpdateBlocks was not changed
+  1 - func_UpdateBlocks was set according (algo) value.
+*/
+
+BoolInt Sha1_SetFunction(CSha1 *p, unsigned algo);
+
+void Sha1_InitState(CSha1 *p);
+void Sha1_Init(CSha1 *p);
+void Sha1_Update(CSha1 *p, const Byte *data, size_t size);
+void Sha1_Final(CSha1 *p, Byte *digest);
+
+void Sha1_PrepareBlock(const CSha1 *p, Byte *block, unsigned size);
+void Sha1_GetBlockDigest(const CSha1 *p, const Byte *data, Byte *destDigest);
+
+// void Z7_FASTCALL Sha1_UpdateBlocks(UInt32 state[5], const Byte *data, size_t numBlocks);
+
+/*
+call Sha1Prepare() once at program start.
+It prepares all supported implementations, and detects the fastest implementation.
+*/
+
+void Sha1Prepare(void);
+
+EXTERN_C_END
+
+#endif

C/Sha1Opt.c

@@ -0,0 +1,386 @@
+/* Sha1Opt.c -- SHA-1 optimized code for SHA-1 hardware instructions
+2023-04-02 : Igor Pavlov : Public domain */
+
+#include "Precomp.h"
+#include "Compiler.h"
+#include "CpuArch.h"
+
+#if defined(_MSC_VER)
+#if (_MSC_VER < 1900) && (_MSC_VER >= 1200)
+// #define USE_MY_MM
+#endif
+#endif
+
+#ifdef MY_CPU_X86_OR_AMD64
+  #if defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 1600) // fix that check
+      #define USE_HW_SHA
+  #elif defined(Z7_LLVM_CLANG_VERSION)  && (Z7_LLVM_CLANG_VERSION  >= 30800) \
+     || defined(Z7_APPLE_CLANG_VERSION) && (Z7_APPLE_CLANG_VERSION >= 50100) \
+     || defined(Z7_GCC_VERSION)         && (Z7_GCC_VERSION         >= 40900)
+      #define USE_HW_SHA
+      #if !defined(_INTEL_COMPILER)
+      // icc defines __GNUC__, but icc doesn't support __attribute__(__target__)
+      #if !defined(__SHA__) || !defined(__SSSE3__)
+        #define ATTRIB_SHA __attribute__((__target__("sha,ssse3")))
+      #endif
+      #endif
+  #elif defined(_MSC_VER)
+    #ifdef USE_MY_MM
+      #define USE_VER_MIN 1300
+    #else
+      #define USE_VER_MIN 1900
+    #endif
+    #if (_MSC_VER >= USE_VER_MIN)
+      #define USE_HW_SHA
+    #endif
+  #endif
+// #endif // MY_CPU_X86_OR_AMD64
+
+#ifdef USE_HW_SHA
+
+// #pragma message("Sha1 HW")
+
+// sse/sse2/ssse3:
+#include <tmmintrin.h>
+// sha*:
+#include <immintrin.h>
+
+#if defined (__clang__) && defined(_MSC_VER)
+  // #if !defined(__SSSE3__)
+  // #endif
+  #if !defined(__SHA__)
+    #include <shaintrin.h>
+  #endif
+#else
+
+#ifdef USE_MY_MM
+#include "My_mm.h"
+#endif
+
+#endif
+
+/*
+SHA1 uses:
+SSE2:
+  _mm_loadu_si128
+  _mm_storeu_si128
+  _mm_set_epi32
+  _mm_add_epi32
+  _mm_shuffle_epi32 / pshufd
+  _mm_xor_si128
+  _mm_cvtsi128_si32
+  _mm_cvtsi32_si128
+SSSE3:
+  _mm_shuffle_epi8 / pshufb
+
+SHA:
+  _mm_sha1*
+*/
+
+
+#define XOR_SI128(dest, src)      dest = _mm_xor_si128(dest, src);
+#define SHUFFLE_EPI8(dest, mask)  dest = _mm_shuffle_epi8(dest, mask);
+#define SHUFFLE_EPI32(dest, mask) dest = _mm_shuffle_epi32(dest, mask);
+#ifdef __clang__
+#define SHA1_RNDS4_RET_TYPE_CAST (__m128i)
+#else
+#define SHA1_RNDS4_RET_TYPE_CAST
+#endif
+#define SHA1_RND4(abcd, e0, f)    abcd = SHA1_RNDS4_RET_TYPE_CAST _mm_sha1rnds4_epu32(abcd, e0, f);
+#define SHA1_NEXTE(e, m)          e = _mm_sha1nexte_epu32(e, m);
+#define ADD_EPI32(dest, src)      dest = _mm_add_epi32(dest, src);
+#define SHA1_MSG1(dest, src)      dest = _mm_sha1msg1_epu32(dest, src);
+#define SHA1_MSG2(dest, src)      dest = _mm_sha1msg2_epu32(dest, src);
+
+
+#define LOAD_SHUFFLE(m, k) \
+    m = _mm_loadu_si128((const __m128i *)(const void *)(data + (k) * 16)); \
+    SHUFFLE_EPI8(m, mask) \
+
+#define SM1(m0, m1, m2, m3) \
+    SHA1_MSG1(m0, m1) \
+
+#define SM2(m0, m1, m2, m3) \
+    XOR_SI128(m3, m1) \
+    SHA1_MSG2(m3, m2) \
+
+#define SM3(m0, m1, m2, m3) \
+    XOR_SI128(m3, m1) \
+    SM1(m0, m1, m2, m3) \
+    SHA1_MSG2(m3, m2) \
+
+#define NNN(m0, m1, m2, m3)
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+#define R4(k, e0, e1, m0, m1, m2, m3, OP) \
+    e1 = abcd; \
+    SHA1_RND4(abcd, e0, (k) / 5) \
+    SHA1_NEXTE(e1, m1) \
+    OP(m0, m1, m2, m3) \
+
+#define R16(k, mx, OP0, OP1, OP2, OP3) \
+    R4 ( (k)*4+0, e0,e1, m0,m1,m2,m3, OP0 ) \
+    R4 ( (k)*4+1, e1,e0, m1,m2,m3,m0, OP1 ) \
+    R4 ( (k)*4+2, e0,e1, m2,m3,m0,m1, OP2 ) \
+    R4 ( (k)*4+3, e1,e0, m3,mx,m1,m2, OP3 ) \
+
+#define PREPARE_STATE \
+    SHUFFLE_EPI32 (abcd, 0x1B) \
+    SHUFFLE_EPI32 (e0,   0x1B) \
+
+
+
+
+
+void Z7_FASTCALL Sha1_UpdateBlocks_HW(UInt32 state[5], const Byte *data, size_t numBlocks);
+#ifdef ATTRIB_SHA
+ATTRIB_SHA
+#endif
+void Z7_FASTCALL Sha1_UpdateBlocks_HW(UInt32 state[5], const Byte *data, size_t numBlocks)
+{
+  const __m128i mask = _mm_set_epi32(0x00010203, 0x04050607, 0x08090a0b, 0x0c0d0e0f);
+
+  __m128i abcd, e0;
+  
+  if (numBlocks == 0)
+    return;
+  
+  abcd = _mm_loadu_si128((const __m128i *) (const void *) &state[0]); // dbca
+  e0 = _mm_cvtsi32_si128((int)state[4]); // 000e
+  
+  PREPARE_STATE
+  
+  do
+  {
+    __m128i abcd_save, e2;
+    __m128i m0, m1, m2, m3;
+    __m128i e1;
+    
+
+    abcd_save = abcd;
+    e2 = e0;
+
+    LOAD_SHUFFLE (m0, 0)
+    LOAD_SHUFFLE (m1, 1)
+    LOAD_SHUFFLE (m2, 2)
+    LOAD_SHUFFLE (m3, 3)
+
+    ADD_EPI32(e0, m0)
+    
+    R16 ( 0, m0, SM1, SM3, SM3, SM3 )
+    R16 ( 1, m0, SM3, SM3, SM3, SM3 )
+    R16 ( 2, m0, SM3, SM3, SM3, SM3 )
+    R16 ( 3, m0, SM3, SM3, SM3, SM3 )
+    R16 ( 4, e2, SM2, NNN, NNN, NNN )
+    
+    ADD_EPI32(abcd, abcd_save)
+    
+    data += 64;
+  }
+  while (--numBlocks);
+
+  PREPARE_STATE
+
+  _mm_storeu_si128((__m128i *) (void *) state, abcd);
+  *(state+4) = (UInt32)_mm_cvtsi128_si32(e0);
+}
+
+#endif // USE_HW_SHA
+
+#elif defined(MY_CPU_ARM_OR_ARM64)
+
+  #if defined(__clang__)
+    #if (__clang_major__ >= 8) // fix that check
+      #define USE_HW_SHA
+    #endif
+  #elif defined(__GNUC__)
+    #if (__GNUC__ >= 6) // fix that check
+      #define USE_HW_SHA
+    #endif
+  #elif defined(_MSC_VER)
+    #if _MSC_VER >= 1910
+      #define USE_HW_SHA
+    #endif
+  #endif
+
+#ifdef USE_HW_SHA
+
+// #pragma message("=== Sha1 HW === ")
+
+#if defined(__clang__) || defined(__GNUC__)
+  #ifdef MY_CPU_ARM64
+    #define ATTRIB_SHA __attribute__((__target__("+crypto")))
+  #else
+    #define ATTRIB_SHA __attribute__((__target__("fpu=crypto-neon-fp-armv8")))
+  #endif
+#else
+  // _MSC_VER
+  // for arm32
+  #define _ARM_USE_NEW_NEON_INTRINSICS
+#endif
+
+#if defined(_MSC_VER) && defined(MY_CPU_ARM64)
+#include <arm64_neon.h>
+#else
+#include <arm_neon.h>
+#endif
+
+typedef uint32x4_t v128;
+// typedef __n128 v128; // MSVC
+
+#ifdef MY_CPU_BE
+  #define MY_rev32_for_LE(x)
+#else
+  #define MY_rev32_for_LE(x) x = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(x)))
+#endif
+
+#define LOAD_128(_p)      (*(const v128 *)(const void *)(_p))
+#define STORE_128(_p, _v) *(v128 *)(void *)(_p) = (_v)
+
+#define LOAD_SHUFFLE(m, k) \
+    m = LOAD_128((data + (k) * 16)); \
+    MY_rev32_for_LE(m); \
+
+#define SU0(dest, src2, src3) dest = vsha1su0q_u32(dest, src2, src3);
+#define SU1(dest, src)        dest = vsha1su1q_u32(dest, src);
+#define C(e)                  abcd = vsha1cq_u32(abcd, e, t);
+#define P(e)                  abcd = vsha1pq_u32(abcd, e, t);
+#define M(e)                  abcd = vsha1mq_u32(abcd, e, t);
+#define H(e)                  e = vsha1h_u32(vgetq_lane_u32(abcd, 0))
+#define T(m, c)               t = vaddq_u32(m, c)
+
+void Z7_FASTCALL Sha1_UpdateBlocks_HW(UInt32 state[8], const Byte *data, size_t numBlocks);
+#ifdef ATTRIB_SHA
+ATTRIB_SHA
+#endif
+void Z7_FASTCALL Sha1_UpdateBlocks_HW(UInt32 state[8], const Byte *data, size_t numBlocks)
+{
+  v128 abcd;
+  v128 c0, c1, c2, c3;
+  uint32_t e0;
+
+  if (numBlocks == 0)
+    return;
+
+  c0 = vdupq_n_u32(0x5a827999);
+  c1 = vdupq_n_u32(0x6ed9eba1);
+  c2 = vdupq_n_u32(0x8f1bbcdc);
+  c3 = vdupq_n_u32(0xca62c1d6);
+
+  abcd = LOAD_128(&state[0]);
+  e0 = state[4];
+  
+  do
+  {
+    v128 abcd_save;
+    v128 m0, m1, m2, m3;
+    v128 t;
+    uint32_t e0_save, e1;
+
+    abcd_save = abcd;
+    e0_save = e0;
+    
+    LOAD_SHUFFLE (m0, 0)
+    LOAD_SHUFFLE (m1, 1)
+    LOAD_SHUFFLE (m2, 2)
+    LOAD_SHUFFLE (m3, 3)
+                     
+    T(m0, c0);                                  H(e1); C(e0);
+    T(m1, c0);  SU0(m0, m1, m2);                H(e0); C(e1);
+    T(m2, c0);  SU0(m1, m2, m3);  SU1(m0, m3);  H(e1); C(e0);
+    T(m3, c0);  SU0(m2, m3, m0);  SU1(m1, m0);  H(e0); C(e1);
+    T(m0, c0);  SU0(m3, m0, m1);  SU1(m2, m1);  H(e1); C(e0);
+    T(m1, c1);  SU0(m0, m1, m2);  SU1(m3, m2);  H(e0); P(e1);
+    T(m2, c1);  SU0(m1, m2, m3);  SU1(m0, m3);  H(e1); P(e0);
+    T(m3, c1);  SU0(m2, m3, m0);  SU1(m1, m0);  H(e0); P(e1);
+    T(m0, c1);  SU0(m3, m0, m1);  SU1(m2, m1);  H(e1); P(e0);
+    T(m1, c1);  SU0(m0, m1, m2);  SU1(m3, m2);  H(e0); P(e1);
+    T(m2, c2);  SU0(m1, m2, m3);  SU1(m0, m3);  H(e1); M(e0);
+    T(m3, c2);  SU0(m2, m3, m0);  SU1(m1, m0);  H(e0); M(e1);
+    T(m0, c2);  SU0(m3, m0, m1);  SU1(m2, m1);  H(e1); M(e0);
+    T(m1, c2);  SU0(m0, m1, m2);  SU1(m3, m2);  H(e0); M(e1);
+    T(m2, c2);  SU0(m1, m2, m3);  SU1(m0, m3);  H(e1); M(e0);
+    T(m3, c3);  SU0(m2, m3, m0);  SU1(m1, m0);  H(e0); P(e1);
+    T(m0, c3);  SU0(m3, m0, m1);  SU1(m2, m1);  H(e1); P(e0);
+    T(m1, c3);                    SU1(m3, m2);  H(e0); P(e1);
+    T(m2, c3);                                  H(e1); P(e0);
+    T(m3, c3);                                  H(e0); P(e1);
+                                                                                                                     
+    abcd = vaddq_u32(abcd, abcd_save);
+    e0 += e0_save;
+    
+    data += 64;
+  }
+  while (--numBlocks);
+
+  STORE_128(&state[0], abcd);
+  state[4] = e0;
+}
+
+#endif // USE_HW_SHA
+
+#endif // MY_CPU_ARM_OR_ARM64
+
+
+#ifndef USE_HW_SHA
+
+// #error Stop_Compiling_UNSUPPORTED_SHA
+// #include <stdlib.h>
+
+// #include "Sha1.h"
+void Z7_FASTCALL Sha1_UpdateBlocks(UInt32 state[5], const Byte *data, size_t numBlocks);
+
+#pragma message("Sha1   HW-SW stub was used")
+
+void Z7_FASTCALL Sha1_UpdateBlocks_HW(UInt32 state[5], const Byte *data, size_t numBlocks);
+void Z7_FASTCALL Sha1_UpdateBlocks_HW(UInt32 state[5], const Byte *data, size_t numBlocks)
+{
+  Sha1_UpdateBlocks(state, data, numBlocks);
+  /*
+  UNUSED_VAR(state);
+  UNUSED_VAR(data);
+  UNUSED_VAR(numBlocks);
+  exit(1);
+  return;
+  */
+}
+
+#endif
+
+#undef SU0
+#undef SU1
+#undef C
+#undef P
+#undef M
+#undef H
+#undef T
+#undef MY_rev32_for_LE
+#undef NNN
+#undef LOAD_128
+#undef STORE_128
+#undef LOAD_SHUFFLE
+#undef SM1
+#undef SM2
+#undef SM3
+#undef NNN
+#undef R4
+#undef R16
+#undef PREPARE_STATE
+#undef USE_HW_SHA
+#undef ATTRIB_SHA
+#undef USE_VER_MIN

C/Sha256.c

@@ -1,16 +1,117 @@
-/* Crypto/Sha256.c -- SHA-256 Hash function
-2008-11-06 : Igor Pavlov : Public domain
+/* Sha256.c -- SHA-256 Hash
+2023-04-02 : Igor Pavlov : Public domain
 This code is based on public domain code from Wei Dai's Crypto++ library. */
 
-#include "Sha256.h"
+#include "Precomp.h"
+
+#include <string.h>
+
+#include "CpuArch.h"
 #include "RotateDefs.h"
+#include "Sha256.h"
+
+#if defined(_MSC_VER) && (_MSC_VER < 1900)
+// #define USE_MY_MM
+#endif
+
+#ifdef MY_CPU_X86_OR_AMD64
+  #ifdef _MSC_VER
+    #if _MSC_VER >= 1200
+      #define Z7_COMPILER_SHA256_SUPPORTED
+    #endif
+  #elif defined(__clang__)
+    #if (__clang_major__ >= 8) // fix that check
+      #define Z7_COMPILER_SHA256_SUPPORTED
+    #endif
+  #elif defined(__GNUC__)
+    #if (__GNUC__ >= 8) // fix that check
+      #define Z7_COMPILER_SHA256_SUPPORTED
+    #endif
+  #elif defined(__INTEL_COMPILER)
+    #if (__INTEL_COMPILER >= 1800) // fix that check
+      #define Z7_COMPILER_SHA256_SUPPORTED
+    #endif
+  #endif
+#elif defined(MY_CPU_ARM_OR_ARM64)
+  #ifdef _MSC_VER
+    #if _MSC_VER >= 1910
+      #define Z7_COMPILER_SHA256_SUPPORTED
+    #endif
+  #elif defined(__clang__)
+    #if (__clang_major__ >= 8) // fix that check
+      #define Z7_COMPILER_SHA256_SUPPORTED
+    #endif
+  #elif defined(__GNUC__)
+    #if (__GNUC__ >= 6) // fix that check
+      #define Z7_COMPILER_SHA256_SUPPORTED
+    #endif
+  #endif
+#endif
+
+void Z7_FASTCALL Sha256_UpdateBlocks(UInt32 state[8], const Byte *data, size_t numBlocks);
+
+#ifdef Z7_COMPILER_SHA256_SUPPORTED
+  void Z7_FASTCALL Sha256_UpdateBlocks_HW(UInt32 state[8], const Byte *data, size_t numBlocks);
+
+  static SHA256_FUNC_UPDATE_BLOCKS g_SHA256_FUNC_UPDATE_BLOCKS = Sha256_UpdateBlocks;
+  static SHA256_FUNC_UPDATE_BLOCKS g_SHA256_FUNC_UPDATE_BLOCKS_HW;
+
+  #define SHA256_UPDATE_BLOCKS(p) p->func_UpdateBlocks
+#else
+  #define SHA256_UPDATE_BLOCKS(p) Sha256_UpdateBlocks
+#endif
+
+
+BoolInt Sha256_SetFunction(CSha256 *p, unsigned algo)
+{
+  SHA256_FUNC_UPDATE_BLOCKS func = Sha256_UpdateBlocks;
+  
+  #ifdef Z7_COMPILER_SHA256_SUPPORTED
+    if (algo != SHA256_ALGO_SW)
+    {
+      if (algo == SHA256_ALGO_DEFAULT)
+        func = g_SHA256_FUNC_UPDATE_BLOCKS;
+      else
+      {
+        if (algo != SHA256_ALGO_HW)
+          return False;
+        func = g_SHA256_FUNC_UPDATE_BLOCKS_HW;
+        if (!func)
+          return False;
+      }
+    }
+  #else
+    if (algo > 1)
+      return False;
+  #endif
+
+  p->func_UpdateBlocks = func;
+  return True;
+}
+
 
 /* define it for speed optimization */
-/* #define _SHA256_UNROLL */
-/* #define _SHA256_UNROLL2 */
 
-void Sha256_Init(CSha256 *p)
+#ifdef Z7_SFX
+  #define STEP_PRE 1
+  #define STEP_MAIN 1
+#else
+  #define STEP_PRE 2
+  #define STEP_MAIN 4
+  // #define Z7_SHA256_UNROLL
+#endif
+
+#undef Z7_SHA256_BIG_W
+#if STEP_MAIN != 16
+  #define Z7_SHA256_BIG_W
+#endif
+
+
+
+
+void Sha256_InitState(CSha256 *p)
 {
+  p->count = 0;
   p->state[0] = 0x6a09e667;
   p->state[1] = 0xbb67ae85;
   p->state[2] = 0x3c6ef372;
@@ -19,7 +120,17 @@ void Sha256_Init(CSha256 *p)
   p->state[5] = 0x9b05688c;
   p->state[6] = 0x1f83d9ab;
   p->state[7] = 0x5be0cd19;
-  p->count = 0;
+}
+
+void Sha256_Init(CSha256 *p)
+{
+  p->func_UpdateBlocks =
+  #ifdef Z7_COMPILER_SHA256_SUPPORTED
+      g_SHA256_FUNC_UPDATE_BLOCKS;
+  #else
+      NULL;
+  #endif
+  Sha256_InitState(p);
 }
 
 #define S0(x) (rotrFixed(x, 2) ^ rotrFixed(x,13) ^ rotrFixed(x, 22))
@@ -27,51 +138,100 @@ void Sha256_Init(CSha256 *p)
 #define s0(x) (rotrFixed(x, 7) ^ rotrFixed(x,18) ^ (x >> 3))
 #define s1(x) (rotrFixed(x,17) ^ rotrFixed(x,19) ^ (x >> 10))
 
-#define blk0(i) (W[i] = data[i])
-#define blk2(i) (W[i&15] += s1(W[(i-2)&15]) + W[(i-7)&15] + s0(W[(i-15)&15]))
-
 #define Ch(x,y,z) (z^(x&(y^z)))
 #define Maj(x,y,z) ((x&y)|(z&(x|y)))
 
-#define a(i) T[(0-(i))&7]
-#define b(i) T[(1-(i))&7]
-#define c(i) T[(2-(i))&7]
-#define d(i) T[(3-(i))&7]
-#define e(i) T[(4-(i))&7]
-#define f(i) T[(5-(i))&7]
-#define g(i) T[(6-(i))&7]
-#define h(i) T[(7-(i))&7]
 
+#define W_PRE(i) (W[(i) + (size_t)(j)] = GetBe32(data + ((size_t)(j) + i) * 4))
 
-#ifdef _SHA256_UNROLL2
-
-#define R(a,b,c,d,e,f,g,h, i) h += S1(e) + Ch(e,f,g) + K[i+j] + (j?blk2(i):blk0(i));\
-  d += h; h += S0(a) + Maj(a, b, c)
-
-#define RX_8(i) \
-  R(a,b,c,d,e,f,g,h, i); \
-  R(h,a,b,c,d,e,f,g, i+1); \
-  R(g,h,a,b,c,d,e,f, i+2); \
-  R(f,g,h,a,b,c,d,e, i+3); \
-  R(e,f,g,h,a,b,c,d, i+4); \
-  R(d,e,f,g,h,a,b,c, i+5); \
-  R(c,d,e,f,g,h,a,b, i+6); \
-  R(b,c,d,e,f,g,h,a, i+7)
+#define blk2_main(j, i)  s1(w(j, (i)-2)) + w(j, (i)-7) + s0(w(j, (i)-15))
 
+#ifdef Z7_SHA256_BIG_W
+    // we use +i instead of +(i) to change the order to solve CLANG compiler warning for signed/unsigned.
+    #define w(j, i)     W[(size_t)(j) + i]
+    #define blk2(j, i)  (w(j, i) = w(j, (i)-16) + blk2_main(j, i))
 #else
+    #if STEP_MAIN == 16
+        #define w(j, i)  W[(i) & 15]
+    #else
+        #define w(j, i)  W[((size_t)(j) + (i)) & 15]
+    #endif
+    #define blk2(j, i)  (w(j, i) += blk2_main(j, i))
+#endif
 
-#define R(i) h(i) += S1(e(i)) + Ch(e(i),f(i),g(i)) + K[i+j] + (j?blk2(i):blk0(i));\
-  d(i) += h(i); h(i) += S0(a(i)) + Maj(a(i), b(i), c(i))
+#define W_MAIN(i)  blk2(j, i)
 
-#ifdef _SHA256_UNROLL
 
-#define RX_8(i) R(i+0); R(i+1); R(i+2); R(i+3); R(i+4); R(i+5); R(i+6); R(i+7);
+#define T1(wx, i) \
+    tmp = h + S1(e) + Ch(e,f,g) + K[(i)+(size_t)(j)] + wx(i); \
+    h = g; \
+    g = f; \
+    f = e; \
+    e = d + tmp; \
+    tmp += S0(a) + Maj(a, b, c); \
+    d = c; \
+    c = b; \
+    b = a; \
+    a = tmp; \
+
+#define R1_PRE(i)  T1( W_PRE, i)
+#define R1_MAIN(i) T1( W_MAIN, i)
+
+#if (!defined(Z7_SHA256_UNROLL) || STEP_MAIN < 8) && (STEP_MAIN >= 4)
+#define R2_MAIN(i) \
+    R1_MAIN(i) \
+    R1_MAIN(i + 1) \
 
 #endif
 
+
+
+#if defined(Z7_SHA256_UNROLL) && STEP_MAIN >= 8
+
+#define T4( a,b,c,d,e,f,g,h, wx, i) \
+    h += S1(e) + Ch(e,f,g) + K[(i)+(size_t)(j)] + wx(i); \
+    tmp = h; \
+    h += d; \
+    d = tmp + S0(a) + Maj(a, b, c); \
+
+#define R4( wx, i) \
+    T4 ( a,b,c,d,e,f,g,h, wx, (i  )); \
+    T4 ( d,a,b,c,h,e,f,g, wx, (i+1)); \
+    T4 ( c,d,a,b,g,h,e,f, wx, (i+2)); \
+    T4 ( b,c,d,a,f,g,h,e, wx, (i+3)); \
+
+#define R4_PRE(i)  R4( W_PRE, i)
+#define R4_MAIN(i) R4( W_MAIN, i)
+
+
+#define T8( a,b,c,d,e,f,g,h, wx, i) \
+    h += S1(e) + Ch(e,f,g) + K[(i)+(size_t)(j)] + wx(i); \
+    d += h; \
+    h += S0(a) + Maj(a, b, c); \
+
+#define R8( wx, i) \
+    T8 ( a,b,c,d,e,f,g,h, wx, i  ); \
+    T8 ( h,a,b,c,d,e,f,g, wx, i+1); \
+    T8 ( g,h,a,b,c,d,e,f, wx, i+2); \
+    T8 ( f,g,h,a,b,c,d,e, wx, i+3); \
+    T8 ( e,f,g,h,a,b,c,d, wx, i+4); \
+    T8 ( d,e,f,g,h,a,b,c, wx, i+5); \
+    T8 ( c,d,e,f,g,h,a,b, wx, i+6); \
+    T8 ( b,c,d,e,f,g,h,a, wx, i+7); \
+
+#define R8_PRE(i)  R8( W_PRE, i)
+#define R8_MAIN(i) R8( W_MAIN, i)
+
 #endif
 
-const UInt32 K[64] = {
+void Z7_FASTCALL Sha256_UpdateBlocks_HW(UInt32 state[8], const Byte *data, size_t numBlocks);
+
+// static
+extern MY_ALIGN(64)
+const UInt32 SHA256_K_ARRAY[64];
+
+MY_ALIGN(64)
+const UInt32 SHA256_K_ARRAY[64] = {
   0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5,
   0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,
   0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3,
@@ -90,12 +250,27 @@ const UInt32 K[64] = {
   0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2
 };
 
-static void Sha256_Transform(UInt32 *state, const UInt32 *data)
+#define K SHA256_K_ARRAY
+
+
+Z7_NO_INLINE
+void Z7_FASTCALL Sha256_UpdateBlocks(UInt32 state[8], const Byte *data, size_t numBlocks)
 {
-  UInt32 W[16];
+  UInt32 W
+  #ifdef Z7_SHA256_BIG_W
+      [64];
+  #else
+      [16];
+  #endif
+
   unsigned j;
-  #ifdef _SHA256_UNROLL2
+
   UInt32 a,b,c,d,e,f,g,h;
+
+  #if !defined(Z7_SHA256_UNROLL) || (STEP_MAIN <= 4) || (STEP_PRE <= 4)
+  UInt32 tmp;
+  #endif
+  
   a = state[0];
   b = state[1];
   c = state[2];
@@ -104,101 +279,238 @@ static void Sha256_Transform(UInt32 *state, const UInt32 *data)
   f = state[5];
   g = state[6];
   h = state[7];
-  #else
-  UInt32 T[8];
-  for (j = 0; j < 8; j++)
-    T[j] = state[j];
-  #endif
 
-  for (j = 0; j < 64; j += 16)
+  while (numBlocks)
   {
-    #if defined(_SHA256_UNROLL) || defined(_SHA256_UNROLL2)
-    RX_8(0); RX_8(8);
+
+  for (j = 0; j < 16; j += STEP_PRE)
+  {
+    #if STEP_PRE > 4
+
+      #if STEP_PRE < 8
+      R4_PRE(0);
+      #else
+      R8_PRE(0);
+      #if STEP_PRE == 16
+      R8_PRE(8);
+      #endif
+      #endif
+
     #else
-    unsigned i;
-    for (i = 0; i < 16; i++) { R(i); }
+
+      R1_PRE(0)
+      #if STEP_PRE >= 2
+      R1_PRE(1)
+      #if STEP_PRE >= 4
+      R1_PRE(2)
+      R1_PRE(3)
+      #endif
+      #endif
+    
     #endif
   }
 
-  #ifdef _SHA256_UNROLL2
-  state[0] += a;
-  state[1] += b;
-  state[2] += c;
-  state[3] += d;
-  state[4] += e;
-  state[5] += f;
-  state[6] += g;
-  state[7] += h;
-  #else
-  for (j = 0; j < 8; j++)
-    state[j] += T[j];
-  #endif
-  
+  for (j = 16; j < 64; j += STEP_MAIN)
+  {
+    #if defined(Z7_SHA256_UNROLL) && STEP_MAIN >= 8
+
+      #if STEP_MAIN < 8
+      R4_MAIN(0)
+      #else
+      R8_MAIN(0)
+      #if STEP_MAIN == 16
+      R8_MAIN(8)
+      #endif
+      #endif
+
+    #else
+      
+      R1_MAIN(0)
+      #if STEP_MAIN >= 2
+      R1_MAIN(1)
+      #if STEP_MAIN >= 4
+      R2_MAIN(2)
+      #if STEP_MAIN >= 8
+      R2_MAIN(4)
+      R2_MAIN(6)
+      #if STEP_MAIN >= 16
+      R2_MAIN(8)
+      R2_MAIN(10)
+      R2_MAIN(12)
+      R2_MAIN(14)
+      #endif
+      #endif
+      #endif
+      #endif
+    #endif
+  }
+
+  a += state[0]; state[0] = a;
+  b += state[1]; state[1] = b;
+  c += state[2]; state[2] = c;
+  d += state[3]; state[3] = d;
+  e += state[4]; state[4] = e;
+  f += state[5]; state[5] = f;
+  g += state[6]; state[6] = g;
+  h += state[7]; state[7] = h;
+
+  data += 64;
+  numBlocks--;
+  }
+
   /* Wipe variables */
   /* memset(W, 0, sizeof(W)); */
-  /* memset(T, 0, sizeof(T)); */
 }
 
 #undef S0
 #undef S1
 #undef s0
 #undef s1
+#undef K
 
-static void Sha256_WriteByteBlock(CSha256 *p)
-{
-  UInt32 data32[16];
-  unsigned i;
-  for (i = 0; i < 16; i++)
-    data32[i] =
-      ((UInt32)(p->buffer[i * 4    ]) << 24) +
-      ((UInt32)(p->buffer[i * 4 + 1]) << 16) +
-      ((UInt32)(p->buffer[i * 4 + 2]) <<  8) +
-      ((UInt32)(p->buffer[i * 4 + 3]));
-  Sha256_Transform(p->state, data32);
-}
+#define Sha256_UpdateBlock(p) SHA256_UPDATE_BLOCKS(p)(p->state, p->buffer, 1)
 
 void Sha256_Update(CSha256 *p, const Byte *data, size_t size)
 {
-  UInt32 curBufferPos = (UInt32)p->count & 0x3F;
-  while (size > 0)
+  if (size == 0)
+    return;
+
   {
-    p->buffer[curBufferPos++] = *data++;
-    p->count++;
-    size--;
-    if (curBufferPos == 64)
+    unsigned pos = (unsigned)p->count & 0x3F;
+    unsigned num;
+    
+    p->count += size;
+    
+    num = 64 - pos;
+    if (num > size)
     {
-      curBufferPos = 0;
-      Sha256_WriteByteBlock(p);
+      memcpy(p->buffer + pos, data, size);
+      return;
+    }
+    
+    if (pos != 0)
+    {
+      size -= num;
+      memcpy(p->buffer + pos, data, num);
+      data += num;
+      Sha256_UpdateBlock(p);
     }
   }
+  {
+    size_t numBlocks = size >> 6;
+    SHA256_UPDATE_BLOCKS(p)(p->state, data, numBlocks);
+    size &= 0x3F;
+    if (size == 0)
+      return;
+    data += (numBlocks << 6);
+    memcpy(p->buffer, data, size);
+  }
 }
 
+
 void Sha256_Final(CSha256 *p, Byte *digest)
 {
-  UInt64 lenInBits = (p->count << 3);
-  UInt32 curBufferPos = (UInt32)p->count & 0x3F;
+  unsigned pos = (unsigned)p->count & 0x3F;
   unsigned i;
-  p->buffer[curBufferPos++] = 0x80;
-  while (curBufferPos != (64 - 8))
+  
+  p->buffer[pos++] = 0x80;
+  
+  if (pos > (64 - 8))
   {
-    curBufferPos &= 0x3F;
-    if (curBufferPos == 0)
-      Sha256_WriteByteBlock(p);
-    p->buffer[curBufferPos++] = 0;
+    while (pos != 64) { p->buffer[pos++] = 0; }
+    // memset(&p->buf.buffer[pos], 0, 64 - pos);
+    Sha256_UpdateBlock(p);
+    pos = 0;
   }
-  for (i = 0; i < 8; i++)
-  {
-    p->buffer[curBufferPos++] = (Byte)(lenInBits >> 56);
-    lenInBits <<= 8;
-  }
-  Sha256_WriteByteBlock(p);
 
-  for (i = 0; i < 8; i++)
+  /*
+  if (pos & 3)
   {
-    *digest++ = (Byte)(p->state[i] >> 24);
-    *digest++ = (Byte)(p->state[i] >> 16);
-    *digest++ = (Byte)(p->state[i] >> 8);
-    *digest++ = (Byte)(p->state[i]);
+    p->buffer[pos] = 0;
+    p->buffer[pos + 1] = 0;
+    p->buffer[pos + 2] = 0;
+    pos += 3;
+    pos &= ~3;
   }
-  Sha256_Init(p);
+  {
+    for (; pos < 64 - 8; pos += 4)
+      *(UInt32 *)(&p->buffer[pos]) = 0;
+  }
+  */
+
+  memset(&p->buffer[pos], 0, (64 - 8) - pos);
+
+  {
+    UInt64 numBits = (p->count << 3);
+    SetBe32(p->buffer + 64 - 8, (UInt32)(numBits >> 32))
+    SetBe32(p->buffer + 64 - 4, (UInt32)(numBits))
+  }
+  
+  Sha256_UpdateBlock(p);
+
+  for (i = 0; i < 8; i += 2)
+  {
+    UInt32 v0 = p->state[i];
+    UInt32 v1 = p->state[(size_t)i + 1];
+    SetBe32(digest    , v0)
+    SetBe32(digest + 4, v1)
+    digest += 8;
+  }
+  
+  Sha256_InitState(p);
 }
+
+
+void Sha256Prepare(void)
+{
+  #ifdef Z7_COMPILER_SHA256_SUPPORTED
+  SHA256_FUNC_UPDATE_BLOCKS f, f_hw;
+  f = Sha256_UpdateBlocks;
+  f_hw = NULL;
+  #ifdef MY_CPU_X86_OR_AMD64
+  #ifndef USE_MY_MM
+  if (CPU_IsSupported_SHA()
+      && CPU_IsSupported_SSSE3()
+      // && CPU_IsSupported_SSE41()
+      )
+  #endif
+  #else
+  if (CPU_IsSupported_SHA2())
+  #endif
+  {
+    // printf("\n========== HW SHA256 ======== \n");
+    f = f_hw = Sha256_UpdateBlocks_HW;
+  }
+  g_SHA256_FUNC_UPDATE_BLOCKS    = f;
+  g_SHA256_FUNC_UPDATE_BLOCKS_HW = f_hw;
+  #endif
+}
+
+#undef S0
+#undef S1
+#undef s0
+#undef s1
+#undef Ch
+#undef Maj
+#undef W_MAIN
+#undef W_PRE
+#undef w
+#undef blk2_main
+#undef blk2
+#undef T1
+#undef T4
+#undef T8
+#undef R1_PRE
+#undef R1_MAIN
+#undef R2_MAIN
+#undef R4
+#undef R4_PRE
+#undef R4_MAIN
+#undef R8
+#undef R8_PRE
+#undef R8_MAIN
+#undef STEP_PRE
+#undef STEP_MAIN
+#undef Z7_SHA256_BIG_W
+#undef Z7_SHA256_UNROLL
+#undef Z7_COMPILER_SHA256_SUPPORTED

C/Sha256.h

@@ -1,30 +1,76 @@
 /* Sha256.h -- SHA-256 Hash
-2009-02-07 : Igor Pavlov : Public domain */
+2023-04-02 : Igor Pavlov : Public domain */
 
-#ifndef __CRYPTO_SHA256_H
-#define __CRYPTO_SHA256_H
+#ifndef ZIP7_INC_SHA256_H
+#define ZIP7_INC_SHA256_H
 
-#include "Types.h"
+#include "7zTypes.h"
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+EXTERN_C_BEGIN
 
-#define SHA256_DIGEST_SIZE 32
+#define SHA256_NUM_BLOCK_WORDS  16
+#define SHA256_NUM_DIGEST_WORDS  8
+
+#define SHA256_BLOCK_SIZE   (SHA256_NUM_BLOCK_WORDS * 4)
+#define SHA256_DIGEST_SIZE  (SHA256_NUM_DIGEST_WORDS * 4)
+
+typedef void (Z7_FASTCALL *SHA256_FUNC_UPDATE_BLOCKS)(UInt32 state[8], const Byte *data, size_t numBlocks);
+
+/*
+  if (the system supports different SHA256 code implementations)
+  {
+    (CSha256::func_UpdateBlocks) will be used
+    (CSha256::func_UpdateBlocks) can be set by
+       Sha256_Init()        - to default (fastest)
+       Sha256_SetFunction() - to any algo
+  }
+  else
+  {
+    (CSha256::func_UpdateBlocks) is ignored.
+  }
+*/
 
 typedef struct
 {
-  UInt32 state[8];
+  SHA256_FUNC_UPDATE_BLOCKS func_UpdateBlocks;
   UInt64 count;
-  Byte buffer[64];
+  UInt64 _pad_2[2];
+  UInt32 state[SHA256_NUM_DIGEST_WORDS];
+
+  Byte buffer[SHA256_BLOCK_SIZE];
 } CSha256;
 
+
+#define SHA256_ALGO_DEFAULT 0
+#define SHA256_ALGO_SW      1
+#define SHA256_ALGO_HW      2
+
+/*
+Sha256_SetFunction()
+return:
+  0 - (algo) value is not supported, and func_UpdateBlocks was not changed
+  1 - func_UpdateBlocks was set according (algo) value.
+*/
+
+BoolInt Sha256_SetFunction(CSha256 *p, unsigned algo);
+
+void Sha256_InitState(CSha256 *p);
 void Sha256_Init(CSha256 *p);
 void Sha256_Update(CSha256 *p, const Byte *data, size_t size);
 void Sha256_Final(CSha256 *p, Byte *digest);
 
-#ifdef __cplusplus
-}
-#endif
+
+
+
+// void Z7_FASTCALL Sha256_UpdateBlocks(UInt32 state[8], const Byte *data, size_t numBlocks);
+
+/*
+call Sha256Prepare() once at program start.
+It prepares all supported implementations, and detects the fastest implementation.
+*/
+
+void Sha256Prepare(void);
+
+EXTERN_C_END
 
 #endif