24.07

Asm/x86/LzFindOpt.asm

@@ -1,5 +1,5 @@
 ; LzFindOpt.asm -- ASM version of GetMatchesSpecN_2() function
-; 2021-07-21: Igor Pavlov : Public domain
+; 2024-06-18: Igor Pavlov : Public domain
 ;
 
 ifndef x64
@@ -11,10 +11,31 @@ include 7zAsm.asm
 
 MY_ASM_START
 
-_TEXT$LZFINDOPT SEGMENT ALIGN(64) 'CODE'
+ifndef Z7_LZ_FIND_OPT_ASM_USE_SEGMENT
+if (IS_LINUX gt 0)
+  Z7_LZ_FIND_OPT_ASM_USE_SEGMENT equ 1
+else
+  Z7_LZ_FIND_OPT_ASM_USE_SEGMENT equ 1
+endif
+endif
 
+ifdef Z7_LZ_FIND_OPT_ASM_USE_SEGMENT
+_TEXT$LZFINDOPT SEGMENT ALIGN(64) 'CODE'
 MY_ALIGN macro num:req
         align  num
+        ; align  16
+endm
+else
+MY_ALIGN macro num:req
+        ; We expect that ".text" is aligned for 16-bytes.
+        ; So we don't need large alignment inside our function.
+        align  16
+endm
+endif
+
+
+MY_ALIGN_16 macro
+        MY_ALIGN 16
 endm
 
 MY_ALIGN_32 macro
@@ -136,7 +157,11 @@ COPY_VAR_64 macro dest_var, src_var
 endm
 
 
+ifdef Z7_LZ_FIND_OPT_ASM_USE_SEGMENT
 ; MY_ALIGN_64
+else
+  MY_ALIGN_16
+endif
 MY_PROC GetMatchesSpecN_2, 13
 MY_PUSH_PRESERVED_ABI_REGS
         mov     r0, RSP
@@ -508,6 +533,8 @@ fin:
 MY_POP_PRESERVED_ABI_REGS
 MY_ENDP
 
+ifdef Z7_LZ_FIND_OPT_ASM_USE_SEGMENT
 _TEXT$LZFINDOPT ENDS
+endif
 
 end

Asm/x86/LzmaDecOpt.asm

@@ -1,5 +1,5 @@
 ; LzmaDecOpt.asm -- ASM version of LzmaDec_DecodeReal_3() function
-; 2021-02-23: Igor Pavlov : Public domain
+; 2024-06-18: Igor Pavlov : Public domain
 ;
 ; 3 - is the code compatibility version of LzmaDec_DecodeReal_*()
 ; function for check at link time.
@@ -17,11 +17,41 @@ include 7zAsm.asm
 
 MY_ASM_START
 
-_TEXT$LZMADECOPT SEGMENT ALIGN(64) 'CODE'
+; if Z7_LZMA_DEC_OPT_ASM_USE_SEGMENT is     defined, we use additional SEGMENT with 64-byte alignment.
+; if Z7_LZMA_DEC_OPT_ASM_USE_SEGMENT is not defined, we use default SEGMENT (where default 16-byte alignment of segment is expected).
+; The performance is almost identical in our tests.
+; But the performance can depend from position of lzmadec code inside instruction cache
+; or micro-op cache line (depending from low address bits in 32-byte/64-byte cache lines).
+; And 64-byte alignment provides a more consistent speed regardless
+; of the code's position in the executable.
+; But also it's possible that code without Z7_LZMA_DEC_OPT_ASM_USE_SEGMENT can be
+; slightly faster than 64-bytes aligned code in some cases, if offset of lzmadec
+; code in 64-byte block after compilation provides better speed by some reason.
+; Note that Z7_LZMA_DEC_OPT_ASM_USE_SEGMENT adds an extra section to the ELF file.
+; If you don't want to get that extra section, do not define Z7_LZMA_DEC_OPT_ASM_USE_SEGMENT.
+
+ifndef Z7_LZMA_DEC_OPT_ASM_USE_SEGMENT
+if (IS_LINUX gt 0)
+  Z7_LZMA_DEC_OPT_ASM_USE_SEGMENT equ 1
+else
+  Z7_LZMA_DEC_OPT_ASM_USE_SEGMENT equ 1
+endif
+endif
 
+ifdef Z7_LZMA_DEC_OPT_ASM_USE_SEGMENT
+_TEXT$LZMADECOPT SEGMENT ALIGN(64) 'CODE'
 MY_ALIGN macro num:req
         align  num
+        ; align  16
 endm
+else
+MY_ALIGN macro num:req
+        ; We expect that ".text" is aligned for 16-bytes.
+        ; So we don't need large alignment inside out function.
+        align  16
+endm
+endif
+
 
 MY_ALIGN_16 macro
         MY_ALIGN 16
@@ -610,7 +640,11 @@ PARAM_lzma      equ REG_ABI_PARAM_0
 PARAM_limit     equ REG_ABI_PARAM_1
 PARAM_bufLimit  equ REG_ABI_PARAM_2
 
+ifdef Z7_LZMA_DEC_OPT_ASM_USE_SEGMENT
 ; MY_ALIGN_64
+else
+  MY_ALIGN_16
+endif
 MY_PROC LzmaDec_DecodeReal_3, 3
 MY_PUSH_PRESERVED_ABI_REGS
 
@@ -1298,6 +1332,8 @@ fin:
 MY_POP_PRESERVED_ABI_REGS
 MY_ENDP
 
+ifdef Z7_LZMA_DEC_OPT_ASM_USE_SEGMENT
 _TEXT$LZMADECOPT ENDS
+endif
 
 end

Asm/x86/Sha1Opt.asm

@@ -1,5 +1,5 @@
 ; Sha1Opt.asm -- SHA-1 optimized code for SHA-1 x86 hardware instructions
-; 2021-03-10 : Igor Pavlov : Public domain
+; 2024-06-16 : Igor Pavlov : Public domain
 
 include 7zAsm.asm
 
@@ -20,7 +20,7 @@ MY_ASM_START
 
 
 
-CONST   SEGMENT
+CONST   SEGMENT READONLY
 
 align 16
 Reverse_Endian_Mask db 15,14,13,12, 11,10,9,8, 7,6,5,4, 3,2,1,0

Asm/x86/Sha256Opt.asm

@@ -1,5 +1,5 @@
 ; Sha256Opt.asm -- SHA-256 optimized code for SHA-256 x86 hardware instructions
-; 2022-04-17 : Igor Pavlov : Public domain
+; 2024-06-16 : Igor Pavlov : Public domain
 
 include 7zAsm.asm
 
@@ -20,7 +20,7 @@ endif
 EXTRN   K_CONST:xmmword
 @
 
-CONST   SEGMENT
+CONST   SEGMENT READONLY
 
 align 16
 Reverse_Endian_Mask db 3,2,1,0, 7,6,5,4, 11,10,9,8, 15,14,13,12

C/7zVersion.h

@@ -1,7 +1,7 @@
 #define MY_VER_MAJOR 24
-#define MY_VER_MINOR 06
+#define MY_VER_MINOR 07
 #define MY_VER_BUILD 0
-#define MY_VERSION_NUMBERS "24.06"
+#define MY_VERSION_NUMBERS "24.07"
 #define MY_VERSION MY_VERSION_NUMBERS
 
 #ifdef MY_CPU_NAME
@@ -10,7 +10,7 @@
   #define MY_VERSION_CPU MY_VERSION
 #endif
 
-#define MY_DATE "2024-05-26"
+#define MY_DATE "2024-06-19"
 #undef MY_COPYRIGHT
 #undef MY_VERSION_COPYRIGHT_DATE
 #define MY_AUTHOR_NAME "Igor Pavlov"

C/CpuArch.h

@@ -1,5 +1,5 @@
 /* CpuArch.h -- CPU specific code
-2024-05-18 : Igor Pavlov : Public domain */
+2024-06-17 : Igor Pavlov : Public domain */
 
 #ifndef ZIP7_INC_CPU_ARCH_H
 #define ZIP7_INC_CPU_ARCH_H
@@ -564,13 +564,15 @@ problem-4 : performace:
 #define SetBe32a(p, v)   { *(UInt32 *)(void *)(p) = (v); }
 #define SetBe16a(p, v)   { *(UInt16 *)(void *)(p) = (v); }
 
+#define GetUi64a(p)      GetUi64(p)
 #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 GetUi64a(p)      (*(const UInt64 *)(const void *)(p))
 #define GetUi32a(p)      (*(const UInt32 *)(const void *)(p))
 #define GetUi16a(p)      (*(const UInt16 *)(const void *)(p))
 #define SetUi32a(p, v)   { *(UInt32 *)(void *)(p) = (v); }

C/ZstdDec.c

@@ -1,5 +1,5 @@
 /* ZstdDec.c -- Zstd Decoder
-2024-05-26 : the code was developed by Igor Pavlov, using Zstandard format
+2024-06-18 : the code was developed by Igor Pavlov, using Zstandard format
              specification and original zstd decoder code as reference code.
 original zstd decoder code: Copyright (c) Facebook, Inc. All rights reserved.
 This source code is licensed under BSD 3-Clause License.
@@ -1308,8 +1308,10 @@ FSE_Decode_SeqTable(CFseRecord * const table,
   in->len--;
   {
     const Byte *ptr = in->ptr;
-    const Byte sym = ptr[0];
+    const unsigned sym = ptr[0];
     in->ptr = ptr + 1;
+    if (sym >= numSymbolsMax)
+      return SZ_ERROR_DATA;
     table[0] = (FastInt32)sym
       #if defined(Z7_ZSTD_DEC_USE_ML_PLUS3)
         + (numSymbolsMax == NUM_ML_SYMBOLS ? MATCH_LEN_MIN : 0)

CPP/7zip/7zip_gcc.mak

@@ -220,6 +220,9 @@ endif
 
 all: $(O) $(PROGPATH) $(STATIC_TARGET)
 
+# we need $(O) as order-only-prerequisites:
+$(OBJS): | $(O)
+
 $(O):
 	$(MY_MKDIR) $(O)
 

CPP/7zip/Archive/7z/7zUpdate.cpp

@@ -219,6 +219,14 @@ static int Parse_EXE(const Byte *buf, size_t size, CFilterMode *filterMode)
 }
 
 
+/*
+  Filters don't improve the compression ratio for relocatable object files (".o").
+  But we can get compression ratio gain, if we compress object
+  files and executables in same solid block.
+  So we use filters for relocatable object files (".o"):
+*/
+// #define Z7_7Z_CREATE_ARC_DISABLE_FILTER_FOR_OBJ
+
 /* ---------- ELF ---------- */
 
 #define ELF_SIG 0x464C457F
@@ -258,6 +266,12 @@ static int Parse_ELF(const Byte *buf, size_t size, CFilterMode *filterMode)
     default: return 0;
   }
 
+#ifdef Z7_7Z_CREATE_ARC_DISABLE_FILTER_FOR_OBJ
+#define ELF_ET_REL  1
+  if (Get16(buf + 0x10, be) == ELF_ET_REL)
+    return 0;
+#endif
+
   switch (Get16(buf + 0x12, be))
   {
     case 3:
@@ -318,6 +332,12 @@ static unsigned Parse_MACH(const Byte *buf, size_t size, CFilterMode *filterMode
     default: return 0;
   }
 
+#ifdef Z7_7Z_CREATE_ARC_DISABLE_FILTER_FOR_OBJ
+#define MACH_TYPE_OBJECT 1
+  if (Get32(buf + 0xC, be) == MACH_TYPE_OBJECT)
+      return 0;
+#endif
+
   switch (Get32(buf + 4, be))
   {
     case MACH_MACHINE_386: