24.0724.07

4 files changed, 71 insertions, 8 deletions

diff --git a/Asm/x86/LzFindOpt.asm b/Asm/x86/LzFindOpt.asm
index 42e10bd..94c5c76 100644
--- a/Asm/x86/LzFindOpt.asm
+++ b/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
diff --git a/Asm/x86/LzmaDecOpt.asm b/Asm/x86/LzmaDecOpt.asm
index f2818e7..7c568df 100644
--- a/Asm/x86/LzmaDecOpt.asm
+++ b/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
diff --git a/Asm/x86/Sha1Opt.asm b/Asm/x86/Sha1Opt.asm
index 3495fd1..0b63aeb 100644
--- a/Asm/x86/Sha1Opt.asm
+++ b/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
diff --git a/Asm/x86/Sha256Opt.asm b/Asm/x86/Sha256Opt.asm
index 3e9f6ed..bc2f9da 100644
--- a/Asm/x86/Sha256Opt.asm
+++ b/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