;uInt longest_match_x64( ; deflate_state *s, ; IPos cur_match); /* current match */ ; gvmat64.asm -- Asm portion of the optimized longest_match for 32 bits x86_64 ; (AMD64 on Athlon 64, Opteron, Phenom ; and Intel EM64T on Pentium 4 with EM64T, Pentium D, Core 2 Duo, Core I5/I7) ; Copyright (C) 1995-2010 Jean-loup Gailly, Brian Raiter and Gilles Vollant. ; ; File written by Gilles Vollant, by converting to assembly the longest_match ; from Jean-loup Gailly in deflate.c of zLib and infoZip zip. ; ; and by taking inspiration on asm686 with masm, optimised assembly code ; from Brian Raiter, written 1998 ; ; This software is provided 'as-is', without any express or implied ; warranty. In no event will the authors be held liable for any damages ; arising from the use of this software. ; ; Permission is granted to anyone to use this software for any purpose, ; including commercial applications, and to alter it and redistribute it ; freely, subject to the following restrictions: ; ; 1. The origin of this software must not be misrepresented; you must not ; claim that you wrote the original software. If you use this software ; in a product, an acknowledgment in the product documentation would be ; appreciated but is not required. ; 2. Altered source versions must be plainly marked as such, and must not be ; misrepresented as being the original software ; 3. This notice may not be removed or altered from any source distribution. ; ; ; ; http://www.zlib.net ; http://www.winimage.com/zLibDll ; http://www.muppetlabs.com/~breadbox/software/assembly.html ; ; to compile this file for infozip Zip, I use option: ; ml64.exe /Flgvmat64 /c /Zi /DINFOZIP gvmat64.asm ; ; to compile this file for zLib, I use option: ; ml64.exe /Flgvmat64 /c /Zi gvmat64.asm ; Be carrefull to adapt zlib1222add below to your version of zLib ; (if you use a version of zLib before 1.0.4 or after 1.2.2.2, change ; value of zlib1222add later) ; ; This file compile with Microsoft Macro Assembler (x64) for AMD64 ; ; ml64.exe is given with Visual Studio 2005/2008/2010 and Windows WDK ; ; (you can get Windows WDK with ml64 for AMD64 from ; http://www.microsoft.com/whdc/Devtools/wdk/default.mspx for low price) ; ;uInt longest_match(s, cur_match) ; deflate_state *s; ; IPos cur_match; /* current match */ .code longest_match PROC ;LocalVarsSize equ 88 LocalVarsSize equ 72 ; register used : rax,rbx,rcx,rdx,rsi,rdi,r8,r9,r10,r11,r12 ; free register : r14,r15 ; register can be saved : rsp chainlenwmask equ rsp + 8 - LocalVarsSize ; high word: current chain len ; low word: s->wmask ;window equ rsp + xx - LocalVarsSize ; local copy of s->window ; stored in r10 ;windowbestlen equ rsp + xx - LocalVarsSize ; s->window + bestlen , use r10+r11 ;scanstart equ rsp + xx - LocalVarsSize ; first two bytes of string ; stored in r12w ;scanend equ rsp + xx - LocalVarsSize ; last two bytes of string use ebx ;scanalign equ rsp + xx - LocalVarsSize ; dword-misalignment of string r13 ;bestlen equ rsp + xx - LocalVarsSize ; size of best match so far -> r11d ;scan equ rsp + xx - LocalVarsSize ; ptr to string wanting match -> r9 IFDEF INFOZIP ELSE nicematch equ (rsp + 16 - LocalVarsSize) ; a good enough match size ENDIF save_rdi equ rsp + 24 - LocalVarsSize save_rsi equ rsp + 32 - LocalVarsSize save_rbx equ rsp + 40 - LocalVarsSize save_rbp equ rsp + 48 - LocalVarsSize save_r12 equ rsp + 56 - LocalVarsSize save_r13 equ rsp + 64 - LocalVarsSize ;save_r14 equ rsp + 72 - LocalVarsSize ;save_r15 equ rsp + 80 - LocalVarsSize ; summary of register usage ; scanend ebx ; scanendw bx ; chainlenwmask edx ; curmatch rsi ; curmatchd esi ; windowbestlen r8 ; scanalign r9 ; scanalignd r9d ; window r10 ; bestlen r11 ; bestlend r11d ; scanstart r12d ; scanstartw r12w ; scan r13 ; nicematch r14d ; limit r15 ; limitd r15d ; prev rcx ; all the +4 offsets are due to the addition of pending_buf_size (in zlib ; in the deflate_state structure since the asm code was first written ; (if you compile with zlib 1.0.4 or older, remove the +4). ; Note : these value are good with a 8 bytes boundary pack structure MAX_MATCH equ 258 MIN_MATCH equ 3 MIN_LOOKAHEAD equ (MAX_MATCH+MIN_MATCH+1) ;;; Offsets for fields in the deflate_state structure. These numbers ;;; are calculated from the definition of deflate_state, with the ;;; assumption that the compiler will dword-align the fields. (Thus, ;;; changing the definition of deflate_state could easily cause this ;;; program to crash horribly, without so much as a warning at ;;; compile time. Sigh.) ; all the +zlib1222add offsets are due to the addition of fields ; in zlib in the deflate_state structure since the asm code was first written ; (if you compile with zlib 1.0.4 or older, use "zlib1222add equ (-4)"). ; (if you compile with zlib between 1.0.5 and 1.2.2.1, use "zlib1222add equ 0"). ; if you compile with zlib 1.2.2.2 or later , use "zlib1222add equ 8"). IFDEF INFOZIP _DATA SEGMENT COMM window_size:DWORD ; WMask ; 7fff COMM window:BYTE:010040H COMM prev:WORD:08000H ; MatchLen : unused ; PrevMatch : unused COMM strstart:DWORD COMM match_start:DWORD ; Lookahead : ignore COMM prev_length:DWORD ; PrevLen COMM max_chain_length:DWORD COMM good_match:DWORD COMM nice_match:DWORD prev_ad equ OFFSET prev window_ad equ OFFSET window nicematch equ nice_match _DATA ENDS WMask equ 07fffh ELSE IFNDEF zlib1222add zlib1222add equ 8 ENDIF dsWSize equ 56+zlib1222add+(zlib1222add/2) dsWMask equ 64+zlib1222add+(zlib1222add/2) dsWindow equ 72+zlib1222add dsPrev equ 88+zlib1222add dsMatchLen equ 128+zlib1222add dsPrevMatch equ 132+zlib1222add dsStrStart equ 140+zlib1222add dsMatchStart equ 144+zlib1222add dsLookahead equ 148+zlib1222add dsPrevLen equ 152+zlib1222add dsMaxChainLen equ 156+zlib1222add dsGoodMatch equ 172+zlib1222add dsNiceMatch equ 176+zlib1222add window_size equ [ rcx + dsWSize] WMask equ [ rcx + dsWMask] window_ad equ [ rcx + dsWindow] prev_ad equ [ rcx + dsPrev] strstart equ [ rcx + dsStrStart] match_start equ [ rcx + dsMatchStart] Lookahead equ [ rcx + dsLookahead] ; 0ffffffffh on infozip prev_length equ [ rcx + dsPrevLen] max_chain_length equ [ rcx + dsMaxChainLen] good_match equ [ rcx + dsGoodMatch] nice_match equ [ rcx + dsNiceMatch] ENDIF ; parameter 1 in r8(deflate state s), param 2 in rdx (cur match) ; see http://weblogs.asp.net/oldnewthing/archive/2004/01/14/58579.aspx and ; http://msdn.microsoft.com/library/en-us/kmarch/hh/kmarch/64bitAMD_8e951dd2-ee77-4728-8702-55ce4b5dd24a.xml.asp ; ; All registers must be preserved across the call, except for ; rax, rcx, rdx, r8, r9, r10, and r11, which are scratch. ;;; Save registers that the compiler may be using, and adjust esp to ;;; make room for our stack frame. ;;; Retrieve the function arguments. r8d will hold cur_match ;;; throughout the entire function. edx will hold the pointer to the ;;; deflate_state structure during the function's setup (before ;;; entering the main loop. ; parameter 1 in rcx (deflate_state* s), param 2 in edx -> r8 (cur match) ; this clear high 32 bits of r8, which can be garbage in both r8 and rdx mov [save_rdi],rdi mov [save_rsi],rsi mov [save_rbx],rbx mov [save_rbp],rbp IFDEF INFOZIP mov r8d,ecx ELSE mov r8d,edx ENDIF mov [save_r12],r12 mov [save_r13],r13 ; mov [save_r14],r14 ; mov [save_r15],r15 ;;; uInt wmask = s->w_mask; ;;; unsigned chain_length = s->max_chain_length; ;;; if (s->prev_length >= s->good_match) { ;;; chain_length >>= 2; ;;; } mov edi, prev_length mov esi, good_match mov eax, WMask mov ebx, max_chain_length cmp edi, esi jl LastMatchGood shr ebx, 2 LastMatchGood: ;;; chainlen is decremented once beforehand so that the function can ;;; use the sign flag instead of the zero flag for the exit test. ;;; It is then shifted into the high word, to make room for the wmask ;;; value, which it will always accompany. dec ebx shl ebx, 16 or ebx, eax ;;; on zlib only ;;; if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; IFDEF INFOZIP mov [chainlenwmask], ebx ; on infozip nice_match = [nice_match] ELSE mov eax, nice_match mov [chainlenwmask], ebx mov r10d, Lookahead cmp r10d, eax cmovnl r10d, eax mov [nicematch],r10d ENDIF ;;; register Bytef *scan = s->window + s->strstart; mov r10, window_ad mov ebp, strstart lea r13, [r10 + rbp] ;;; Determine how many bytes the scan ptr is off from being ;;; dword-aligned. mov r9,r13 neg r13 and r13,3 ;;; IPos limit = s->strstart > (IPos)MAX_DIST(s) ? ;;; s->strstart - (IPos)MAX_DIST(s) : NIL; IFDEF INFOZIP mov eax,07efah ; MAX_DIST = (WSIZE-MIN_LOOKAHEAD) (0x8000-(3+8+1)) ELSE mov eax, window_size sub eax, MIN_LOOKAHEAD ENDIF xor edi,edi sub ebp, eax mov r11d, prev_length cmovng ebp,edi ;;; int best_len = s->prev_length; ;;; Store the sum of s->window + best_len in esi locally, and in esi. lea rsi,[r10+r11] ;;; register ush scan_start = *(ushf*)scan; ;;; register ush scan_end = *(ushf*)(scan+best_len-1); ;;; Posf *prev = s->prev; movzx r12d,word ptr [r9] movzx ebx, word ptr [r9 + r11 - 1] mov rdi, prev_ad ;;; Jump into the main loop. mov edx, [chainlenwmask] cmp bx,word ptr [rsi + r8 - 1] jz LookupLoopIsZero LookupLoop1: and r8d, edx movzx r8d, word ptr [rdi + r8*2] cmp r8d, ebp jbe LeaveNow sub edx, 00010000h js LeaveNow LoopEntry1: cmp bx,word ptr [rsi + r8 - 1] jz LookupLoopIsZero LookupLoop2: and r8d, edx movzx r8d, word ptr [rdi + r8*2] cmp r8d, ebp jbe LeaveNow sub edx, 00010000h js LeaveNow LoopEntry2: cmp bx,word ptr [rsi + r8 - 1] jz LookupLoopIsZero LookupLoop4: and r8d, edx movzx r8d, word ptr [rdi + r8*2] cmp r8d, ebp jbe LeaveNow sub edx, 00010000h js LeaveNow LoopEntry4: cmp bx,word ptr [rsi + r8 - 1] jnz LookupLoop1 jmp LookupLoopIsZero ;;; do { ;;; match = s->window + cur_match; ;;; if (*(ushf*)(match+best_len-1) != scan_end || ;;; *(ushf*)match != scan_start) continue; ;;; [...] ;;; } while ((cur_match = prev[cur_match & wmask]) > limit ;;; && --chain_length != 0); ;;; ;;; Here is the inner loop of the function. The function will spend the ;;; majority of its time in this loop, and majority of that time will ;;; be spent in the first ten instructions. ;;; ;;; Within this loop: ;;; ebx = scanend ;;; r8d = curmatch ;;; edx = chainlenwmask - i.e., ((chainlen << 16) | wmask) ;;; esi = windowbestlen - i.e., (window + bestlen) ;;; edi = prev ;;; ebp = limit LookupLoop: and r8d, edx movzx r8d, word ptr [rdi + r8*2] cmp r8d, ebp jbe LeaveNow sub edx, 00010000h js LeaveNow LoopEntry: cmp bx,word ptr [rsi + r8 - 1] jnz LookupLoop1 LookupLoopIsZero: cmp r12w, word ptr [r10 + r8] jnz LookupLoop1 ;;; Store the current value of chainlen. mov [chainlenwmask], edx ;;; Point edi to the string under scrutiny, and esi to the string we ;;; are hoping to match it up with. In actuality, esi and edi are ;;; both pointed (MAX_MATCH_8 - scanalign) bytes ahead, and edx is ;;; initialized to -(MAX_MATCH_8 - scanalign). lea rsi,[r8+r10] mov rdx, 0fffffffffffffef8h; -(MAX_MATCH_8) lea rsi, [rsi + r13 + 0108h] ;MAX_MATCH_8] lea rdi, [r9 + r13 + 0108h] ;MAX_MATCH_8] prefetcht1 [rsi+rdx] prefetcht1 [rdi+rdx] ;;; Test the strings for equality, 8 bytes at a time. At the end, ;;; adjust rdx so that it is offset to the exact byte that mismatched. ;;; ;;; We already know at this point that the first three bytes of the ;;; strings match each other, and they can be safely passed over before ;;; starting the compare loop. So what this code does is skip over 0-3 ;;; bytes, as much as necessary in order to dword-align the edi ;;; pointer. (rsi will still be misaligned three times out of four.) ;;; ;;; It should be confessed that this loop usually does not represent ;;; much of the total running time. Replacing it with a more ;;; straightforward "rep cmpsb" would not drastically degrade ;;; performance. LoopCmps: mov rax, [rsi + rdx] xor rax, [rdi + rdx] jnz LeaveLoopCmps mov rax, [rsi + rdx + 8] xor rax, [rdi + rdx + 8] jnz LeaveLoopCmps8 mov rax, [rsi + rdx + 8+8] xor rax, [rdi + rdx + 8+8] jnz LeaveLoopCmps16 add rdx,8+8+8 jnz short LoopCmps jmp short LenMaximum LeaveLoopCmps16: add rdx,8 LeaveLoopCmps8: add rdx,8 LeaveLoopCmps: test eax, 0000FFFFh jnz LenLower test eax,0ffffffffh jnz LenLower32 add rdx,4 shr rax,32 or ax,ax jnz LenLower LenLower32: shr eax,16 add rdx,2 LenLower: sub al, 1 adc rdx, 0 ;;; Calculate the length of the match. If it is longer than MAX_MATCH, ;;; then automatically accept it as the best possible match and leave. lea rax, [rdi + rdx] sub rax, r9 cmp eax, MAX_MATCH jge LenMaximum ;;; If the length of the match is not longer than the best match we ;;; have so far, then forget it and return to the lookup loop. ;/////////////////////////////////// cmp eax, r11d jg LongerMatch lea rsi,[r10+r11] mov rdi, prev_ad mov edx, [chainlenwmask] jmp LookupLoop ;;; s->match_start = cur_match; ;;; best_len = len; ;;; if (len >= nice_match) break; ;;; scan_end = *(ushf*)(scan+best_len-1); LongerMatch: mov r11d, eax mov match_start, r8d cmp eax, [nicematch] jge LeaveNow lea rsi,[r10+rax] movzx ebx, word ptr [r9 + rax - 1] mov rdi, prev_ad mov edx, [chainlenwmask] jmp LookupLoop ;;; Accept the current string, with the maximum possible length. LenMaximum: mov r11d,MAX_MATCH mov match_start, r8d ;;; if ((uInt)best_len <= s->lookahead) return (uInt)best_len; ;;; return s->lookahead; LeaveNow: IFDEF INFOZIP mov eax,r11d ELSE mov eax, Lookahead cmp r11d, eax cmovng eax, r11d ENDIF ;;; Restore the stack and return from whence we came. mov rsi,[save_rsi] mov rdi,[save_rdi] mov rbx,[save_rbx] mov rbp,[save_rbp] mov r12,[save_r12] mov r13,[save_r13] ; mov r14,[save_r14] ; mov r15,[save_r15] ret 0 ; please don't remove this string ! ; Your can freely use gvmat64 in any free or commercial app ; but it is far better don't remove the string in the binary! db 0dh,0ah,"asm686 with masm, optimised assembly code from Brian Raiter, written 1998, converted to amd 64 by Gilles Vollant 2005",0dh,0ah,0 longest_match ENDP match_init PROC ret 0 match_init ENDP END