/* ;uInt longest_match_x64( ; deflate_state *s, ; IPos cur_match); // current match
; gvmat64.S -- 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) ; this file is translation from gvmat64.asm to GCC 4.x (for Linux, Mac XCode) ; 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 zLib, I use option: ; gcc -c -arch x86_64 gvmat64.S
;uInt longest_match(s, cur_match) ; deflate_state *s; ; IPos cur_match; // current match / ; ; with XCode for Mac, I had strange error with some jump on intel syntax ; this is why BEFORE_JMP and AFTER_JMP are used
*/
/* ; 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
*/
/* ;;; 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").
*/
;;; 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.
; ms: parameter 1 in rcx (deflate_state* s), param 2 in edx -> r8 (cur match) ; mac: param 1 in rdi, param 2 rsi ; this clear high 32 bits of r8, which can be garbage in both r8 and rdx
*/
mov [save_rbx],rbx
mov [save_rbp],rbp
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;
cmp bx,word ptr [rsi + r8 - 1]
BEFORE_JMP
jnz LookupLoop1
jmp LookupLoopIsZero
AFTER_JMP /* ;;; 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
*/
.balign 16
LookupLoop:
and r8d, edx
//;;; 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, 0xfffffffffffffef8 //; -(MAX_MATCH_8)
lea rsi, [rsi + r13 + 0x0108] //;MAX_MATCH_8]
lea rdi, [r9 + r13 + 0x0108] //;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.
*/
add rdx,4
shr rax,32
or ax,ax
BEFORE_JMP
jnz LenLower
AFTER_JMP
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
BEFORE_JMP
jge LenMaximum
AFTER_JMP /* ;;; 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
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
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.