summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorjan.nijtmans <nijtmans@users.sourceforge.net>2012-05-04 21:11:58 (GMT)
committerjan.nijtmans <nijtmans@users.sourceforge.net>2012-05-04 21:11:58 (GMT)
commit14452102b5028ba9b20ceae48237a751f97f3b43 (patch)
tree5c2d242d388629c51debe71a52c47c2625303522
parentc65a6131bf900e600e7f3b32f602e2cd3d555be3 (diff)
downloadtcl-14452102b5028ba9b20ceae48237a751f97f3b43.zip
tcl-14452102b5028ba9b20ceae48237a751f97f3b43.tar.gz
tcl-14452102b5028ba9b20ceae48237a751f97f3b43.tar.bz2
some more removals from zlib1.2.7
-rw-r--r--compat/zlib/contrib/asm586/README.58643
-rw-r--r--compat/zlib/contrib/asm586/match.S364
-rw-r--r--compat/zlib/contrib/masm686/match.asm413
3 files changed, 0 insertions, 820 deletions
diff --git a/compat/zlib/contrib/asm586/README.586 b/compat/zlib/contrib/asm586/README.586
deleted file mode 100644
index 6bb78f3..0000000
--- a/compat/zlib/contrib/asm586/README.586
+++ /dev/null
@@ -1,43 +0,0 @@
-This is a patched version of zlib modified to use
-Pentium-optimized assembly code in the deflation algorithm. The files
-changed/added by this patch are:
-
-README.586
-match.S
-
-The effectiveness of these modifications is a bit marginal, as the the
-program's bottleneck seems to be mostly L1-cache contention, for which
-there is no real way to work around without rewriting the basic
-algorithm. The speedup on average is around 5-10% (which is generally
-less than the amount of variance between subsequent executions).
-However, when used at level 9 compression, the cache contention can
-drop enough for the assembly version to achieve 10-20% speedup (and
-sometimes more, depending on the amount of overall redundancy in the
-files). Even here, though, cache contention can still be the limiting
-factor, depending on the nature of the program using the zlib library.
-This may also mean that better improvements will be seen on a Pentium
-with MMX, which suffers much less from L1-cache contention, but I have
-not yet verified this.
-
-Note that this code has been tailored for the Pentium in particular,
-and will not perform well on the Pentium Pro (due to the use of a
-partial register in the inner loop).
-
-If you are using an assembler other than GNU as, you will have to
-translate match.S to use your assembler's syntax. (Have fun.)
-
-Brian Raiter
-breadbox@muppetlabs.com
-April, 1998
-
-
-Added for zlib 1.1.3:
-
-The patches come from
-http://www.muppetlabs.com/~breadbox/software/assembly.html
-
-To compile zlib with this asm file, copy match.S to the zlib directory
-then do:
-
-CFLAGS="-O3 -DASMV" ./configure
-make OBJA=match.o
diff --git a/compat/zlib/contrib/asm586/match.S b/compat/zlib/contrib/asm586/match.S
deleted file mode 100644
index 0368b35..0000000
--- a/compat/zlib/contrib/asm586/match.S
+++ /dev/null
@@ -1,364 +0,0 @@
-/* match.s -- Pentium-optimized version of longest_match()
- * Written for zlib 1.1.2
- * Copyright (C) 1998 Brian Raiter <breadbox@muppetlabs.com>
- *
- * This is free software; you can redistribute it and/or modify it
- * under the terms of the GNU General Public License.
- */
-
-#ifndef NO_UNDERLINE
-#define match_init _match_init
-#define longest_match _longest_match
-#endif
-
-#define MAX_MATCH (258)
-#define MIN_MATCH (3)
-#define MIN_LOOKAHEAD (MAX_MATCH + MIN_MATCH + 1)
-#define MAX_MATCH_8 ((MAX_MATCH + 7) & ~7)
-
-/* stack frame offsets */
-
-#define wmask 0 /* local copy of s->wmask */
-#define window 4 /* local copy of s->window */
-#define windowbestlen 8 /* s->window + bestlen */
-#define chainlenscanend 12 /* high word: current chain len */
- /* low word: last bytes sought */
-#define scanstart 16 /* first two bytes of string */
-#define scanalign 20 /* dword-misalignment of string */
-#define nicematch 24 /* a good enough match size */
-#define bestlen 28 /* size of best match so far */
-#define scan 32 /* ptr to string wanting match */
-
-#define LocalVarsSize (36)
-/* saved ebx 36 */
-/* saved edi 40 */
-/* saved esi 44 */
-/* saved ebp 48 */
-/* return address 52 */
-#define deflatestate 56 /* the function arguments */
-#define curmatch 60
-
-/* 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").
- */
-
-#define zlib1222add (8)
-
-#define dsWSize (36+zlib1222add)
-#define dsWMask (44+zlib1222add)
-#define dsWindow (48+zlib1222add)
-#define dsPrev (56+zlib1222add)
-#define dsMatchLen (88+zlib1222add)
-#define dsPrevMatch (92+zlib1222add)
-#define dsStrStart (100+zlib1222add)
-#define dsMatchStart (104+zlib1222add)
-#define dsLookahead (108+zlib1222add)
-#define dsPrevLen (112+zlib1222add)
-#define dsMaxChainLen (116+zlib1222add)
-#define dsGoodMatch (132+zlib1222add)
-#define dsNiceMatch (136+zlib1222add)
-
-
-.file "match.S"
-
-.globl match_init, longest_match
-
-.text
-
-/* uInt longest_match(deflate_state *deflatestate, IPos curmatch) */
-
-longest_match:
-
-/* Save registers that the compiler may be using, and adjust %esp to */
-/* make room for our stack frame. */
-
- pushl %ebp
- pushl %edi
- pushl %esi
- pushl %ebx
- subl $LocalVarsSize, %esp
-
-/* Retrieve the function arguments. %ecx 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). */
-
- movl deflatestate(%esp), %edx
- movl curmatch(%esp), %ecx
-
-/* if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; */
-
- movl dsNiceMatch(%edx), %eax
- movl dsLookahead(%edx), %ebx
- cmpl %eax, %ebx
- jl LookaheadLess
- movl %eax, %ebx
-LookaheadLess: movl %ebx, nicematch(%esp)
-
-/* register Bytef *scan = s->window + s->strstart; */
-
- movl dsWindow(%edx), %esi
- movl %esi, window(%esp)
- movl dsStrStart(%edx), %ebp
- lea (%esi,%ebp), %edi
- movl %edi, scan(%esp)
-
-/* Determine how many bytes the scan ptr is off from being */
-/* dword-aligned. */
-
- movl %edi, %eax
- negl %eax
- andl $3, %eax
- movl %eax, scanalign(%esp)
-
-/* IPos limit = s->strstart > (IPos)MAX_DIST(s) ? */
-/* s->strstart - (IPos)MAX_DIST(s) : NIL; */
-
- movl dsWSize(%edx), %eax
- subl $MIN_LOOKAHEAD, %eax
- subl %eax, %ebp
- jg LimitPositive
- xorl %ebp, %ebp
-LimitPositive:
-
-/* unsigned chain_length = s->max_chain_length; */
-/* if (s->prev_length >= s->good_match) { */
-/* chain_length >>= 2; */
-/* } */
-
- movl dsPrevLen(%edx), %eax
- movl dsGoodMatch(%edx), %ebx
- cmpl %ebx, %eax
- movl dsMaxChainLen(%edx), %ebx
- jl LastMatchGood
- shrl $2, %ebx
-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 scanend */
-/* scanend value, which it will always accompany. */
-
- decl %ebx
- shll $16, %ebx
-
-/* int best_len = s->prev_length; */
-
- movl dsPrevLen(%edx), %eax
- movl %eax, bestlen(%esp)
-
-/* Store the sum of s->window + best_len in %esi locally, and in %esi. */
-
- addl %eax, %esi
- movl %esi, windowbestlen(%esp)
-
-/* register ush scan_start = *(ushf*)scan; */
-/* register ush scan_end = *(ushf*)(scan+best_len-1); */
-
- movw (%edi), %bx
- movw %bx, scanstart(%esp)
- movw -1(%edi,%eax), %bx
- movl %ebx, chainlenscanend(%esp)
-
-/* Posf *prev = s->prev; */
-/* uInt wmask = s->w_mask; */
-
- movl dsPrev(%edx), %edi
- movl dsWMask(%edx), %edx
- mov %edx, wmask(%esp)
-
-/* Jump into the main loop. */
-
- jmp LoopEntry
-
-.balign 16
-
-/* 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 = chainlenscanend - i.e., ((chainlen << 16) | scanend)
- * %ecx = curmatch
- * %edx = curmatch & wmask
- * %esi = windowbestlen - i.e., (window + bestlen)
- * %edi = prev
- * %ebp = limit
- *
- * Two optimization notes on the choice of instructions:
- *
- * The first instruction uses a 16-bit address, which costs an extra,
- * unpairable cycle. This is cheaper than doing a 32-bit access and
- * zeroing the high word, due to the 3-cycle misalignment penalty which
- * would occur half the time. This also turns out to be cheaper than
- * doing two separate 8-bit accesses, as the memory is so rarely in the
- * L1 cache.
- *
- * The window buffer, however, apparently spends a lot of time in the
- * cache, and so it is faster to retrieve the word at the end of the
- * match string with two 8-bit loads. The instructions that test the
- * word at the beginning of the match string, however, are executed
- * much less frequently, and there it was cheaper to use 16-bit
- * instructions, which avoided the necessity of saving off and
- * subsequently reloading one of the other registers.
- */
-LookupLoop:
- /* 1 U & V */
- movw (%edi,%edx,2), %cx /* 2 U pipe */
- movl wmask(%esp), %edx /* 2 V pipe */
- cmpl %ebp, %ecx /* 3 U pipe */
- jbe LeaveNow /* 3 V pipe */
- subl $0x00010000, %ebx /* 4 U pipe */
- js LeaveNow /* 4 V pipe */
-LoopEntry: movb -1(%esi,%ecx), %al /* 5 U pipe */
- andl %ecx, %edx /* 5 V pipe */
- cmpb %bl, %al /* 6 U pipe */
- jnz LookupLoop /* 6 V pipe */
- movb (%esi,%ecx), %ah
- cmpb %bh, %ah
- jnz LookupLoop
- movl window(%esp), %eax
- movw (%eax,%ecx), %ax
- cmpw scanstart(%esp), %ax
- jnz LookupLoop
-
-/* Store the current value of chainlen. */
-
- movl %ebx, chainlenscanend(%esp)
-
-/* 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). */
-
- movl window(%esp), %esi
- movl scan(%esp), %edi
- addl %ecx, %esi
- movl scanalign(%esp), %eax
- movl $(-MAX_MATCH_8), %edx
- lea MAX_MATCH_8(%edi,%eax), %edi
- lea MAX_MATCH_8(%esi,%eax), %esi
-
-/* Test the strings for equality, 8 bytes at a time. At the end,
- * adjust %edx 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. (%esi 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:
- movl (%esi,%edx), %eax
- movl (%edi,%edx), %ebx
- xorl %ebx, %eax
- jnz LeaveLoopCmps
- movl 4(%esi,%edx), %eax
- movl 4(%edi,%edx), %ebx
- xorl %ebx, %eax
- jnz LeaveLoopCmps4
- addl $8, %edx
- jnz LoopCmps
- jmp LenMaximum
-LeaveLoopCmps4: addl $4, %edx
-LeaveLoopCmps: testl $0x0000FFFF, %eax
- jnz LenLower
- addl $2, %edx
- shrl $16, %eax
-LenLower: subb $1, %al
- adcl $0, %edx
-
-/* 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 (%edi,%edx), %eax
- movl scan(%esp), %edi
- subl %edi, %eax
- cmpl $MAX_MATCH, %eax
- 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. */
-
- movl deflatestate(%esp), %edx
- movl bestlen(%esp), %ebx
- cmpl %ebx, %eax
- jg LongerMatch
- movl chainlenscanend(%esp), %ebx
- movl windowbestlen(%esp), %esi
- movl dsPrev(%edx), %edi
- movl wmask(%esp), %edx
- andl %ecx, %edx
- jmp LookupLoop
-
-/* s->match_start = cur_match; */
-/* best_len = len; */
-/* if (len >= nice_match) break; */
-/* scan_end = *(ushf*)(scan+best_len-1); */
-
-LongerMatch: movl nicematch(%esp), %ebx
- movl %eax, bestlen(%esp)
- movl %ecx, dsMatchStart(%edx)
- cmpl %ebx, %eax
- jge LeaveNow
- movl window(%esp), %esi
- addl %eax, %esi
- movl %esi, windowbestlen(%esp)
- movl chainlenscanend(%esp), %ebx
- movw -1(%edi,%eax), %bx
- movl dsPrev(%edx), %edi
- movl %ebx, chainlenscanend(%esp)
- movl wmask(%esp), %edx
- andl %ecx, %edx
- jmp LookupLoop
-
-/* Accept the current string, with the maximum possible length. */
-
-LenMaximum: movl deflatestate(%esp), %edx
- movl $MAX_MATCH, bestlen(%esp)
- movl %ecx, dsMatchStart(%edx)
-
-/* if ((uInt)best_len <= s->lookahead) return (uInt)best_len; */
-/* return s->lookahead; */
-
-LeaveNow:
- movl deflatestate(%esp), %edx
- movl bestlen(%esp), %ebx
- movl dsLookahead(%edx), %eax
- cmpl %eax, %ebx
- jg LookaheadRet
- movl %ebx, %eax
-LookaheadRet:
-
-/* Restore the stack and return from whence we came. */
-
- addl $LocalVarsSize, %esp
- popl %ebx
- popl %esi
- popl %edi
- popl %ebp
-match_init: ret
diff --git a/compat/zlib/contrib/masm686/match.asm b/compat/zlib/contrib/masm686/match.asm
deleted file mode 100644
index 4b03a71..0000000
--- a/compat/zlib/contrib/masm686/match.asm
+++ /dev/null
@@ -1,413 +0,0 @@
-
-; match.asm -- Pentium-Pro optimized version of longest_match()
-;
-; Updated for zlib 1.1.3 and converted to MASM 6.1x
-; Copyright (C) 2000 Dan Higdon <hdan@kinesoft.com>
-; and Chuck Walbourn <chuckw@kinesoft.com>
-; Corrections by Cosmin Truta <cosmint@cs.ubbcluj.ro>
-;
-; This is free software; you can redistribute it and/or modify it
-; under the terms of the GNU General Public License.
-
-; Based on match.S
-; Written for zlib 1.1.2
-; Copyright (C) 1998 Brian Raiter <breadbox@muppetlabs.com>
-;
-; Modified by Gilles Vollant (2005) for add gzhead and gzindex
-
- .686P
- .MODEL FLAT
-
-;===========================================================================
-; EQUATES
-;===========================================================================
-
-MAX_MATCH EQU 258
-MIN_MATCH EQU 3
-MIN_LOOKAHEAD EQU (MAX_MATCH + MIN_MATCH + 1)
-MAX_MATCH_8 EQU ((MAX_MATCH + 7) AND (NOT 7))
-
-;===========================================================================
-; STRUCTURES
-;===========================================================================
-
-; This STRUCT assumes a 4-byte alignment
-
-DEFLATE_STATE STRUCT
-ds_strm dd ?
-ds_status dd ?
-ds_pending_buf dd ?
-ds_pending_buf_size dd ?
-ds_pending_out dd ?
-ds_pending dd ?
-ds_wrap dd ?
-; gzhead and gzindex are added in zlib 1.2.2.2 (see deflate.h)
-ds_gzhead dd ?
-ds_gzindex dd ?
-ds_data_type db ?
-ds_method db ?
- db ? ; padding
- db ? ; padding
-ds_last_flush dd ?
-ds_w_size dd ? ; used
-ds_w_bits dd ?
-ds_w_mask dd ? ; used
-ds_window dd ? ; used
-ds_window_size dd ?
-ds_prev dd ? ; used
-ds_head dd ?
-ds_ins_h dd ?
-ds_hash_size dd ?
-ds_hash_bits dd ?
-ds_hash_mask dd ?
-ds_hash_shift dd ?
-ds_block_start dd ?
-ds_match_length dd ? ; used
-ds_prev_match dd ? ; used
-ds_match_available dd ?
-ds_strstart dd ? ; used
-ds_match_start dd ? ; used
-ds_lookahead dd ? ; used
-ds_prev_length dd ? ; used
-ds_max_chain_length dd ? ; used
-ds_max_laxy_match dd ?
-ds_level dd ?
-ds_strategy dd ?
-ds_good_match dd ? ; used
-ds_nice_match dd ? ; used
-
-; Don't need anymore of the struct for match
-DEFLATE_STATE ENDS
-
-;===========================================================================
-; CODE
-;===========================================================================
-_TEXT SEGMENT
-
-;---------------------------------------------------------------------------
-; match_init
-;---------------------------------------------------------------------------
- ALIGN 4
-PUBLIC _match_init
-_match_init PROC
- ; no initialization needed
- ret
-_match_init ENDP
-
-;---------------------------------------------------------------------------
-; uInt longest_match(deflate_state *deflatestate, IPos curmatch)
-;---------------------------------------------------------------------------
- ALIGN 4
-
-PUBLIC _longest_match
-_longest_match PROC
-
-; Since this code uses EBP for a scratch register, the stack frame must
-; be manually constructed and referenced relative to the ESP register.
-
-; Stack image
-; Variables
-chainlenwmask = 0 ; high word: current chain len
- ; low word: s->wmask
-window = 4 ; local copy of s->window
-windowbestlen = 8 ; s->window + bestlen
-scanend = 12 ; last two bytes of string
-scanstart = 16 ; first two bytes of string
-scanalign = 20 ; dword-misalignment of string
-nicematch = 24 ; a good enough match size
-bestlen = 28 ; size of best match so far
-scan = 32 ; ptr to string wanting match
-varsize = 36 ; number of bytes (also offset to last saved register)
-
-; Saved Registers (actually pushed into place)
-ebx_save = 36
-edi_save = 40
-esi_save = 44
-ebp_save = 48
-
-; Parameters
-retaddr = 52
-deflatestate = 56
-curmatch = 60
-
-; Save registers that the compiler may be using
- push ebp
- push edi
- push esi
- push ebx
-
-; Allocate local variable space
- sub esp,varsize
-
-; Retrieve the function arguments. ecx 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).
-
- mov edx, [esp+deflatestate]
-ASSUME edx:PTR DEFLATE_STATE
-
- mov ecx, [esp+curmatch]
-
-; uInt wmask = s->w_mask;
-; unsigned chain_length = s->max_chain_length;
-; if (s->prev_length >= s->good_match) {
-; chain_length >>= 2;
-; }
-
- mov eax, [edx].ds_prev_length
- mov ebx, [edx].ds_good_match
- cmp eax, ebx
- mov eax, [edx].ds_w_mask
- mov ebx, [edx].ds_max_chain_length
- jl SHORT 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
- mov [esp+chainlenwmask], ebx
-
-; if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
-
- mov eax, [edx].ds_nice_match
- mov ebx, [edx].ds_lookahead
- cmp ebx, eax
- jl SHORT LookaheadLess
- mov ebx, eax
-LookaheadLess:
- mov [esp+nicematch], ebx
-
-;/* register Bytef *scan = s->window + s->strstart; */
-
- mov esi, [edx].ds_window
- mov [esp+window], esi
- mov ebp, [edx].ds_strstart
- lea edi, [esi+ebp]
- mov [esp+scan],edi
-
-;/* Determine how many bytes the scan ptr is off from being */
-;/* dword-aligned. */
-
- mov eax, edi
- neg eax
- and eax, 3
- mov [esp+scanalign], eax
-
-;/* IPos limit = s->strstart > (IPos)MAX_DIST(s) ? */
-;/* s->strstart - (IPos)MAX_DIST(s) : NIL; */
-
- mov eax, [edx].ds_w_size
- sub eax, MIN_LOOKAHEAD
- sub ebp, eax
- jg SHORT LimitPositive
- xor ebp, ebp
-LimitPositive:
-
-;/* int best_len = s->prev_length; */
-
- mov eax, [edx].ds_prev_length
- mov [esp+bestlen], eax
-
-;/* Store the sum of s->window + best_len in %esi locally, and in %esi. */
-
- add esi, eax
- mov [esp+windowbestlen], esi
-
-;/* register ush scan_start = *(ushf*)scan; */
-;/* register ush scan_end = *(ushf*)(scan+best_len-1); */
-;/* Posf *prev = s->prev; */
-
- movzx ebx, WORD PTR[edi]
- mov [esp+scanstart], ebx
- movzx ebx, WORD PTR[eax+edi-1]
- mov [esp+scanend], ebx
- mov edi, [edx].ds_prev
-
-;/* Jump into the main loop. */
-
- mov edx, [esp+chainlenwmask]
- jmp SHORT LoopEntry
-
-;/* 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
-; * %ecx = curmatch
-; * %edx = chainlenwmask - i.e., ((chainlen << 16) | wmask)
-; * %esi = windowbestlen - i.e., (window + bestlen)
-; * %edi = prev
-; * %ebp = limit
-; */
-
- ALIGN 4
-LookupLoop:
- and ecx, edx
- movzx ecx, WORD PTR[edi+ecx*2]
- cmp ecx, ebp
- jbe LeaveNow
- sub edx, 000010000H
- js LeaveNow
-
-LoopEntry:
- movzx eax, WORD PTR[esi+ecx-1]
- cmp eax, ebx
- jnz SHORT LookupLoop
-
- mov eax, [esp+window]
- movzx eax, WORD PTR[eax+ecx]
- cmp eax, [esp+scanstart]
- jnz SHORT LookupLoop
-
-;/* Store the current value of chainlen. */
-
- mov [esp+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). */
-
- mov esi, [esp+window]
- mov edi, [esp+scan]
- add esi, ecx
- mov eax, [esp+scanalign]
- mov edx, -MAX_MATCH_8
- lea edi, [edi+eax+MAX_MATCH_8]
- lea esi, [esi+eax+MAX_MATCH_8]
-
-;/* Test the strings for equality, 8 bytes at a time. At the end,
-; * adjust %edx 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. (%esi 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 eax, DWORD PTR[esi+edx]
- xor eax, DWORD PTR[edi+edx]
- jnz SHORT LeaveLoopCmps
-
- mov eax, DWORD PTR[esi+edx+4]
- xor eax, DWORD PTR[edi+edx+4]
- jnz SHORT LeaveLoopCmps4
-
- add edx, 8
- jnz SHORT LoopCmps
- jmp LenMaximum
- ALIGN 4
-
-LeaveLoopCmps4:
- add edx, 4
-
-LeaveLoopCmps:
- test eax, 00000FFFFH
- jnz SHORT LenLower
-
- add edx, 2
- shr eax, 16
-
-LenLower:
- sub al, 1
- adc edx, 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 eax, [edi+edx]
- mov edi, [esp+scan]
- sub eax, edi
- cmp eax, MAX_MATCH
- jge SHORT 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. */
-
- mov edx, [esp+deflatestate]
- mov ebx, [esp+bestlen]
- cmp eax, ebx
- jg SHORT LongerMatch
- mov esi, [esp+windowbestlen]
- mov edi, [edx].ds_prev
- mov ebx, [esp+scanend]
- mov edx, [esp+chainlenwmask]
- jmp LookupLoop
- ALIGN 4
-
-;/* s->match_start = cur_match; */
-;/* best_len = len; */
-;/* if (len >= nice_match) break; */
-;/* scan_end = *(ushf*)(scan+best_len-1); */
-
-LongerMatch:
- mov ebx, [esp+nicematch]
- mov [esp+bestlen], eax
- mov [edx].ds_match_start, ecx
- cmp eax, ebx
- jge SHORT LeaveNow
- mov esi, [esp+window]
- add esi, eax
- mov [esp+windowbestlen], esi
- movzx ebx, WORD PTR[edi+eax-1]
- mov edi, [edx].ds_prev
- mov [esp+scanend], ebx
- mov edx, [esp+chainlenwmask]
- jmp LookupLoop
- ALIGN 4
-
-;/* Accept the current string, with the maximum possible length. */
-
-LenMaximum:
- mov edx, [esp+deflatestate]
- mov DWORD PTR[esp+bestlen], MAX_MATCH
- mov [edx].ds_match_start, ecx
-
-;/* if ((uInt)best_len <= s->lookahead) return (uInt)best_len; */
-;/* return s->lookahead; */
-
-LeaveNow:
- mov edx, [esp+deflatestate]
- mov ebx, [esp+bestlen]
- mov eax, [edx].ds_lookahead
- cmp ebx, eax
- jg SHORT LookaheadRet
- mov eax, ebx
-LookaheadRet:
-
-; Restore the stack and return from whence we came.
-
- add esp, varsize
- pop ebx
- pop esi
- pop edi
- pop ebp
- ret
-
-_longest_match ENDP
-
-_TEXT ENDS
-END