--- /dev/null
+/*
+ * match.S -- optimized version of longest_match()
+ * based on the similar work by Gilles Vollant, and Brian Raiter, written 1998
+ *
+ * This is free software; you can redistribute it and/or modify it
+ * under the terms of the BSD License. Use by owners of Che Guevarra
+ * parafernalia is prohibited, where possible, and highly discouraged
+ * elsewhere.
+ */
+
+#ifndef NO_UNDERLINE
+# define match_init _match_init
+# define longest_match _longest_match
+#endif
+
+#define scanend ebx
+#define scanendw bx
+#define chainlenwmask edx /* high word: current chain len low word: s->wmask */
+#define curmatch rsi
+#define curmatchd esi
+#define windowbestlen r8
+#define scanalign r9
+#define scanalignd r9d
+#define window r10
+#define bestlen r11
+#define bestlend r11d
+#define scanstart r12d
+#define scanstartw r12w
+#define scan r13
+#define nicematch r14d
+#define limit r15
+#define limitd r15d
+#define prev rcx
+
+/*
+ * The 258 is a "magic number, not a parameter -- changing it
+ * breaks the hell loose
+ */
+#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 LocalVarsSize (112)
+#define _chainlenwmask ( 8-LocalVarsSize)(%rsp)
+#define _windowbestlen (16-LocalVarsSize)(%rsp)
+#define save_r14 (24-LocalVarsSize)(%rsp)
+#define save_rsi (32-LocalVarsSize)(%rsp)
+#define save_rbx (40-LocalVarsSize)(%rsp)
+#define save_r12 (56-LocalVarsSize)(%rsp)
+#define save_r13 (64-LocalVarsSize)(%rsp)
+#define save_r15 (80-LocalVarsSize)(%rsp)
+
+
+.globl match_init, longest_match
+
+/*
+ * On AMD64 the first argument of a function (in our case -- the pointer to
+ * deflate_state structure) is passed in %rdi, hence our offsets below are
+ * all off of that.
+ */
+
+/* you can check the structure offset by running
+
+#include <stdlib.h>
+#include <stdio.h>
+#include "deflate.h"
+
+void print_depl()
+{
+deflate_state ds;
+deflate_state *s=&ds;
+printf("size pointer=%u\n",(int)sizeof(void*));
+
+printf("#define dsWSize (%3u)(%%rdi)\n",(int)(((char*)&(s->w_size))-((char*)s)));
+printf("#define dsWMask (%3u)(%%rdi)\n",(int)(((char*)&(s->w_mask))-((char*)s)));
+printf("#define dsWindow (%3u)(%%rdi)\n",(int)(((char*)&(s->window))-((char*)s)));
+printf("#define dsPrev (%3u)(%%rdi)\n",(int)(((char*)&(s->prev))-((char*)s)));
+printf("#define dsMatchLen (%3u)(%%rdi)\n",(int)(((char*)&(s->match_length))-((char*)s)));
+printf("#define dsPrevMatch (%3u)(%%rdi)\n",(int)(((char*)&(s->prev_match))-((char*)s)));
+printf("#define dsStrStart (%3u)(%%rdi)\n",(int)(((char*)&(s->strstart))-((char*)s)));
+printf("#define dsMatchStart (%3u)(%%rdi)\n",(int)(((char*)&(s->match_start))-((char*)s)));
+printf("#define dsLookahead (%3u)(%%rdi)\n",(int)(((char*)&(s->lookahead))-((char*)s)));
+printf("#define dsPrevLen (%3u)(%%rdi)\n",(int)(((char*)&(s->prev_length))-((char*)s)));
+printf("#define dsMaxChainLen (%3u)(%%rdi)\n",(int)(((char*)&(s->max_chain_length))-((char*)s)));
+printf("#define dsGoodMatch (%3u)(%%rdi)\n",(int)(((char*)&(s->good_match))-((char*)s)));
+printf("#define dsNiceMatch (%3u)(%%rdi)\n",(int)(((char*)&(s->nice_match))-((char*)s)));
+}
+
+*/
+
+
+/*
+ to compile for XCode 3.2 on MacOSX x86_64
+ - run "gcc -g -c -DXCODE_MAC_X64_STRUCTURE amd64-match.S"
+ */
+
+
+#ifndef CURRENT_LINX_XCODE_MAC_X64_STRUCTURE
+#define dsWSize ( 68)(%rdi)
+#define dsWMask ( 76)(%rdi)
+#define dsWindow ( 80)(%rdi)
+#define dsPrev ( 96)(%rdi)
+#define dsMatchLen (144)(%rdi)
+#define dsPrevMatch (148)(%rdi)
+#define dsStrStart (156)(%rdi)
+#define dsMatchStart (160)(%rdi)
+#define dsLookahead (164)(%rdi)
+#define dsPrevLen (168)(%rdi)
+#define dsMaxChainLen (172)(%rdi)
+#define dsGoodMatch (188)(%rdi)
+#define dsNiceMatch (192)(%rdi)
+
+#else
+
+#ifndef STRUCT_OFFSET
+# define STRUCT_OFFSET (0)
+#endif
+
+
+#define dsWSize ( 56 + STRUCT_OFFSET)(%rdi)
+#define dsWMask ( 64 + STRUCT_OFFSET)(%rdi)
+#define dsWindow ( 72 + STRUCT_OFFSET)(%rdi)
+#define dsPrev ( 88 + STRUCT_OFFSET)(%rdi)
+#define dsMatchLen (136 + STRUCT_OFFSET)(%rdi)
+#define dsPrevMatch (140 + STRUCT_OFFSET)(%rdi)
+#define dsStrStart (148 + STRUCT_OFFSET)(%rdi)
+#define dsMatchStart (152 + STRUCT_OFFSET)(%rdi)
+#define dsLookahead (156 + STRUCT_OFFSET)(%rdi)
+#define dsPrevLen (160 + STRUCT_OFFSET)(%rdi)
+#define dsMaxChainLen (164 + STRUCT_OFFSET)(%rdi)
+#define dsGoodMatch (180 + STRUCT_OFFSET)(%rdi)
+#define dsNiceMatch (184 + STRUCT_OFFSET)(%rdi)
+
+#endif
+
+
+
+
+.text
+
+/* uInt longest_match(deflate_state *deflatestate, IPos curmatch) */
+
+longest_match:
+/*
+ * Retrieve the function arguments. %curmatch will hold cur_match
+ * throughout the entire function (passed via rsi on amd64).
+ * rdi will hold the pointer to the deflate_state (first arg on amd64)
+ */
+ mov %rsi, save_rsi
+ mov %rbx, save_rbx
+ mov %r12, save_r12
+ mov %r13, save_r13
+ mov %r14, save_r14
+ mov %r15, save_r15
+
+/* uInt wmask = s->w_mask; */
+/* unsigned chain_length = s->max_chain_length; */
+/* if (s->prev_length >= s->good_match) { */
+/* chain_length >>= 2; */
+/* } */
+
+ movl dsPrevLen, %eax
+ movl dsGoodMatch, %ebx
+ cmpl %ebx, %eax
+ movl dsWMask, %eax
+ movl dsMaxChainLen, %chainlenwmask
+ jl LastMatchGood
+ shrl $2, %chainlenwmask
+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. */
+
+ decl %chainlenwmask
+ shll $16, %chainlenwmask
+ orl %eax, %chainlenwmask
+
+/* if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; */
+
+ movl dsNiceMatch, %eax
+ movl dsLookahead, %ebx
+ cmpl %eax, %ebx
+ jl LookaheadLess
+ movl %eax, %ebx
+LookaheadLess: movl %ebx, %nicematch
+
+/* register Bytef *scan = s->window + s->strstart; */
+
+ mov dsWindow, %window
+ movl dsStrStart, %limitd
+ lea (%limit, %window), %scan
+
+/* Determine how many bytes the scan ptr is off from being */
+/* dword-aligned. */
+
+ mov %scan, %scanalign
+ negl %scanalignd
+ andl $3, %scanalignd
+
+/* IPos limit = s->strstart > (IPos)MAX_DIST(s) ? */
+/* s->strstart - (IPos)MAX_DIST(s) : NIL; */
+
+ movl dsWSize, %eax
+ subl $MIN_LOOKAHEAD, %eax
+ xorl %ecx, %ecx
+ subl %eax, %limitd
+ cmovng %ecx, %limitd
+
+/* int best_len = s->prev_length; */
+
+ movl dsPrevLen, %bestlend
+
+/* Store the sum of s->window + best_len in %windowbestlen locally, and in memory. */
+
+ lea (%window, %bestlen), %windowbestlen
+ mov %windowbestlen, _windowbestlen
+
+/* register ush scan_start = *(ushf*)scan; */
+/* register ush scan_end = *(ushf*)(scan+best_len-1); */
+/* Posf *prev = s->prev; */
+
+ movzwl (%scan), %scanstart
+ movzwl -1(%scan, %bestlen), %scanend
+ mov dsPrev, %prev
+
+/* Jump into the main loop. */
+
+ movl %chainlenwmask, _chainlenwmask
+ 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.
+ */
+LookupLoop:
+ andl %chainlenwmask, %curmatchd
+ movzwl (%prev, %curmatch, 2), %curmatchd
+ cmpl %limitd, %curmatchd
+ jbe LeaveNow
+ subl $0x00010000, %chainlenwmask
+ js LeaveNow
+LoopEntry: cmpw -1(%windowbestlen, %curmatch), %scanendw
+ jne LookupLoop
+ cmpw %scanstartw, (%window, %curmatch)
+ jne LookupLoop
+
+/* Store the current value of chainlen. */
+ movl %chainlenwmask, _chainlenwmask
+
+/* %scan is the string under scrutiny, and %prev 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 $(-MAX_MATCH_8), %rdx
+ lea (%curmatch, %window), %windowbestlen
+ lea MAX_MATCH_8(%windowbestlen, %scanalign), %windowbestlen
+ lea MAX_MATCH_8(%scan, %scanalign), %prev
+
+/* the prefetching below makes very little difference... */
+ prefetcht1 (%windowbestlen, %rdx)
+ prefetcht1 (%prev, %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.
+ *
+ * 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 -- unrolling it, for example, makes no difference.
+ */
+
+#undef USE_SSE /* works, but is 6-7% slower, than non-SSE... */
+
+LoopCmps:
+#ifdef USE_SSE
+ /* Preload the SSE registers */
+ movdqu (%windowbestlen, %rdx), %xmm1
+ movdqu (%prev, %rdx), %xmm2
+ pcmpeqb %xmm2, %xmm1
+ movdqu 16(%windowbestlen, %rdx), %xmm3
+ movdqu 16(%prev, %rdx), %xmm4
+ pcmpeqb %xmm4, %xmm3
+ movdqu 32(%windowbestlen, %rdx), %xmm5
+ movdqu 32(%prev, %rdx), %xmm6
+ pcmpeqb %xmm6, %xmm5
+ movdqu 48(%windowbestlen, %rdx), %xmm7
+ movdqu 48(%prev, %rdx), %xmm8
+ pcmpeqb %xmm8, %xmm7
+
+ /* Check the comparisions' results */
+ pmovmskb %xmm1, %rax
+ notw %ax
+ bsfw %ax, %ax
+ jnz LeaveLoopCmps
+
+ /* this is the only iteration of the loop with a possibility of having
+ incremented rdx by 0x108 (each loop iteration add 16*4 = 0x40
+ and (0x40*4)+8=0x108 */
+ add $8, %rdx
+ jz LenMaximum
+ add $8, %rdx
+
+
+ pmovmskb %xmm3, %rax
+ notw %ax
+ bsfw %ax, %ax
+ jnz LeaveLoopCmps
+
+
+ add $16, %rdx
+
+
+ pmovmskb %xmm5, %rax
+ notw %ax
+ bsfw %ax, %ax
+ jnz LeaveLoopCmps
+
+ add $16, %rdx
+
+
+ pmovmskb %xmm7, %rax
+ notw %ax
+ bsfw %ax, %ax
+ jnz LeaveLoopCmps
+
+ add $16, %rdx
+
+ jmp LoopCmps
+LeaveLoopCmps: add %rax, %rdx
+#else
+ mov (%windowbestlen, %rdx), %rax
+ xor (%prev, %rdx), %rax
+ jnz LeaveLoopCmps
+
+ mov 8(%windowbestlen, %rdx), %rax
+ xor 8(%prev, %rdx), %rax
+ jnz LeaveLoopCmps8
+
+ mov 16(%windowbestlen, %rdx), %rax
+ xor 16(%prev, %rdx), %rax
+ jnz LeaveLoopCmps16
+
+ add $24, %rdx
+ jnz LoopCmps
+ jmp LenMaximum
+# if 0
+/*
+ * This three-liner is tantalizingly simple, but bsf is a slow instruction,
+ * and the complicated alternative down below is quite a bit faster. Sad...
+ */
+
+LeaveLoopCmps: bsf %rax, %rax /* find the first non-zero bit */
+ shrl $3, %eax /* divide by 8 to get the byte */
+ add %rax, %rdx
+# else
+LeaveLoopCmps16:
+ add $8, %rdx
+LeaveLoopCmps8:
+ add $8, %rdx
+LeaveLoopCmps: testl $0xFFFFFFFF, %eax /* Check the first 4 bytes */
+ jnz Check16
+ add $4, %rdx
+ shr $32, %rax
+Check16: testw $0xFFFF, %ax
+ jnz LenLower
+ add $2, %rdx
+ shrl $16, %eax
+LenLower: subb $1, %al
+ adc $0, %rdx
+# endif
+#endif
+
+/* 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 (%prev, %rdx), %rax
+ sub %scan, %rax
+ 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. */
+
+ cmpl %bestlend, %eax
+ jg LongerMatch
+ mov _windowbestlen, %windowbestlen
+ mov dsPrev, %prev
+ movl _chainlenwmask, %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 %eax, %bestlend
+ movl %curmatchd, dsMatchStart
+ cmpl %nicematch, %eax
+ jge LeaveNow
+
+ lea (%window, %bestlen), %windowbestlen
+ mov %windowbestlen, _windowbestlen
+
+ movzwl -1(%scan, %rax), %scanend
+ mov dsPrev, %prev
+ movl _chainlenwmask, %chainlenwmask
+ jmp LookupLoop
+
+/* Accept the current string, with the maximum possible length. */
+
+LenMaximum:
+ movl $MAX_MATCH, %bestlend
+ movl %curmatchd, dsMatchStart
+
+/* if ((uInt)best_len <= s->lookahead) return (uInt)best_len; */
+/* return s->lookahead; */
+
+LeaveNow:
+ movl dsLookahead, %eax
+ cmpl %eax, %bestlend
+ cmovngl %bestlend, %eax
+LookaheadRet:
+
+/* Restore the registers and return from whence we came. */
+
+ mov save_rsi, %rsi
+ mov save_rbx, %rbx
+ mov save_r12, %r12
+ mov save_r13, %r13
+ mov save_r14, %r14
+ mov save_r15, %r15
+
+ ret
+
+match_init: ret
projects / ossec-hids.git / blobdiff