diff options
| author | Mark Adler <zlib@madler.net> | 2017-10-12 20:08:53 -0700 |
|---|---|---|
| committer | Mark Adler <zlib@madler.net> | 2017-10-12 20:27:14 -0700 |
| commit | 288f1080317b954b6bdca33708631c011549c008 (patch) | |
| tree | 9629f01104722ba8e490f04a0790c56513ba989a | |
| parent | a5773513942b1c57d0eff51fcb2ebac72796ed95 (diff) | |
| download | zlib-288f1080317b954b6bdca33708631c011549c008.tar.gz zlib-288f1080317b954b6bdca33708631c011549c008.tar.bz2 zlib-288f1080317b954b6bdca33708631c011549c008.zip | |
Remove old assembler code in which bugs have manifested.
In addition, there is not sufficient gain from the inflate
assembler code to warrant its inclusion.
| -rw-r--r-- | contrib/README.contrib | 21 | ||||
| -rw-r--r-- | contrib/amd64/amd64-match.S | 452 | ||||
| -rw-r--r-- | contrib/asm686/README.686 | 51 | ||||
| -rw-r--r-- | contrib/asm686/match.S | 357 | ||||
| -rw-r--r-- | contrib/inflate86/inffas86.c | 1157 | ||||
| -rw-r--r-- | contrib/inflate86/inffast.S | 1368 | ||||
| -rw-r--r-- | contrib/masmx64/bld_ml64.bat | 2 | ||||
| -rw-r--r-- | contrib/masmx64/gvmat64.asm | 553 | ||||
| -rw-r--r-- | contrib/masmx64/inffas8664.c | 186 | ||||
| -rw-r--r-- | contrib/masmx64/inffasx64.asm | 396 | ||||
| -rw-r--r-- | contrib/masmx64/readme.txt | 31 | ||||
| -rw-r--r-- | contrib/masmx86/bld_ml32.bat | 2 | ||||
| -rw-r--r-- | contrib/masmx86/inffas32.asm | 1080 | ||||
| -rw-r--r-- | contrib/masmx86/match686.asm | 479 | ||||
| -rw-r--r-- | contrib/masmx86/readme.txt | 27 | ||||
| -rw-r--r-- | win32/Makefile.bor | 1 | ||||
| -rw-r--r-- | win32/Makefile.gcc | 5 | ||||
| -rw-r--r-- | win32/Makefile.msc | 4 |
18 files changed, 0 insertions, 6172 deletions
diff --git a/contrib/README.contrib b/contrib/README.contrib index a411d5c..335e435 100644 --- a/contrib/README.contrib +++ b/contrib/README.contrib | |||
| @@ -8,14 +8,6 @@ ada/ by Dmitriy Anisimkov <anisimkov@yahoo.com> | |||
| 8 | Support for Ada | 8 | Support for Ada |
| 9 | See http://zlib-ada.sourceforge.net/ | 9 | See http://zlib-ada.sourceforge.net/ |
| 10 | 10 | ||
| 11 | amd64/ by Mikhail Teterin <mi@ALDAN.algebra.com> | ||
| 12 | asm code for AMD64 | ||
| 13 | See patch at http://www.freebsd.org/cgi/query-pr.cgi?pr=bin/96393 | ||
| 14 | |||
| 15 | asm686/ by Brian Raiter <breadbox@muppetlabs.com> | ||
| 16 | asm code for Pentium and PPro/PII, using the AT&T (GNU as) syntax | ||
| 17 | See http://www.muppetlabs.com/~breadbox/software/assembly.html | ||
| 18 | |||
| 19 | blast/ by Mark Adler <madler@alumni.caltech.edu> | 11 | blast/ by Mark Adler <madler@alumni.caltech.edu> |
| 20 | Decompressor for output of PKWare Data Compression Library (DCL) | 12 | Decompressor for output of PKWare Data Compression Library (DCL) |
| 21 | 13 | ||
| @@ -32,9 +24,6 @@ gcc_gvmat64/by Gilles Vollant <info@winimage.com> | |||
| 32 | infback9/ by Mark Adler <madler@alumni.caltech.edu> | 24 | infback9/ by Mark Adler <madler@alumni.caltech.edu> |
| 33 | Unsupported diffs to infback to decode the deflate64 format | 25 | Unsupported diffs to infback to decode the deflate64 format |
| 34 | 26 | ||
| 35 | inflate86/ by Chris Anderson <christop@charm.net> | ||
| 36 | Tuned x86 gcc asm code to replace inflate_fast() | ||
| 37 | |||
| 38 | iostream/ by Kevin Ruland <kevin@rodin.wustl.edu> | 27 | iostream/ by Kevin Ruland <kevin@rodin.wustl.edu> |
| 39 | A C++ I/O streams interface to the zlib gz* functions | 28 | A C++ I/O streams interface to the zlib gz* functions |
| 40 | 29 | ||
| @@ -45,16 +34,6 @@ iostream3/ by Ludwig Schwardt <schwardt@sun.ac.za> | |||
| 45 | and Kevin Ruland <kevin@rodin.wustl.edu> | 34 | and Kevin Ruland <kevin@rodin.wustl.edu> |
| 46 | Yet another C++ I/O streams interface | 35 | Yet another C++ I/O streams interface |
| 47 | 36 | ||
| 48 | masmx64/ by Gilles Vollant <info@winimage.com> | ||
| 49 | x86 64-bit (AMD64 and Intel EM64t) code for x64 assembler to | ||
| 50 | replace longest_match() and inflate_fast(), also masm x86 | ||
| 51 | 64-bits translation of Chris Anderson inflate_fast() | ||
| 52 | |||
| 53 | masmx86/ by Gilles Vollant <info@winimage.com> | ||
| 54 | x86 asm code to replace longest_match() and inflate_fast(), | ||
| 55 | for Visual C++ and MASM (32 bits). | ||
| 56 | Based on Brian Raiter (asm686) and Chris Anderson (inflate86) | ||
| 57 | |||
| 58 | minizip/ by Gilles Vollant <info@winimage.com> | 37 | minizip/ by Gilles Vollant <info@winimage.com> |
| 59 | Mini zip and unzip based on zlib | 38 | Mini zip and unzip based on zlib |
| 60 | Includes Zip64 support by Mathias Svensson <mathias@result42.com> | 39 | Includes Zip64 support by Mathias Svensson <mathias@result42.com> |
diff --git a/contrib/amd64/amd64-match.S b/contrib/amd64/amd64-match.S deleted file mode 100644 index 81d4a1c..0000000 --- a/contrib/amd64/amd64-match.S +++ /dev/null | |||
| @@ -1,452 +0,0 @@ | |||
| 1 | /* | ||
| 2 | * match.S -- optimized version of longest_match() | ||
| 3 | * based on the similar work by Gilles Vollant, and Brian Raiter, written 1998 | ||
| 4 | * | ||
| 5 | * This is free software; you can redistribute it and/or modify it | ||
| 6 | * under the terms of the BSD License. Use by owners of Che Guevarra | ||
| 7 | * parafernalia is prohibited, where possible, and highly discouraged | ||
| 8 | * elsewhere. | ||
| 9 | */ | ||
| 10 | |||
| 11 | #ifndef NO_UNDERLINE | ||
| 12 | # define match_init _match_init | ||
| 13 | # define longest_match _longest_match | ||
| 14 | #endif | ||
| 15 | |||
| 16 | #define scanend ebx | ||
| 17 | #define scanendw bx | ||
| 18 | #define chainlenwmask edx /* high word: current chain len low word: s->wmask */ | ||
| 19 | #define curmatch rsi | ||
| 20 | #define curmatchd esi | ||
| 21 | #define windowbestlen r8 | ||
| 22 | #define scanalign r9 | ||
| 23 | #define scanalignd r9d | ||
| 24 | #define window r10 | ||
| 25 | #define bestlen r11 | ||
| 26 | #define bestlend r11d | ||
| 27 | #define scanstart r12d | ||
| 28 | #define scanstartw r12w | ||
| 29 | #define scan r13 | ||
| 30 | #define nicematch r14d | ||
| 31 | #define limit r15 | ||
| 32 | #define limitd r15d | ||
| 33 | #define prev rcx | ||
| 34 | |||
| 35 | /* | ||
| 36 | * The 258 is a "magic number, not a parameter -- changing it | ||
| 37 | * breaks the hell loose | ||
| 38 | */ | ||
| 39 | #define MAX_MATCH (258) | ||
| 40 | #define MIN_MATCH (3) | ||
| 41 | #define MIN_LOOKAHEAD (MAX_MATCH + MIN_MATCH + 1) | ||
| 42 | #define MAX_MATCH_8 ((MAX_MATCH + 7) & ~7) | ||
| 43 | |||
| 44 | /* stack frame offsets */ | ||
| 45 | #define LocalVarsSize (112) | ||
| 46 | #define _chainlenwmask ( 8-LocalVarsSize)(%rsp) | ||
| 47 | #define _windowbestlen (16-LocalVarsSize)(%rsp) | ||
| 48 | #define save_r14 (24-LocalVarsSize)(%rsp) | ||
| 49 | #define save_rsi (32-LocalVarsSize)(%rsp) | ||
| 50 | #define save_rbx (40-LocalVarsSize)(%rsp) | ||
| 51 | #define save_r12 (56-LocalVarsSize)(%rsp) | ||
| 52 | #define save_r13 (64-LocalVarsSize)(%rsp) | ||
| 53 | #define save_r15 (80-LocalVarsSize)(%rsp) | ||
| 54 | |||
| 55 | |||
| 56 | .globl match_init, longest_match | ||
| 57 | |||
| 58 | /* | ||
| 59 | * On AMD64 the first argument of a function (in our case -- the pointer to | ||
| 60 | * deflate_state structure) is passed in %rdi, hence our offsets below are | ||
| 61 | * all off of that. | ||
| 62 | */ | ||
| 63 | |||
| 64 | /* you can check the structure offset by running | ||
| 65 | |||
| 66 | #include <stdlib.h> | ||
| 67 | #include <stdio.h> | ||
| 68 | #include "deflate.h" | ||
| 69 | |||
| 70 | void print_depl() | ||
| 71 | { | ||
| 72 | deflate_state ds; | ||
| 73 | deflate_state *s=&ds; | ||
| 74 | printf("size pointer=%u\n",(int)sizeof(void*)); | ||
| 75 | |||
| 76 | printf("#define dsWSize (%3u)(%%rdi)\n",(int)(((char*)&(s->w_size))-((char*)s))); | ||
| 77 | printf("#define dsWMask (%3u)(%%rdi)\n",(int)(((char*)&(s->w_mask))-((char*)s))); | ||
| 78 | printf("#define dsWindow (%3u)(%%rdi)\n",(int)(((char*)&(s->window))-((char*)s))); | ||
| 79 | printf("#define dsPrev (%3u)(%%rdi)\n",(int)(((char*)&(s->prev))-((char*)s))); | ||
| 80 | printf("#define dsMatchLen (%3u)(%%rdi)\n",(int)(((char*)&(s->match_length))-((char*)s))); | ||
| 81 | printf("#define dsPrevMatch (%3u)(%%rdi)\n",(int)(((char*)&(s->prev_match))-((char*)s))); | ||
| 82 | printf("#define dsStrStart (%3u)(%%rdi)\n",(int)(((char*)&(s->strstart))-((char*)s))); | ||
| 83 | printf("#define dsMatchStart (%3u)(%%rdi)\n",(int)(((char*)&(s->match_start))-((char*)s))); | ||
| 84 | printf("#define dsLookahead (%3u)(%%rdi)\n",(int)(((char*)&(s->lookahead))-((char*)s))); | ||
| 85 | printf("#define dsPrevLen (%3u)(%%rdi)\n",(int)(((char*)&(s->prev_length))-((char*)s))); | ||
| 86 | printf("#define dsMaxChainLen (%3u)(%%rdi)\n",(int)(((char*)&(s->max_chain_length))-((char*)s))); | ||
| 87 | printf("#define dsGoodMatch (%3u)(%%rdi)\n",(int)(((char*)&(s->good_match))-((char*)s))); | ||
| 88 | printf("#define dsNiceMatch (%3u)(%%rdi)\n",(int)(((char*)&(s->nice_match))-((char*)s))); | ||
| 89 | } | ||
| 90 | |||
| 91 | */ | ||
| 92 | |||
| 93 | |||
| 94 | /* | ||
| 95 | to compile for XCode 3.2 on MacOSX x86_64 | ||
| 96 | - run "gcc -g -c -DXCODE_MAC_X64_STRUCTURE amd64-match.S" | ||
| 97 | */ | ||
| 98 | |||
| 99 | |||
| 100 | #ifndef CURRENT_LINX_XCODE_MAC_X64_STRUCTURE | ||
| 101 | #define dsWSize ( 68)(%rdi) | ||
| 102 | #define dsWMask ( 76)(%rdi) | ||
| 103 | #define dsWindow ( 80)(%rdi) | ||
| 104 | #define dsPrev ( 96)(%rdi) | ||
| 105 | #define dsMatchLen (144)(%rdi) | ||
| 106 | #define dsPrevMatch (148)(%rdi) | ||
| 107 | #define dsStrStart (156)(%rdi) | ||
| 108 | #define dsMatchStart (160)(%rdi) | ||
| 109 | #define dsLookahead (164)(%rdi) | ||
| 110 | #define dsPrevLen (168)(%rdi) | ||
| 111 | #define dsMaxChainLen (172)(%rdi) | ||
| 112 | #define dsGoodMatch (188)(%rdi) | ||
| 113 | #define dsNiceMatch (192)(%rdi) | ||
| 114 | |||
| 115 | #else | ||
| 116 | |||
| 117 | #ifndef STRUCT_OFFSET | ||
| 118 | # define STRUCT_OFFSET (0) | ||
| 119 | #endif | ||
| 120 | |||
| 121 | |||
| 122 | #define dsWSize ( 56 + STRUCT_OFFSET)(%rdi) | ||
| 123 | #define dsWMask ( 64 + STRUCT_OFFSET)(%rdi) | ||
| 124 | #define dsWindow ( 72 + STRUCT_OFFSET)(%rdi) | ||
| 125 | #define dsPrev ( 88 + STRUCT_OFFSET)(%rdi) | ||
| 126 | #define dsMatchLen (136 + STRUCT_OFFSET)(%rdi) | ||
| 127 | #define dsPrevMatch (140 + STRUCT_OFFSET)(%rdi) | ||
| 128 | #define dsStrStart (148 + STRUCT_OFFSET)(%rdi) | ||
| 129 | #define dsMatchStart (152 + STRUCT_OFFSET)(%rdi) | ||
| 130 | #define dsLookahead (156 + STRUCT_OFFSET)(%rdi) | ||
| 131 | #define dsPrevLen (160 + STRUCT_OFFSET)(%rdi) | ||
| 132 | #define dsMaxChainLen (164 + STRUCT_OFFSET)(%rdi) | ||
| 133 | #define dsGoodMatch (180 + STRUCT_OFFSET)(%rdi) | ||
| 134 | #define dsNiceMatch (184 + STRUCT_OFFSET)(%rdi) | ||
| 135 | |||
| 136 | #endif | ||
| 137 | |||
| 138 | |||
| 139 | |||
| 140 | |||
| 141 | .text | ||
| 142 | |||
| 143 | /* uInt longest_match(deflate_state *deflatestate, IPos curmatch) */ | ||
| 144 | |||
| 145 | longest_match: | ||
| 146 | /* | ||
| 147 | * Retrieve the function arguments. %curmatch will hold cur_match | ||
| 148 | * throughout the entire function (passed via rsi on amd64). | ||
| 149 | * rdi will hold the pointer to the deflate_state (first arg on amd64) | ||
| 150 | */ | ||
| 151 | mov %rsi, save_rsi | ||
| 152 | mov %rbx, save_rbx | ||
| 153 | mov %r12, save_r12 | ||
| 154 | mov %r13, save_r13 | ||
| 155 | mov %r14, save_r14 | ||
| 156 | mov %r15, save_r15 | ||
| 157 | |||
| 158 | /* uInt wmask = s->w_mask; */ | ||
| 159 | /* unsigned chain_length = s->max_chain_length; */ | ||
| 160 | /* if (s->prev_length >= s->good_match) { */ | ||
| 161 | /* chain_length >>= 2; */ | ||
| 162 | /* } */ | ||
| 163 | |||
| 164 | movl dsPrevLen, %eax | ||
| 165 | movl dsGoodMatch, %ebx | ||
| 166 | cmpl %ebx, %eax | ||
| 167 | movl dsWMask, %eax | ||
| 168 | movl dsMaxChainLen, %chainlenwmask | ||
| 169 | jl LastMatchGood | ||
| 170 | shrl $2, %chainlenwmask | ||
| 171 | LastMatchGood: | ||
| 172 | |||
| 173 | /* chainlen is decremented once beforehand so that the function can */ | ||
| 174 | /* use the sign flag instead of the zero flag for the exit test. */ | ||
| 175 | /* It is then shifted into the high word, to make room for the wmask */ | ||
| 176 | /* value, which it will always accompany. */ | ||
| 177 | |||
| 178 | decl %chainlenwmask | ||
| 179 | shll $16, %chainlenwmask | ||
| 180 | orl %eax, %chainlenwmask | ||
| 181 | |||
| 182 | /* if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; */ | ||
| 183 | |||
| 184 | movl dsNiceMatch, %eax | ||
| 185 | movl dsLookahead, %ebx | ||
| 186 | cmpl %eax, %ebx | ||
| 187 | jl LookaheadLess | ||
| 188 | movl %eax, %ebx | ||
| 189 | LookaheadLess: movl %ebx, %nicematch | ||
| 190 | |||
| 191 | /* register Bytef *scan = s->window + s->strstart; */ | ||
| 192 | |||
| 193 | mov dsWindow, %window | ||
| 194 | movl dsStrStart, %limitd | ||
| 195 | lea (%limit, %window), %scan | ||
| 196 | |||
| 197 | /* Determine how many bytes the scan ptr is off from being */ | ||
| 198 | /* dword-aligned. */ | ||
| 199 | |||
| 200 | mov %scan, %scanalign | ||
| 201 | negl %scanalignd | ||
| 202 | andl $3, %scanalignd | ||
| 203 | |||
| 204 | /* IPos limit = s->strstart > (IPos)MAX_DIST(s) ? */ | ||
| 205 | /* s->strstart - (IPos)MAX_DIST(s) : NIL; */ | ||
| 206 | |||
| 207 | movl dsWSize, %eax | ||
| 208 | subl $MIN_LOOKAHEAD, %eax | ||
| 209 | xorl %ecx, %ecx | ||
| 210 | subl %eax, %limitd | ||
| 211 | cmovng %ecx, %limitd | ||
| 212 | |||
| 213 | /* int best_len = s->prev_length; */ | ||
| 214 | |||
| 215 | movl dsPrevLen, %bestlend | ||
| 216 | |||
| 217 | /* Store the sum of s->window + best_len in %windowbestlen locally, and in memory. */ | ||
| 218 | |||
| 219 | lea (%window, %bestlen), %windowbestlen | ||
| 220 | mov %windowbestlen, _windowbestlen | ||
| 221 | |||
| 222 | /* register ush scan_start = *(ushf*)scan; */ | ||
| 223 | /* register ush scan_end = *(ushf*)(scan+best_len-1); */ | ||
| 224 | /* Posf *prev = s->prev; */ | ||
| 225 | |||
| 226 | movzwl (%scan), %scanstart | ||
| 227 | movzwl -1(%scan, %bestlen), %scanend | ||
| 228 | mov dsPrev, %prev | ||
| 229 | |||
| 230 | /* Jump into the main loop. */ | ||
| 231 | |||
| 232 | movl %chainlenwmask, _chainlenwmask | ||
| 233 | jmp LoopEntry | ||
| 234 | |||
| 235 | .balign 16 | ||
| 236 | |||
| 237 | /* do { | ||
| 238 | * match = s->window + cur_match; | ||
| 239 | * if (*(ushf*)(match+best_len-1) != scan_end || | ||
| 240 | * *(ushf*)match != scan_start) continue; | ||
| 241 | * [...] | ||
| 242 | * } while ((cur_match = prev[cur_match & wmask]) > limit | ||
| 243 | * && --chain_length != 0); | ||
| 244 | * | ||
| 245 | * Here is the inner loop of the function. The function will spend the | ||
| 246 | * majority of its time in this loop, and majority of that time will | ||
| 247 | * be spent in the first ten instructions. | ||
| 248 | */ | ||
| 249 | LookupLoop: | ||
| 250 | andl %chainlenwmask, %curmatchd | ||
| 251 | movzwl (%prev, %curmatch, 2), %curmatchd | ||
| 252 | cmpl %limitd, %curmatchd | ||
| 253 | jbe LeaveNow | ||
| 254 | subl $0x00010000, %chainlenwmask | ||
| 255 | js LeaveNow | ||
| 256 | LoopEntry: cmpw -1(%windowbestlen, %curmatch), %scanendw | ||
| 257 | jne LookupLoop | ||
| 258 | cmpw %scanstartw, (%window, %curmatch) | ||
| 259 | jne LookupLoop | ||
| 260 | |||
| 261 | /* Store the current value of chainlen. */ | ||
| 262 | movl %chainlenwmask, _chainlenwmask | ||
| 263 | |||
| 264 | /* %scan is the string under scrutiny, and %prev to the string we */ | ||
| 265 | /* are hoping to match it up with. In actuality, %esi and %edi are */ | ||
| 266 | /* both pointed (MAX_MATCH_8 - scanalign) bytes ahead, and %edx is */ | ||
| 267 | /* initialized to -(MAX_MATCH_8 - scanalign). */ | ||
| 268 | |||
| 269 | mov $(-MAX_MATCH_8), %rdx | ||
| 270 | lea (%curmatch, %window), %windowbestlen | ||
| 271 | lea MAX_MATCH_8(%windowbestlen, %scanalign), %windowbestlen | ||
| 272 | lea MAX_MATCH_8(%scan, %scanalign), %prev | ||
| 273 | |||
| 274 | /* the prefetching below makes very little difference... */ | ||
| 275 | prefetcht1 (%windowbestlen, %rdx) | ||
| 276 | prefetcht1 (%prev, %rdx) | ||
| 277 | |||
| 278 | /* | ||
| 279 | * Test the strings for equality, 8 bytes at a time. At the end, | ||
| 280 | * adjust %rdx so that it is offset to the exact byte that mismatched. | ||
| 281 | * | ||
| 282 | * It should be confessed that this loop usually does not represent | ||
| 283 | * much of the total running time. Replacing it with a more | ||
| 284 | * straightforward "rep cmpsb" would not drastically degrade | ||
| 285 | * performance -- unrolling it, for example, makes no difference. | ||
| 286 | */ | ||
| 287 | |||
| 288 | #undef USE_SSE /* works, but is 6-7% slower, than non-SSE... */ | ||
| 289 | |||
| 290 | LoopCmps: | ||
| 291 | #ifdef USE_SSE | ||
| 292 | /* Preload the SSE registers */ | ||
| 293 | movdqu (%windowbestlen, %rdx), %xmm1 | ||
| 294 | movdqu (%prev, %rdx), %xmm2 | ||
| 295 | pcmpeqb %xmm2, %xmm1 | ||
| 296 | movdqu 16(%windowbestlen, %rdx), %xmm3 | ||
| 297 | movdqu 16(%prev, %rdx), %xmm4 | ||
| 298 | pcmpeqb %xmm4, %xmm3 | ||
| 299 | movdqu 32(%windowbestlen, %rdx), %xmm5 | ||
| 300 | movdqu 32(%prev, %rdx), %xmm6 | ||
| 301 | pcmpeqb %xmm6, %xmm5 | ||
| 302 | movdqu 48(%windowbestlen, %rdx), %xmm7 | ||
| 303 | movdqu 48(%prev, %rdx), %xmm8 | ||
| 304 | pcmpeqb %xmm8, %xmm7 | ||
| 305 | |||
| 306 | /* Check the comparisions' results */ | ||
| 307 | pmovmskb %xmm1, %rax | ||
| 308 | notw %ax | ||
| 309 | bsfw %ax, %ax | ||
| 310 | jnz LeaveLoopCmps | ||
| 311 | |||
| 312 | /* this is the only iteration of the loop with a possibility of having | ||
| 313 | incremented rdx by 0x108 (each loop iteration add 16*4 = 0x40 | ||
| 314 | and (0x40*4)+8=0x108 */ | ||
| 315 | add $8, %rdx | ||
| 316 | jz LenMaximum | ||
| 317 | add $8, %rdx | ||
| 318 | |||
| 319 | |||
| 320 | pmovmskb %xmm3, %rax | ||
| 321 | notw %ax | ||
| 322 | bsfw %ax, %ax | ||
| 323 | jnz LeaveLoopCmps | ||
| 324 | |||
| 325 | |||
| 326 | add $16, %rdx | ||
| 327 | |||
| 328 | |||
| 329 | pmovmskb %xmm5, %rax | ||
| 330 | notw %ax | ||
| 331 | bsfw %ax, %ax | ||
| 332 | jnz LeaveLoopCmps | ||
| 333 | |||
| 334 | add $16, %rdx | ||
| 335 | |||
| 336 | |||
| 337 | pmovmskb %xmm7, %rax | ||
| 338 | notw %ax | ||
| 339 | bsfw %ax, %ax | ||
| 340 | jnz LeaveLoopCmps | ||
| 341 | |||
| 342 | add $16, %rdx | ||
| 343 | |||
| 344 | jmp LoopCmps | ||
| 345 | LeaveLoopCmps: add %rax, %rdx | ||
| 346 | #else | ||
| 347 | mov (%windowbestlen, %rdx), %rax | ||
| 348 | xor (%prev, %rdx), %rax | ||
| 349 | jnz LeaveLoopCmps | ||
| 350 | |||
| 351 | mov 8(%windowbestlen, %rdx), %rax | ||
| 352 | xor 8(%prev, %rdx), %rax | ||
| 353 | jnz LeaveLoopCmps8 | ||
| 354 | |||
| 355 | mov 16(%windowbestlen, %rdx), %rax | ||
| 356 | xor 16(%prev, %rdx), %rax | ||
| 357 | jnz LeaveLoopCmps16 | ||
| 358 | |||
| 359 | add $24, %rdx | ||
| 360 | jnz LoopCmps | ||
| 361 | jmp LenMaximum | ||
| 362 | # if 0 | ||
| 363 | /* | ||
| 364 | * This three-liner is tantalizingly simple, but bsf is a slow instruction, | ||
| 365 | * and the complicated alternative down below is quite a bit faster. Sad... | ||
| 366 | */ | ||
| 367 | |||
| 368 | LeaveLoopCmps: bsf %rax, %rax /* find the first non-zero bit */ | ||
| 369 | shrl $3, %eax /* divide by 8 to get the byte */ | ||
| 370 | add %rax, %rdx | ||
| 371 | # else | ||
| 372 | LeaveLoopCmps16: | ||
| 373 | add $8, %rdx | ||
| 374 | LeaveLoopCmps8: | ||
| 375 | add $8, %rdx | ||
| 376 | LeaveLoopCmps: testl $0xFFFFFFFF, %eax /* Check the first 4 bytes */ | ||
| 377 | jnz Check16 | ||
| 378 | add $4, %rdx | ||
| 379 | shr $32, %rax | ||
| 380 | Check16: testw $0xFFFF, %ax | ||
| 381 | jnz LenLower | ||
| 382 | add $2, %rdx | ||
| 383 | shrl $16, %eax | ||
| 384 | LenLower: subb $1, %al | ||
| 385 | adc $0, %rdx | ||
| 386 | # endif | ||
| 387 | #endif | ||
| 388 | |||
| 389 | /* Calculate the length of the match. If it is longer than MAX_MATCH, */ | ||
| 390 | /* then automatically accept it as the best possible match and leave. */ | ||
| 391 | |||
| 392 | lea (%prev, %rdx), %rax | ||
| 393 | sub %scan, %rax | ||
| 394 | cmpl $MAX_MATCH, %eax | ||
| 395 | jge LenMaximum | ||
| 396 | |||
| 397 | /* If the length of the match is not longer than the best match we */ | ||
| 398 | /* have so far, then forget it and return to the lookup loop. */ | ||
| 399 | |||
| 400 | cmpl %bestlend, %eax | ||
| 401 | jg LongerMatch | ||
| 402 | mov _windowbestlen, %windowbestlen | ||
| 403 | mov dsPrev, %prev | ||
| 404 | movl _chainlenwmask, %edx | ||
| 405 | jmp LookupLoop | ||
| 406 | |||
| 407 | /* s->match_start = cur_match; */ | ||
| 408 | /* best_len = len; */ | ||
| 409 | /* if (len >= nice_match) break; */ | ||
| 410 | /* scan_end = *(ushf*)(scan+best_len-1); */ | ||
| 411 | |||
| 412 | LongerMatch: | ||
| 413 | movl %eax, %bestlend | ||
| 414 | movl %curmatchd, dsMatchStart | ||
| 415 | cmpl %nicematch, %eax | ||
| 416 | jge LeaveNow | ||
| 417 | |||
| 418 | lea (%window, %bestlen), %windowbestlen | ||
| 419 | mov %windowbestlen, _windowbestlen | ||
| 420 | |||
| 421 | movzwl -1(%scan, %rax), %scanend | ||
| 422 | mov dsPrev, %prev | ||
| 423 | movl _chainlenwmask, %chainlenwmask | ||
| 424 | jmp LookupLoop | ||
| 425 | |||
| 426 | /* Accept the current string, with the maximum possible length. */ | ||
| 427 | |||
| 428 | LenMaximum: | ||
| 429 | movl $MAX_MATCH, %bestlend | ||
| 430 | movl %curmatchd, dsMatchStart | ||
| 431 | |||
| 432 | /* if ((uInt)best_len <= s->lookahead) return (uInt)best_len; */ | ||
| 433 | /* return s->lookahead; */ | ||
| 434 | |||
| 435 | LeaveNow: | ||
| 436 | movl dsLookahead, %eax | ||
| 437 | cmpl %eax, %bestlend | ||
| 438 | cmovngl %bestlend, %eax | ||
| 439 | LookaheadRet: | ||
| 440 | |||
| 441 | /* Restore the registers and return from whence we came. */ | ||
| 442 | |||
| 443 | mov save_rsi, %rsi | ||
| 444 | mov save_rbx, %rbx | ||
| 445 | mov save_r12, %r12 | ||
| 446 | mov save_r13, %r13 | ||
| 447 | mov save_r14, %r14 | ||
| 448 | mov save_r15, %r15 | ||
| 449 | |||
| 450 | ret | ||
| 451 | |||
| 452 | match_init: ret | ||
diff --git a/contrib/asm686/README.686 b/contrib/asm686/README.686 deleted file mode 100644 index a0bf3be..0000000 --- a/contrib/asm686/README.686 +++ /dev/null | |||
| @@ -1,51 +0,0 @@ | |||
| 1 | This is a patched version of zlib, modified to use | ||
| 2 | Pentium-Pro-optimized assembly code in the deflation algorithm. The | ||
| 3 | files changed/added by this patch are: | ||
| 4 | |||
| 5 | README.686 | ||
| 6 | match.S | ||
| 7 | |||
| 8 | The speedup that this patch provides varies, depending on whether the | ||
| 9 | compiler used to build the original version of zlib falls afoul of the | ||
| 10 | PPro's speed traps. My own tests show a speedup of around 10-20% at | ||
| 11 | the default compression level, and 20-30% using -9, against a version | ||
| 12 | compiled using gcc 2.7.2.3. Your mileage may vary. | ||
| 13 | |||
| 14 | Note that this code has been tailored for the PPro/PII in particular, | ||
| 15 | and will not perform particuarly well on a Pentium. | ||
| 16 | |||
| 17 | If you are using an assembler other than GNU as, you will have to | ||
| 18 | translate match.S to use your assembler's syntax. (Have fun.) | ||
| 19 | |||
| 20 | Brian Raiter | ||
| 21 | breadbox@muppetlabs.com | ||
| 22 | April, 1998 | ||
| 23 | |||
| 24 | |||
| 25 | Added for zlib 1.1.3: | ||
| 26 | |||
| 27 | The patches come from | ||
| 28 | http://www.muppetlabs.com/~breadbox/software/assembly.html | ||
| 29 | |||
| 30 | To compile zlib with this asm file, copy match.S to the zlib directory | ||
| 31 | then do: | ||
| 32 | |||
| 33 | CFLAGS="-O3 -DASMV" ./configure | ||
| 34 | make OBJA=match.o | ||
| 35 | |||
| 36 | |||
| 37 | Update: | ||
| 38 | |||
| 39 | I've been ignoring these assembly routines for years, believing that | ||
| 40 | gcc's generated code had caught up with it sometime around gcc 2.95 | ||
| 41 | and the major rearchitecting of the Pentium 4. However, I recently | ||
| 42 | learned that, despite what I believed, this code still has some life | ||
| 43 | in it. On the Pentium 4 and AMD64 chips, it continues to run about 8% | ||
| 44 | faster than the code produced by gcc 4.1. | ||
| 45 | |||
| 46 | In acknowledgement of its continuing usefulness, I've altered the | ||
| 47 | license to match that of the rest of zlib. Share and Enjoy! | ||
| 48 | |||
| 49 | Brian Raiter | ||
| 50 | breadbox@muppetlabs.com | ||
| 51 | April, 2007 | ||
diff --git a/contrib/asm686/match.S b/contrib/asm686/match.S deleted file mode 100644 index fa42109..0000000 --- a/contrib/asm686/match.S +++ /dev/null | |||
| @@ -1,357 +0,0 @@ | |||
| 1 | /* match.S -- x86 assembly version of the zlib longest_match() function. | ||
| 2 | * Optimized for the Intel 686 chips (PPro and later). | ||
| 3 | * | ||
| 4 | * Copyright (C) 1998, 2007 Brian Raiter <breadbox@muppetlabs.com> | ||
| 5 | * | ||
| 6 | * This software is provided 'as-is', without any express or implied | ||
| 7 | * warranty. In no event will the author be held liable for any damages | ||
| 8 | * arising from the use of this software. | ||
| 9 | * | ||
| 10 | * Permission is granted to anyone to use this software for any purpose, | ||
| 11 | * including commercial applications, and to alter it and redistribute it | ||
| 12 | * freely, subject to the following restrictions: | ||
| 13 | * | ||
| 14 | * 1. The origin of this software must not be misrepresented; you must not | ||
| 15 | * claim that you wrote the original software. If you use this software | ||
| 16 | * in a product, an acknowledgment in the product documentation would be | ||
| 17 | * appreciated but is not required. | ||
| 18 | * 2. Altered source versions must be plainly marked as such, and must not be | ||
| 19 | * misrepresented as being the original software. | ||
| 20 | * 3. This notice may not be removed or altered from any source distribution. | ||
| 21 | */ | ||
| 22 | |||
| 23 | #ifndef NO_UNDERLINE | ||
| 24 | #define match_init _match_init | ||
| 25 | #define longest_match _longest_match | ||
| 26 | #endif | ||
| 27 | |||
| 28 | #define MAX_MATCH (258) | ||
| 29 | #define MIN_MATCH (3) | ||
| 30 | #define MIN_LOOKAHEAD (MAX_MATCH + MIN_MATCH + 1) | ||
| 31 | #define MAX_MATCH_8 ((MAX_MATCH + 7) & ~7) | ||
| 32 | |||
| 33 | /* stack frame offsets */ | ||
| 34 | |||
| 35 | #define chainlenwmask 0 /* high word: current chain len */ | ||
| 36 | /* low word: s->wmask */ | ||
| 37 | #define window 4 /* local copy of s->window */ | ||
| 38 | #define windowbestlen 8 /* s->window + bestlen */ | ||
| 39 | #define scanstart 16 /* first two bytes of string */ | ||
| 40 | #define scanend 12 /* last two bytes of string */ | ||
| 41 | #define scanalign 20 /* dword-misalignment of string */ | ||
| 42 | #define nicematch 24 /* a good enough match size */ | ||
| 43 | #define bestlen 28 /* size of best match so far */ | ||
| 44 | #define scan 32 /* ptr to string wanting match */ | ||
| 45 | |||
| 46 | #define LocalVarsSize (36) | ||
| 47 | /* saved ebx 36 */ | ||
| 48 | /* saved edi 40 */ | ||
| 49 | /* saved esi 44 */ | ||
| 50 | /* saved ebp 48 */ | ||
| 51 | /* return address 52 */ | ||
| 52 | #define deflatestate 56 /* the function arguments */ | ||
| 53 | #define curmatch 60 | ||
| 54 | |||
| 55 | /* All the +zlib1222add offsets are due to the addition of fields | ||
| 56 | * in zlib in the deflate_state structure since the asm code was first written | ||
| 57 | * (if you compile with zlib 1.0.4 or older, use "zlib1222add equ (-4)"). | ||
| 58 | * (if you compile with zlib between 1.0.5 and 1.2.2.1, use "zlib1222add equ 0"). | ||
| 59 | * if you compile with zlib 1.2.2.2 or later , use "zlib1222add equ 8"). | ||
| 60 | */ | ||
| 61 | |||
| 62 | #define zlib1222add (8) | ||
| 63 | |||
| 64 | #define dsWSize (36+zlib1222add) | ||
| 65 | #define dsWMask (44+zlib1222add) | ||
| 66 | #define dsWindow (48+zlib1222add) | ||
| 67 | #define dsPrev (56+zlib1222add) | ||
| 68 | #define dsMatchLen (88+zlib1222add) | ||
| 69 | #define dsPrevMatch (92+zlib1222add) | ||
| 70 | #define dsStrStart (100+zlib1222add) | ||
| 71 | #define dsMatchStart (104+zlib1222add) | ||
| 72 | #define dsLookahead (108+zlib1222add) | ||
| 73 | #define dsPrevLen (112+zlib1222add) | ||
| 74 | #define dsMaxChainLen (116+zlib1222add) | ||
| 75 | #define dsGoodMatch (132+zlib1222add) | ||
| 76 | #define dsNiceMatch (136+zlib1222add) | ||
| 77 | |||
| 78 | |||
| 79 | .file "match.S" | ||
| 80 | |||
| 81 | .globl match_init, longest_match | ||
| 82 | |||
| 83 | .text | ||
| 84 | |||
| 85 | /* uInt longest_match(deflate_state *deflatestate, IPos curmatch) */ | ||
| 86 | .cfi_sections .debug_frame | ||
| 87 | |||
| 88 | longest_match: | ||
| 89 | |||
| 90 | .cfi_startproc | ||
| 91 | /* Save registers that the compiler may be using, and adjust %esp to */ | ||
| 92 | /* make room for our stack frame. */ | ||
| 93 | |||
| 94 | pushl %ebp | ||
| 95 | .cfi_def_cfa_offset 8 | ||
| 96 | .cfi_offset ebp, -8 | ||
| 97 | pushl %edi | ||
| 98 | .cfi_def_cfa_offset 12 | ||
| 99 | pushl %esi | ||
| 100 | .cfi_def_cfa_offset 16 | ||
| 101 | pushl %ebx | ||
| 102 | .cfi_def_cfa_offset 20 | ||
| 103 | subl $LocalVarsSize, %esp | ||
| 104 | .cfi_def_cfa_offset LocalVarsSize+20 | ||
| 105 | |||
| 106 | /* Retrieve the function arguments. %ecx will hold cur_match */ | ||
| 107 | /* throughout the entire function. %edx will hold the pointer to the */ | ||
| 108 | /* deflate_state structure during the function's setup (before */ | ||
| 109 | /* entering the main loop). */ | ||
| 110 | |||
| 111 | movl deflatestate(%esp), %edx | ||
| 112 | movl curmatch(%esp), %ecx | ||
| 113 | |||
| 114 | /* uInt wmask = s->w_mask; */ | ||
| 115 | /* unsigned chain_length = s->max_chain_length; */ | ||
| 116 | /* if (s->prev_length >= s->good_match) { */ | ||
| 117 | /* chain_length >>= 2; */ | ||
| 118 | /* } */ | ||
| 119 | |||
| 120 | movl dsPrevLen(%edx), %eax | ||
| 121 | movl dsGoodMatch(%edx), %ebx | ||
| 122 | cmpl %ebx, %eax | ||
| 123 | movl dsWMask(%edx), %eax | ||
| 124 | movl dsMaxChainLen(%edx), %ebx | ||
| 125 | jl LastMatchGood | ||
| 126 | shrl $2, %ebx | ||
| 127 | LastMatchGood: | ||
| 128 | |||
| 129 | /* chainlen is decremented once beforehand so that the function can */ | ||
| 130 | /* use the sign flag instead of the zero flag for the exit test. */ | ||
| 131 | /* It is then shifted into the high word, to make room for the wmask */ | ||
| 132 | /* value, which it will always accompany. */ | ||
| 133 | |||
| 134 | decl %ebx | ||
| 135 | shll $16, %ebx | ||
| 136 | orl %eax, %ebx | ||
| 137 | movl %ebx, chainlenwmask(%esp) | ||
| 138 | |||
| 139 | /* if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; */ | ||
| 140 | |||
| 141 | movl dsNiceMatch(%edx), %eax | ||
| 142 | movl dsLookahead(%edx), %ebx | ||
| 143 | cmpl %eax, %ebx | ||
| 144 | jl LookaheadLess | ||
| 145 | movl %eax, %ebx | ||
| 146 | LookaheadLess: movl %ebx, nicematch(%esp) | ||
| 147 | |||
| 148 | /* register Bytef *scan = s->window + s->strstart; */ | ||
| 149 | |||
| 150 | movl dsWindow(%edx), %esi | ||
| 151 | movl %esi, window(%esp) | ||
| 152 | movl dsStrStart(%edx), %ebp | ||
| 153 | lea (%esi,%ebp), %edi | ||
| 154 | movl %edi, scan(%esp) | ||
| 155 | |||
| 156 | /* Determine how many bytes the scan ptr is off from being */ | ||
| 157 | /* dword-aligned. */ | ||
| 158 | |||
| 159 | movl %edi, %eax | ||
| 160 | negl %eax | ||
| 161 | andl $3, %eax | ||
| 162 | movl %eax, scanalign(%esp) | ||
| 163 | |||
| 164 | /* IPos limit = s->strstart > (IPos)MAX_DIST(s) ? */ | ||
| 165 | /* s->strstart - (IPos)MAX_DIST(s) : NIL; */ | ||
| 166 | |||
| 167 | movl dsWSize(%edx), %eax | ||
| 168 | subl $MIN_LOOKAHEAD, %eax | ||
| 169 | subl %eax, %ebp | ||
| 170 | jg LimitPositive | ||
| 171 | xorl %ebp, %ebp | ||
| 172 | LimitPositive: | ||
| 173 | |||
| 174 | /* int best_len = s->prev_length; */ | ||
| 175 | |||
| 176 | movl dsPrevLen(%edx), %eax | ||
| 177 | movl %eax, bestlen(%esp) | ||
| 178 | |||
| 179 | /* Store the sum of s->window + best_len in %esi locally, and in %esi. */ | ||
| 180 | |||
| 181 | addl %eax, %esi | ||
| 182 | movl %esi, windowbestlen(%esp) | ||
| 183 | |||
| 184 | /* register ush scan_start = *(ushf*)scan; */ | ||
| 185 | /* register ush scan_end = *(ushf*)(scan+best_len-1); */ | ||
| 186 | /* Posf *prev = s->prev; */ | ||
| 187 | |||
| 188 | movzwl (%edi), %ebx | ||
| 189 | movl %ebx, scanstart(%esp) | ||
| 190 | movzwl -1(%edi,%eax), %ebx | ||
| 191 | movl %ebx, scanend(%esp) | ||
| 192 | movl dsPrev(%edx), %edi | ||
| 193 | |||
| 194 | /* Jump into the main loop. */ | ||
| 195 | |||
| 196 | movl chainlenwmask(%esp), %edx | ||
| 197 | jmp LoopEntry | ||
| 198 | |||
| 199 | .balign 16 | ||
| 200 | |||
| 201 | /* do { | ||
| 202 | * match = s->window + cur_match; | ||
| 203 | * if (*(ushf*)(match+best_len-1) != scan_end || | ||
| 204 | * *(ushf*)match != scan_start) continue; | ||
| 205 | * [...] | ||
| 206 | * } while ((cur_match = prev[cur_match & wmask]) > limit | ||
| 207 | * && --chain_length != 0); | ||
| 208 | * | ||
| 209 | * Here is the inner loop of the function. The function will spend the | ||
| 210 | * majority of its time in this loop, and majority of that time will | ||
| 211 | * be spent in the first ten instructions. | ||
| 212 | * | ||
| 213 | * Within this loop: | ||
| 214 | * %ebx = scanend | ||
| 215 | * %ecx = curmatch | ||
| 216 | * %edx = chainlenwmask - i.e., ((chainlen << 16) | wmask) | ||
| 217 | * %esi = windowbestlen - i.e., (window + bestlen) | ||
| 218 | * %edi = prev | ||
| 219 | * %ebp = limit | ||
| 220 | */ | ||
| 221 | LookupLoop: | ||
| 222 | andl %edx, %ecx | ||
| 223 | movzwl (%edi,%ecx,2), %ecx | ||
| 224 | cmpl %ebp, %ecx | ||
| 225 | jbe LeaveNow | ||
| 226 | subl $0x00010000, %edx | ||
| 227 | js LeaveNow | ||
| 228 | LoopEntry: movzwl -1(%esi,%ecx), %eax | ||
| 229 | cmpl %ebx, %eax | ||
| 230 | jnz LookupLoop | ||
| 231 | movl window(%esp), %eax | ||
| 232 | movzwl (%eax,%ecx), %eax | ||
| 233 | cmpl scanstart(%esp), %eax | ||
| 234 | jnz LookupLoop | ||
| 235 | |||
| 236 | /* Store the current value of chainlen. */ | ||
| 237 | |||
| 238 | movl %edx, chainlenwmask(%esp) | ||
| 239 | |||
| 240 | /* Point %edi to the string under scrutiny, and %esi to the string we */ | ||
| 241 | /* are hoping to match it up with. In actuality, %esi and %edi are */ | ||
| 242 | /* both pointed (MAX_MATCH_8 - scanalign) bytes ahead, and %edx is */ | ||
| 243 | /* initialized to -(MAX_MATCH_8 - scanalign). */ | ||
| 244 | |||
| 245 | movl window(%esp), %esi | ||
| 246 | movl scan(%esp), %edi | ||
| 247 | addl %ecx, %esi | ||
| 248 | movl scanalign(%esp), %eax | ||
| 249 | movl $(-MAX_MATCH_8), %edx | ||
| 250 | lea MAX_MATCH_8(%edi,%eax), %edi | ||
| 251 | lea MAX_MATCH_8(%esi,%eax), %esi | ||
| 252 | |||
| 253 | /* Test the strings for equality, 8 bytes at a time. At the end, | ||
| 254 | * adjust %edx so that it is offset to the exact byte that mismatched. | ||
| 255 | * | ||
| 256 | * We already know at this point that the first three bytes of the | ||
| 257 | * strings match each other, and they can be safely passed over before | ||
| 258 | * starting the compare loop. So what this code does is skip over 0-3 | ||
| 259 | * bytes, as much as necessary in order to dword-align the %edi | ||
| 260 | * pointer. (%esi will still be misaligned three times out of four.) | ||
| 261 | * | ||
| 262 | * It should be confessed that this loop usually does not represent | ||
| 263 | * much of the total running time. Replacing it with a more | ||
| 264 | * straightforward "rep cmpsb" would not drastically degrade | ||
| 265 | * performance. | ||
| 266 | */ | ||
| 267 | LoopCmps: | ||
| 268 | movl (%esi,%edx), %eax | ||
| 269 | xorl (%edi,%edx), %eax | ||
| 270 | jnz LeaveLoopCmps | ||
| 271 | movl 4(%esi,%edx), %eax | ||
| 272 | xorl 4(%edi,%edx), %eax | ||
| 273 | jnz LeaveLoopCmps4 | ||
| 274 | addl $8, %edx | ||
| 275 | jnz LoopCmps | ||
| 276 | jmp LenMaximum | ||
| 277 | LeaveLoopCmps4: addl $4, %edx | ||
| 278 | LeaveLoopCmps: testl $0x0000FFFF, %eax | ||
| 279 | jnz LenLower | ||
| 280 | addl $2, %edx | ||
| 281 | shrl $16, %eax | ||
| 282 | LenLower: subb $1, %al | ||
| 283 | adcl $0, %edx | ||
| 284 | |||
| 285 | /* Calculate the length of the match. If it is longer than MAX_MATCH, */ | ||
| 286 | /* then automatically accept it as the best possible match and leave. */ | ||
| 287 | |||
| 288 | lea (%edi,%edx), %eax | ||
| 289 | movl scan(%esp), %edi | ||
| 290 | subl %edi, %eax | ||
| 291 | cmpl $MAX_MATCH, %eax | ||
| 292 | jge LenMaximum | ||
| 293 | |||
| 294 | /* If the length of the match is not longer than the best match we */ | ||
| 295 | /* have so far, then forget it and return to the lookup loop. */ | ||
| 296 | |||
| 297 | movl deflatestate(%esp), %edx | ||
| 298 | movl bestlen(%esp), %ebx | ||
| 299 | cmpl %ebx, %eax | ||
| 300 | jg LongerMatch | ||
| 301 | movl windowbestlen(%esp), %esi | ||
| 302 | movl dsPrev(%edx), %edi | ||
| 303 | movl scanend(%esp), %ebx | ||
| 304 | movl chainlenwmask(%esp), %edx | ||
| 305 | jmp LookupLoop | ||
| 306 | |||
| 307 | /* s->match_start = cur_match; */ | ||
| 308 | /* best_len = len; */ | ||
| 309 | /* if (len >= nice_match) break; */ | ||
| 310 | /* scan_end = *(ushf*)(scan+best_len-1); */ | ||
| 311 | |||
| 312 | LongerMatch: movl nicematch(%esp), %ebx | ||
| 313 | movl %eax, bestlen(%esp) | ||
| 314 | movl %ecx, dsMatchStart(%edx) | ||
| 315 | cmpl %ebx, %eax | ||
| 316 | jge LeaveNow | ||
| 317 | movl window(%esp), %esi | ||
| 318 | addl %eax, %esi | ||
| 319 | movl %esi, windowbestlen(%esp) | ||
| 320 | movzwl -1(%edi,%eax), %ebx | ||
| 321 | movl dsPrev(%edx), %edi | ||
| 322 | movl %ebx, scanend(%esp) | ||
| 323 | movl chainlenwmask(%esp), %edx | ||
| 324 | jmp LookupLoop | ||
| 325 | |||
| 326 | /* Accept the current string, with the maximum possible length. */ | ||
| 327 | |||
| 328 | LenMaximum: movl deflatestate(%esp), %edx | ||
| 329 | movl $MAX_MATCH, bestlen(%esp) | ||
| 330 | movl %ecx, dsMatchStart(%edx) | ||
| 331 | |||
| 332 | /* if ((uInt)best_len <= s->lookahead) return (uInt)best_len; */ | ||
| 333 | /* return s->lookahead; */ | ||
| 334 | |||
| 335 | LeaveNow: | ||
| 336 | movl deflatestate(%esp), %edx | ||
| 337 | movl bestlen(%esp), %ebx | ||
| 338 | movl dsLookahead(%edx), %eax | ||
| 339 | cmpl %eax, %ebx | ||
| 340 | jg LookaheadRet | ||
| 341 | movl %ebx, %eax | ||
| 342 | LookaheadRet: | ||
| 343 | |||
| 344 | /* Restore the stack and return from whence we came. */ | ||
| 345 | |||
| 346 | addl $LocalVarsSize, %esp | ||
| 347 | .cfi_def_cfa_offset 20 | ||
| 348 | popl %ebx | ||
| 349 | .cfi_def_cfa_offset 16 | ||
| 350 | popl %esi | ||
| 351 | .cfi_def_cfa_offset 12 | ||
| 352 | popl %edi | ||
| 353 | .cfi_def_cfa_offset 8 | ||
| 354 | popl %ebp | ||
| 355 | .cfi_def_cfa_offset 4 | ||
| 356 | .cfi_endproc | ||
| 357 | match_init: ret | ||
diff --git a/contrib/inflate86/inffas86.c b/contrib/inflate86/inffas86.c deleted file mode 100644 index 7292f67..0000000 --- a/contrib/inflate86/inffas86.c +++ /dev/null | |||
| @@ -1,1157 +0,0 @@ | |||
| 1 | /* inffas86.c is a hand tuned assembler version of | ||
| 2 | * | ||
| 3 | * inffast.c -- fast decoding | ||
| 4 | * Copyright (C) 1995-2003 Mark Adler | ||
| 5 | * For conditions of distribution and use, see copyright notice in zlib.h | ||
| 6 | * | ||
| 7 | * Copyright (C) 2003 Chris Anderson <christop@charm.net> | ||
| 8 | * Please use the copyright conditions above. | ||
| 9 | * | ||
| 10 | * Dec-29-2003 -- I added AMD64 inflate asm support. This version is also | ||
| 11 | * slightly quicker on x86 systems because, instead of using rep movsb to copy | ||
| 12 | * data, it uses rep movsw, which moves data in 2-byte chunks instead of single | ||
| 13 | * bytes. I've tested the AMD64 code on a Fedora Core 1 + the x86_64 updates | ||
| 14 | * from http://fedora.linux.duke.edu/fc1_x86_64 | ||
| 15 | * which is running on an Athlon 64 3000+ / Gigabyte GA-K8VT800M system with | ||
| 16 | * 1GB ram. The 64-bit version is about 4% faster than the 32-bit version, | ||
| 17 | * when decompressing mozilla-source-1.3.tar.gz. | ||
| 18 | * | ||
| 19 | * Mar-13-2003 -- Most of this is derived from inffast.S which is derived from | ||
| 20 | * the gcc -S output of zlib-1.2.0/inffast.c. Zlib-1.2.0 is in beta release at | ||
| 21 | * the moment. I have successfully compiled and tested this code with gcc2.96, | ||
| 22 | * gcc3.2, icc5.0, msvc6.0. It is very close to the speed of inffast.S | ||
| 23 | * compiled with gcc -DNO_MMX, but inffast.S is still faster on the P3 with MMX | ||
| 24 | * enabled. I will attempt to merge the MMX code into this version. Newer | ||
| 25 | * versions of this and inffast.S can be found at | ||
| 26 | * http://www.eetbeetee.com/zlib/ and http://www.charm.net/~christop/zlib/ | ||
| 27 | */ | ||
| 28 | |||
| 29 | #include "zutil.h" | ||
| 30 | #include "inftrees.h" | ||
| 31 | #include "inflate.h" | ||
| 32 | #include "inffast.h" | ||
| 33 | |||
| 34 | /* Mark Adler's comments from inffast.c: */ | ||
| 35 | |||
| 36 | /* | ||
| 37 | Decode literal, length, and distance codes and write out the resulting | ||
| 38 | literal and match bytes until either not enough input or output is | ||
| 39 | available, an end-of-block is encountered, or a data error is encountered. | ||
| 40 | When large enough input and output buffers are supplied to inflate(), for | ||
| 41 | example, a 16K input buffer and a 64K output buffer, more than 95% of the | ||
| 42 | inflate execution time is spent in this routine. | ||
| 43 | |||
| 44 | Entry assumptions: | ||
| 45 | |||
| 46 | state->mode == LEN | ||
| 47 | strm->avail_in >= 6 | ||
| 48 | strm->avail_out >= 258 | ||
| 49 | start >= strm->avail_out | ||
| 50 | state->bits < 8 | ||
| 51 | |||
| 52 | On return, state->mode is one of: | ||
| 53 | |||
| 54 | LEN -- ran out of enough output space or enough available input | ||
| 55 | TYPE -- reached end of block code, inflate() to interpret next block | ||
| 56 | BAD -- error in block data | ||
| 57 | |||
| 58 | Notes: | ||
| 59 | |||
| 60 | - The maximum input bits used by a length/distance pair is 15 bits for the | ||
| 61 | length code, 5 bits for the length extra, 15 bits for the distance code, | ||
| 62 | and 13 bits for the distance extra. This totals 48 bits, or six bytes. | ||
| 63 | Therefore if strm->avail_in >= 6, then there is enough input to avoid | ||
| 64 | checking for available input while decoding. | ||
| 65 | |||
| 66 | - The maximum bytes that a single length/distance pair can output is 258 | ||
| 67 | bytes, which is the maximum length that can be coded. inflate_fast() | ||
| 68 | requires strm->avail_out >= 258 for each loop to avoid checking for | ||
| 69 | output space. | ||
| 70 | */ | ||
| 71 | void inflate_fast(strm, start) | ||
| 72 | z_streamp strm; | ||
| 73 | unsigned start; /* inflate()'s starting value for strm->avail_out */ | ||
| 74 | { | ||
| 75 | struct inflate_state FAR *state; | ||
| 76 | struct inffast_ar { | ||
| 77 | /* 64 32 x86 x86_64 */ | ||
| 78 | /* ar offset register */ | ||
| 79 | /* 0 0 */ void *esp; /* esp save */ | ||
| 80 | /* 8 4 */ void *ebp; /* ebp save */ | ||
| 81 | /* 16 8 */ unsigned char FAR *in; /* esi rsi local strm->next_in */ | ||
| 82 | /* 24 12 */ unsigned char FAR *last; /* r9 while in < last */ | ||
| 83 | /* 32 16 */ unsigned char FAR *out; /* edi rdi local strm->next_out */ | ||
| 84 | /* 40 20 */ unsigned char FAR *beg; /* inflate()'s init next_out */ | ||
| 85 | /* 48 24 */ unsigned char FAR *end; /* r10 while out < end */ | ||
| 86 | /* 56 28 */ unsigned char FAR *window;/* size of window, wsize!=0 */ | ||
| 87 | /* 64 32 */ code const FAR *lcode; /* ebp rbp local strm->lencode */ | ||
| 88 | /* 72 36 */ code const FAR *dcode; /* r11 local strm->distcode */ | ||
| 89 | /* 80 40 */ unsigned long hold; /* edx rdx local strm->hold */ | ||
| 90 | /* 88 44 */ unsigned bits; /* ebx rbx local strm->bits */ | ||
| 91 | /* 92 48 */ unsigned wsize; /* window size */ | ||
| 92 | /* 96 52 */ unsigned write; /* window write index */ | ||
| 93 | /*100 56 */ unsigned lmask; /* r12 mask for lcode */ | ||
| 94 | /*104 60 */ unsigned dmask; /* r13 mask for dcode */ | ||
| 95 | /*108 64 */ unsigned len; /* r14 match length */ | ||
| 96 | /*112 68 */ unsigned dist; /* r15 match distance */ | ||
| 97 | /*116 72 */ unsigned status; /* set when state chng*/ | ||
| 98 | } ar; | ||
| 99 | |||
| 100 | #if defined( __GNUC__ ) && defined( __amd64__ ) && ! defined( __i386 ) | ||
| 101 | #define PAD_AVAIL_IN 6 | ||
| 102 | #define PAD_AVAIL_OUT 258 | ||
| 103 | #else | ||
| 104 | #define PAD_AVAIL_IN 5 | ||
| 105 | #define PAD_AVAIL_OUT 257 | ||
| 106 | #endif | ||
| 107 | |||
| 108 | /* copy state to local variables */ | ||
| 109 | state = (struct inflate_state FAR *)strm->state; | ||
| 110 | ar.in = strm->next_in; | ||
| 111 | ar.last = ar.in + (strm->avail_in - PAD_AVAIL_IN); | ||
| 112 | ar.out = strm->next_out; | ||
| 113 | ar.beg = ar.out - (start - strm->avail_out); | ||
| 114 | ar.end = ar.out + (strm->avail_out - PAD_AVAIL_OUT); | ||
| 115 | ar.wsize = state->wsize; | ||
| 116 | ar.write = state->wnext; | ||
| 117 | ar.window = state->window; | ||
| 118 | ar.hold = state->hold; | ||
| 119 | ar.bits = state->bits; | ||
| 120 | ar.lcode = state->lencode; | ||
| 121 | ar.dcode = state->distcode; | ||
| 122 | ar.lmask = (1U << state->lenbits) - 1; | ||
| 123 | ar.dmask = (1U << state->distbits) - 1; | ||
| 124 | |||
| 125 | /* decode literals and length/distances until end-of-block or not enough | ||
| 126 | input data or output space */ | ||
| 127 | |||
| 128 | /* align in on 1/2 hold size boundary */ | ||
| 129 | while (((unsigned long)(void *)ar.in & (sizeof(ar.hold) / 2 - 1)) != 0) { | ||
| 130 | ar.hold += (unsigned long)*ar.in++ << ar.bits; | ||
| 131 | ar.bits += 8; | ||
| 132 | } | ||
| 133 | |||
| 134 | #if defined( __GNUC__ ) && defined( __amd64__ ) && ! defined( __i386 ) | ||
| 135 | __asm__ __volatile__ ( | ||
| 136 | " leaq %0, %%rax\n" | ||
| 137 | " movq %%rbp, 8(%%rax)\n" /* save regs rbp and rsp */ | ||
| 138 | " movq %%rsp, (%%rax)\n" | ||
| 139 | " movq %%rax, %%rsp\n" /* make rsp point to &ar */ | ||
| 140 | " movq 16(%%rsp), %%rsi\n" /* rsi = in */ | ||
| 141 | " movq 32(%%rsp), %%rdi\n" /* rdi = out */ | ||
| 142 | " movq 24(%%rsp), %%r9\n" /* r9 = last */ | ||
| 143 | " movq 48(%%rsp), %%r10\n" /* r10 = end */ | ||
| 144 | " movq 64(%%rsp), %%rbp\n" /* rbp = lcode */ | ||
| 145 | " movq 72(%%rsp), %%r11\n" /* r11 = dcode */ | ||
| 146 | " movq 80(%%rsp), %%rdx\n" /* rdx = hold */ | ||
| 147 | " movl 88(%%rsp), %%ebx\n" /* ebx = bits */ | ||
| 148 | " movl 100(%%rsp), %%r12d\n" /* r12d = lmask */ | ||
| 149 | " movl 104(%%rsp), %%r13d\n" /* r13d = dmask */ | ||
| 150 | /* r14d = len */ | ||
| 151 | /* r15d = dist */ | ||
| 152 | " cld\n" | ||
| 153 | " cmpq %%rdi, %%r10\n" | ||
| 154 | " je .L_one_time\n" /* if only one decode left */ | ||
| 155 | " cmpq %%rsi, %%r9\n" | ||
| 156 | " je .L_one_time\n" | ||
| 157 | " jmp .L_do_loop\n" | ||
| 158 | |||
| 159 | ".L_one_time:\n" | ||
| 160 | " movq %%r12, %%r8\n" /* r8 = lmask */ | ||
| 161 | " cmpb $32, %%bl\n" | ||
| 162 | " ja .L_get_length_code_one_time\n" | ||
| 163 | |||
| 164 | " lodsl\n" /* eax = *(uint *)in++ */ | ||
| 165 | " movb %%bl, %%cl\n" /* cl = bits, needs it for shifting */ | ||
| 166 | " addb $32, %%bl\n" /* bits += 32 */ | ||
| 167 | " shlq %%cl, %%rax\n" | ||
| 168 | " orq %%rax, %%rdx\n" /* hold |= *((uint *)in)++ << bits */ | ||
| 169 | " jmp .L_get_length_code_one_time\n" | ||
| 170 | |||
| 171 | ".align 32,0x90\n" | ||
| 172 | ".L_while_test:\n" | ||
| 173 | " cmpq %%rdi, %%r10\n" | ||
| 174 | " jbe .L_break_loop\n" | ||
| 175 | " cmpq %%rsi, %%r9\n" | ||
| 176 | " jbe .L_break_loop\n" | ||
| 177 | |||
| 178 | ".L_do_loop:\n" | ||
| 179 | " movq %%r12, %%r8\n" /* r8 = lmask */ | ||
| 180 | " cmpb $32, %%bl\n" | ||
| 181 | " ja .L_get_length_code\n" /* if (32 < bits) */ | ||
| 182 | |||
| 183 | " lodsl\n" /* eax = *(uint *)in++ */ | ||
| 184 | " movb %%bl, %%cl\n" /* cl = bits, needs it for shifting */ | ||
| 185 | " addb $32, %%bl\n" /* bits += 32 */ | ||
| 186 | " shlq %%cl, %%rax\n" | ||
| 187 | " orq %%rax, %%rdx\n" /* hold |= *((uint *)in)++ << bits */ | ||
| 188 | |||
| 189 | ".L_get_length_code:\n" | ||
| 190 | " andq %%rdx, %%r8\n" /* r8 &= hold */ | ||
| 191 | " movl (%%rbp,%%r8,4), %%eax\n" /* eax = lcode[hold & lmask] */ | ||
| 192 | |||
| 193 | " movb %%ah, %%cl\n" /* cl = this.bits */ | ||
| 194 | " subb %%ah, %%bl\n" /* bits -= this.bits */ | ||
| 195 | " shrq %%cl, %%rdx\n" /* hold >>= this.bits */ | ||
| 196 | |||
| 197 | " testb %%al, %%al\n" | ||
| 198 | " jnz .L_test_for_length_base\n" /* if (op != 0) 45.7% */ | ||
| 199 | |||
| 200 | " movq %%r12, %%r8\n" /* r8 = lmask */ | ||
| 201 | " shrl $16, %%eax\n" /* output this.val char */ | ||
| 202 | " stosb\n" | ||
| 203 | |||
| 204 | ".L_get_length_code_one_time:\n" | ||
| 205 | " andq %%rdx, %%r8\n" /* r8 &= hold */ | ||
| 206 | " movl (%%rbp,%%r8,4), %%eax\n" /* eax = lcode[hold & lmask] */ | ||
| 207 | |||
| 208 | ".L_dolen:\n" | ||
| 209 | " movb %%ah, %%cl\n" /* cl = this.bits */ | ||
| 210 | " subb %%ah, %%bl\n" /* bits -= this.bits */ | ||
| 211 | " shrq %%cl, %%rdx\n" /* hold >>= this.bits */ | ||
| 212 | |||
| 213 | " testb %%al, %%al\n" | ||
| 214 | " jnz .L_test_for_length_base\n" /* if (op != 0) 45.7% */ | ||
| 215 | |||
| 216 | " shrl $16, %%eax\n" /* output this.val char */ | ||
| 217 | " stosb\n" | ||
| 218 | " jmp .L_while_test\n" | ||
| 219 | |||
| 220 | ".align 32,0x90\n" | ||
| 221 | ".L_test_for_length_base:\n" | ||
| 222 | " movl %%eax, %%r14d\n" /* len = this */ | ||
| 223 | " shrl $16, %%r14d\n" /* len = this.val */ | ||
| 224 | " movb %%al, %%cl\n" | ||
| 225 | |||
| 226 | " testb $16, %%al\n" | ||
| 227 | " jz .L_test_for_second_level_length\n" /* if ((op & 16) == 0) 8% */ | ||
| 228 | " andb $15, %%cl\n" /* op &= 15 */ | ||
| 229 | " jz .L_decode_distance\n" /* if (!op) */ | ||
| 230 | |||
| 231 | ".L_add_bits_to_len:\n" | ||
| 232 | " subb %%cl, %%bl\n" | ||
| 233 | " xorl %%eax, %%eax\n" | ||
| 234 | " incl %%eax\n" | ||
| 235 | " shll %%cl, %%eax\n" | ||
| 236 | " decl %%eax\n" | ||
| 237 | " andl %%edx, %%eax\n" /* eax &= hold */ | ||
| 238 | " shrq %%cl, %%rdx\n" | ||
| 239 | " addl %%eax, %%r14d\n" /* len += hold & mask[op] */ | ||
| 240 | |||
| 241 | ".L_decode_distance:\n" | ||
| 242 | " movq %%r13, %%r8\n" /* r8 = dmask */ | ||
| 243 | " cmpb $32, %%bl\n" | ||
| 244 | " ja .L_get_distance_code\n" /* if (32 < bits) */ | ||
| 245 | |||
| 246 | " lodsl\n" /* eax = *(uint *)in++ */ | ||
| 247 | " movb %%bl, %%cl\n" /* cl = bits, needs it for shifting */ | ||
| 248 | " addb $32, %%bl\n" /* bits += 32 */ | ||
| 249 | " shlq %%cl, %%rax\n" | ||
| 250 | " orq %%rax, %%rdx\n" /* hold |= *((uint *)in)++ << bits */ | ||
| 251 | |||
| 252 | ".L_get_distance_code:\n" | ||
| 253 | " andq %%rdx, %%r8\n" /* r8 &= hold */ | ||
| 254 | " movl (%%r11,%%r8,4), %%eax\n" /* eax = dcode[hold & dmask] */ | ||
| 255 | |||
| 256 | ".L_dodist:\n" | ||
| 257 | " movl %%eax, %%r15d\n" /* dist = this */ | ||
| 258 | " shrl $16, %%r15d\n" /* dist = this.val */ | ||
| 259 | " movb %%ah, %%cl\n" | ||
| 260 | " subb %%ah, %%bl\n" /* bits -= this.bits */ | ||
| 261 | " shrq %%cl, %%rdx\n" /* hold >>= this.bits */ | ||
| 262 | " movb %%al, %%cl\n" /* cl = this.op */ | ||
| 263 | |||
| 264 | " testb $16, %%al\n" /* if ((op & 16) == 0) */ | ||
| 265 | " jz .L_test_for_second_level_dist\n" | ||
| 266 | " andb $15, %%cl\n" /* op &= 15 */ | ||
| 267 | " jz .L_check_dist_one\n" | ||
| 268 | |||
| 269 | ".L_add_bits_to_dist:\n" | ||
| 270 | " subb %%cl, %%bl\n" | ||
| 271 | " xorl %%eax, %%eax\n" | ||
| 272 | " incl %%eax\n" | ||
| 273 | " shll %%cl, %%eax\n" | ||
| 274 | " decl %%eax\n" /* (1 << op) - 1 */ | ||
| 275 | " andl %%edx, %%eax\n" /* eax &= hold */ | ||
| 276 | " shrq %%cl, %%rdx\n" | ||
| 277 | " addl %%eax, %%r15d\n" /* dist += hold & ((1 << op) - 1) */ | ||
| 278 | |||
| 279 | ".L_check_window:\n" | ||
| 280 | " movq %%rsi, %%r8\n" /* save in so from can use it's reg */ | ||
| 281 | " movq %%rdi, %%rax\n" | ||
| 282 | " subq 40(%%rsp), %%rax\n" /* nbytes = out - beg */ | ||
| 283 | |||
| 284 | " cmpl %%r15d, %%eax\n" | ||
| 285 | " jb .L_clip_window\n" /* if (dist > nbytes) 4.2% */ | ||
| 286 | |||
| 287 | " movl %%r14d, %%ecx\n" /* ecx = len */ | ||
| 288 | " movq %%rdi, %%rsi\n" | ||
| 289 | " subq %%r15, %%rsi\n" /* from = out - dist */ | ||
| 290 | |||
| 291 | " sarl %%ecx\n" | ||
| 292 | " jnc .L_copy_two\n" /* if len % 2 == 0 */ | ||
| 293 | |||
| 294 | " rep movsw\n" | ||
| 295 | " movb (%%rsi), %%al\n" | ||
| 296 | " movb %%al, (%%rdi)\n" | ||
| 297 | " incq %%rdi\n" | ||
| 298 | |||
| 299 | " movq %%r8, %%rsi\n" /* move in back to %rsi, toss from */ | ||
| 300 | " jmp .L_while_test\n" | ||
| 301 | |||
| 302 | ".L_copy_two:\n" | ||
| 303 | " rep movsw\n" | ||
| 304 | " movq %%r8, %%rsi\n" /* move in back to %rsi, toss from */ | ||
| 305 | " jmp .L_while_test\n" | ||
| 306 | |||
| 307 | ".align 32,0x90\n" | ||
| 308 | ".L_check_dist_one:\n" | ||
| 309 | " cmpl $1, %%r15d\n" /* if dist 1, is a memset */ | ||
| 310 | " jne .L_check_window\n" | ||
| 311 | " cmpq %%rdi, 40(%%rsp)\n" /* if out == beg, outside window */ | ||
| 312 | " je .L_check_window\n" | ||
| 313 | |||
| 314 | " movl %%r14d, %%ecx\n" /* ecx = len */ | ||
| 315 | " movb -1(%%rdi), %%al\n" | ||
| 316 | " movb %%al, %%ah\n" | ||
| 317 | |||
| 318 | " sarl %%ecx\n" | ||
| 319 | " jnc .L_set_two\n" | ||
| 320 | " movb %%al, (%%rdi)\n" | ||
| 321 | " incq %%rdi\n" | ||
| 322 | |||
| 323 | ".L_set_two:\n" | ||
| 324 | " rep stosw\n" | ||
| 325 | " jmp .L_while_test\n" | ||
| 326 | |||
| 327 | ".align 32,0x90\n" | ||
| 328 | ".L_test_for_second_level_length:\n" | ||
| 329 | " testb $64, %%al\n" | ||
| 330 | " jnz .L_test_for_end_of_block\n" /* if ((op & 64) != 0) */ | ||
| 331 | |||
| 332 | " xorl %%eax, %%eax\n" | ||
| 333 | " incl %%eax\n" | ||
| 334 | " shll %%cl, %%eax\n" | ||
| 335 | " decl %%eax\n" | ||
| 336 | " andl %%edx, %%eax\n" /* eax &= hold */ | ||
| 337 | " addl %%r14d, %%eax\n" /* eax += len */ | ||
| 338 | " movl (%%rbp,%%rax,4), %%eax\n" /* eax = lcode[val+(hold&mask[op])]*/ | ||
| 339 | " jmp .L_dolen\n" | ||
| 340 | |||
| 341 | ".align 32,0x90\n" | ||
| 342 | ".L_test_for_second_level_dist:\n" | ||
| 343 | " testb $64, %%al\n" | ||
| 344 | " jnz .L_invalid_distance_code\n" /* if ((op & 64) != 0) */ | ||
| 345 | |||
| 346 | " xorl %%eax, %%eax\n" | ||
| 347 | " incl %%eax\n" | ||
| 348 | " shll %%cl, %%eax\n" | ||
| 349 | " decl %%eax\n" | ||
| 350 | " andl %%edx, %%eax\n" /* eax &= hold */ | ||
| 351 | " addl %%r15d, %%eax\n" /* eax += dist */ | ||
| 352 | " movl (%%r11,%%rax,4), %%eax\n" /* eax = dcode[val+(hold&mask[op])]*/ | ||
| 353 | " jmp .L_dodist\n" | ||
| 354 | |||
| 355 | ".align 32,0x90\n" | ||
| 356 | ".L_clip_window:\n" | ||
| 357 | " movl %%eax, %%ecx\n" /* ecx = nbytes */ | ||
| 358 | " movl 92(%%rsp), %%eax\n" /* eax = wsize, prepare for dist cmp */ | ||
| 359 | " negl %%ecx\n" /* nbytes = -nbytes */ | ||
| 360 | |||
| 361 | " cmpl %%r15d, %%eax\n" | ||
| 362 | " jb .L_invalid_distance_too_far\n" /* if (dist > wsize) */ | ||
| 363 | |||
| 364 | " addl %%r15d, %%ecx\n" /* nbytes = dist - nbytes */ | ||
| 365 | " cmpl $0, 96(%%rsp)\n" | ||
| 366 | " jne .L_wrap_around_window\n" /* if (write != 0) */ | ||
| 367 | |||
| 368 | " movq 56(%%rsp), %%rsi\n" /* from = window */ | ||
| 369 | " subl %%ecx, %%eax\n" /* eax -= nbytes */ | ||
| 370 | " addq %%rax, %%rsi\n" /* from += wsize - nbytes */ | ||
| 371 | |||
| 372 | " movl %%r14d, %%eax\n" /* eax = len */ | ||
| 373 | " cmpl %%ecx, %%r14d\n" | ||
| 374 | " jbe .L_do_copy\n" /* if (nbytes >= len) */ | ||
| 375 | |||
| 376 | " subl %%ecx, %%eax\n" /* eax -= nbytes */ | ||
| 377 | " rep movsb\n" | ||
| 378 | " movq %%rdi, %%rsi\n" | ||
| 379 | " subq %%r15, %%rsi\n" /* from = &out[ -dist ] */ | ||
| 380 | " jmp .L_do_copy\n" | ||
| 381 | |||
| 382 | ".align 32,0x90\n" | ||
| 383 | ".L_wrap_around_window:\n" | ||
| 384 | " movl 96(%%rsp), %%eax\n" /* eax = write */ | ||
| 385 | " cmpl %%eax, %%ecx\n" | ||
| 386 | " jbe .L_contiguous_in_window\n" /* if (write >= nbytes) */ | ||
| 387 | |||
| 388 | " movl 92(%%rsp), %%esi\n" /* from = wsize */ | ||
| 389 | " addq 56(%%rsp), %%rsi\n" /* from += window */ | ||
| 390 | " addq %%rax, %%rsi\n" /* from += write */ | ||
| 391 | " subq %%rcx, %%rsi\n" /* from -= nbytes */ | ||
| 392 | " subl %%eax, %%ecx\n" /* nbytes -= write */ | ||
| 393 | |||
| 394 | " movl %%r14d, %%eax\n" /* eax = len */ | ||
| 395 | " cmpl %%ecx, %%eax\n" | ||
| 396 | " jbe .L_do_copy\n" /* if (nbytes >= len) */ | ||
| 397 | |||
| 398 | " subl %%ecx, %%eax\n" /* len -= nbytes */ | ||
| 399 | " rep movsb\n" | ||
| 400 | " movq 56(%%rsp), %%rsi\n" /* from = window */ | ||
| 401 | " movl 96(%%rsp), %%ecx\n" /* nbytes = write */ | ||
| 402 | " cmpl %%ecx, %%eax\n" | ||
| 403 | " jbe .L_do_copy\n" /* if (nbytes >= len) */ | ||
| 404 | |||
| 405 | " subl %%ecx, %%eax\n" /* len -= nbytes */ | ||
| 406 | " rep movsb\n" | ||
| 407 | " movq %%rdi, %%rsi\n" | ||
| 408 | " subq %%r15, %%rsi\n" /* from = out - dist */ | ||
| 409 | " jmp .L_do_copy\n" | ||
| 410 | |||
| 411 | ".align 32,0x90\n" | ||
| 412 | ".L_contiguous_in_window:\n" | ||
| 413 | " movq 56(%%rsp), %%rsi\n" /* rsi = window */ | ||
| 414 | " addq %%rax, %%rsi\n" | ||
| 415 | " subq %%rcx, %%rsi\n" /* from += write - nbytes */ | ||
| 416 | |||
| 417 | " movl %%r14d, %%eax\n" /* eax = len */ | ||
| 418 | " cmpl %%ecx, %%eax\n" | ||
| 419 | " jbe .L_do_copy\n" /* if (nbytes >= len) */ | ||
| 420 | |||
| 421 | " subl %%ecx, %%eax\n" /* len -= nbytes */ | ||
| 422 | " rep movsb\n" | ||
| 423 | " movq %%rdi, %%rsi\n" | ||
| 424 | " subq %%r15, %%rsi\n" /* from = out - dist */ | ||
| 425 | " jmp .L_do_copy\n" /* if (nbytes >= len) */ | ||
| 426 | |||
| 427 | ".align 32,0x90\n" | ||
| 428 | ".L_do_copy:\n" | ||
| 429 | " movl %%eax, %%ecx\n" /* ecx = len */ | ||
| 430 | " rep movsb\n" | ||
| 431 | |||
| 432 | " movq %%r8, %%rsi\n" /* move in back to %esi, toss from */ | ||
| 433 | " jmp .L_while_test\n" | ||
| 434 | |||
| 435 | ".L_test_for_end_of_block:\n" | ||
| 436 | " testb $32, %%al\n" | ||
| 437 | " jz .L_invalid_literal_length_code\n" | ||
| 438 | " movl $1, 116(%%rsp)\n" | ||
| 439 | " jmp .L_break_loop_with_status\n" | ||
| 440 | |||
| 441 | ".L_invalid_literal_length_code:\n" | ||
| 442 | " movl $2, 116(%%rsp)\n" | ||
| 443 | " jmp .L_break_loop_with_status\n" | ||
| 444 | |||
| 445 | ".L_invalid_distance_code:\n" | ||
| 446 | " movl $3, 116(%%rsp)\n" | ||
| 447 | " jmp .L_break_loop_with_status\n" | ||
| 448 | |||
| 449 | ".L_invalid_distance_too_far:\n" | ||
| 450 | " movl $4, 116(%%rsp)\n" | ||
| 451 | " jmp .L_break_loop_with_status\n" | ||
| 452 | |||
| 453 | ".L_break_loop:\n" | ||
| 454 | " movl $0, 116(%%rsp)\n" | ||
| 455 | |||
| 456 | ".L_break_loop_with_status:\n" | ||
| 457 | /* put in, out, bits, and hold back into ar and pop esp */ | ||
| 458 | " movq %%rsi, 16(%%rsp)\n" /* in */ | ||
| 459 | " movq %%rdi, 32(%%rsp)\n" /* out */ | ||
| 460 | " movl %%ebx, 88(%%rsp)\n" /* bits */ | ||
| 461 | " movq %%rdx, 80(%%rsp)\n" /* hold */ | ||
| 462 | " movq (%%rsp), %%rax\n" /* restore rbp and rsp */ | ||
| 463 | " movq 8(%%rsp), %%rbp\n" | ||
| 464 | " movq %%rax, %%rsp\n" | ||
| 465 | : | ||
| 466 | : "m" (ar) | ||
| 467 | : "memory", "%rax", "%rbx", "%rcx", "%rdx", "%rsi", "%rdi", | ||
| 468 | "%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15" | ||
| 469 | ); | ||
| 470 | #elif ( defined( __GNUC__ ) || defined( __ICC ) ) && defined( __i386 ) | ||
| 471 | __asm__ __volatile__ ( | ||
| 472 | " leal %0, %%eax\n" | ||
| 473 | " movl %%esp, (%%eax)\n" /* save esp, ebp */ | ||
| 474 | " movl %%ebp, 4(%%eax)\n" | ||
| 475 | " movl %%eax, %%esp\n" | ||
| 476 | " movl 8(%%esp), %%esi\n" /* esi = in */ | ||
| 477 | " movl 16(%%esp), %%edi\n" /* edi = out */ | ||
| 478 | " movl 40(%%esp), %%edx\n" /* edx = hold */ | ||
| 479 | " movl 44(%%esp), %%ebx\n" /* ebx = bits */ | ||
| 480 | " movl 32(%%esp), %%ebp\n" /* ebp = lcode */ | ||
| 481 | |||
| 482 | " cld\n" | ||
| 483 | " jmp .L_do_loop\n" | ||
| 484 | |||
| 485 | ".align 32,0x90\n" | ||
| 486 | ".L_while_test:\n" | ||
| 487 | " cmpl %%edi, 24(%%esp)\n" /* out < end */ | ||
| 488 | " jbe .L_break_loop\n" | ||
| 489 | " cmpl %%esi, 12(%%esp)\n" /* in < last */ | ||
| 490 | " jbe .L_break_loop\n" | ||
| 491 | |||
| 492 | ".L_do_loop:\n" | ||
| 493 | " cmpb $15, %%bl\n" | ||
| 494 | " ja .L_get_length_code\n" /* if (15 < bits) */ | ||
| 495 | |||
| 496 | " xorl %%eax, %%eax\n" | ||
| 497 | " lodsw\n" /* al = *(ushort *)in++ */ | ||
| 498 | " movb %%bl, %%cl\n" /* cl = bits, needs it for shifting */ | ||
| 499 | " addb $16, %%bl\n" /* bits += 16 */ | ||
| 500 | " shll %%cl, %%eax\n" | ||
| 501 | " orl %%eax, %%edx\n" /* hold |= *((ushort *)in)++ << bits */ | ||
| 502 | |||
| 503 | ".L_get_length_code:\n" | ||
| 504 | " movl 56(%%esp), %%eax\n" /* eax = lmask */ | ||
| 505 | " andl %%edx, %%eax\n" /* eax &= hold */ | ||
| 506 | " movl (%%ebp,%%eax,4), %%eax\n" /* eax = lcode[hold & lmask] */ | ||
| 507 | |||
| 508 | ".L_dolen:\n" | ||
| 509 | " movb %%ah, %%cl\n" /* cl = this.bits */ | ||
| 510 | " subb %%ah, %%bl\n" /* bits -= this.bits */ | ||
| 511 | " shrl %%cl, %%edx\n" /* hold >>= this.bits */ | ||
| 512 | |||
| 513 | " testb %%al, %%al\n" | ||
| 514 | " jnz .L_test_for_length_base\n" /* if (op != 0) 45.7% */ | ||
| 515 | |||
| 516 | " shrl $16, %%eax\n" /* output this.val char */ | ||
| 517 | " stosb\n" | ||
| 518 | " jmp .L_while_test\n" | ||
| 519 | |||
| 520 | ".align 32,0x90\n" | ||
| 521 | ".L_test_for_length_base:\n" | ||
| 522 | " movl %%eax, %%ecx\n" /* len = this */ | ||
| 523 | " shrl $16, %%ecx\n" /* len = this.val */ | ||
| 524 | " movl %%ecx, 64(%%esp)\n" /* save len */ | ||
| 525 | " movb %%al, %%cl\n" | ||
| 526 | |||
| 527 | " testb $16, %%al\n" | ||
| 528 | " jz .L_test_for_second_level_length\n" /* if ((op & 16) == 0) 8% */ | ||
| 529 | " andb $15, %%cl\n" /* op &= 15 */ | ||
| 530 | " jz .L_decode_distance\n" /* if (!op) */ | ||
| 531 | " cmpb %%cl, %%bl\n" | ||
| 532 | " jae .L_add_bits_to_len\n" /* if (op <= bits) */ | ||
| 533 | |||
| 534 | " movb %%cl, %%ch\n" /* stash op in ch, freeing cl */ | ||
| 535 | " xorl %%eax, %%eax\n" | ||
| 536 | " lodsw\n" /* al = *(ushort *)in++ */ | ||
| 537 | " movb %%bl, %%cl\n" /* cl = bits, needs it for shifting */ | ||
| 538 | " addb $16, %%bl\n" /* bits += 16 */ | ||
| 539 | " shll %%cl, %%eax\n" | ||
| 540 | " orl %%eax, %%edx\n" /* hold |= *((ushort *)in)++ << bits */ | ||
| 541 | " movb %%ch, %%cl\n" /* move op back to ecx */ | ||
| 542 | |||
| 543 | ".L_add_bits_to_len:\n" | ||
| 544 | " subb %%cl, %%bl\n" | ||
| 545 | " xorl %%eax, %%eax\n" | ||
| 546 | " incl %%eax\n" | ||
| 547 | " shll %%cl, %%eax\n" | ||
| 548 | " decl %%eax\n" | ||
| 549 | " andl %%edx, %%eax\n" /* eax &= hold */ | ||
| 550 | " shrl %%cl, %%edx\n" | ||
| 551 | " addl %%eax, 64(%%esp)\n" /* len += hold & mask[op] */ | ||
| 552 | |||
| 553 | ".L_decode_distance:\n" | ||
| 554 | " cmpb $15, %%bl\n" | ||
| 555 | " ja .L_get_distance_code\n" /* if (15 < bits) */ | ||
| 556 | |||
| 557 | " xorl %%eax, %%eax\n" | ||
| 558 | " lodsw\n" /* al = *(ushort *)in++ */ | ||
| 559 | " movb %%bl, %%cl\n" /* cl = bits, needs it for shifting */ | ||
| 560 | " addb $16, %%bl\n" /* bits += 16 */ | ||
| 561 | " shll %%cl, %%eax\n" | ||
| 562 | " orl %%eax, %%edx\n" /* hold |= *((ushort *)in)++ << bits */ | ||
| 563 | |||
| 564 | ".L_get_distance_code:\n" | ||
| 565 | " movl 60(%%esp), %%eax\n" /* eax = dmask */ | ||
| 566 | " movl 36(%%esp), %%ecx\n" /* ecx = dcode */ | ||
| 567 | " andl %%edx, %%eax\n" /* eax &= hold */ | ||
| 568 | " movl (%%ecx,%%eax,4), %%eax\n"/* eax = dcode[hold & dmask] */ | ||
| 569 | |||
| 570 | ".L_dodist:\n" | ||
| 571 | " movl %%eax, %%ebp\n" /* dist = this */ | ||
| 572 | " shrl $16, %%ebp\n" /* dist = this.val */ | ||
| 573 | " movb %%ah, %%cl\n" | ||
| 574 | " subb %%ah, %%bl\n" /* bits -= this.bits */ | ||
| 575 | " shrl %%cl, %%edx\n" /* hold >>= this.bits */ | ||
| 576 | " movb %%al, %%cl\n" /* cl = this.op */ | ||
| 577 | |||
| 578 | " testb $16, %%al\n" /* if ((op & 16) == 0) */ | ||
| 579 | " jz .L_test_for_second_level_dist\n" | ||
| 580 | " andb $15, %%cl\n" /* op &= 15 */ | ||
| 581 | " jz .L_check_dist_one\n" | ||
| 582 | " cmpb %%cl, %%bl\n" | ||
| 583 | " jae .L_add_bits_to_dist\n" /* if (op <= bits) 97.6% */ | ||
| 584 | |||
| 585 | " movb %%cl, %%ch\n" /* stash op in ch, freeing cl */ | ||
| 586 | " xorl %%eax, %%eax\n" | ||
| 587 | " lodsw\n" /* al = *(ushort *)in++ */ | ||
| 588 | " movb %%bl, %%cl\n" /* cl = bits, needs it for shifting */ | ||
| 589 | " addb $16, %%bl\n" /* bits += 16 */ | ||
| 590 | " shll %%cl, %%eax\n" | ||
| 591 | " orl %%eax, %%edx\n" /* hold |= *((ushort *)in)++ << bits */ | ||
| 592 | " movb %%ch, %%cl\n" /* move op back to ecx */ | ||
| 593 | |||
| 594 | ".L_add_bits_to_dist:\n" | ||
| 595 | " subb %%cl, %%bl\n" | ||
| 596 | " xorl %%eax, %%eax\n" | ||
| 597 | " incl %%eax\n" | ||
| 598 | " shll %%cl, %%eax\n" | ||
| 599 | " decl %%eax\n" /* (1 << op) - 1 */ | ||
| 600 | " andl %%edx, %%eax\n" /* eax &= hold */ | ||
| 601 | " shrl %%cl, %%edx\n" | ||
| 602 | " addl %%eax, %%ebp\n" /* dist += hold & ((1 << op) - 1) */ | ||
| 603 | |||
| 604 | ".L_check_window:\n" | ||
| 605 | " movl %%esi, 8(%%esp)\n" /* save in so from can use it's reg */ | ||
| 606 | " movl %%edi, %%eax\n" | ||
| 607 | " subl 20(%%esp), %%eax\n" /* nbytes = out - beg */ | ||
| 608 | |||
| 609 | " cmpl %%ebp, %%eax\n" | ||
| 610 | " jb .L_clip_window\n" /* if (dist > nbytes) 4.2% */ | ||
| 611 | |||
| 612 | " movl 64(%%esp), %%ecx\n" /* ecx = len */ | ||
| 613 | " movl %%edi, %%esi\n" | ||
| 614 | " subl %%ebp, %%esi\n" /* from = out - dist */ | ||
| 615 | |||
| 616 | " sarl %%ecx\n" | ||
| 617 | " jnc .L_copy_two\n" /* if len % 2 == 0 */ | ||
| 618 | |||
| 619 | " rep movsw\n" | ||
| 620 | " movb (%%esi), %%al\n" | ||
| 621 | " movb %%al, (%%edi)\n" | ||
| 622 | " incl %%edi\n" | ||
| 623 | |||
| 624 | " movl 8(%%esp), %%esi\n" /* move in back to %esi, toss from */ | ||
| 625 | " movl 32(%%esp), %%ebp\n" /* ebp = lcode */ | ||
| 626 | " jmp .L_while_test\n" | ||
| 627 | |||
| 628 | ".L_copy_two:\n" | ||
| 629 | " rep movsw\n" | ||
| 630 | " movl 8(%%esp), %%esi\n" /* move in back to %esi, toss from */ | ||
| 631 | " movl 32(%%esp), %%ebp\n" /* ebp = lcode */ | ||
| 632 | " jmp .L_while_test\n" | ||
| 633 | |||
| 634 | ".align 32,0x90\n" | ||
| 635 | ".L_check_dist_one:\n" | ||
| 636 | " cmpl $1, %%ebp\n" /* if dist 1, is a memset */ | ||
| 637 | " jne .L_check_window\n" | ||
| 638 | " cmpl %%edi, 20(%%esp)\n" | ||
| 639 | " je .L_check_window\n" /* out == beg, if outside window */ | ||
| 640 | |||
| 641 | " movl 64(%%esp), %%ecx\n" /* ecx = len */ | ||
| 642 | " movb -1(%%edi), %%al\n" | ||
| 643 | " movb %%al, %%ah\n" | ||
| 644 | |||
| 645 | " sarl %%ecx\n" | ||
| 646 | " jnc .L_set_two\n" | ||
| 647 | " movb %%al, (%%edi)\n" | ||
| 648 | " incl %%edi\n" | ||
| 649 | |||
| 650 | ".L_set_two:\n" | ||
| 651 | " rep stosw\n" | ||
| 652 | " movl 32(%%esp), %%ebp\n" /* ebp = lcode */ | ||
| 653 | " jmp .L_while_test\n" | ||
| 654 | |||
| 655 | ".align 32,0x90\n" | ||
| 656 | ".L_test_for_second_level_length:\n" | ||
| 657 | " testb $64, %%al\n" | ||
| 658 | " jnz .L_test_for_end_of_block\n" /* if ((op & 64) != 0) */ | ||
| 659 | |||
| 660 | " xorl %%eax, %%eax\n" | ||
| 661 | " incl %%eax\n" | ||
| 662 | " shll %%cl, %%eax\n" | ||
| 663 | " decl %%eax\n" | ||
| 664 | " andl %%edx, %%eax\n" /* eax &= hold */ | ||
| 665 | " addl 64(%%esp), %%eax\n" /* eax += len */ | ||
| 666 | " movl (%%ebp,%%eax,4), %%eax\n" /* eax = lcode[val+(hold&mask[op])]*/ | ||
| 667 | " jmp .L_dolen\n" | ||
| 668 | |||
| 669 | ".align 32,0x90\n" | ||
| 670 | ".L_test_for_second_level_dist:\n" | ||
| 671 | " testb $64, %%al\n" | ||
| 672 | " jnz .L_invalid_distance_code\n" /* if ((op & 64) != 0) */ | ||
| 673 | |||
| 674 | " xorl %%eax, %%eax\n" | ||
| 675 | " incl %%eax\n" | ||
| 676 | " shll %%cl, %%eax\n" | ||
| 677 | " decl %%eax\n" | ||
| 678 | " andl %%edx, %%eax\n" /* eax &= hold */ | ||
| 679 | " addl %%ebp, %%eax\n" /* eax += dist */ | ||
| 680 | " movl 36(%%esp), %%ecx\n" /* ecx = dcode */ | ||
| 681 | " movl (%%ecx,%%eax,4), %%eax\n" /* eax = dcode[val+(hold&mask[op])]*/ | ||
| 682 | " jmp .L_dodist\n" | ||
| 683 | |||
| 684 | ".align 32,0x90\n" | ||
| 685 | ".L_clip_window:\n" | ||
| 686 | " movl %%eax, %%ecx\n" | ||
| 687 | " movl 48(%%esp), %%eax\n" /* eax = wsize */ | ||
| 688 | " negl %%ecx\n" /* nbytes = -nbytes */ | ||
| 689 | " movl 28(%%esp), %%esi\n" /* from = window */ | ||
| 690 | |||
| 691 | " cmpl %%ebp, %%eax\n" | ||
| 692 | " jb .L_invalid_distance_too_far\n" /* if (dist > wsize) */ | ||
| 693 | |||
| 694 | " addl %%ebp, %%ecx\n" /* nbytes = dist - nbytes */ | ||
| 695 | " cmpl $0, 52(%%esp)\n" | ||
| 696 | " jne .L_wrap_around_window\n" /* if (write != 0) */ | ||
| 697 | |||
| 698 | " subl %%ecx, %%eax\n" | ||
| 699 | " addl %%eax, %%esi\n" /* from += wsize - nbytes */ | ||
| 700 | |||
| 701 | " movl 64(%%esp), %%eax\n" /* eax = len */ | ||
| 702 | " cmpl %%ecx, %%eax\n" | ||
| 703 | " jbe .L_do_copy\n" /* if (nbytes >= len) */ | ||
| 704 | |||
| 705 | " subl %%ecx, %%eax\n" /* len -= nbytes */ | ||
| 706 | " rep movsb\n" | ||
| 707 | " movl %%edi, %%esi\n" | ||
| 708 | " subl %%ebp, %%esi\n" /* from = out - dist */ | ||
| 709 | " jmp .L_do_copy\n" | ||
| 710 | |||
| 711 | ".align 32,0x90\n" | ||
| 712 | ".L_wrap_around_window:\n" | ||
| 713 | " movl 52(%%esp), %%eax\n" /* eax = write */ | ||
| 714 | " cmpl %%eax, %%ecx\n" | ||
| 715 | " jbe .L_contiguous_in_window\n" /* if (write >= nbytes) */ | ||
| 716 | |||
| 717 | " addl 48(%%esp), %%esi\n" /* from += wsize */ | ||
| 718 | " addl %%eax, %%esi\n" /* from += write */ | ||
| 719 | " subl %%ecx, %%esi\n" /* from -= nbytes */ | ||
| 720 | " subl %%eax, %%ecx\n" /* nbytes -= write */ | ||
| 721 | |||
| 722 | " movl 64(%%esp), %%eax\n" /* eax = len */ | ||
| 723 | " cmpl %%ecx, %%eax\n" | ||
| 724 | " jbe .L_do_copy\n" /* if (nbytes >= len) */ | ||
| 725 | |||
| 726 | " subl %%ecx, %%eax\n" /* len -= nbytes */ | ||
| 727 | " rep movsb\n" | ||
| 728 | " movl 28(%%esp), %%esi\n" /* from = window */ | ||
| 729 | " movl 52(%%esp), %%ecx\n" /* nbytes = write */ | ||
| 730 | " cmpl %%ecx, %%eax\n" | ||
| 731 | " jbe .L_do_copy\n" /* if (nbytes >= len) */ | ||
| 732 | |||
| 733 | " subl %%ecx, %%eax\n" /* len -= nbytes */ | ||
| 734 | " rep movsb\n" | ||
| 735 | " movl %%edi, %%esi\n" | ||
| 736 | " subl %%ebp, %%esi\n" /* from = out - dist */ | ||
| 737 | " jmp .L_do_copy\n" | ||
| 738 | |||
| 739 | ".align 32,0x90\n" | ||
| 740 | ".L_contiguous_in_window:\n" | ||
| 741 | " addl %%eax, %%esi\n" | ||
| 742 | " subl %%ecx, %%esi\n" /* from += write - nbytes */ | ||
| 743 | |||
| 744 | " movl 64(%%esp), %%eax\n" /* eax = len */ | ||
| 745 | " cmpl %%ecx, %%eax\n" | ||
| 746 | " jbe .L_do_copy\n" /* if (nbytes >= len) */ | ||
| 747 | |||
| 748 | " subl %%ecx, %%eax\n" /* len -= nbytes */ | ||
| 749 | " rep movsb\n" | ||
| 750 | " movl %%edi, %%esi\n" | ||
| 751 | " subl %%ebp, %%esi\n" /* from = out - dist */ | ||
| 752 | " jmp .L_do_copy\n" /* if (nbytes >= len) */ | ||
| 753 | |||
| 754 | ".align 32,0x90\n" | ||
| 755 | ".L_do_copy:\n" | ||
| 756 | " movl %%eax, %%ecx\n" | ||
| 757 | " rep movsb\n" | ||
| 758 | |||
| 759 | " movl 8(%%esp), %%esi\n" /* move in back to %esi, toss from */ | ||
| 760 | " movl 32(%%esp), %%ebp\n" /* ebp = lcode */ | ||
| 761 | " jmp .L_while_test\n" | ||
| 762 | |||
| 763 | ".L_test_for_end_of_block:\n" | ||
| 764 | " testb $32, %%al\n" | ||
| 765 | " jz .L_invalid_literal_length_code\n" | ||
| 766 | " movl $1, 72(%%esp)\n" | ||
| 767 | " jmp .L_break_loop_with_status\n" | ||
| 768 | |||
| 769 | ".L_invalid_literal_length_code:\n" | ||
| 770 | " movl $2, 72(%%esp)\n" | ||
| 771 | " jmp .L_break_loop_with_status\n" | ||
| 772 | |||
| 773 | ".L_invalid_distance_code:\n" | ||
| 774 | " movl $3, 72(%%esp)\n" | ||
| 775 | " jmp .L_break_loop_with_status\n" | ||
| 776 | |||
| 777 | ".L_invalid_distance_too_far:\n" | ||
| 778 | " movl 8(%%esp), %%esi\n" | ||
| 779 | " movl $4, 72(%%esp)\n" | ||
| 780 | " jmp .L_break_loop_with_status\n" | ||
| 781 | |||
| 782 | ".L_break_loop:\n" | ||
| 783 | " movl $0, 72(%%esp)\n" | ||
| 784 | |||
| 785 | ".L_break_loop_with_status:\n" | ||
| 786 | /* put in, out, bits, and hold back into ar and pop esp */ | ||
| 787 | " movl %%esi, 8(%%esp)\n" /* save in */ | ||
| 788 | " movl %%edi, 16(%%esp)\n" /* save out */ | ||
| 789 | " movl %%ebx, 44(%%esp)\n" /* save bits */ | ||
| 790 | " movl %%edx, 40(%%esp)\n" /* save hold */ | ||
| 791 | " movl 4(%%esp), %%ebp\n" /* restore esp, ebp */ | ||
| 792 | " movl (%%esp), %%esp\n" | ||
| 793 | : | ||
| 794 | : "m" (ar) | ||
| 795 | : "memory", "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi" | ||
| 796 | ); | ||
| 797 | #elif defined( _MSC_VER ) && ! defined( _M_AMD64 ) | ||
| 798 | __asm { | ||
| 799 | lea eax, ar | ||
| 800 | mov [eax], esp /* save esp, ebp */ | ||
| 801 | mov [eax+4], ebp | ||
| 802 | mov esp, eax | ||
| 803 | mov esi, [esp+8] /* esi = in */ | ||
| 804 | mov edi, [esp+16] /* edi = out */ | ||
| 805 | mov edx, [esp+40] /* edx = hold */ | ||
| 806 | mov ebx, [esp+44] /* ebx = bits */ | ||
| 807 | mov ebp, [esp+32] /* ebp = lcode */ | ||
| 808 | |||
| 809 | cld | ||
| 810 | jmp L_do_loop | ||
| 811 | |||
| 812 | ALIGN 4 | ||
| 813 | L_while_test: | ||
| 814 | cmp [esp+24], edi | ||
| 815 | jbe L_break_loop | ||
| 816 | cmp [esp+12], esi | ||
| 817 | jbe L_break_loop | ||
| 818 | |||
| 819 | L_do_loop: | ||
| 820 | cmp bl, 15 | ||
| 821 | ja L_get_length_code /* if (15 < bits) */ | ||
| 822 | |||
| 823 | xor eax, eax | ||
| 824 | lodsw /* al = *(ushort *)in++ */ | ||
| 825 | mov cl, bl /* cl = bits, needs it for shifting */ | ||
| 826 | add bl, 16 /* bits += 16 */ | ||
| 827 | shl eax, cl | ||
| 828 | or edx, eax /* hold |= *((ushort *)in)++ << bits */ | ||
| 829 | |||
| 830 | L_get_length_code: | ||
| 831 | mov eax, [esp+56] /* eax = lmask */ | ||
| 832 | and eax, edx /* eax &= hold */ | ||
| 833 | mov eax, [ebp+eax*4] /* eax = lcode[hold & lmask] */ | ||
| 834 | |||
| 835 | L_dolen: | ||
| 836 | mov cl, ah /* cl = this.bits */ | ||
| 837 | sub bl, ah /* bits -= this.bits */ | ||
| 838 | shr edx, cl /* hold >>= this.bits */ | ||
| 839 | |||
| 840 | test al, al | ||
| 841 | jnz L_test_for_length_base /* if (op != 0) 45.7% */ | ||
| 842 | |||
| 843 | shr eax, 16 /* output this.val char */ | ||
| 844 | stosb | ||
| 845 | jmp L_while_test | ||
| 846 | |||
| 847 | ALIGN 4 | ||
| 848 | L_test_for_length_base: | ||
| 849 | mov ecx, eax /* len = this */ | ||
| 850 | shr ecx, 16 /* len = this.val */ | ||
| 851 | mov [esp+64], ecx /* save len */ | ||
| 852 | mov cl, al | ||
| 853 | |||
| 854 | test al, 16 | ||
| 855 | jz L_test_for_second_level_length /* if ((op & 16) == 0) 8% */ | ||
| 856 | and cl, 15 /* op &= 15 */ | ||
| 857 | jz L_decode_distance /* if (!op) */ | ||
| 858 | cmp bl, cl | ||
| 859 | jae L_add_bits_to_len /* if (op <= bits) */ | ||
| 860 | |||
| 861 | mov ch, cl /* stash op in ch, freeing cl */ | ||
| 862 | xor eax, eax | ||
| 863 | lodsw /* al = *(ushort *)in++ */ | ||
| 864 | mov cl, bl /* cl = bits, needs it for shifting */ | ||
| 865 | add bl, 16 /* bits += 16 */ | ||
| 866 | shl eax, cl | ||
| 867 | or edx, eax /* hold |= *((ushort *)in)++ << bits */ | ||
| 868 | mov cl, ch /* move op back to ecx */ | ||
| 869 | |||
| 870 | L_add_bits_to_len: | ||
| 871 | sub bl, cl | ||
| 872 | xor eax, eax | ||
| 873 | inc eax | ||
| 874 | shl eax, cl | ||
| 875 | dec eax | ||
| 876 | and eax, edx /* eax &= hold */ | ||
| 877 | shr edx, cl | ||
| 878 | add [esp+64], eax /* len += hold & mask[op] */ | ||
| 879 | |||
| 880 | L_decode_distance: | ||
| 881 | cmp bl, 15 | ||
| 882 | ja L_get_distance_code /* if (15 < bits) */ | ||
| 883 | |||
| 884 | xor eax, eax | ||
| 885 | lodsw /* al = *(ushort *)in++ */ | ||
| 886 | mov cl, bl /* cl = bits, needs it for shifting */ | ||
| 887 | add bl, 16 /* bits += 16 */ | ||
| 888 | shl eax, cl | ||
| 889 | or edx, eax /* hold |= *((ushort *)in)++ << bits */ | ||
| 890 | |||
| 891 | L_get_distance_code: | ||
| 892 | mov eax, [esp+60] /* eax = dmask */ | ||
| 893 | mov ecx, [esp+36] /* ecx = dcode */ | ||
| 894 | and eax, edx /* eax &= hold */ | ||
| 895 | mov eax, [ecx+eax*4]/* eax = dcode[hold & dmask] */ | ||
| 896 | |||
| 897 | L_dodist: | ||
| 898 | mov ebp, eax /* dist = this */ | ||
| 899 | shr ebp, 16 /* dist = this.val */ | ||
| 900 | mov cl, ah | ||
| 901 | sub bl, ah /* bits -= this.bits */ | ||
| 902 | shr edx, cl /* hold >>= this.bits */ | ||
| 903 | mov cl, al /* cl = this.op */ | ||
| 904 | |||
| 905 | test al, 16 /* if ((op & 16) == 0) */ | ||
| 906 | jz L_test_for_second_level_dist | ||
| 907 | and cl, 15 /* op &= 15 */ | ||
| 908 | jz L_check_dist_one | ||
| 909 | cmp bl, cl | ||
| 910 | jae L_add_bits_to_dist /* if (op <= bits) 97.6% */ | ||
| 911 | |||
| 912 | mov ch, cl /* stash op in ch, freeing cl */ | ||
| 913 | xor eax, eax | ||
| 914 | lodsw /* al = *(ushort *)in++ */ | ||
| 915 | mov cl, bl /* cl = bits, needs it for shifting */ | ||
| 916 | add bl, 16 /* bits += 16 */ | ||
| 917 | shl eax, cl | ||
| 918 | or edx, eax /* hold |= *((ushort *)in)++ << bits */ | ||
| 919 | mov cl, ch /* move op back to ecx */ | ||
| 920 | |||
| 921 | L_add_bits_to_dist: | ||
| 922 | sub bl, cl | ||
| 923 | xor eax, eax | ||
| 924 | inc eax | ||
| 925 | shl eax, cl | ||
| 926 | dec eax /* (1 << op) - 1 */ | ||
| 927 | and eax, edx /* eax &= hold */ | ||
| 928 | shr edx, cl | ||
| 929 | add ebp, eax /* dist += hold & ((1 << op) - 1) */ | ||
| 930 | |||
| 931 | L_check_window: | ||
| 932 | mov [esp+8], esi /* save in so from can use it's reg */ | ||
| 933 | mov eax, edi | ||
| 934 | sub eax, [esp+20] /* nbytes = out - beg */ | ||
| 935 | |||
| 936 | cmp eax, ebp | ||
| 937 | jb L_clip_window /* if (dist > nbytes) 4.2% */ | ||
| 938 | |||
| 939 | mov ecx, [esp+64] /* ecx = len */ | ||
| 940 | mov esi, edi | ||
| 941 | sub esi, ebp /* from = out - dist */ | ||
| 942 | |||
| 943 | sar ecx, 1 | ||
| 944 | jnc L_copy_two | ||
| 945 | |||
| 946 | rep movsw | ||
| 947 | mov al, [esi] | ||
| 948 | mov [edi], al | ||
| 949 | inc edi | ||
| 950 | |||
| 951 | mov esi, [esp+8] /* move in back to %esi, toss from */ | ||
| 952 | mov ebp, [esp+32] /* ebp = lcode */ | ||
| 953 | jmp L_while_test | ||
| 954 | |||
| 955 | L_copy_two: | ||
| 956 | rep movsw | ||
| 957 | mov esi, [esp+8] /* move in back to %esi, toss from */ | ||
| 958 | mov ebp, [esp+32] /* ebp = lcode */ | ||
| 959 | jmp L_while_test | ||
| 960 | |||
| 961 | ALIGN 4 | ||
| 962 | L_check_dist_one: | ||
| 963 | cmp ebp, 1 /* if dist 1, is a memset */ | ||
| 964 | jne L_check_window | ||
| 965 | cmp [esp+20], edi | ||
| 966 | je L_check_window /* out == beg, if outside window */ | ||
| 967 | |||
| 968 | mov ecx, [esp+64] /* ecx = len */ | ||
| 969 | mov al, [edi-1] | ||
| 970 | mov ah, al | ||
| 971 | |||
| 972 | sar ecx, 1 | ||
| 973 | jnc L_set_two | ||
| 974 | mov [edi], al /* memset out with from[-1] */ | ||
| 975 | inc edi | ||
| 976 | |||
| 977 | L_set_two: | ||
| 978 | rep stosw | ||
| 979 | mov ebp, [esp+32] /* ebp = lcode */ | ||
| 980 | jmp L_while_test | ||
| 981 | |||
| 982 | ALIGN 4 | ||
| 983 | L_test_for_second_level_length: | ||
| 984 | test al, 64 | ||
| 985 | jnz L_test_for_end_of_block /* if ((op & 64) != 0) */ | ||
| 986 | |||
| 987 | xor eax, eax | ||
| 988 | inc eax | ||
| 989 | shl eax, cl | ||
| 990 | dec eax | ||
| 991 | and eax, edx /* eax &= hold */ | ||
| 992 | add eax, [esp+64] /* eax += len */ | ||
| 993 | mov eax, [ebp+eax*4] /* eax = lcode[val+(hold&mask[op])]*/ | ||
| 994 | jmp L_dolen | ||
| 995 | |||
| 996 | ALIGN 4 | ||
| 997 | L_test_for_second_level_dist: | ||
| 998 | test al, 64 | ||
| 999 | jnz L_invalid_distance_code /* if ((op & 64) != 0) */ | ||
| 1000 | |||
| 1001 | xor eax, eax | ||
| 1002 | inc eax | ||
| 1003 | shl eax, cl | ||
| 1004 | dec eax | ||
| 1005 | and eax, edx /* eax &= hold */ | ||
| 1006 | add eax, ebp /* eax += dist */ | ||
| 1007 | mov ecx, [esp+36] /* ecx = dcode */ | ||
| 1008 | mov eax, [ecx+eax*4] /* eax = dcode[val+(hold&mask[op])]*/ | ||
| 1009 | jmp L_dodist | ||
| 1010 | |||
| 1011 | ALIGN 4 | ||
| 1012 | L_clip_window: | ||
| 1013 | mov ecx, eax | ||
| 1014 | mov eax, [esp+48] /* eax = wsize */ | ||
| 1015 | neg ecx /* nbytes = -nbytes */ | ||
| 1016 | mov esi, [esp+28] /* from = window */ | ||
| 1017 | |||
| 1018 | cmp eax, ebp | ||
| 1019 | jb L_invalid_distance_too_far /* if (dist > wsize) */ | ||
| 1020 | |||
| 1021 | add ecx, ebp /* nbytes = dist - nbytes */ | ||
| 1022 | cmp dword ptr [esp+52], 0 | ||
| 1023 | jne L_wrap_around_window /* if (write != 0) */ | ||
| 1024 | |||
| 1025 | sub eax, ecx | ||
| 1026 | add esi, eax /* from += wsize - nbytes */ | ||
| 1027 | |||
| 1028 | mov eax, [esp+64] /* eax = len */ | ||
| 1029 | cmp eax, ecx | ||
| 1030 | jbe L_do_copy /* if (nbytes >= len) */ | ||
| 1031 | |||
| 1032 | sub eax, ecx /* len -= nbytes */ | ||
| 1033 | rep movsb | ||
| 1034 | mov esi, edi | ||
| 1035 | sub esi, ebp /* from = out - dist */ | ||
| 1036 | jmp L_do_copy | ||
| 1037 | |||
| 1038 | ALIGN 4 | ||
| 1039 | L_wrap_around_window: | ||
| 1040 | mov eax, [esp+52] /* eax = write */ | ||
| 1041 | cmp ecx, eax | ||
| 1042 | jbe L_contiguous_in_window /* if (write >= nbytes) */ | ||
| 1043 | |||
| 1044 | add esi, [esp+48] /* from += wsize */ | ||
| 1045 | add esi, eax /* from += write */ | ||
| 1046 | sub esi, ecx /* from -= nbytes */ | ||
| 1047 | sub ecx, eax /* nbytes -= write */ | ||
| 1048 | |||
| 1049 | mov eax, [esp+64] /* eax = len */ | ||
| 1050 | cmp eax, ecx | ||
| 1051 | jbe L_do_copy /* if (nbytes >= len) */ | ||
| 1052 | |||
| 1053 | sub eax, ecx /* len -= nbytes */ | ||
| 1054 | rep movsb | ||
| 1055 | mov esi, [esp+28] /* from = window */ | ||
| 1056 | mov ecx, [esp+52] /* nbytes = write */ | ||
| 1057 | cmp eax, ecx | ||
| 1058 | jbe L_do_copy /* if (nbytes >= len) */ | ||
| 1059 | |||
| 1060 | sub eax, ecx /* len -= nbytes */ | ||
| 1061 | rep movsb | ||
| 1062 | mov esi, edi | ||
| 1063 | sub esi, ebp /* from = out - dist */ | ||
| 1064 | jmp L_do_copy | ||
| 1065 | |||
| 1066 | ALIGN 4 | ||
| 1067 | L_contiguous_in_window: | ||
| 1068 | add esi, eax | ||
| 1069 | sub esi, ecx /* from += write - nbytes */ | ||
| 1070 | |||
| 1071 | mov eax, [esp+64] /* eax = len */ | ||
| 1072 | cmp eax, ecx | ||
| 1073 | jbe L_do_copy /* if (nbytes >= len) */ | ||
| 1074 | |||
| 1075 | sub eax, ecx /* len -= nbytes */ | ||
| 1076 | rep movsb | ||
| 1077 | mov esi, edi | ||
| 1078 | sub esi, ebp /* from = out - dist */ | ||
| 1079 | jmp L_do_copy | ||
| 1080 | |||
| 1081 | ALIGN 4 | ||
| 1082 | L_do_copy: | ||
| 1083 | mov ecx, eax | ||
| 1084 | rep movsb | ||
| 1085 | |||
| 1086 | mov esi, [esp+8] /* move in back to %esi, toss from */ | ||
| 1087 | mov ebp, [esp+32] /* ebp = lcode */ | ||
| 1088 | jmp L_while_test | ||
| 1089 | |||
| 1090 | L_test_for_end_of_block: | ||
| 1091 | test al, 32 | ||
| 1092 | jz L_invalid_literal_length_code | ||
| 1093 | mov dword ptr [esp+72], 1 | ||
| 1094 | jmp L_break_loop_with_status | ||
| 1095 | |||
| 1096 | L_invalid_literal_length_code: | ||
| 1097 | mov dword ptr [esp+72], 2 | ||
| 1098 | jmp L_break_loop_with_status | ||
| 1099 | |||
| 1100 | L_invalid_distance_code: | ||
| 1101 | mov dword ptr [esp+72], 3 | ||
| 1102 | jmp L_break_loop_with_status | ||
| 1103 | |||
| 1104 | L_invalid_distance_too_far: | ||
| 1105 | mov esi, [esp+4] | ||
| 1106 | mov dword ptr [esp+72], 4 | ||
| 1107 | jmp L_break_loop_with_status | ||
| 1108 | |||
| 1109 | L_break_loop: | ||
| 1110 | mov dword ptr [esp+72], 0 | ||
| 1111 | |||
| 1112 | L_break_loop_with_status: | ||
| 1113 | /* put in, out, bits, and hold back into ar and pop esp */ | ||
| 1114 | mov [esp+8], esi /* save in */ | ||
| 1115 | mov [esp+16], edi /* save out */ | ||
| 1116 | mov [esp+44], ebx /* save bits */ | ||
| 1117 | mov [esp+40], edx /* save hold */ | ||
| 1118 | mov ebp, [esp+4] /* restore esp, ebp */ | ||
| 1119 | mov esp, [esp] | ||
| 1120 | } | ||
| 1121 | #else | ||
| 1122 | #error "x86 architecture not defined" | ||
| 1123 | #endif | ||
| 1124 | |||
| 1125 | if (ar.status > 1) { | ||
| 1126 | if (ar.status == 2) | ||
| 1127 | strm->msg = "invalid literal/length code"; | ||
| 1128 | else if (ar.status == 3) | ||
| 1129 | strm->msg = "invalid distance code"; | ||
| 1130 | else | ||
| 1131 | strm->msg = "invalid distance too far back"; | ||
| 1132 | state->mode = BAD; | ||
| 1133 | } | ||
| 1134 | else if ( ar.status == 1 ) { | ||
| 1135 | state->mode = TYPE; | ||
| 1136 | } | ||
| 1137 | |||
| 1138 | /* return unused bytes (on entry, bits < 8, so in won't go too far back) */ | ||
| 1139 | ar.len = ar.bits >> 3; | ||
| 1140 | ar.in -= ar.len; | ||
| 1141 | ar.bits -= ar.len << 3; | ||
| 1142 | ar.hold &= (1U << ar.bits) - 1; | ||
| 1143 | |||
| 1144 | /* update state and return */ | ||
| 1145 | strm->next_in = ar.in; | ||
| 1146 | strm->next_out = ar.out; | ||
| 1147 | strm->avail_in = (unsigned)(ar.in < ar.last ? | ||
| 1148 | PAD_AVAIL_IN + (ar.last - ar.in) : | ||
| 1149 | PAD_AVAIL_IN - (ar.in - ar.last)); | ||
| 1150 | strm->avail_out = (unsigned)(ar.out < ar.end ? | ||
| 1151 | PAD_AVAIL_OUT + (ar.end - ar.out) : | ||
| 1152 | PAD_AVAIL_OUT - (ar.out - ar.end)); | ||
| 1153 | state->hold = ar.hold; | ||
| 1154 | state->bits = ar.bits; | ||
| 1155 | return; | ||
| 1156 | } | ||
| 1157 | |||
diff --git a/contrib/inflate86/inffast.S b/contrib/inflate86/inffast.S deleted file mode 100644 index 2245a29..0000000 --- a/contrib/inflate86/inffast.S +++ /dev/null | |||
| @@ -1,1368 +0,0 @@ | |||
| 1 | /* | ||
| 2 | * inffast.S is a hand tuned assembler version of: | ||
| 3 | * | ||
| 4 | * inffast.c -- fast decoding | ||
| 5 | * Copyright (C) 1995-2003 Mark Adler | ||
| 6 | * For conditions of distribution and use, see copyright notice in zlib.h | ||
| 7 | * | ||
| 8 | * Copyright (C) 2003 Chris Anderson <christop@charm.net> | ||
| 9 | * Please use the copyright conditions above. | ||
| 10 | * | ||
| 11 | * This version (Jan-23-2003) of inflate_fast was coded and tested under | ||
| 12 | * GNU/Linux on a pentium 3, using the gcc-3.2 compiler distribution. On that | ||
| 13 | * machine, I found that gzip style archives decompressed about 20% faster than | ||
| 14 | * the gcc-3.2 -O3 -fomit-frame-pointer compiled version. Your results will | ||
| 15 | * depend on how large of a buffer is used for z_stream.next_in & next_out | ||
| 16 | * (8K-32K worked best for my 256K cpu cache) and how much overhead there is in | ||
| 17 | * stream processing I/O and crc32/addler32. In my case, this routine used | ||
| 18 | * 70% of the cpu time and crc32 used 20%. | ||
| 19 | * | ||
| 20 | * I am confident that this version will work in the general case, but I have | ||
| 21 | * not tested a wide variety of datasets or a wide variety of platforms. | ||
| 22 | * | ||
| 23 | * Jan-24-2003 -- Added -DUSE_MMX define for slightly faster inflating. | ||
| 24 | * It should be a runtime flag instead of compile time flag... | ||
| 25 | * | ||
| 26 | * Jan-26-2003 -- Added runtime check for MMX support with cpuid instruction. | ||
| 27 | * With -DUSE_MMX, only MMX code is compiled. With -DNO_MMX, only non-MMX code | ||
| 28 | * is compiled. Without either option, runtime detection is enabled. Runtime | ||
| 29 | * detection should work on all modern cpus and the recomended algorithm (flip | ||
| 30 | * ID bit on eflags and then use the cpuid instruction) is used in many | ||
| 31 | * multimedia applications. Tested under win2k with gcc-2.95 and gas-2.12 | ||
| 32 | * distributed with cygwin3. Compiling with gcc-2.95 -c inffast.S -o | ||
| 33 | * inffast.obj generates a COFF object which can then be linked with MSVC++ | ||
| 34 | * compiled code. Tested under FreeBSD 4.7 with gcc-2.95. | ||
| 35 | * | ||
| 36 | * Jan-28-2003 -- Tested Athlon XP... MMX mode is slower than no MMX (and | ||
| 37 | * slower than compiler generated code). Adjusted cpuid check to use the MMX | ||
| 38 | * code only for Pentiums < P4 until I have more data on the P4. Speed | ||
| 39 | * improvment is only about 15% on the Athlon when compared with code generated | ||
| 40 | * with MSVC++. Not sure yet, but I think the P4 will also be slower using the | ||
| 41 | * MMX mode because many of it's x86 ALU instructions execute in .5 cycles and | ||
| 42 | * have less latency than MMX ops. Added code to buffer the last 11 bytes of | ||
| 43 | * the input stream since the MMX code grabs bits in chunks of 32, which | ||
| 44 | * differs from the inffast.c algorithm. I don't think there would have been | ||
| 45 | * read overruns where a page boundary was crossed (a segfault), but there | ||
| 46 | * could have been overruns when next_in ends on unaligned memory (unintialized | ||
| 47 | * memory read). | ||
| 48 | * | ||
| 49 | * Mar-13-2003 -- P4 MMX is slightly slower than P4 NO_MMX. I created a C | ||
| 50 | * version of the non-MMX code so that it doesn't depend on zstrm and zstate | ||
| 51 | * structure offsets which are hard coded in this file. This was last tested | ||
| 52 | * with zlib-1.2.0 which is currently in beta testing, newer versions of this | ||
| 53 | * and inffas86.c can be found at http://www.eetbeetee.com/zlib/ and | ||
| 54 | * http://www.charm.net/~christop/zlib/ | ||
| 55 | */ | ||
| 56 | |||
| 57 | |||
| 58 | /* | ||
| 59 | * if you have underscore linking problems (_inflate_fast undefined), try | ||
| 60 | * using -DGAS_COFF | ||
| 61 | */ | ||
| 62 | #if ! defined( GAS_COFF ) && ! defined( GAS_ELF ) | ||
| 63 | |||
| 64 | #if defined( WIN32 ) || defined( __CYGWIN__ ) | ||
| 65 | #define GAS_COFF /* windows object format */ | ||
| 66 | #else | ||
| 67 | #define GAS_ELF | ||
| 68 | #endif | ||
| 69 | |||
| 70 | #endif /* ! GAS_COFF && ! GAS_ELF */ | ||
| 71 | |||
| 72 | |||
| 73 | #if defined( GAS_COFF ) | ||
| 74 | |||
| 75 | /* coff externals have underscores */ | ||
| 76 | #define inflate_fast _inflate_fast | ||
| 77 | #define inflate_fast_use_mmx _inflate_fast_use_mmx | ||
| 78 | |||
| 79 | #endif /* GAS_COFF */ | ||
| 80 | |||
| 81 | |||
| 82 | .file "inffast.S" | ||
| 83 | |||
| 84 | .globl inflate_fast | ||
| 85 | |||
| 86 | .text | ||
| 87 | .align 4,0 | ||
| 88 | .L_invalid_literal_length_code_msg: | ||
| 89 | .string "invalid literal/length code" | ||
| 90 | |||
| 91 | .align 4,0 | ||
| 92 | .L_invalid_distance_code_msg: | ||
| 93 | .string "invalid distance code" | ||
| 94 | |||
| 95 | .align 4,0 | ||
| 96 | .L_invalid_distance_too_far_msg: | ||
| 97 | .string "invalid distance too far back" | ||
| 98 | |||
| 99 | #if ! defined( NO_MMX ) | ||
| 100 | .align 4,0 | ||
| 101 | .L_mask: /* mask[N] = ( 1 << N ) - 1 */ | ||
| 102 | .long 0 | ||
| 103 | .long 1 | ||
| 104 | .long 3 | ||
| 105 | .long 7 | ||
| 106 | .long 15 | ||
| 107 | .long 31 | ||
| 108 | .long 63 | ||
| 109 | .long 127 | ||
| 110 | .long 255 | ||
| 111 | .long 511 | ||
| 112 | .long 1023 | ||
| 113 | .long 2047 | ||
| 114 | .long 4095 | ||
| 115 | .long 8191 | ||
| 116 | .long 16383 | ||
| 117 | .long 32767 | ||
| 118 | .long 65535 | ||
| 119 | .long 131071 | ||
| 120 | .long 262143 | ||
| 121 | .long 524287 | ||
| 122 | .long 1048575 | ||
| 123 | .long 2097151 | ||
| 124 | .long 4194303 | ||
| 125 | .long 8388607 | ||
| 126 | .long 16777215 | ||
| 127 | .long 33554431 | ||
| 128 | .long 67108863 | ||
| 129 | .long 134217727 | ||
| 130 | .long 268435455 | ||
| 131 | .long 536870911 | ||
| 132 | .long 1073741823 | ||
| 133 | .long 2147483647 | ||
| 134 | .long 4294967295 | ||
| 135 | #endif /* NO_MMX */ | ||
| 136 | |||
| 137 | .text | ||
| 138 | |||
| 139 | /* | ||
| 140 | * struct z_stream offsets, in zlib.h | ||
| 141 | */ | ||
| 142 | #define next_in_strm 0 /* strm->next_in */ | ||
| 143 | #define avail_in_strm 4 /* strm->avail_in */ | ||
| 144 | #define next_out_strm 12 /* strm->next_out */ | ||
| 145 | #define avail_out_strm 16 /* strm->avail_out */ | ||
| 146 | #define msg_strm 24 /* strm->msg */ | ||
| 147 | #define state_strm 28 /* strm->state */ | ||
| 148 | |||
| 149 | /* | ||
| 150 | * struct inflate_state offsets, in inflate.h | ||
| 151 | */ | ||
| 152 | #define mode_state 0 /* state->mode */ | ||
| 153 | #define wsize_state 32 /* state->wsize */ | ||
| 154 | #define write_state 40 /* state->write */ | ||
| 155 | #define window_state 44 /* state->window */ | ||
| 156 | #define hold_state 48 /* state->hold */ | ||
| 157 | #define bits_state 52 /* state->bits */ | ||
| 158 | #define lencode_state 68 /* state->lencode */ | ||
| 159 | #define distcode_state 72 /* state->distcode */ | ||
| 160 | #define lenbits_state 76 /* state->lenbits */ | ||
| 161 | #define distbits_state 80 /* state->distbits */ | ||
| 162 | |||
| 163 | /* | ||
| 164 | * inflate_fast's activation record | ||
| 165 | */ | ||
| 166 | #define local_var_size 64 /* how much local space for vars */ | ||
| 167 | #define strm_sp 88 /* first arg: z_stream * (local_var_size + 24) */ | ||
| 168 | #define start_sp 92 /* second arg: unsigned int (local_var_size + 28) */ | ||
| 169 | |||
| 170 | /* | ||
| 171 | * offsets for local vars on stack | ||
| 172 | */ | ||
| 173 | #define out 60 /* unsigned char* */ | ||
| 174 | #define window 56 /* unsigned char* */ | ||
| 175 | #define wsize 52 /* unsigned int */ | ||
| 176 | #define write 48 /* unsigned int */ | ||
| 177 | #define in 44 /* unsigned char* */ | ||
| 178 | #define beg 40 /* unsigned char* */ | ||
| 179 | #define buf 28 /* char[ 12 ] */ | ||
| 180 | #define len 24 /* unsigned int */ | ||
| 181 | #define last 20 /* unsigned char* */ | ||
| 182 | #define end 16 /* unsigned char* */ | ||
| 183 | #define dcode 12 /* code* */ | ||
| 184 | #define lcode 8 /* code* */ | ||
| 185 | #define dmask 4 /* unsigned int */ | ||
| 186 | #define lmask 0 /* unsigned int */ | ||
| 187 | |||
| 188 | /* | ||
| 189 | * typedef enum inflate_mode consts, in inflate.h | ||
| 190 | */ | ||
| 191 | #define INFLATE_MODE_TYPE 11 /* state->mode flags enum-ed in inflate.h */ | ||
| 192 | #define INFLATE_MODE_BAD 26 | ||
| 193 | |||
| 194 | |||
| 195 | #if ! defined( USE_MMX ) && ! defined( NO_MMX ) | ||
| 196 | |||
| 197 | #define RUN_TIME_MMX | ||
| 198 | |||
| 199 | #define CHECK_MMX 1 | ||
| 200 | #define DO_USE_MMX 2 | ||
| 201 | #define DONT_USE_MMX 3 | ||
| 202 | |||
| 203 | .globl inflate_fast_use_mmx | ||
| 204 | |||
| 205 | .data | ||
| 206 | |||
| 207 | .align 4,0 | ||
| 208 | inflate_fast_use_mmx: /* integer flag for run time control 1=check,2=mmx,3=no */ | ||
| 209 | .long CHECK_MMX | ||
| 210 | |||
| 211 | #if defined( GAS_ELF ) | ||
| 212 | /* elf info */ | ||
| 213 | .type inflate_fast_use_mmx,@object | ||
| 214 | .size inflate_fast_use_mmx,4 | ||
| 215 | #endif | ||
| 216 | |||
| 217 | #endif /* RUN_TIME_MMX */ | ||
| 218 | |||
| 219 | #if defined( GAS_COFF ) | ||
| 220 | /* coff info: scl 2 = extern, type 32 = function */ | ||
| 221 | .def inflate_fast; .scl 2; .type 32; .endef | ||
| 222 | #endif | ||
| 223 | |||
| 224 | .text | ||
| 225 | |||
| 226 | .align 32,0x90 | ||
| 227 | inflate_fast: | ||
| 228 | pushl %edi | ||
| 229 | pushl %esi | ||
| 230 | pushl %ebp | ||
| 231 | pushl %ebx | ||
| 232 | pushf /* save eflags (strm_sp, state_sp assumes this is 32 bits) */ | ||
| 233 | subl $local_var_size, %esp | ||
| 234 | cld | ||
| 235 | |||
| 236 | #define strm_r %esi | ||
| 237 | #define state_r %edi | ||
| 238 | |||
| 239 | movl strm_sp(%esp), strm_r | ||
| 240 | movl state_strm(strm_r), state_r | ||
| 241 | |||
| 242 | /* in = strm->next_in; | ||
| 243 | * out = strm->next_out; | ||
| 244 | * last = in + strm->avail_in - 11; | ||
| 245 | * beg = out - (start - strm->avail_out); | ||
| 246 | * end = out + (strm->avail_out - 257); | ||
| 247 | */ | ||
| 248 | movl avail_in_strm(strm_r), %edx | ||
| 249 | movl next_in_strm(strm_r), %eax | ||
| 250 | |||
| 251 | addl %eax, %edx /* avail_in += next_in */ | ||
| 252 | subl $11, %edx /* avail_in -= 11 */ | ||
| 253 | |||
| 254 | movl %eax, in(%esp) | ||
| 255 | movl %edx, last(%esp) | ||
| 256 | |||
| 257 | movl start_sp(%esp), %ebp | ||
| 258 | movl avail_out_strm(strm_r), %ecx | ||
| 259 | movl next_out_strm(strm_r), %ebx | ||
| 260 | |||
| 261 | subl %ecx, %ebp /* start -= avail_out */ | ||
| 262 | negl %ebp /* start = -start */ | ||
| 263 | addl %ebx, %ebp /* start += next_out */ | ||
| 264 | |||
| 265 | subl $257, %ecx /* avail_out -= 257 */ | ||
| 266 | addl %ebx, %ecx /* avail_out += out */ | ||
| 267 | |||
| 268 | movl %ebx, out(%esp) | ||
| 269 | movl %ebp, beg(%esp) | ||
| 270 | movl %ecx, end(%esp) | ||
| 271 | |||
| 272 | /* wsize = state->wsize; | ||
| 273 | * write = state->write; | ||
| 274 | * window = state->window; | ||
| 275 | * hold = state->hold; | ||
| 276 | * bits = state->bits; | ||
| 277 | * lcode = state->lencode; | ||
| 278 | * dcode = state->distcode; | ||
| 279 | * lmask = ( 1 << state->lenbits ) - 1; | ||
| 280 | * dmask = ( 1 << state->distbits ) - 1; | ||
| 281 | */ | ||
| 282 | |||
| 283 | movl lencode_state(state_r), %eax | ||
| 284 | movl distcode_state(state_r), %ecx | ||
| 285 | |||
| 286 | movl %eax, lcode(%esp) | ||
| 287 | movl %ecx, dcode(%esp) | ||
| 288 | |||
| 289 | movl $1, %eax | ||
| 290 | movl lenbits_state(state_r), %ecx | ||
| 291 | shll %cl, %eax | ||
| 292 | decl %eax | ||
| 293 | movl %eax, lmask(%esp) | ||
| 294 | |||
| 295 | movl $1, %eax | ||
| 296 | movl distbits_state(state_r), %ecx | ||
| 297 | shll %cl, %eax | ||
| 298 | decl %eax | ||
| 299 | movl %eax, dmask(%esp) | ||
| 300 | |||
| 301 | movl wsize_state(state_r), %eax | ||
| 302 | movl write_state(state_r), %ecx | ||
| 303 | movl window_state(state_r), %edx | ||
| 304 | |||
| 305 | movl %eax, wsize(%esp) | ||
| 306 | movl %ecx, write(%esp) | ||
| 307 | movl %edx, window(%esp) | ||
| 308 | |||
| 309 | movl hold_state(state_r), %ebp | ||
| 310 | movl bits_state(state_r), %ebx | ||
| 311 | |||
| 312 | #undef strm_r | ||
| 313 | #undef state_r | ||
| 314 | |||
| 315 | #define in_r %esi | ||
| 316 | #define from_r %esi | ||
| 317 | #define out_r %edi | ||
| 318 | |||
| 319 | movl in(%esp), in_r | ||
| 320 | movl last(%esp), %ecx | ||
| 321 | cmpl in_r, %ecx | ||
| 322 | ja .L_align_long /* if in < last */ | ||
| 323 | |||
| 324 | addl $11, %ecx /* ecx = &in[ avail_in ] */ | ||
| 325 | subl in_r, %ecx /* ecx = avail_in */ | ||
| 326 | movl $12, %eax | ||
| 327 | subl %ecx, %eax /* eax = 12 - avail_in */ | ||
| 328 | leal buf(%esp), %edi | ||
| 329 | rep movsb /* memcpy( buf, in, avail_in ) */ | ||
| 330 | movl %eax, %ecx | ||
| 331 | xorl %eax, %eax | ||
| 332 | rep stosb /* memset( &buf[ avail_in ], 0, 12 - avail_in ) */ | ||
| 333 | leal buf(%esp), in_r /* in = buf */ | ||
| 334 | movl in_r, last(%esp) /* last = in, do just one iteration */ | ||
| 335 | jmp .L_is_aligned | ||
| 336 | |||
| 337 | /* align in_r on long boundary */ | ||
| 338 | .L_align_long: | ||
| 339 | testl $3, in_r | ||
| 340 | jz .L_is_aligned | ||
| 341 | xorl %eax, %eax | ||
| 342 | movb (in_r), %al | ||
| 343 | incl in_r | ||
| 344 | movl %ebx, %ecx | ||
| 345 | addl $8, %ebx | ||
| 346 | shll %cl, %eax | ||
| 347 | orl %eax, %ebp | ||
| 348 | jmp .L_align_long | ||
| 349 | |||
| 350 | .L_is_aligned: | ||
| 351 | movl out(%esp), out_r | ||
| 352 | |||
| 353 | #if defined( NO_MMX ) | ||
| 354 | jmp .L_do_loop | ||
| 355 | #endif | ||
| 356 | |||
| 357 | #if defined( USE_MMX ) | ||
| 358 | jmp .L_init_mmx | ||
| 359 | #endif | ||
| 360 | |||
| 361 | /*** Runtime MMX check ***/ | ||
| 362 | |||
| 363 | #if defined( RUN_TIME_MMX ) | ||
| 364 | .L_check_mmx: | ||
| 365 | cmpl $DO_USE_MMX, inflate_fast_use_mmx | ||
| 366 | je .L_init_mmx | ||
| 367 | ja .L_do_loop /* > 2 */ | ||
| 368 | |||
| 369 | pushl %eax | ||
| 370 | pushl %ebx | ||
| 371 | pushl %ecx | ||
| 372 | pushl %edx | ||
| 373 | pushf | ||
| 374 | movl (%esp), %eax /* copy eflags to eax */ | ||
| 375 | xorl $0x200000, (%esp) /* try toggling ID bit of eflags (bit 21) | ||
| 376 | * to see if cpu supports cpuid... | ||
| 377 | * ID bit method not supported by NexGen but | ||
| 378 | * bios may load a cpuid instruction and | ||
| 379 | * cpuid may be disabled on Cyrix 5-6x86 */ | ||
| 380 | popf | ||
| 381 | pushf | ||
| 382 | popl %edx /* copy new eflags to edx */ | ||
| 383 | xorl %eax, %edx /* test if ID bit is flipped */ | ||
| 384 | jz .L_dont_use_mmx /* not flipped if zero */ | ||
| 385 | xorl %eax, %eax | ||
| 386 | cpuid | ||
| 387 | cmpl $0x756e6547, %ebx /* check for GenuineIntel in ebx,ecx,edx */ | ||
| 388 | jne .L_dont_use_mmx | ||
| 389 | cmpl $0x6c65746e, %ecx | ||
| 390 | jne .L_dont_use_mmx | ||
| 391 | cmpl $0x49656e69, %edx | ||
| 392 | jne .L_dont_use_mmx | ||
| 393 | movl $1, %eax | ||
| 394 | cpuid /* get cpu features */ | ||
| 395 | shrl $8, %eax | ||
| 396 | andl $15, %eax | ||
| 397 | cmpl $6, %eax /* check for Pentium family, is 0xf for P4 */ | ||
| 398 | jne .L_dont_use_mmx | ||
| 399 | testl $0x800000, %edx /* test if MMX feature is set (bit 23) */ | ||
| 400 | jnz .L_use_mmx | ||
| 401 | jmp .L_dont_use_mmx | ||
| 402 | .L_use_mmx: | ||
| 403 | movl $DO_USE_MMX, inflate_fast_use_mmx | ||
| 404 | jmp .L_check_mmx_pop | ||
| 405 | .L_dont_use_mmx: | ||
| 406 | movl $DONT_USE_MMX, inflate_fast_use_mmx | ||
| 407 | .L_check_mmx_pop: | ||
| 408 | popl %edx | ||
| 409 | popl %ecx | ||
| 410 | popl %ebx | ||
| 411 | popl %eax | ||
| 412 | jmp .L_check_mmx | ||
| 413 | #endif | ||
| 414 | |||
| 415 | |||
| 416 | /*** Non-MMX code ***/ | ||
| 417 | |||
| 418 | #if defined ( NO_MMX ) || defined( RUN_TIME_MMX ) | ||
| 419 | |||
| 420 | #define hold_r %ebp | ||
| 421 | #define bits_r %bl | ||
| 422 | #define bitslong_r %ebx | ||
| 423 | |||
| 424 | .align 32,0x90 | ||
| 425 | .L_while_test: | ||
| 426 | /* while (in < last && out < end) | ||
| 427 | */ | ||
| 428 | cmpl out_r, end(%esp) | ||
| 429 | jbe .L_break_loop /* if (out >= end) */ | ||
| 430 | |||
| 431 | cmpl in_r, last(%esp) | ||
| 432 | jbe .L_break_loop | ||
| 433 | |||
| 434 | .L_do_loop: | ||
| 435 | /* regs: %esi = in, %ebp = hold, %bl = bits, %edi = out | ||
| 436 | * | ||
| 437 | * do { | ||
| 438 | * if (bits < 15) { | ||
| 439 | * hold |= *((unsigned short *)in)++ << bits; | ||
| 440 | * bits += 16 | ||
| 441 | * } | ||
| 442 | * this = lcode[hold & lmask] | ||
| 443 | */ | ||
| 444 | cmpb $15, bits_r | ||
| 445 | ja .L_get_length_code /* if (15 < bits) */ | ||
| 446 | |||
| 447 | xorl %eax, %eax | ||
| 448 | lodsw /* al = *(ushort *)in++ */ | ||
| 449 | movb bits_r, %cl /* cl = bits, needs it for shifting */ | ||
| 450 | addb $16, bits_r /* bits += 16 */ | ||
| 451 | shll %cl, %eax | ||
| 452 | orl %eax, hold_r /* hold |= *((ushort *)in)++ << bits */ | ||
| 453 | |||
| 454 | .L_get_length_code: | ||
| 455 | movl lmask(%esp), %edx /* edx = lmask */ | ||
| 456 | movl lcode(%esp), %ecx /* ecx = lcode */ | ||
| 457 | andl hold_r, %edx /* edx &= hold */ | ||
| 458 | movl (%ecx,%edx,4), %eax /* eax = lcode[hold & lmask] */ | ||
| 459 | |||
| 460 | .L_dolen: | ||
| 461 | /* regs: %esi = in, %ebp = hold, %bl = bits, %edi = out | ||
| 462 | * | ||
| 463 | * dolen: | ||
| 464 | * bits -= this.bits; | ||
| 465 | * hold >>= this.bits | ||
| 466 | */ | ||
| 467 | movb %ah, %cl /* cl = this.bits */ | ||
| 468 | subb %ah, bits_r /* bits -= this.bits */ | ||
| 469 | shrl %cl, hold_r /* hold >>= this.bits */ | ||
| 470 | |||
| 471 | /* check if op is a literal | ||
| 472 | * if (op == 0) { | ||
| 473 | * PUP(out) = this.val; | ||
| 474 | * } | ||
| 475 | */ | ||
| 476 | testb %al, %al | ||
| 477 | jnz .L_test_for_length_base /* if (op != 0) 45.7% */ | ||
| 478 | |||
| 479 | shrl $16, %eax /* output this.val char */ | ||
| 480 | stosb | ||
| 481 | jmp .L_while_test | ||
| 482 | |||
| 483 | .L_test_for_length_base: | ||
| 484 | /* regs: %esi = in, %ebp = hold, %bl = bits, %edi = out, %edx = len | ||
| 485 | * | ||
| 486 | * else if (op & 16) { | ||
| 487 | * len = this.val | ||
| 488 | * op &= 15 | ||
| 489 | * if (op) { | ||
| 490 | * if (op > bits) { | ||
| 491 | * hold |= *((unsigned short *)in)++ << bits; | ||
| 492 | * bits += 16 | ||
| 493 | * } | ||
| 494 | * len += hold & mask[op]; | ||
| 495 | * bits -= op; | ||
| 496 | * hold >>= op; | ||
| 497 | * } | ||
| 498 | */ | ||
| 499 | #define len_r %edx | ||
| 500 | movl %eax, len_r /* len = this */ | ||
| 501 | shrl $16, len_r /* len = this.val */ | ||
| 502 | movb %al, %cl | ||
| 503 | |||
| 504 | testb $16, %al | ||
| 505 | jz .L_test_for_second_level_length /* if ((op & 16) == 0) 8% */ | ||
| 506 | andb $15, %cl /* op &= 15 */ | ||
| 507 | jz .L_save_len /* if (!op) */ | ||
| 508 | cmpb %cl, bits_r | ||
| 509 | jae .L_add_bits_to_len /* if (op <= bits) */ | ||
| 510 | |||
| 511 | movb %cl, %ch /* stash op in ch, freeing cl */ | ||
| 512 | xorl %eax, %eax | ||
| 513 | lodsw /* al = *(ushort *)in++ */ | ||
| 514 | movb bits_r, %cl /* cl = bits, needs it for shifting */ | ||
| 515 | addb $16, bits_r /* bits += 16 */ | ||
| 516 | shll %cl, %eax | ||
| 517 | orl %eax, hold_r /* hold |= *((ushort *)in)++ << bits */ | ||
| 518 | movb %ch, %cl /* move op back to ecx */ | ||
| 519 | |||
| 520 | .L_add_bits_to_len: | ||
| 521 | movl $1, %eax | ||
| 522 | shll %cl, %eax | ||
| 523 | decl %eax | ||
| 524 | subb %cl, bits_r | ||
| 525 | andl hold_r, %eax /* eax &= hold */ | ||
| 526 | shrl %cl, hold_r | ||
| 527 | addl %eax, len_r /* len += hold & mask[op] */ | ||
| 528 | |||
| 529 | .L_save_len: | ||
| 530 | movl len_r, len(%esp) /* save len */ | ||
| 531 | #undef len_r | ||
| 532 | |||
| 533 | .L_decode_distance: | ||
| 534 | /* regs: %esi = in, %ebp = hold, %bl = bits, %edi = out, %edx = dist | ||
| 535 | * | ||
| 536 | * if (bits < 15) { | ||
| 537 | * hold |= *((unsigned short *)in)++ << bits; | ||
| 538 | * bits += 16 | ||
| 539 | * } | ||
| 540 | * this = dcode[hold & dmask]; | ||
| 541 | * dodist: | ||
| 542 | * bits -= this.bits; | ||
| 543 | * hold >>= this.bits; | ||
| 544 | * op = this.op; | ||
| 545 | */ | ||
| 546 | |||
| 547 | cmpb $15, bits_r | ||
| 548 | ja .L_get_distance_code /* if (15 < bits) */ | ||
| 549 | |||
| 550 | xorl %eax, %eax | ||
| 551 | lodsw /* al = *(ushort *)in++ */ | ||
| 552 | movb bits_r, %cl /* cl = bits, needs it for shifting */ | ||
| 553 | addb $16, bits_r /* bits += 16 */ | ||
| 554 | shll %cl, %eax | ||
| 555 | orl %eax, hold_r /* hold |= *((ushort *)in)++ << bits */ | ||
| 556 | |||
| 557 | .L_get_distance_code: | ||
| 558 | movl dmask(%esp), %edx /* edx = dmask */ | ||
| 559 | movl dcode(%esp), %ecx /* ecx = dcode */ | ||
| 560 | andl hold_r, %edx /* edx &= hold */ | ||
| 561 | movl (%ecx,%edx,4), %eax /* eax = dcode[hold & dmask] */ | ||
| 562 | |||
| 563 | #define dist_r %edx | ||
| 564 | .L_dodist: | ||
| 565 | movl %eax, dist_r /* dist = this */ | ||
| 566 | shrl $16, dist_r /* dist = this.val */ | ||
| 567 | movb %ah, %cl | ||
| 568 | subb %ah, bits_r /* bits -= this.bits */ | ||
| 569 | shrl %cl, hold_r /* hold >>= this.bits */ | ||
| 570 | |||
| 571 | /* if (op & 16) { | ||
| 572 | * dist = this.val | ||
| 573 | * op &= 15 | ||
| 574 | * if (op > bits) { | ||
| 575 | * hold |= *((unsigned short *)in)++ << bits; | ||
| 576 | * bits += 16 | ||
| 577 | * } | ||
| 578 | * dist += hold & mask[op]; | ||
| 579 | * bits -= op; | ||
| 580 | * hold >>= op; | ||
| 581 | */ | ||
| 582 | movb %al, %cl /* cl = this.op */ | ||
| 583 | |||
| 584 | testb $16, %al /* if ((op & 16) == 0) */ | ||
| 585 | jz .L_test_for_second_level_dist | ||
| 586 | andb $15, %cl /* op &= 15 */ | ||
| 587 | jz .L_check_dist_one | ||
| 588 | cmpb %cl, bits_r | ||
| 589 | jae .L_add_bits_to_dist /* if (op <= bits) 97.6% */ | ||
| 590 | |||
| 591 | movb %cl, %ch /* stash op in ch, freeing cl */ | ||
| 592 | xorl %eax, %eax | ||
| 593 | lodsw /* al = *(ushort *)in++ */ | ||
| 594 | movb bits_r, %cl /* cl = bits, needs it for shifting */ | ||
| 595 | addb $16, bits_r /* bits += 16 */ | ||
| 596 | shll %cl, %eax | ||
| 597 | orl %eax, hold_r /* hold |= *((ushort *)in)++ << bits */ | ||
| 598 | movb %ch, %cl /* move op back to ecx */ | ||
| 599 | |||
| 600 | .L_add_bits_to_dist: | ||
| 601 | movl $1, %eax | ||
| 602 | shll %cl, %eax | ||
| 603 | decl %eax /* (1 << op) - 1 */ | ||
| 604 | subb %cl, bits_r | ||
| 605 | andl hold_r, %eax /* eax &= hold */ | ||
| 606 | shrl %cl, hold_r | ||
| 607 | addl %eax, dist_r /* dist += hold & ((1 << op) - 1) */ | ||
| 608 | jmp .L_check_window | ||
| 609 | |||
| 610 | .L_check_window: | ||
| 611 | /* regs: %esi = from, %ebp = hold, %bl = bits, %edi = out, %edx = dist | ||
| 612 | * %ecx = nbytes | ||
| 613 | * | ||
| 614 | * nbytes = out - beg; | ||
| 615 | * if (dist <= nbytes) { | ||
| 616 | * from = out - dist; | ||
| 617 | * do { | ||
| 618 | * PUP(out) = PUP(from); | ||
| 619 | * } while (--len > 0) { | ||
| 620 | * } | ||
| 621 | */ | ||
| 622 | |||
| 623 | movl in_r, in(%esp) /* save in so from can use it's reg */ | ||
| 624 | movl out_r, %eax | ||
| 625 | subl beg(%esp), %eax /* nbytes = out - beg */ | ||
| 626 | |||
| 627 | cmpl dist_r, %eax | ||
| 628 | jb .L_clip_window /* if (dist > nbytes) 4.2% */ | ||
| 629 | |||
| 630 | movl len(%esp), %ecx | ||
| 631 | movl out_r, from_r | ||
| 632 | subl dist_r, from_r /* from = out - dist */ | ||
| 633 | |||
| 634 | subl $3, %ecx | ||
| 635 | movb (from_r), %al | ||
| 636 | movb %al, (out_r) | ||
| 637 | movb 1(from_r), %al | ||
| 638 | movb 2(from_r), %dl | ||
| 639 | addl $3, from_r | ||
| 640 | movb %al, 1(out_r) | ||
| 641 | movb %dl, 2(out_r) | ||
| 642 | addl $3, out_r | ||
| 643 | rep movsb | ||
| 644 | |||
| 645 | movl in(%esp), in_r /* move in back to %esi, toss from */ | ||
| 646 | jmp .L_while_test | ||
| 647 | |||
| 648 | .align 16,0x90 | ||
| 649 | .L_check_dist_one: | ||
| 650 | cmpl $1, dist_r | ||
| 651 | jne .L_check_window | ||
| 652 | cmpl out_r, beg(%esp) | ||
| 653 | je .L_check_window | ||
| 654 | |||
| 655 | decl out_r | ||
| 656 | movl len(%esp), %ecx | ||
| 657 | movb (out_r), %al | ||
| 658 | subl $3, %ecx | ||
| 659 | |||
| 660 | movb %al, 1(out_r) | ||
| 661 | movb %al, 2(out_r) | ||
| 662 | movb %al, 3(out_r) | ||
| 663 | addl $4, out_r | ||
| 664 | rep stosb | ||
| 665 | |||
| 666 | jmp .L_while_test | ||
| 667 | |||
| 668 | .align 16,0x90 | ||
| 669 | .L_test_for_second_level_length: | ||
| 670 | /* else if ((op & 64) == 0) { | ||
| 671 | * this = lcode[this.val + (hold & mask[op])]; | ||
| 672 | * } | ||
| 673 | */ | ||
| 674 | testb $64, %al | ||
| 675 | jnz .L_test_for_end_of_block /* if ((op & 64) != 0) */ | ||
| 676 | |||
| 677 | movl $1, %eax | ||
| 678 | shll %cl, %eax | ||
| 679 | decl %eax | ||
| 680 | andl hold_r, %eax /* eax &= hold */ | ||
| 681 | addl %edx, %eax /* eax += this.val */ | ||
| 682 | movl lcode(%esp), %edx /* edx = lcode */ | ||
| 683 | movl (%edx,%eax,4), %eax /* eax = lcode[val + (hold&mask[op])] */ | ||
| 684 | jmp .L_dolen | ||
| 685 | |||
| 686 | .align 16,0x90 | ||
| 687 | .L_test_for_second_level_dist: | ||
| 688 | /* else if ((op & 64) == 0) { | ||
| 689 | * this = dcode[this.val + (hold & mask[op])]; | ||
| 690 | * } | ||
| 691 | */ | ||
| 692 | testb $64, %al | ||
| 693 | jnz .L_invalid_distance_code /* if ((op & 64) != 0) */ | ||
| 694 | |||
| 695 | movl $1, %eax | ||
| 696 | shll %cl, %eax | ||
| 697 | decl %eax | ||
| 698 | andl hold_r, %eax /* eax &= hold */ | ||
| 699 | addl %edx, %eax /* eax += this.val */ | ||
| 700 | movl dcode(%esp), %edx /* edx = dcode */ | ||
| 701 | movl (%edx,%eax,4), %eax /* eax = dcode[val + (hold&mask[op])] */ | ||
| 702 | jmp .L_dodist | ||
| 703 | |||
| 704 | .align 16,0x90 | ||
| 705 | .L_clip_window: | ||
| 706 | /* regs: %esi = from, %ebp = hold, %bl = bits, %edi = out, %edx = dist | ||
| 707 | * %ecx = nbytes | ||
| 708 | * | ||
| 709 | * else { | ||
| 710 | * if (dist > wsize) { | ||
| 711 | * invalid distance | ||
| 712 | * } | ||
| 713 | * from = window; | ||
| 714 | * nbytes = dist - nbytes; | ||
| 715 | * if (write == 0) { | ||
| 716 | * from += wsize - nbytes; | ||
| 717 | */ | ||
| 718 | #define nbytes_r %ecx | ||
| 719 | movl %eax, nbytes_r | ||
| 720 | movl wsize(%esp), %eax /* prepare for dist compare */ | ||
| 721 | negl nbytes_r /* nbytes = -nbytes */ | ||
| 722 | movl window(%esp), from_r /* from = window */ | ||
| 723 | |||
| 724 | cmpl dist_r, %eax | ||
| 725 | jb .L_invalid_distance_too_far /* if (dist > wsize) */ | ||
| 726 | |||
| 727 | addl dist_r, nbytes_r /* nbytes = dist - nbytes */ | ||
| 728 | cmpl $0, write(%esp) | ||
| 729 | jne .L_wrap_around_window /* if (write != 0) */ | ||
| 730 | |||
| 731 | subl nbytes_r, %eax | ||
| 732 | addl %eax, from_r /* from += wsize - nbytes */ | ||
| 733 | |||
| 734 | /* regs: %esi = from, %ebp = hold, %bl = bits, %edi = out, %edx = dist | ||
| 735 | * %ecx = nbytes, %eax = len | ||
| 736 | * | ||
| 737 | * if (nbytes < len) { | ||
| 738 | * len -= nbytes; | ||
| 739 | * do { | ||
| 740 | * PUP(out) = PUP(from); | ||
| 741 | * } while (--nbytes); | ||
| 742 | * from = out - dist; | ||
| 743 | * } | ||
| 744 | * } | ||
| 745 | */ | ||
| 746 | #define len_r %eax | ||
| 747 | movl len(%esp), len_r | ||
| 748 | cmpl nbytes_r, len_r | ||
| 749 | jbe .L_do_copy1 /* if (nbytes >= len) */ | ||
| 750 | |||
| 751 | subl nbytes_r, len_r /* len -= nbytes */ | ||
| 752 | rep movsb | ||
| 753 | movl out_r, from_r | ||
| 754 | subl dist_r, from_r /* from = out - dist */ | ||
| 755 | jmp .L_do_copy1 | ||
| 756 | |||
| 757 | cmpl nbytes_r, len_r | ||
| 758 | jbe .L_do_copy1 /* if (nbytes >= len) */ | ||
| 759 | |||
| 760 | subl nbytes_r, len_r /* len -= nbytes */ | ||
| 761 | rep movsb | ||
| 762 | movl out_r, from_r | ||
| 763 | subl dist_r, from_r /* from = out - dist */ | ||
| 764 | jmp .L_do_copy1 | ||
| 765 | |||
| 766 | .L_wrap_around_window: | ||
| 767 | /* regs: %esi = from, %ebp = hold, %bl = bits, %edi = out, %edx = dist | ||
| 768 | * %ecx = nbytes, %eax = write, %eax = len | ||
| 769 | * | ||
| 770 | * else if (write < nbytes) { | ||
| 771 | * from += wsize + write - nbytes; | ||
| 772 | * nbytes -= write; | ||
| 773 | * if (nbytes < len) { | ||
| 774 | * len -= nbytes; | ||
| 775 | * do { | ||
| 776 | * PUP(out) = PUP(from); | ||
| 777 | * } while (--nbytes); | ||
| 778 | * from = window; | ||
| 779 | * nbytes = write; | ||
| 780 | * if (nbytes < len) { | ||
| 781 | * len -= nbytes; | ||
| 782 | * do { | ||
| 783 | * PUP(out) = PUP(from); | ||
| 784 | * } while(--nbytes); | ||
| 785 | * from = out - dist; | ||
| 786 | * } | ||
| 787 | * } | ||
| 788 | * } | ||
| 789 | */ | ||
| 790 | #define write_r %eax | ||
| 791 | movl write(%esp), write_r | ||
| 792 | cmpl write_r, nbytes_r | ||
| 793 | jbe .L_contiguous_in_window /* if (write >= nbytes) */ | ||
| 794 | |||
| 795 | addl wsize(%esp), from_r | ||
| 796 | addl write_r, from_r | ||
| 797 | subl nbytes_r, from_r /* from += wsize + write - nbytes */ | ||
| 798 | subl write_r, nbytes_r /* nbytes -= write */ | ||
| 799 | #undef write_r | ||
| 800 | |||
| 801 | movl len(%esp), len_r | ||
| 802 | cmpl nbytes_r, len_r | ||
| 803 | jbe .L_do_copy1 /* if (nbytes >= len) */ | ||
| 804 | |||
| 805 | subl nbytes_r, len_r /* len -= nbytes */ | ||
| 806 | rep movsb | ||
| 807 | movl window(%esp), from_r /* from = window */ | ||
| 808 | movl write(%esp), nbytes_r /* nbytes = write */ | ||
| 809 | cmpl nbytes_r, len_r | ||
| 810 | jbe .L_do_copy1 /* if (nbytes >= len) */ | ||
| 811 | |||
| 812 | subl nbytes_r, len_r /* len -= nbytes */ | ||
| 813 | rep movsb | ||
| 814 | movl out_r, from_r | ||
| 815 | subl dist_r, from_r /* from = out - dist */ | ||
| 816 | jmp .L_do_copy1 | ||
| 817 | |||
| 818 | .L_contiguous_in_window: | ||
| 819 | /* regs: %esi = from, %ebp = hold, %bl = bits, %edi = out, %edx = dist | ||
| 820 | * %ecx = nbytes, %eax = write, %eax = len | ||
| 821 | * | ||
| 822 | * else { | ||
| 823 | * from += write - nbytes; | ||
| 824 | * if (nbytes < len) { | ||
| 825 | * len -= nbytes; | ||
| 826 | * do { | ||
| 827 | * PUP(out) = PUP(from); | ||
| 828 | * } while (--nbytes); | ||
| 829 | * from = out - dist; | ||
| 830 | * } | ||
| 831 | * } | ||
| 832 | */ | ||
| 833 | #define write_r %eax | ||
| 834 | addl write_r, from_r | ||
| 835 | subl nbytes_r, from_r /* from += write - nbytes */ | ||
| 836 | #undef write_r | ||
| 837 | |||
| 838 | movl len(%esp), len_r | ||
| 839 | cmpl nbytes_r, len_r | ||
| 840 | jbe .L_do_copy1 /* if (nbytes >= len) */ | ||
| 841 | |||
| 842 | subl nbytes_r, len_r /* len -= nbytes */ | ||
| 843 | rep movsb | ||
| 844 | movl out_r, from_r | ||
| 845 | subl dist_r, from_r /* from = out - dist */ | ||
| 846 | |||
| 847 | .L_do_copy1: | ||
| 848 | /* regs: %esi = from, %esi = in, %ebp = hold, %bl = bits, %edi = out | ||
| 849 | * %eax = len | ||
| 850 | * | ||
| 851 | * while (len > 0) { | ||
| 852 | * PUP(out) = PUP(from); | ||
| 853 | * len--; | ||
| 854 | * } | ||
| 855 | * } | ||
| 856 | * } while (in < last && out < end); | ||
| 857 | */ | ||
| 858 | #undef nbytes_r | ||
| 859 | #define in_r %esi | ||
| 860 | movl len_r, %ecx | ||
| 861 | rep movsb | ||
| 862 | |||
| 863 | movl in(%esp), in_r /* move in back to %esi, toss from */ | ||
| 864 | jmp .L_while_test | ||
| 865 | |||
| 866 | #undef len_r | ||
| 867 | #undef dist_r | ||
| 868 | |||
| 869 | #endif /* NO_MMX || RUN_TIME_MMX */ | ||
| 870 | |||
| 871 | |||
| 872 | /*** MMX code ***/ | ||
| 873 | |||
| 874 | #if defined( USE_MMX ) || defined( RUN_TIME_MMX ) | ||
| 875 | |||
| 876 | .align 32,0x90 | ||
| 877 | .L_init_mmx: | ||
| 878 | emms | ||
| 879 | |||
| 880 | #undef bits_r | ||
| 881 | #undef bitslong_r | ||
| 882 | #define bitslong_r %ebp | ||
| 883 | #define hold_mm %mm0 | ||
| 884 | movd %ebp, hold_mm | ||
| 885 | movl %ebx, bitslong_r | ||
| 886 | |||
| 887 | #define used_mm %mm1 | ||
| 888 | #define dmask2_mm %mm2 | ||
| 889 | #define lmask2_mm %mm3 | ||
| 890 | #define lmask_mm %mm4 | ||
| 891 | #define dmask_mm %mm5 | ||
| 892 | #define tmp_mm %mm6 | ||
| 893 | |||
| 894 | movd lmask(%esp), lmask_mm | ||
| 895 | movq lmask_mm, lmask2_mm | ||
| 896 | movd dmask(%esp), dmask_mm | ||
| 897 | movq dmask_mm, dmask2_mm | ||
| 898 | pxor used_mm, used_mm | ||
| 899 | movl lcode(%esp), %ebx /* ebx = lcode */ | ||
| 900 | jmp .L_do_loop_mmx | ||
| 901 | |||
| 902 | .align 32,0x90 | ||
| 903 | .L_while_test_mmx: | ||
| 904 | /* while (in < last && out < end) | ||
| 905 | */ | ||
| 906 | cmpl out_r, end(%esp) | ||
| 907 | jbe .L_break_loop /* if (out >= end) */ | ||
| 908 | |||
| 909 | cmpl in_r, last(%esp) | ||
| 910 | jbe .L_break_loop | ||
| 911 | |||
| 912 | .L_do_loop_mmx: | ||
| 913 | psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ | ||
| 914 | |||
| 915 | cmpl $32, bitslong_r | ||
| 916 | ja .L_get_length_code_mmx /* if (32 < bits) */ | ||
| 917 | |||
| 918 | movd bitslong_r, tmp_mm | ||
| 919 | movd (in_r), %mm7 | ||
| 920 | addl $4, in_r | ||
| 921 | psllq tmp_mm, %mm7 | ||
| 922 | addl $32, bitslong_r | ||
| 923 | por %mm7, hold_mm /* hold_mm |= *((uint *)in)++ << bits */ | ||
| 924 | |||
| 925 | .L_get_length_code_mmx: | ||
| 926 | pand hold_mm, lmask_mm | ||
| 927 | movd lmask_mm, %eax | ||
| 928 | movq lmask2_mm, lmask_mm | ||
| 929 | movl (%ebx,%eax,4), %eax /* eax = lcode[hold & lmask] */ | ||
| 930 | |||
| 931 | .L_dolen_mmx: | ||
| 932 | movzbl %ah, %ecx /* ecx = this.bits */ | ||
| 933 | movd %ecx, used_mm | ||
| 934 | subl %ecx, bitslong_r /* bits -= this.bits */ | ||
| 935 | |||
| 936 | testb %al, %al | ||
| 937 | jnz .L_test_for_length_base_mmx /* if (op != 0) 45.7% */ | ||
| 938 | |||
| 939 | shrl $16, %eax /* output this.val char */ | ||
| 940 | stosb | ||
| 941 | jmp .L_while_test_mmx | ||
| 942 | |||
| 943 | .L_test_for_length_base_mmx: | ||
| 944 | #define len_r %edx | ||
| 945 | movl %eax, len_r /* len = this */ | ||
| 946 | shrl $16, len_r /* len = this.val */ | ||
| 947 | |||
| 948 | testb $16, %al | ||
| 949 | jz .L_test_for_second_level_length_mmx /* if ((op & 16) == 0) 8% */ | ||
| 950 | andl $15, %eax /* op &= 15 */ | ||
| 951 | jz .L_decode_distance_mmx /* if (!op) */ | ||
| 952 | |||
| 953 | psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ | ||
| 954 | movd %eax, used_mm | ||
| 955 | movd hold_mm, %ecx | ||
| 956 | subl %eax, bitslong_r | ||
| 957 | andl .L_mask(,%eax,4), %ecx | ||
| 958 | addl %ecx, len_r /* len += hold & mask[op] */ | ||
| 959 | |||
| 960 | .L_decode_distance_mmx: | ||
| 961 | psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ | ||
| 962 | |||
| 963 | cmpl $32, bitslong_r | ||
| 964 | ja .L_get_dist_code_mmx /* if (32 < bits) */ | ||
| 965 | |||
| 966 | movd bitslong_r, tmp_mm | ||
| 967 | movd (in_r), %mm7 | ||
| 968 | addl $4, in_r | ||
| 969 | psllq tmp_mm, %mm7 | ||
| 970 | addl $32, bitslong_r | ||
| 971 | por %mm7, hold_mm /* hold_mm |= *((uint *)in)++ << bits */ | ||
| 972 | |||
| 973 | .L_get_dist_code_mmx: | ||
| 974 | movl dcode(%esp), %ebx /* ebx = dcode */ | ||
| 975 | pand hold_mm, dmask_mm | ||
| 976 | movd dmask_mm, %eax | ||
| 977 | movq dmask2_mm, dmask_mm | ||
| 978 | movl (%ebx,%eax,4), %eax /* eax = dcode[hold & lmask] */ | ||
| 979 | |||
| 980 | .L_dodist_mmx: | ||
| 981 | #define dist_r %ebx | ||
| 982 | movzbl %ah, %ecx /* ecx = this.bits */ | ||
| 983 | movl %eax, dist_r | ||
| 984 | shrl $16, dist_r /* dist = this.val */ | ||
| 985 | subl %ecx, bitslong_r /* bits -= this.bits */ | ||
| 986 | movd %ecx, used_mm | ||
| 987 | |||
| 988 | testb $16, %al /* if ((op & 16) == 0) */ | ||
| 989 | jz .L_test_for_second_level_dist_mmx | ||
| 990 | andl $15, %eax /* op &= 15 */ | ||
| 991 | jz .L_check_dist_one_mmx | ||
| 992 | |||
| 993 | .L_add_bits_to_dist_mmx: | ||
| 994 | psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ | ||
| 995 | movd %eax, used_mm /* save bit length of current op */ | ||
| 996 | movd hold_mm, %ecx /* get the next bits on input stream */ | ||
| 997 | subl %eax, bitslong_r /* bits -= op bits */ | ||
| 998 | andl .L_mask(,%eax,4), %ecx /* ecx = hold & mask[op] */ | ||
| 999 | addl %ecx, dist_r /* dist += hold & mask[op] */ | ||
| 1000 | |||
| 1001 | .L_check_window_mmx: | ||
| 1002 | movl in_r, in(%esp) /* save in so from can use it's reg */ | ||
| 1003 | movl out_r, %eax | ||
| 1004 | subl beg(%esp), %eax /* nbytes = out - beg */ | ||
| 1005 | |||
| 1006 | cmpl dist_r, %eax | ||
| 1007 | jb .L_clip_window_mmx /* if (dist > nbytes) 4.2% */ | ||
| 1008 | |||
| 1009 | movl len_r, %ecx | ||
| 1010 | movl out_r, from_r | ||
| 1011 | subl dist_r, from_r /* from = out - dist */ | ||
| 1012 | |||
| 1013 | subl $3, %ecx | ||
| 1014 | movb (from_r), %al | ||
| 1015 | movb %al, (out_r) | ||
| 1016 | movb 1(from_r), %al | ||
| 1017 | movb 2(from_r), %dl | ||
| 1018 | addl $3, from_r | ||
| 1019 | movb %al, 1(out_r) | ||
| 1020 | movb %dl, 2(out_r) | ||
| 1021 | addl $3, out_r | ||
| 1022 | rep movsb | ||
| 1023 | |||
| 1024 | movl in(%esp), in_r /* move in back to %esi, toss from */ | ||
| 1025 | movl lcode(%esp), %ebx /* move lcode back to %ebx, toss dist */ | ||
| 1026 | jmp .L_while_test_mmx | ||
| 1027 | |||
| 1028 | .align 16,0x90 | ||
| 1029 | .L_check_dist_one_mmx: | ||
| 1030 | cmpl $1, dist_r | ||
| 1031 | jne .L_check_window_mmx | ||
| 1032 | cmpl out_r, beg(%esp) | ||
| 1033 | je .L_check_window_mmx | ||
| 1034 | |||
| 1035 | decl out_r | ||
| 1036 | movl len_r, %ecx | ||
| 1037 | movb (out_r), %al | ||
| 1038 | subl $3, %ecx | ||
| 1039 | |||
| 1040 | movb %al, 1(out_r) | ||
| 1041 | movb %al, 2(out_r) | ||
| 1042 | movb %al, 3(out_r) | ||
| 1043 | addl $4, out_r | ||
| 1044 | rep stosb | ||
| 1045 | |||
| 1046 | movl lcode(%esp), %ebx /* move lcode back to %ebx, toss dist */ | ||
| 1047 | jmp .L_while_test_mmx | ||
| 1048 | |||
| 1049 | .align 16,0x90 | ||
| 1050 | .L_test_for_second_level_length_mmx: | ||
| 1051 | testb $64, %al | ||
| 1052 | jnz .L_test_for_end_of_block /* if ((op & 64) != 0) */ | ||
| 1053 | |||
| 1054 | andl $15, %eax | ||
| 1055 | psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ | ||
| 1056 | movd hold_mm, %ecx | ||
| 1057 | andl .L_mask(,%eax,4), %ecx | ||
| 1058 | addl len_r, %ecx | ||
| 1059 | movl (%ebx,%ecx,4), %eax /* eax = lcode[hold & lmask] */ | ||
| 1060 | jmp .L_dolen_mmx | ||
| 1061 | |||
| 1062 | .align 16,0x90 | ||
| 1063 | .L_test_for_second_level_dist_mmx: | ||
| 1064 | testb $64, %al | ||
| 1065 | jnz .L_invalid_distance_code /* if ((op & 64) != 0) */ | ||
| 1066 | |||
| 1067 | andl $15, %eax | ||
| 1068 | psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ | ||
| 1069 | movd hold_mm, %ecx | ||
| 1070 | andl .L_mask(,%eax,4), %ecx | ||
| 1071 | movl dcode(%esp), %eax /* ecx = dcode */ | ||
| 1072 | addl dist_r, %ecx | ||
| 1073 | movl (%eax,%ecx,4), %eax /* eax = lcode[hold & lmask] */ | ||
| 1074 | jmp .L_dodist_mmx | ||
| 1075 | |||
| 1076 | .align 16,0x90 | ||
| 1077 | .L_clip_window_mmx: | ||
| 1078 | #define nbytes_r %ecx | ||
| 1079 | movl %eax, nbytes_r | ||
| 1080 | movl wsize(%esp), %eax /* prepare for dist compare */ | ||
| 1081 | negl nbytes_r /* nbytes = -nbytes */ | ||
| 1082 | movl window(%esp), from_r /* from = window */ | ||
| 1083 | |||
| 1084 | cmpl dist_r, %eax | ||
| 1085 | jb .L_invalid_distance_too_far /* if (dist > wsize) */ | ||
| 1086 | |||
| 1087 | addl dist_r, nbytes_r /* nbytes = dist - nbytes */ | ||
| 1088 | cmpl $0, write(%esp) | ||
| 1089 | jne .L_wrap_around_window_mmx /* if (write != 0) */ | ||
| 1090 | |||
| 1091 | subl nbytes_r, %eax | ||
| 1092 | addl %eax, from_r /* from += wsize - nbytes */ | ||
| 1093 | |||
| 1094 | cmpl nbytes_r, len_r | ||
| 1095 | jbe .L_do_copy1_mmx /* if (nbytes >= len) */ | ||
| 1096 | |||
| 1097 | subl nbytes_r, len_r /* len -= nbytes */ | ||
| 1098 | rep movsb | ||
| 1099 | movl out_r, from_r | ||
| 1100 | subl dist_r, from_r /* from = out - dist */ | ||
| 1101 | jmp .L_do_copy1_mmx | ||
| 1102 | |||
| 1103 | cmpl nbytes_r, len_r | ||
| 1104 | jbe .L_do_copy1_mmx /* if (nbytes >= len) */ | ||
| 1105 | |||
| 1106 | subl nbytes_r, len_r /* len -= nbytes */ | ||
| 1107 | rep movsb | ||
| 1108 | movl out_r, from_r | ||
| 1109 | subl dist_r, from_r /* from = out - dist */ | ||
| 1110 | jmp .L_do_copy1_mmx | ||
| 1111 | |||
| 1112 | .L_wrap_around_window_mmx: | ||
| 1113 | #define write_r %eax | ||
| 1114 | movl write(%esp), write_r | ||
| 1115 | cmpl write_r, nbytes_r | ||
| 1116 | jbe .L_contiguous_in_window_mmx /* if (write >= nbytes) */ | ||
| 1117 | |||
| 1118 | addl wsize(%esp), from_r | ||
| 1119 | addl write_r, from_r | ||
| 1120 | subl nbytes_r, from_r /* from += wsize + write - nbytes */ | ||
| 1121 | subl write_r, nbytes_r /* nbytes -= write */ | ||
| 1122 | #undef write_r | ||
| 1123 | |||
| 1124 | cmpl nbytes_r, len_r | ||
| 1125 | jbe .L_do_copy1_mmx /* if (nbytes >= len) */ | ||
| 1126 | |||
| 1127 | subl nbytes_r, len_r /* len -= nbytes */ | ||
| 1128 | rep movsb | ||
| 1129 | movl window(%esp), from_r /* from = window */ | ||
| 1130 | movl write(%esp), nbytes_r /* nbytes = write */ | ||
| 1131 | cmpl nbytes_r, len_r | ||
| 1132 | jbe .L_do_copy1_mmx /* if (nbytes >= len) */ | ||
| 1133 | |||
| 1134 | subl nbytes_r, len_r /* len -= nbytes */ | ||
| 1135 | rep movsb | ||
| 1136 | movl out_r, from_r | ||
| 1137 | subl dist_r, from_r /* from = out - dist */ | ||
| 1138 | jmp .L_do_copy1_mmx | ||
| 1139 | |||
| 1140 | .L_contiguous_in_window_mmx: | ||
| 1141 | #define write_r %eax | ||
| 1142 | addl write_r, from_r | ||
| 1143 | subl nbytes_r, from_r /* from += write - nbytes */ | ||
| 1144 | #undef write_r | ||
| 1145 | |||
| 1146 | cmpl nbytes_r, len_r | ||
| 1147 | jbe .L_do_copy1_mmx /* if (nbytes >= len) */ | ||
| 1148 | |||
| 1149 | subl nbytes_r, len_r /* len -= nbytes */ | ||
| 1150 | rep movsb | ||
| 1151 | movl out_r, from_r | ||
| 1152 | subl dist_r, from_r /* from = out - dist */ | ||
| 1153 | |||
| 1154 | .L_do_copy1_mmx: | ||
| 1155 | #undef nbytes_r | ||
| 1156 | #define in_r %esi | ||
| 1157 | movl len_r, %ecx | ||
| 1158 | rep movsb | ||
| 1159 | |||
| 1160 | movl in(%esp), in_r /* move in back to %esi, toss from */ | ||
| 1161 | movl lcode(%esp), %ebx /* move lcode back to %ebx, toss dist */ | ||
| 1162 | jmp .L_while_test_mmx | ||
| 1163 | |||
| 1164 | #undef hold_r | ||
| 1165 | #undef bitslong_r | ||
| 1166 | |||
| 1167 | #endif /* USE_MMX || RUN_TIME_MMX */ | ||
| 1168 | |||
| 1169 | |||
| 1170 | /*** USE_MMX, NO_MMX, and RUNTIME_MMX from here on ***/ | ||
| 1171 | |||
| 1172 | .L_invalid_distance_code: | ||
| 1173 | /* else { | ||
| 1174 | * strm->msg = "invalid distance code"; | ||
| 1175 | * state->mode = BAD; | ||
| 1176 | * } | ||
| 1177 | */ | ||
| 1178 | movl $.L_invalid_distance_code_msg, %ecx | ||
| 1179 | movl $INFLATE_MODE_BAD, %edx | ||
| 1180 | jmp .L_update_stream_state | ||
| 1181 | |||
| 1182 | .L_test_for_end_of_block: | ||
| 1183 | /* else if (op & 32) { | ||
| 1184 | * state->mode = TYPE; | ||
| 1185 | * break; | ||
| 1186 | * } | ||
| 1187 | */ | ||
| 1188 | testb $32, %al | ||
| 1189 | jz .L_invalid_literal_length_code /* if ((op & 32) == 0) */ | ||
| 1190 | |||
| 1191 | movl $0, %ecx | ||
| 1192 | movl $INFLATE_MODE_TYPE, %edx | ||
| 1193 | jmp .L_update_stream_state | ||
| 1194 | |||
| 1195 | .L_invalid_literal_length_code: | ||
| 1196 | /* else { | ||
| 1197 | * strm->msg = "invalid literal/length code"; | ||
| 1198 | * state->mode = BAD; | ||
| 1199 | * } | ||
| 1200 | */ | ||
| 1201 | movl $.L_invalid_literal_length_code_msg, %ecx | ||
| 1202 | movl $INFLATE_MODE_BAD, %edx | ||
| 1203 | jmp .L_update_stream_state | ||
| 1204 | |||
| 1205 | .L_invalid_distance_too_far: | ||
| 1206 | /* strm->msg = "invalid distance too far back"; | ||
| 1207 | * state->mode = BAD; | ||
| 1208 | */ | ||
| 1209 | movl in(%esp), in_r /* from_r has in's reg, put in back */ | ||
| 1210 | movl $.L_invalid_distance_too_far_msg, %ecx | ||
| 1211 | movl $INFLATE_MODE_BAD, %edx | ||
| 1212 | jmp .L_update_stream_state | ||
| 1213 | |||
| 1214 | .L_update_stream_state: | ||
| 1215 | /* set strm->msg = %ecx, strm->state->mode = %edx */ | ||
| 1216 | movl strm_sp(%esp), %eax | ||
| 1217 | testl %ecx, %ecx /* if (msg != NULL) */ | ||
| 1218 | jz .L_skip_msg | ||
| 1219 | movl %ecx, msg_strm(%eax) /* strm->msg = msg */ | ||
| 1220 | .L_skip_msg: | ||
| 1221 | movl state_strm(%eax), %eax /* state = strm->state */ | ||
| 1222 | movl %edx, mode_state(%eax) /* state->mode = edx (BAD | TYPE) */ | ||
| 1223 | jmp .L_break_loop | ||
| 1224 | |||
| 1225 | .align 32,0x90 | ||
| 1226 | .L_break_loop: | ||
| 1227 | |||
| 1228 | /* | ||
| 1229 | * Regs: | ||
| 1230 | * | ||
| 1231 | * bits = %ebp when mmx, and in %ebx when non-mmx | ||
| 1232 | * hold = %hold_mm when mmx, and in %ebp when non-mmx | ||
| 1233 | * in = %esi | ||
| 1234 | * out = %edi | ||
| 1235 | */ | ||
| 1236 | |||
| 1237 | #if defined( USE_MMX ) || defined( RUN_TIME_MMX ) | ||
| 1238 | |||
| 1239 | #if defined( RUN_TIME_MMX ) | ||
| 1240 | |||
| 1241 | cmpl $DO_USE_MMX, inflate_fast_use_mmx | ||
| 1242 | jne .L_update_next_in | ||
| 1243 | |||
| 1244 | #endif /* RUN_TIME_MMX */ | ||
| 1245 | |||
| 1246 | movl %ebp, %ebx | ||
| 1247 | |||
| 1248 | .L_update_next_in: | ||
| 1249 | |||
| 1250 | #endif | ||
| 1251 | |||
| 1252 | #define strm_r %eax | ||
| 1253 | #define state_r %edx | ||
| 1254 | |||
| 1255 | /* len = bits >> 3; | ||
| 1256 | * in -= len; | ||
| 1257 | * bits -= len << 3; | ||
| 1258 | * hold &= (1U << bits) - 1; | ||
| 1259 | * state->hold = hold; | ||
| 1260 | * state->bits = bits; | ||
| 1261 | * strm->next_in = in; | ||
| 1262 | * strm->next_out = out; | ||
| 1263 | */ | ||
| 1264 | movl strm_sp(%esp), strm_r | ||
| 1265 | movl %ebx, %ecx | ||
| 1266 | movl state_strm(strm_r), state_r | ||
| 1267 | shrl $3, %ecx | ||
| 1268 | subl %ecx, in_r | ||
| 1269 | shll $3, %ecx | ||
| 1270 | subl %ecx, %ebx | ||
| 1271 | movl out_r, next_out_strm(strm_r) | ||
| 1272 | movl %ebx, bits_state(state_r) | ||
| 1273 | movl %ebx, %ecx | ||
| 1274 | |||
| 1275 | leal buf(%esp), %ebx | ||
| 1276 | cmpl %ebx, last(%esp) | ||
| 1277 | jne .L_buf_not_used /* if buf != last */ | ||
| 1278 | |||
| 1279 | subl %ebx, in_r /* in -= buf */ | ||
| 1280 | movl next_in_strm(strm_r), %ebx | ||
| 1281 | movl %ebx, last(%esp) /* last = strm->next_in */ | ||
| 1282 | addl %ebx, in_r /* in += strm->next_in */ | ||
| 1283 | movl avail_in_strm(strm_r), %ebx | ||
| 1284 | subl $11, %ebx | ||
| 1285 | addl %ebx, last(%esp) /* last = &strm->next_in[ avail_in - 11 ] */ | ||
| 1286 | |||
| 1287 | .L_buf_not_used: | ||
| 1288 | movl in_r, next_in_strm(strm_r) | ||
| 1289 | |||
| 1290 | movl $1, %ebx | ||
| 1291 | shll %cl, %ebx | ||
| 1292 | decl %ebx | ||
| 1293 | |||
| 1294 | #if defined( USE_MMX ) || defined( RUN_TIME_MMX ) | ||
| 1295 | |||
| 1296 | #if defined( RUN_TIME_MMX ) | ||
| 1297 | |||
| 1298 | cmpl $DO_USE_MMX, inflate_fast_use_mmx | ||
| 1299 | jne .L_update_hold | ||
| 1300 | |||
| 1301 | #endif /* RUN_TIME_MMX */ | ||
| 1302 | |||
| 1303 | psrlq used_mm, hold_mm /* hold_mm >>= last bit length */ | ||
| 1304 | movd hold_mm, %ebp | ||
| 1305 | |||
| 1306 | emms | ||
| 1307 | |||
| 1308 | .L_update_hold: | ||
| 1309 | |||
| 1310 | #endif /* USE_MMX || RUN_TIME_MMX */ | ||
| 1311 | |||
| 1312 | andl %ebx, %ebp | ||
| 1313 | movl %ebp, hold_state(state_r) | ||
| 1314 | |||
| 1315 | #define last_r %ebx | ||
| 1316 | |||
| 1317 | /* strm->avail_in = in < last ? 11 + (last - in) : 11 - (in - last) */ | ||
| 1318 | movl last(%esp), last_r | ||
| 1319 | cmpl in_r, last_r | ||
| 1320 | jbe .L_last_is_smaller /* if (in >= last) */ | ||
| 1321 | |||
| 1322 | subl in_r, last_r /* last -= in */ | ||
| 1323 | addl $11, last_r /* last += 11 */ | ||
| 1324 | movl last_r, avail_in_strm(strm_r) | ||
| 1325 | jmp .L_fixup_out | ||
| 1326 | .L_last_is_smaller: | ||
| 1327 | subl last_r, in_r /* in -= last */ | ||
| 1328 | negl in_r /* in = -in */ | ||
| 1329 | addl $11, in_r /* in += 11 */ | ||
| 1330 | movl in_r, avail_in_strm(strm_r) | ||
| 1331 | |||
| 1332 | #undef last_r | ||
| 1333 | #define end_r %ebx | ||
| 1334 | |||
| 1335 | .L_fixup_out: | ||
| 1336 | /* strm->avail_out = out < end ? 257 + (end - out) : 257 - (out - end)*/ | ||
| 1337 | movl end(%esp), end_r | ||
| 1338 | cmpl out_r, end_r | ||
| 1339 | jbe .L_end_is_smaller /* if (out >= end) */ | ||
| 1340 | |||
| 1341 | subl out_r, end_r /* end -= out */ | ||
| 1342 | addl $257, end_r /* end += 257 */ | ||
| 1343 | movl end_r, avail_out_strm(strm_r) | ||
| 1344 | jmp .L_done | ||
| 1345 | .L_end_is_smaller: | ||
| 1346 | subl end_r, out_r /* out -= end */ | ||
| 1347 | negl out_r /* out = -out */ | ||
| 1348 | addl $257, out_r /* out += 257 */ | ||
| 1349 | movl out_r, avail_out_strm(strm_r) | ||
| 1350 | |||
| 1351 | #undef end_r | ||
| 1352 | #undef strm_r | ||
| 1353 | #undef state_r | ||
| 1354 | |||
| 1355 | .L_done: | ||
| 1356 | addl $local_var_size, %esp | ||
| 1357 | popf | ||
| 1358 | popl %ebx | ||
| 1359 | popl %ebp | ||
| 1360 | popl %esi | ||
| 1361 | popl %edi | ||
| 1362 | ret | ||
| 1363 | |||
| 1364 | #if defined( GAS_ELF ) | ||
| 1365 | /* elf info */ | ||
| 1366 | .type inflate_fast,@function | ||
| 1367 | .size inflate_fast,.-inflate_fast | ||
| 1368 | #endif | ||
diff --git a/contrib/masmx64/bld_ml64.bat b/contrib/masmx64/bld_ml64.bat deleted file mode 100644 index 8f9343d..0000000 --- a/contrib/masmx64/bld_ml64.bat +++ /dev/null | |||
| @@ -1,2 +0,0 @@ | |||
| 1 | ml64.exe /Flinffasx64 /c /Zi inffasx64.asm | ||
| 2 | ml64.exe /Flgvmat64 /c /Zi gvmat64.asm | ||
diff --git a/contrib/masmx64/gvmat64.asm b/contrib/masmx64/gvmat64.asm deleted file mode 100644 index 9879c28..0000000 --- a/contrib/masmx64/gvmat64.asm +++ /dev/null | |||
| @@ -1,553 +0,0 @@ | |||
| 1 | ;uInt longest_match_x64( | ||
| 2 | ; deflate_state *s, | ||
| 3 | ; IPos cur_match); /* current match */ | ||
| 4 | |||
| 5 | ; gvmat64.asm -- Asm portion of the optimized longest_match for 32 bits x86_64 | ||
| 6 | ; (AMD64 on Athlon 64, Opteron, Phenom | ||
| 7 | ; and Intel EM64T on Pentium 4 with EM64T, Pentium D, Core 2 Duo, Core I5/I7) | ||
| 8 | ; Copyright (C) 1995-2010 Jean-loup Gailly, Brian Raiter and Gilles Vollant. | ||
| 9 | ; | ||
| 10 | ; File written by Gilles Vollant, by converting to assembly the longest_match | ||
| 11 | ; from Jean-loup Gailly in deflate.c of zLib and infoZip zip. | ||
| 12 | ; | ||
| 13 | ; and by taking inspiration on asm686 with masm, optimised assembly code | ||
| 14 | ; from Brian Raiter, written 1998 | ||
| 15 | ; | ||
| 16 | ; This software is provided 'as-is', without any express or implied | ||
| 17 | ; warranty. In no event will the authors be held liable for any damages | ||
| 18 | ; arising from the use of this software. | ||
| 19 | ; | ||
| 20 | ; Permission is granted to anyone to use this software for any purpose, | ||
| 21 | ; including commercial applications, and to alter it and redistribute it | ||
| 22 | ; freely, subject to the following restrictions: | ||
| 23 | ; | ||
| 24 | ; 1. The origin of this software must not be misrepresented; you must not | ||
| 25 | ; claim that you wrote the original software. If you use this software | ||
| 26 | ; in a product, an acknowledgment in the product documentation would be | ||
| 27 | ; appreciated but is not required. | ||
| 28 | ; 2. Altered source versions must be plainly marked as such, and must not be | ||
| 29 | ; misrepresented as being the original software | ||
| 30 | ; 3. This notice may not be removed or altered from any source distribution. | ||
| 31 | ; | ||
| 32 | ; | ||
| 33 | ; | ||
| 34 | ; http://www.zlib.net | ||
| 35 | ; http://www.winimage.com/zLibDll | ||
| 36 | ; http://www.muppetlabs.com/~breadbox/software/assembly.html | ||
| 37 | ; | ||
| 38 | ; to compile this file for infozip Zip, I use option: | ||
| 39 | ; ml64.exe /Flgvmat64 /c /Zi /DINFOZIP gvmat64.asm | ||
| 40 | ; | ||
| 41 | ; to compile this file for zLib, I use option: | ||
| 42 | ; ml64.exe /Flgvmat64 /c /Zi gvmat64.asm | ||
| 43 | ; Be carrefull to adapt zlib1222add below to your version of zLib | ||
| 44 | ; (if you use a version of zLib before 1.0.4 or after 1.2.2.2, change | ||
| 45 | ; value of zlib1222add later) | ||
| 46 | ; | ||
| 47 | ; This file compile with Microsoft Macro Assembler (x64) for AMD64 | ||
| 48 | ; | ||
| 49 | ; ml64.exe is given with Visual Studio 2005/2008/2010 and Windows WDK | ||
| 50 | ; | ||
| 51 | ; (you can get Windows WDK with ml64 for AMD64 from | ||
| 52 | ; http://www.microsoft.com/whdc/Devtools/wdk/default.mspx for low price) | ||
| 53 | ; | ||
| 54 | |||
| 55 | |||
| 56 | ;uInt longest_match(s, cur_match) | ||
| 57 | ; deflate_state *s; | ||
| 58 | ; IPos cur_match; /* current match */ | ||
| 59 | .code | ||
| 60 | longest_match PROC | ||
| 61 | |||
| 62 | |||
| 63 | ;LocalVarsSize equ 88 | ||
| 64 | LocalVarsSize equ 72 | ||
| 65 | |||
| 66 | ; register used : rax,rbx,rcx,rdx,rsi,rdi,r8,r9,r10,r11,r12 | ||
| 67 | ; free register : r14,r15 | ||
| 68 | ; register can be saved : rsp | ||
| 69 | |||
| 70 | chainlenwmask equ rsp + 8 - LocalVarsSize ; high word: current chain len | ||
| 71 | ; low word: s->wmask | ||
| 72 | ;window equ rsp + xx - LocalVarsSize ; local copy of s->window ; stored in r10 | ||
| 73 | ;windowbestlen equ rsp + xx - LocalVarsSize ; s->window + bestlen , use r10+r11 | ||
| 74 | ;scanstart equ rsp + xx - LocalVarsSize ; first two bytes of string ; stored in r12w | ||
| 75 | ;scanend equ rsp + xx - LocalVarsSize ; last two bytes of string use ebx | ||
| 76 | ;scanalign equ rsp + xx - LocalVarsSize ; dword-misalignment of string r13 | ||
| 77 | ;bestlen equ rsp + xx - LocalVarsSize ; size of best match so far -> r11d | ||
| 78 | ;scan equ rsp + xx - LocalVarsSize ; ptr to string wanting match -> r9 | ||
| 79 | IFDEF INFOZIP | ||
| 80 | ELSE | ||
| 81 | nicematch equ (rsp + 16 - LocalVarsSize) ; a good enough match size | ||
| 82 | ENDIF | ||
| 83 | |||
| 84 | save_rdi equ rsp + 24 - LocalVarsSize | ||
| 85 | save_rsi equ rsp + 32 - LocalVarsSize | ||
| 86 | save_rbx equ rsp + 40 - LocalVarsSize | ||
| 87 | save_rbp equ rsp + 48 - LocalVarsSize | ||
| 88 | save_r12 equ rsp + 56 - LocalVarsSize | ||
| 89 | save_r13 equ rsp + 64 - LocalVarsSize | ||
| 90 | ;save_r14 equ rsp + 72 - LocalVarsSize | ||
| 91 | ;save_r15 equ rsp + 80 - LocalVarsSize | ||
| 92 | |||
| 93 | |||
| 94 | ; summary of register usage | ||
| 95 | ; scanend ebx | ||
| 96 | ; scanendw bx | ||
| 97 | ; chainlenwmask edx | ||
| 98 | ; curmatch rsi | ||
| 99 | ; curmatchd esi | ||
| 100 | ; windowbestlen r8 | ||
| 101 | ; scanalign r9 | ||
| 102 | ; scanalignd r9d | ||
| 103 | ; window r10 | ||
| 104 | ; bestlen r11 | ||
| 105 | ; bestlend r11d | ||
| 106 | ; scanstart r12d | ||
| 107 | ; scanstartw r12w | ||
| 108 | ; scan r13 | ||
| 109 | ; nicematch r14d | ||
| 110 | ; limit r15 | ||
| 111 | ; limitd r15d | ||
| 112 | ; prev rcx | ||
| 113 | |||
| 114 | ; all the +4 offsets are due to the addition of pending_buf_size (in zlib | ||
| 115 | ; in the deflate_state structure since the asm code was first written | ||
| 116 | ; (if you compile with zlib 1.0.4 or older, remove the +4). | ||
| 117 | ; Note : these value are good with a 8 bytes boundary pack structure | ||
| 118 | |||
| 119 | |||
| 120 | MAX_MATCH equ 258 | ||
| 121 | MIN_MATCH equ 3 | ||
| 122 | MIN_LOOKAHEAD equ (MAX_MATCH+MIN_MATCH+1) | ||
| 123 | |||
| 124 | |||
| 125 | ;;; Offsets for fields in the deflate_state structure. These numbers | ||
| 126 | ;;; are calculated from the definition of deflate_state, with the | ||
| 127 | ;;; assumption that the compiler will dword-align the fields. (Thus, | ||
| 128 | ;;; changing the definition of deflate_state could easily cause this | ||
| 129 | ;;; program to crash horribly, without so much as a warning at | ||
| 130 | ;;; compile time. Sigh.) | ||
| 131 | |||
| 132 | ; all the +zlib1222add offsets are due to the addition of fields | ||
| 133 | ; in zlib in the deflate_state structure since the asm code was first written | ||
| 134 | ; (if you compile with zlib 1.0.4 or older, use "zlib1222add equ (-4)"). | ||
| 135 | ; (if you compile with zlib between 1.0.5 and 1.2.2.1, use "zlib1222add equ 0"). | ||
| 136 | ; if you compile with zlib 1.2.2.2 or later , use "zlib1222add equ 8"). | ||
| 137 | |||
| 138 | |||
| 139 | IFDEF INFOZIP | ||
| 140 | |||
| 141 | _DATA SEGMENT | ||
| 142 | COMM window_size:DWORD | ||
| 143 | ; WMask ; 7fff | ||
| 144 | COMM window:BYTE:010040H | ||
| 145 | COMM prev:WORD:08000H | ||
| 146 | ; MatchLen : unused | ||
| 147 | ; PrevMatch : unused | ||
| 148 | COMM strstart:DWORD | ||
| 149 | COMM match_start:DWORD | ||
| 150 | ; Lookahead : ignore | ||
| 151 | COMM prev_length:DWORD ; PrevLen | ||
| 152 | COMM max_chain_length:DWORD | ||
| 153 | COMM good_match:DWORD | ||
| 154 | COMM nice_match:DWORD | ||
| 155 | prev_ad equ OFFSET prev | ||
| 156 | window_ad equ OFFSET window | ||
| 157 | nicematch equ nice_match | ||
| 158 | _DATA ENDS | ||
| 159 | WMask equ 07fffh | ||
| 160 | |||
| 161 | ELSE | ||
| 162 | |||
| 163 | IFNDEF zlib1222add | ||
| 164 | zlib1222add equ 8 | ||
| 165 | ENDIF | ||
| 166 | dsWSize equ 56+zlib1222add+(zlib1222add/2) | ||
| 167 | dsWMask equ 64+zlib1222add+(zlib1222add/2) | ||
| 168 | dsWindow equ 72+zlib1222add | ||
| 169 | dsPrev equ 88+zlib1222add | ||
| 170 | dsMatchLen equ 128+zlib1222add | ||
| 171 | dsPrevMatch equ 132+zlib1222add | ||
| 172 | dsStrStart equ 140+zlib1222add | ||
| 173 | dsMatchStart equ 144+zlib1222add | ||
| 174 | dsLookahead equ 148+zlib1222add | ||
| 175 | dsPrevLen equ 152+zlib1222add | ||
| 176 | dsMaxChainLen equ 156+zlib1222add | ||
| 177 | dsGoodMatch equ 172+zlib1222add | ||
| 178 | dsNiceMatch equ 176+zlib1222add | ||
| 179 | |||
| 180 | window_size equ [ rcx + dsWSize] | ||
| 181 | WMask equ [ rcx + dsWMask] | ||
| 182 | window_ad equ [ rcx + dsWindow] | ||
| 183 | prev_ad equ [ rcx + dsPrev] | ||
| 184 | strstart equ [ rcx + dsStrStart] | ||
| 185 | match_start equ [ rcx + dsMatchStart] | ||
| 186 | Lookahead equ [ rcx + dsLookahead] ; 0ffffffffh on infozip | ||
| 187 | prev_length equ [ rcx + dsPrevLen] | ||
| 188 | max_chain_length equ [ rcx + dsMaxChainLen] | ||
| 189 | good_match equ [ rcx + dsGoodMatch] | ||
| 190 | nice_match equ [ rcx + dsNiceMatch] | ||
| 191 | ENDIF | ||
| 192 | |||
| 193 | ; parameter 1 in r8(deflate state s), param 2 in rdx (cur match) | ||
| 194 | |||
| 195 | ; see http://weblogs.asp.net/oldnewthing/archive/2004/01/14/58579.aspx and | ||
| 196 | ; http://msdn.microsoft.com/library/en-us/kmarch/hh/kmarch/64bitAMD_8e951dd2-ee77-4728-8702-55ce4b5dd24a.xml.asp | ||
| 197 | ; | ||
| 198 | ; All registers must be preserved across the call, except for | ||
| 199 | ; rax, rcx, rdx, r8, r9, r10, and r11, which are scratch. | ||
| 200 | |||
| 201 | |||
| 202 | |||
| 203 | ;;; Save registers that the compiler may be using, and adjust esp to | ||
| 204 | ;;; make room for our stack frame. | ||
| 205 | |||
| 206 | |||
| 207 | ;;; Retrieve the function arguments. r8d will hold cur_match | ||
| 208 | ;;; throughout the entire function. edx will hold the pointer to the | ||
| 209 | ;;; deflate_state structure during the function's setup (before | ||
| 210 | ;;; entering the main loop. | ||
| 211 | |||
| 212 | ; parameter 1 in rcx (deflate_state* s), param 2 in edx -> r8 (cur match) | ||
| 213 | |||
| 214 | ; this clear high 32 bits of r8, which can be garbage in both r8 and rdx | ||
| 215 | |||
| 216 | mov [save_rdi],rdi | ||
| 217 | mov [save_rsi],rsi | ||
| 218 | mov [save_rbx],rbx | ||
| 219 | mov [save_rbp],rbp | ||
| 220 | IFDEF INFOZIP | ||
| 221 | mov r8d,ecx | ||
| 222 | ELSE | ||
| 223 | mov r8d,edx | ||
| 224 | ENDIF | ||
| 225 | mov [save_r12],r12 | ||
| 226 | mov [save_r13],r13 | ||
| 227 | ; mov [save_r14],r14 | ||
| 228 | ; mov [save_r15],r15 | ||
| 229 | |||
| 230 | |||
| 231 | ;;; uInt wmask = s->w_mask; | ||
| 232 | ;;; unsigned chain_length = s->max_chain_length; | ||
| 233 | ;;; if (s->prev_length >= s->good_match) { | ||
| 234 | ;;; chain_length >>= 2; | ||
| 235 | ;;; } | ||
| 236 | |||
| 237 | mov edi, prev_length | ||
| 238 | mov esi, good_match | ||
| 239 | mov eax, WMask | ||
| 240 | mov ebx, max_chain_length | ||
| 241 | cmp edi, esi | ||
| 242 | jl LastMatchGood | ||
| 243 | shr ebx, 2 | ||
| 244 | LastMatchGood: | ||
| 245 | |||
| 246 | ;;; chainlen is decremented once beforehand so that the function can | ||
| 247 | ;;; use the sign flag instead of the zero flag for the exit test. | ||
| 248 | ;;; It is then shifted into the high word, to make room for the wmask | ||
| 249 | ;;; value, which it will always accompany. | ||
| 250 | |||
| 251 | dec ebx | ||
| 252 | shl ebx, 16 | ||
| 253 | or ebx, eax | ||
| 254 | |||
| 255 | ;;; on zlib only | ||
| 256 | ;;; if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; | ||
| 257 | |||
| 258 | IFDEF INFOZIP | ||
| 259 | mov [chainlenwmask], ebx | ||
| 260 | ; on infozip nice_match = [nice_match] | ||
| 261 | ELSE | ||
| 262 | mov eax, nice_match | ||
| 263 | mov [chainlenwmask], ebx | ||
| 264 | mov r10d, Lookahead | ||
| 265 | cmp r10d, eax | ||
| 266 | cmovnl r10d, eax | ||
| 267 | mov [nicematch],r10d | ||
| 268 | ENDIF | ||
| 269 | |||
| 270 | ;;; register Bytef *scan = s->window + s->strstart; | ||
| 271 | mov r10, window_ad | ||
| 272 | mov ebp, strstart | ||
| 273 | lea r13, [r10 + rbp] | ||
| 274 | |||
| 275 | ;;; Determine how many bytes the scan ptr is off from being | ||
| 276 | ;;; dword-aligned. | ||
| 277 | |||
| 278 | mov r9,r13 | ||
| 279 | neg r13 | ||
| 280 | and r13,3 | ||
| 281 | |||
| 282 | ;;; IPos limit = s->strstart > (IPos)MAX_DIST(s) ? | ||
| 283 | ;;; s->strstart - (IPos)MAX_DIST(s) : NIL; | ||
| 284 | IFDEF INFOZIP | ||
| 285 | mov eax,07efah ; MAX_DIST = (WSIZE-MIN_LOOKAHEAD) (0x8000-(3+8+1)) | ||
| 286 | ELSE | ||
| 287 | mov eax, window_size | ||
| 288 | sub eax, MIN_LOOKAHEAD | ||
| 289 | ENDIF | ||
| 290 | xor edi,edi | ||
| 291 | sub ebp, eax | ||
| 292 | |||
| 293 | mov r11d, prev_length | ||
| 294 | |||
| 295 | cmovng ebp,edi | ||
| 296 | |||
| 297 | ;;; int best_len = s->prev_length; | ||
| 298 | |||
| 299 | |||
| 300 | ;;; Store the sum of s->window + best_len in esi locally, and in esi. | ||
| 301 | |||
| 302 | lea rsi,[r10+r11] | ||
| 303 | |||
| 304 | ;;; register ush scan_start = *(ushf*)scan; | ||
| 305 | ;;; register ush scan_end = *(ushf*)(scan+best_len-1); | ||
| 306 | ;;; Posf *prev = s->prev; | ||
| 307 | |||
| 308 | movzx r12d,word ptr [r9] | ||
| 309 | movzx ebx, word ptr [r9 + r11 - 1] | ||
| 310 | |||
| 311 | mov rdi, prev_ad | ||
| 312 | |||
| 313 | ;;; Jump into the main loop. | ||
| 314 | |||
| 315 | mov edx, [chainlenwmask] | ||
| 316 | |||
| 317 | cmp bx,word ptr [rsi + r8 - 1] | ||
| 318 | jz LookupLoopIsZero | ||
| 319 | |||
| 320 | LookupLoop1: | ||
| 321 | and r8d, edx | ||
| 322 | |||
| 323 | movzx r8d, word ptr [rdi + r8*2] | ||
| 324 | cmp r8d, ebp | ||
| 325 | jbe LeaveNow | ||
| 326 | sub edx, 00010000h | ||
| 327 | js LeaveNow | ||
| 328 | |||
| 329 | LoopEntry1: | ||
| 330 | cmp bx,word ptr [rsi + r8 - 1] | ||
| 331 | jz LookupLoopIsZero | ||
| 332 | |||
| 333 | LookupLoop2: | ||
| 334 | and r8d, edx | ||
| 335 | |||
| 336 | movzx r8d, word ptr [rdi + r8*2] | ||
| 337 | cmp r8d, ebp | ||
| 338 | jbe LeaveNow | ||
| 339 | sub edx, 00010000h | ||
| 340 | js LeaveNow | ||
| 341 | |||
| 342 | LoopEntry2: | ||
| 343 | cmp bx,word ptr [rsi + r8 - 1] | ||
| 344 | jz LookupLoopIsZero | ||
| 345 | |||
| 346 | LookupLoop4: | ||
| 347 | and r8d, edx | ||
| 348 | |||
| 349 | movzx r8d, word ptr [rdi + r8*2] | ||
| 350 | cmp r8d, ebp | ||
| 351 | jbe LeaveNow | ||
| 352 | sub edx, 00010000h | ||
| 353 | js LeaveNow | ||
| 354 | |||
| 355 | LoopEntry4: | ||
| 356 | |||
| 357 | cmp bx,word ptr [rsi + r8 - 1] | ||
| 358 | jnz LookupLoop1 | ||
| 359 | jmp LookupLoopIsZero | ||
| 360 | |||
| 361 | |||
| 362 | ;;; do { | ||
| 363 | ;;; match = s->window + cur_match; | ||
| 364 | ;;; if (*(ushf*)(match+best_len-1) != scan_end || | ||
| 365 | ;;; *(ushf*)match != scan_start) continue; | ||
| 366 | ;;; [...] | ||
| 367 | ;;; } while ((cur_match = prev[cur_match & wmask]) > limit | ||
| 368 | ;;; && --chain_length != 0); | ||
| 369 | ;;; | ||
| 370 | ;;; Here is the inner loop of the function. The function will spend the | ||
| 371 | ;;; majority of its time in this loop, and majority of that time will | ||
| 372 | ;;; be spent in the first ten instructions. | ||
| 373 | ;;; | ||
| 374 | ;;; Within this loop: | ||
| 375 | ;;; ebx = scanend | ||
| 376 | ;;; r8d = curmatch | ||
| 377 | ;;; edx = chainlenwmask - i.e., ((chainlen << 16) | wmask) | ||
| 378 | ;;; esi = windowbestlen - i.e., (window + bestlen) | ||
| 379 | ;;; edi = prev | ||
| 380 | ;;; ebp = limit | ||
| 381 | |||
| 382 | LookupLoop: | ||
| 383 | and r8d, edx | ||
| 384 | |||
| 385 | movzx r8d, word ptr [rdi + r8*2] | ||
| 386 | cmp r8d, ebp | ||
| 387 | jbe LeaveNow | ||
| 388 | sub edx, 00010000h | ||
| 389 | js LeaveNow | ||
| 390 | |||
| 391 | LoopEntry: | ||
| 392 | |||
| 393 | cmp bx,word ptr [rsi + r8 - 1] | ||
| 394 | jnz LookupLoop1 | ||
| 395 | LookupLoopIsZero: | ||
| 396 | cmp r12w, word ptr [r10 + r8] | ||
| 397 | jnz LookupLoop1 | ||
| 398 | |||
| 399 | |||
| 400 | ;;; Store the current value of chainlen. | ||
| 401 | mov [chainlenwmask], edx | ||
| 402 | |||
| 403 | ;;; Point edi to the string under scrutiny, and esi to the string we | ||
| 404 | ;;; are hoping to match it up with. In actuality, esi and edi are | ||
| 405 | ;;; both pointed (MAX_MATCH_8 - scanalign) bytes ahead, and edx is | ||
| 406 | ;;; initialized to -(MAX_MATCH_8 - scanalign). | ||
| 407 | |||
| 408 | lea rsi,[r8+r10] | ||
| 409 | mov rdx, 0fffffffffffffef8h; -(MAX_MATCH_8) | ||
| 410 | lea rsi, [rsi + r13 + 0108h] ;MAX_MATCH_8] | ||
| 411 | lea rdi, [r9 + r13 + 0108h] ;MAX_MATCH_8] | ||
| 412 | |||
| 413 | prefetcht1 [rsi+rdx] | ||
| 414 | prefetcht1 [rdi+rdx] | ||
| 415 | |||
| 416 | |||
| 417 | ;;; Test the strings for equality, 8 bytes at a time. At the end, | ||
| 418 | ;;; adjust rdx so that it is offset to the exact byte that mismatched. | ||
| 419 | ;;; | ||
| 420 | ;;; We already know at this point that the first three bytes of the | ||
| 421 | ;;; strings match each other, and they can be safely passed over before | ||
| 422 | ;;; starting the compare loop. So what this code does is skip over 0-3 | ||
| 423 | ;;; bytes, as much as necessary in order to dword-align the edi | ||
| 424 | ;;; pointer. (rsi will still be misaligned three times out of four.) | ||
| 425 | ;;; | ||
| 426 | ;;; It should be confessed that this loop usually does not represent | ||
| 427 | ;;; much of the total running time. Replacing it with a more | ||
| 428 | ;;; straightforward "rep cmpsb" would not drastically degrade | ||
| 429 | ;;; performance. | ||
| 430 | |||
| 431 | |||
| 432 | LoopCmps: | ||
| 433 | mov rax, [rsi + rdx] | ||
| 434 | xor rax, [rdi + rdx] | ||
| 435 | jnz LeaveLoopCmps | ||
| 436 | |||
| 437 | mov rax, [rsi + rdx + 8] | ||
| 438 | xor rax, [rdi + rdx + 8] | ||
| 439 | jnz LeaveLoopCmps8 | ||
| 440 | |||
| 441 | |||
| 442 | mov rax, [rsi + rdx + 8+8] | ||
| 443 | xor rax, [rdi + rdx + 8+8] | ||
| 444 | jnz LeaveLoopCmps16 | ||
| 445 | |||
| 446 | add rdx,8+8+8 | ||
| 447 | |||
| 448 | jnz short LoopCmps | ||
| 449 | jmp short LenMaximum | ||
| 450 | LeaveLoopCmps16: add rdx,8 | ||
| 451 | LeaveLoopCmps8: add rdx,8 | ||
| 452 | LeaveLoopCmps: | ||
| 453 | |||
| 454 | test eax, 0000FFFFh | ||
| 455 | jnz LenLower | ||
| 456 | |||
| 457 | test eax,0ffffffffh | ||
| 458 | |||
| 459 | jnz LenLower32 | ||
| 460 | |||
| 461 | add rdx,4 | ||
| 462 | shr rax,32 | ||
| 463 | or ax,ax | ||
| 464 | jnz LenLower | ||
| 465 | |||
| 466 | LenLower32: | ||
| 467 | shr eax,16 | ||
| 468 | add rdx,2 | ||
| 469 | LenLower: sub al, 1 | ||
| 470 | adc rdx, 0 | ||
| 471 | ;;; Calculate the length of the match. If it is longer than MAX_MATCH, | ||
| 472 | ;;; then automatically accept it as the best possible match and leave. | ||
| 473 | |||
| 474 | lea rax, [rdi + rdx] | ||
| 475 | sub rax, r9 | ||
| 476 | cmp eax, MAX_MATCH | ||
| 477 | jge LenMaximum | ||
| 478 | |||
| 479 | ;;; If the length of the match is not longer than the best match we | ||
| 480 | ;;; have so far, then forget it and return to the lookup loop. | ||
| 481 | ;/////////////////////////////////// | ||
| 482 | |||
| 483 | cmp eax, r11d | ||
| 484 | jg LongerMatch | ||
| 485 | |||
| 486 | lea rsi,[r10+r11] | ||
| 487 | |||
| 488 | mov rdi, prev_ad | ||
| 489 | mov edx, [chainlenwmask] | ||
| 490 | jmp LookupLoop | ||
| 491 | |||
| 492 | ;;; s->match_start = cur_match; | ||
| 493 | ;;; best_len = len; | ||
| 494 | ;;; if (len >= nice_match) break; | ||
| 495 | ;;; scan_end = *(ushf*)(scan+best_len-1); | ||
| 496 | |||
| 497 | LongerMatch: | ||
| 498 | mov r11d, eax | ||
| 499 | mov match_start, r8d | ||
| 500 | cmp eax, [nicematch] | ||
| 501 | jge LeaveNow | ||
| 502 | |||
| 503 | lea rsi,[r10+rax] | ||
| 504 | |||
| 505 | movzx ebx, word ptr [r9 + rax - 1] | ||
| 506 | mov rdi, prev_ad | ||
| 507 | mov edx, [chainlenwmask] | ||
| 508 | jmp LookupLoop | ||
| 509 | |||
| 510 | ;;; Accept the current string, with the maximum possible length. | ||
| 511 | |||
| 512 | LenMaximum: | ||
| 513 | mov r11d,MAX_MATCH | ||
| 514 | mov match_start, r8d | ||
| 515 | |||
| 516 | ;;; if ((uInt)best_len <= s->lookahead) return (uInt)best_len; | ||
| 517 | ;;; return s->lookahead; | ||
| 518 | |||
| 519 | LeaveNow: | ||
| 520 | IFDEF INFOZIP | ||
| 521 | mov eax,r11d | ||
| 522 | ELSE | ||
| 523 | mov eax, Lookahead | ||
| 524 | cmp r11d, eax | ||
| 525 | cmovng eax, r11d | ||
| 526 | ENDIF | ||
| 527 | |||
| 528 | ;;; Restore the stack and return from whence we came. | ||
| 529 | |||
| 530 | |||
| 531 | mov rsi,[save_rsi] | ||
| 532 | mov rdi,[save_rdi] | ||
| 533 | mov rbx,[save_rbx] | ||
| 534 | mov rbp,[save_rbp] | ||
| 535 | mov r12,[save_r12] | ||
| 536 | mov r13,[save_r13] | ||
| 537 | ; mov r14,[save_r14] | ||
| 538 | ; mov r15,[save_r15] | ||
| 539 | |||
| 540 | |||
| 541 | ret 0 | ||
| 542 | ; please don't remove this string ! | ||
| 543 | ; Your can freely use gvmat64 in any free or commercial app | ||
| 544 | ; but it is far better don't remove the string in the binary! | ||
| 545 | 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 | ||
| 546 | longest_match ENDP | ||
| 547 | |||
| 548 | match_init PROC | ||
| 549 | ret 0 | ||
| 550 | match_init ENDP | ||
| 551 | |||
| 552 | |||
| 553 | END | ||
diff --git a/contrib/masmx64/inffas8664.c b/contrib/masmx64/inffas8664.c deleted file mode 100644 index e8af06f..0000000 --- a/contrib/masmx64/inffas8664.c +++ /dev/null | |||
| @@ -1,186 +0,0 @@ | |||
| 1 | /* inffas8664.c is a hand tuned assembler version of inffast.c - fast decoding | ||
| 2 | * version for AMD64 on Windows using Microsoft C compiler | ||
| 3 | * | ||
| 4 | * Copyright (C) 1995-2003 Mark Adler | ||
| 5 | * For conditions of distribution and use, see copyright notice in zlib.h | ||
| 6 | * | ||
| 7 | * Copyright (C) 2003 Chris Anderson <christop@charm.net> | ||
| 8 | * Please use the copyright conditions above. | ||
| 9 | * | ||
| 10 | * 2005 - Adaptation to Microsoft C Compiler for AMD64 by Gilles Vollant | ||
| 11 | * | ||
| 12 | * inffas8664.c call function inffas8664fnc in inffasx64.asm | ||
| 13 | * inffasx64.asm is automatically convert from AMD64 portion of inffas86.c | ||
| 14 | * | ||
| 15 | * Dec-29-2003 -- I added AMD64 inflate asm support. This version is also | ||
| 16 | * slightly quicker on x86 systems because, instead of using rep movsb to copy | ||
| 17 | * data, it uses rep movsw, which moves data in 2-byte chunks instead of single | ||
| 18 | * bytes. I've tested the AMD64 code on a Fedora Core 1 + the x86_64 updates | ||
| 19 | * from http://fedora.linux.duke.edu/fc1_x86_64 | ||
| 20 | * which is running on an Athlon 64 3000+ / Gigabyte GA-K8VT800M system with | ||
| 21 | * 1GB ram. The 64-bit version is about 4% faster than the 32-bit version, | ||
| 22 | * when decompressing mozilla-source-1.3.tar.gz. | ||
| 23 | * | ||
| 24 | * Mar-13-2003 -- Most of this is derived from inffast.S which is derived from | ||
| 25 | * the gcc -S output of zlib-1.2.0/inffast.c. Zlib-1.2.0 is in beta release at | ||
| 26 | * the moment. I have successfully compiled and tested this code with gcc2.96, | ||
| 27 | * gcc3.2, icc5.0, msvc6.0. It is very close to the speed of inffast.S | ||
| 28 | * compiled with gcc -DNO_MMX, but inffast.S is still faster on the P3 with MMX | ||
| 29 | * enabled. I will attempt to merge the MMX code into this version. Newer | ||
| 30 | * versions of this and inffast.S can be found at | ||
| 31 | * http://www.eetbeetee.com/zlib/ and http://www.charm.net/~christop/zlib/ | ||
| 32 | * | ||
| 33 | */ | ||
| 34 | |||
| 35 | #include <stdio.h> | ||
| 36 | #include "zutil.h" | ||
| 37 | #include "inftrees.h" | ||
| 38 | #include "inflate.h" | ||
| 39 | #include "inffast.h" | ||
| 40 | |||
| 41 | /* Mark Adler's comments from inffast.c: */ | ||
| 42 | |||
| 43 | /* | ||
| 44 | Decode literal, length, and distance codes and write out the resulting | ||
| 45 | literal and match bytes until either not enough input or output is | ||
| 46 | available, an end-of-block is encountered, or a data error is encountered. | ||
| 47 | When large enough input and output buffers are supplied to inflate(), for | ||
| 48 | example, a 16K input buffer and a 64K output buffer, more than 95% of the | ||
| 49 | inflate execution time is spent in this routine. | ||
| 50 | |||
| 51 | Entry assumptions: | ||
| 52 | |||
| 53 | state->mode == LEN | ||
| 54 | strm->avail_in >= 6 | ||
| 55 | strm->avail_out >= 258 | ||
| 56 | start >= strm->avail_out | ||
| 57 | state->bits < 8 | ||
| 58 | |||
| 59 | On return, state->mode is one of: | ||
| 60 | |||
| 61 | LEN -- ran out of enough output space or enough available input | ||
| 62 | TYPE -- reached end of block code, inflate() to interpret next block | ||
| 63 | BAD -- error in block data | ||
| 64 | |||
| 65 | Notes: | ||
| 66 | |||
| 67 | - The maximum input bits used by a length/distance pair is 15 bits for the | ||
| 68 | length code, 5 bits for the length extra, 15 bits for the distance code, | ||
| 69 | and 13 bits for the distance extra. This totals 48 bits, or six bytes. | ||
| 70 | Therefore if strm->avail_in >= 6, then there is enough input to avoid | ||
| 71 | checking for available input while decoding. | ||
| 72 | |||
| 73 | - The maximum bytes that a single length/distance pair can output is 258 | ||
| 74 | bytes, which is the maximum length that can be coded. inflate_fast() | ||
| 75 | requires strm->avail_out >= 258 for each loop to avoid checking for | ||
| 76 | output space. | ||
| 77 | */ | ||
| 78 | |||
| 79 | |||
| 80 | |||
| 81 | typedef struct inffast_ar { | ||
| 82 | /* 64 32 x86 x86_64 */ | ||
| 83 | /* ar offset register */ | ||
| 84 | /* 0 0 */ void *esp; /* esp save */ | ||
| 85 | /* 8 4 */ void *ebp; /* ebp save */ | ||
| 86 | /* 16 8 */ unsigned char FAR *in; /* esi rsi local strm->next_in */ | ||
| 87 | /* 24 12 */ unsigned char FAR *last; /* r9 while in < last */ | ||
| 88 | /* 32 16 */ unsigned char FAR *out; /* edi rdi local strm->next_out */ | ||
| 89 | /* 40 20 */ unsigned char FAR *beg; /* inflate()'s init next_out */ | ||
| 90 | /* 48 24 */ unsigned char FAR *end; /* r10 while out < end */ | ||
| 91 | /* 56 28 */ unsigned char FAR *window;/* size of window, wsize!=0 */ | ||
| 92 | /* 64 32 */ code const FAR *lcode; /* ebp rbp local strm->lencode */ | ||
| 93 | /* 72 36 */ code const FAR *dcode; /* r11 local strm->distcode */ | ||
| 94 | /* 80 40 */ size_t /*unsigned long */hold; /* edx rdx local strm->hold */ | ||
| 95 | /* 88 44 */ unsigned bits; /* ebx rbx local strm->bits */ | ||
| 96 | /* 92 48 */ unsigned wsize; /* window size */ | ||
| 97 | /* 96 52 */ unsigned write; /* window write index */ | ||
| 98 | /*100 56 */ unsigned lmask; /* r12 mask for lcode */ | ||
| 99 | /*104 60 */ unsigned dmask; /* r13 mask for dcode */ | ||
| 100 | /*108 64 */ unsigned len; /* r14 match length */ | ||
| 101 | /*112 68 */ unsigned dist; /* r15 match distance */ | ||
| 102 | /*116 72 */ unsigned status; /* set when state chng*/ | ||
| 103 | } type_ar; | ||
| 104 | #ifdef ASMINF | ||
| 105 | |||
| 106 | void inflate_fast(strm, start) | ||
| 107 | z_streamp strm; | ||
| 108 | unsigned start; /* inflate()'s starting value for strm->avail_out */ | ||
| 109 | { | ||
| 110 | struct inflate_state FAR *state; | ||
| 111 | type_ar ar; | ||
| 112 | void inffas8664fnc(struct inffast_ar * par); | ||
| 113 | |||
| 114 | |||
| 115 | |||
| 116 | #if (defined( __GNUC__ ) && defined( __amd64__ ) && ! defined( __i386 )) || (defined(_MSC_VER) && defined(_M_AMD64)) | ||
| 117 | #define PAD_AVAIL_IN 6 | ||
| 118 | #define PAD_AVAIL_OUT 258 | ||
| 119 | #else | ||
| 120 | #define PAD_AVAIL_IN 5 | ||
| 121 | #define PAD_AVAIL_OUT 257 | ||
| 122 | #endif | ||
| 123 | |||
| 124 | /* copy state to local variables */ | ||
| 125 | state = (struct inflate_state FAR *)strm->state; | ||
| 126 | |||
| 127 | ar.in = strm->next_in; | ||
| 128 | ar.last = ar.in + (strm->avail_in - PAD_AVAIL_IN); | ||
| 129 | ar.out = strm->next_out; | ||
| 130 | ar.beg = ar.out - (start - strm->avail_out); | ||
| 131 | ar.end = ar.out + (strm->avail_out - PAD_AVAIL_OUT); | ||
| 132 | ar.wsize = state->wsize; | ||
| 133 | ar.write = state->wnext; | ||
| 134 | ar.window = state->window; | ||
| 135 | ar.hold = state->hold; | ||
| 136 | ar.bits = state->bits; | ||
| 137 | ar.lcode = state->lencode; | ||
| 138 | ar.dcode = state->distcode; | ||
| 139 | ar.lmask = (1U << state->lenbits) - 1; | ||
| 140 | ar.dmask = (1U << state->distbits) - 1; | ||
| 141 | |||
| 142 | /* decode literals and length/distances until end-of-block or not enough | ||
| 143 | input data or output space */ | ||
| 144 | |||
| 145 | /* align in on 1/2 hold size boundary */ | ||
| 146 | while (((size_t)(void *)ar.in & (sizeof(ar.hold) / 2 - 1)) != 0) { | ||
| 147 | ar.hold += (unsigned long)*ar.in++ << ar.bits; | ||
| 148 | ar.bits += 8; | ||
| 149 | } | ||
| 150 | |||
| 151 | inffas8664fnc(&ar); | ||
| 152 | |||
| 153 | if (ar.status > 1) { | ||
| 154 | if (ar.status == 2) | ||
| 155 | strm->msg = "invalid literal/length code"; | ||
| 156 | else if (ar.status == 3) | ||
| 157 | strm->msg = "invalid distance code"; | ||
| 158 | else | ||
| 159 | strm->msg = "invalid distance too far back"; | ||
| 160 | state->mode = BAD; | ||
| 161 | } | ||
| 162 | else if ( ar.status == 1 ) { | ||
| 163 | state->mode = TYPE; | ||
| 164 | } | ||
| 165 | |||
| 166 | /* return unused bytes (on entry, bits < 8, so in won't go too far back) */ | ||
| 167 | ar.len = ar.bits >> 3; | ||
| 168 | ar.in -= ar.len; | ||
| 169 | ar.bits -= ar.len << 3; | ||
| 170 | ar.hold &= (1U << ar.bits) - 1; | ||
| 171 | |||
| 172 | /* update state and return */ | ||
| 173 | strm->next_in = ar.in; | ||
| 174 | strm->next_out = ar.out; | ||
| 175 | strm->avail_in = (unsigned)(ar.in < ar.last ? | ||
| 176 | PAD_AVAIL_IN + (ar.last - ar.in) : | ||
| 177 | PAD_AVAIL_IN - (ar.in - ar.last)); | ||
| 178 | strm->avail_out = (unsigned)(ar.out < ar.end ? | ||
| 179 | PAD_AVAIL_OUT + (ar.end - ar.out) : | ||
| 180 | PAD_AVAIL_OUT - (ar.out - ar.end)); | ||
| 181 | state->hold = (unsigned long)ar.hold; | ||
| 182 | state->bits = ar.bits; | ||
| 183 | return; | ||
| 184 | } | ||
| 185 | |||
| 186 | #endif | ||
diff --git a/contrib/masmx64/inffasx64.asm b/contrib/masmx64/inffasx64.asm deleted file mode 100644 index 60a8d89..0000000 --- a/contrib/masmx64/inffasx64.asm +++ /dev/null | |||
| @@ -1,396 +0,0 @@ | |||
| 1 | ; inffasx64.asm is a hand tuned assembler version of inffast.c - fast decoding | ||
| 2 | ; version for AMD64 on Windows using Microsoft C compiler | ||
| 3 | ; | ||
| 4 | ; inffasx64.asm is automatically convert from AMD64 portion of inffas86.c | ||
| 5 | ; inffasx64.asm is called by inffas8664.c, which contain more info. | ||
| 6 | |||
| 7 | |||
| 8 | ; to compile this file, I use option | ||
| 9 | ; ml64.exe /Flinffasx64 /c /Zi inffasx64.asm | ||
| 10 | ; with Microsoft Macro Assembler (x64) for AMD64 | ||
| 11 | ; | ||
| 12 | |||
| 13 | ; This file compile with Microsoft Macro Assembler (x64) for AMD64 | ||
| 14 | ; | ||
| 15 | ; ml64.exe is given with Visual Studio 2005/2008/2010 and Windows WDK | ||
| 16 | ; | ||
| 17 | ; (you can get Windows WDK with ml64 for AMD64 from | ||
| 18 | ; http://www.microsoft.com/whdc/Devtools/wdk/default.mspx for low price) | ||
| 19 | ; | ||
| 20 | |||
| 21 | |||
| 22 | .code | ||
| 23 | inffas8664fnc PROC | ||
| 24 | |||
| 25 | ; see http://weblogs.asp.net/oldnewthing/archive/2004/01/14/58579.aspx and | ||
| 26 | ; http://msdn.microsoft.com/library/en-us/kmarch/hh/kmarch/64bitAMD_8e951dd2-ee77-4728-8702-55ce4b5dd24a.xml.asp | ||
| 27 | ; | ||
| 28 | ; All registers must be preserved across the call, except for | ||
| 29 | ; rax, rcx, rdx, r8, r-9, r10, and r11, which are scratch. | ||
| 30 | |||
| 31 | |||
| 32 | mov [rsp-8],rsi | ||
| 33 | mov [rsp-16],rdi | ||
| 34 | mov [rsp-24],r12 | ||
| 35 | mov [rsp-32],r13 | ||
| 36 | mov [rsp-40],r14 | ||
| 37 | mov [rsp-48],r15 | ||
| 38 | mov [rsp-56],rbx | ||
| 39 | |||
| 40 | mov rax,rcx | ||
| 41 | |||
| 42 | mov [rax+8], rbp ; /* save regs rbp and rsp */ | ||
| 43 | mov [rax], rsp | ||
| 44 | |||
| 45 | mov rsp, rax ; /* make rsp point to &ar */ | ||
| 46 | |||
| 47 | mov rsi, [rsp+16] ; /* rsi = in */ | ||
| 48 | mov rdi, [rsp+32] ; /* rdi = out */ | ||
| 49 | mov r9, [rsp+24] ; /* r9 = last */ | ||
| 50 | mov r10, [rsp+48] ; /* r10 = end */ | ||
| 51 | mov rbp, [rsp+64] ; /* rbp = lcode */ | ||
| 52 | mov r11, [rsp+72] ; /* r11 = dcode */ | ||
| 53 | mov rdx, [rsp+80] ; /* rdx = hold */ | ||
| 54 | mov ebx, [rsp+88] ; /* ebx = bits */ | ||
| 55 | mov r12d, [rsp+100] ; /* r12d = lmask */ | ||
| 56 | mov r13d, [rsp+104] ; /* r13d = dmask */ | ||
| 57 | ; /* r14d = len */ | ||
| 58 | ; /* r15d = dist */ | ||
| 59 | |||
| 60 | |||
| 61 | cld | ||
| 62 | cmp r10, rdi | ||
| 63 | je L_one_time ; /* if only one decode left */ | ||
| 64 | cmp r9, rsi | ||
| 65 | |||
| 66 | jne L_do_loop | ||
| 67 | |||
| 68 | |||
| 69 | L_one_time: | ||
| 70 | mov r8, r12 ; /* r8 = lmask */ | ||
| 71 | cmp bl, 32 | ||
| 72 | ja L_get_length_code_one_time | ||
| 73 | |||
| 74 | lodsd ; /* eax = *(uint *)in++ */ | ||
| 75 | mov cl, bl ; /* cl = bits, needs it for shifting */ | ||
| 76 | add bl, 32 ; /* bits += 32 */ | ||
| 77 | shl rax, cl | ||
| 78 | or rdx, rax ; /* hold |= *((uint *)in)++ << bits */ | ||
| 79 | jmp L_get_length_code_one_time | ||
| 80 | |||
| 81 | ALIGN 4 | ||
| 82 | L_while_test: | ||
| 83 | cmp r10, rdi | ||
| 84 | jbe L_break_loop | ||
| 85 | cmp r9, rsi | ||
| 86 | jbe L_break_loop | ||
| 87 | |||
| 88 | L_do_loop: | ||
| 89 | mov r8, r12 ; /* r8 = lmask */ | ||
| 90 | cmp bl, 32 | ||
| 91 | ja L_get_length_code ; /* if (32 < bits) */ | ||
| 92 | |||
| 93 | lodsd ; /* eax = *(uint *)in++ */ | ||
| 94 | mov cl, bl ; /* cl = bits, needs it for shifting */ | ||
| 95 | add bl, 32 ; /* bits += 32 */ | ||
| 96 | shl rax, cl | ||
| 97 | or rdx, rax ; /* hold |= *((uint *)in)++ << bits */ | ||
| 98 | |||
| 99 | L_get_length_code: | ||
| 100 | and r8, rdx ; /* r8 &= hold */ | ||
| 101 | mov eax, [rbp+r8*4] ; /* eax = lcode[hold & lmask] */ | ||
| 102 | |||
| 103 | mov cl, ah ; /* cl = this.bits */ | ||
| 104 | sub bl, ah ; /* bits -= this.bits */ | ||
| 105 | shr rdx, cl ; /* hold >>= this.bits */ | ||
| 106 | |||
| 107 | test al, al | ||
| 108 | jnz L_test_for_length_base ; /* if (op != 0) 45.7% */ | ||
| 109 | |||
| 110 | mov r8, r12 ; /* r8 = lmask */ | ||
| 111 | shr eax, 16 ; /* output this.val char */ | ||
| 112 | stosb | ||
| 113 | |||
| 114 | L_get_length_code_one_time: | ||
| 115 | and r8, rdx ; /* r8 &= hold */ | ||
| 116 | mov eax, [rbp+r8*4] ; /* eax = lcode[hold & lmask] */ | ||
| 117 | |||
| 118 | L_dolen: | ||
| 119 | mov cl, ah ; /* cl = this.bits */ | ||
| 120 | sub bl, ah ; /* bits -= this.bits */ | ||
| 121 | shr rdx, cl ; /* hold >>= this.bits */ | ||
| 122 | |||
| 123 | test al, al | ||
| 124 | jnz L_test_for_length_base ; /* if (op != 0) 45.7% */ | ||
| 125 | |||
| 126 | shr eax, 16 ; /* output this.val char */ | ||
| 127 | stosb | ||
| 128 | jmp L_while_test | ||
| 129 | |||
| 130 | ALIGN 4 | ||
| 131 | L_test_for_length_base: | ||
| 132 | mov r14d, eax ; /* len = this */ | ||
| 133 | shr r14d, 16 ; /* len = this.val */ | ||
| 134 | mov cl, al | ||
| 135 | |||
| 136 | test al, 16 | ||
| 137 | jz L_test_for_second_level_length ; /* if ((op & 16) == 0) 8% */ | ||
| 138 | and cl, 15 ; /* op &= 15 */ | ||
| 139 | jz L_decode_distance ; /* if (!op) */ | ||
| 140 | |||
| 141 | L_add_bits_to_len: | ||
| 142 | sub bl, cl | ||
| 143 | xor eax, eax | ||
| 144 | inc eax | ||
| 145 | shl eax, cl | ||
| 146 | dec eax | ||
| 147 | and eax, edx ; /* eax &= hold */ | ||
| 148 | shr rdx, cl | ||
| 149 | add r14d, eax ; /* len += hold & mask[op] */ | ||
| 150 | |||
| 151 | L_decode_distance: | ||
| 152 | mov r8, r13 ; /* r8 = dmask */ | ||
| 153 | cmp bl, 32 | ||
| 154 | ja L_get_distance_code ; /* if (32 < bits) */ | ||
| 155 | |||
| 156 | lodsd ; /* eax = *(uint *)in++ */ | ||
| 157 | mov cl, bl ; /* cl = bits, needs it for shifting */ | ||
| 158 | add bl, 32 ; /* bits += 32 */ | ||
| 159 | shl rax, cl | ||
| 160 | or rdx, rax ; /* hold |= *((uint *)in)++ << bits */ | ||
| 161 | |||
| 162 | L_get_distance_code: | ||
| 163 | and r8, rdx ; /* r8 &= hold */ | ||
| 164 | mov eax, [r11+r8*4] ; /* eax = dcode[hold & dmask] */ | ||
| 165 | |||
| 166 | L_dodist: | ||
| 167 | mov r15d, eax ; /* dist = this */ | ||
| 168 | shr r15d, 16 ; /* dist = this.val */ | ||
| 169 | mov cl, ah | ||
| 170 | sub bl, ah ; /* bits -= this.bits */ | ||
| 171 | shr rdx, cl ; /* hold >>= this.bits */ | ||
| 172 | mov cl, al ; /* cl = this.op */ | ||
| 173 | |||
| 174 | test al, 16 ; /* if ((op & 16) == 0) */ | ||
| 175 | jz L_test_for_second_level_dist | ||
| 176 | and cl, 15 ; /* op &= 15 */ | ||
| 177 | jz L_check_dist_one | ||
| 178 | |||
| 179 | L_add_bits_to_dist: | ||
| 180 | sub bl, cl | ||
| 181 | xor eax, eax | ||
| 182 | inc eax | ||
| 183 | shl eax, cl | ||
| 184 | dec eax ; /* (1 << op) - 1 */ | ||
| 185 | and eax, edx ; /* eax &= hold */ | ||
| 186 | shr rdx, cl | ||
| 187 | add r15d, eax ; /* dist += hold & ((1 << op) - 1) */ | ||
| 188 | |||
| 189 | L_check_window: | ||
| 190 | mov r8, rsi ; /* save in so from can use it's reg */ | ||
| 191 | mov rax, rdi | ||
| 192 | sub rax, [rsp+40] ; /* nbytes = out - beg */ | ||
| 193 | |||
| 194 | cmp eax, r15d | ||
| 195 | jb L_clip_window ; /* if (dist > nbytes) 4.2% */ | ||
| 196 | |||
| 197 | mov ecx, r14d ; /* ecx = len */ | ||
| 198 | mov rsi, rdi | ||
| 199 | sub rsi, r15 ; /* from = out - dist */ | ||
| 200 | |||
| 201 | sar ecx, 1 | ||
| 202 | jnc L_copy_two ; /* if len % 2 == 0 */ | ||
| 203 | |||
| 204 | rep movsw | ||
| 205 | mov al, [rsi] | ||
| 206 | mov [rdi], al | ||
| 207 | inc rdi | ||
| 208 | |||
| 209 | mov rsi, r8 ; /* move in back to %rsi, toss from */ | ||
| 210 | jmp L_while_test | ||
| 211 | |||
| 212 | L_copy_two: | ||
| 213 | rep movsw | ||
| 214 | mov rsi, r8 ; /* move in back to %rsi, toss from */ | ||
| 215 | jmp L_while_test | ||
| 216 | |||
| 217 | ALIGN 4 | ||
| 218 | L_check_dist_one: | ||
| 219 | cmp r15d, 1 ; /* if dist 1, is a memset */ | ||
| 220 | jne L_check_window | ||
| 221 | cmp [rsp+40], rdi ; /* if out == beg, outside window */ | ||
| 222 | je L_check_window | ||
| 223 | |||
| 224 | mov ecx, r14d ; /* ecx = len */ | ||
| 225 | mov al, [rdi-1] | ||
| 226 | mov ah, al | ||
| 227 | |||
| 228 | sar ecx, 1 | ||
| 229 | jnc L_set_two | ||
| 230 | mov [rdi], al | ||
| 231 | inc rdi | ||
| 232 | |||
| 233 | L_set_two: | ||
| 234 | rep stosw | ||
| 235 | jmp L_while_test | ||
| 236 | |||
| 237 | ALIGN 4 | ||
| 238 | L_test_for_second_level_length: | ||
| 239 | test al, 64 | ||
| 240 | jnz L_test_for_end_of_block ; /* if ((op & 64) != 0) */ | ||
| 241 | |||
| 242 | xor eax, eax | ||
| 243 | inc eax | ||
| 244 | shl eax, cl | ||
| 245 | dec eax | ||
| 246 | and eax, edx ; /* eax &= hold */ | ||
| 247 | add eax, r14d ; /* eax += len */ | ||
| 248 | mov eax, [rbp+rax*4] ; /* eax = lcode[val+(hold&mask[op])]*/ | ||
| 249 | jmp L_dolen | ||
| 250 | |||
| 251 | ALIGN 4 | ||
| 252 | L_test_for_second_level_dist: | ||
| 253 | test al, 64 | ||
| 254 | jnz L_invalid_distance_code ; /* if ((op & 64) != 0) */ | ||
| 255 | |||
| 256 | xor eax, eax | ||
| 257 | inc eax | ||
| 258 | shl eax, cl | ||
| 259 | dec eax | ||
| 260 | and eax, edx ; /* eax &= hold */ | ||
| 261 | add eax, r15d ; /* eax += dist */ | ||
| 262 | mov eax, [r11+rax*4] ; /* eax = dcode[val+(hold&mask[op])]*/ | ||
| 263 | jmp L_dodist | ||
| 264 | |||
| 265 | ALIGN 4 | ||
| 266 | L_clip_window: | ||
| 267 | mov ecx, eax ; /* ecx = nbytes */ | ||
| 268 | mov eax, [rsp+92] ; /* eax = wsize, prepare for dist cmp */ | ||
| 269 | neg ecx ; /* nbytes = -nbytes */ | ||
| 270 | |||
| 271 | cmp eax, r15d | ||
| 272 | jb L_invalid_distance_too_far ; /* if (dist > wsize) */ | ||
| 273 | |||
| 274 | add ecx, r15d ; /* nbytes = dist - nbytes */ | ||
| 275 | cmp dword ptr [rsp+96], 0 | ||
| 276 | jne L_wrap_around_window ; /* if (write != 0) */ | ||
| 277 | |||
| 278 | mov rsi, [rsp+56] ; /* from = window */ | ||
| 279 | sub eax, ecx ; /* eax -= nbytes */ | ||
| 280 | add rsi, rax ; /* from += wsize - nbytes */ | ||
| 281 | |||
| 282 | mov eax, r14d ; /* eax = len */ | ||
| 283 | cmp r14d, ecx | ||
| 284 | jbe L_do_copy ; /* if (nbytes >= len) */ | ||
| 285 | |||
| 286 | sub eax, ecx ; /* eax -= nbytes */ | ||
| 287 | rep movsb | ||
| 288 | mov rsi, rdi | ||
| 289 | sub rsi, r15 ; /* from = &out[ -dist ] */ | ||
| 290 | jmp L_do_copy | ||
| 291 | |||
| 292 | ALIGN 4 | ||
| 293 | L_wrap_around_window: | ||
| 294 | mov eax, [rsp+96] ; /* eax = write */ | ||
| 295 | cmp ecx, eax | ||
| 296 | jbe L_contiguous_in_window ; /* if (write >= nbytes) */ | ||
| 297 | |||
| 298 | mov esi, [rsp+92] ; /* from = wsize */ | ||
| 299 | add rsi, [rsp+56] ; /* from += window */ | ||
| 300 | add rsi, rax ; /* from += write */ | ||
| 301 | sub rsi, rcx ; /* from -= nbytes */ | ||
| 302 | sub ecx, eax ; /* nbytes -= write */ | ||
| 303 | |||
| 304 | mov eax, r14d ; /* eax = len */ | ||
| 305 | cmp eax, ecx | ||
| 306 | jbe L_do_copy ; /* if (nbytes >= len) */ | ||
| 307 | |||
| 308 | sub eax, ecx ; /* len -= nbytes */ | ||
| 309 | rep movsb | ||
| 310 | mov rsi, [rsp+56] ; /* from = window */ | ||
| 311 | mov ecx, [rsp+96] ; /* nbytes = write */ | ||
| 312 | cmp eax, ecx | ||
| 313 | jbe L_do_copy ; /* if (nbytes >= len) */ | ||
| 314 | |||
| 315 | sub eax, ecx ; /* len -= nbytes */ | ||
| 316 | rep movsb | ||
| 317 | mov rsi, rdi | ||
| 318 | sub rsi, r15 ; /* from = out - dist */ | ||
| 319 | jmp L_do_copy | ||
| 320 | |||
| 321 | ALIGN 4 | ||
| 322 | L_contiguous_in_window: | ||
| 323 | mov rsi, [rsp+56] ; /* rsi = window */ | ||
| 324 | add rsi, rax | ||
| 325 | sub rsi, rcx ; /* from += write - nbytes */ | ||
| 326 | |||
| 327 | mov eax, r14d ; /* eax = len */ | ||
| 328 | cmp eax, ecx | ||
| 329 | jbe L_do_copy ; /* if (nbytes >= len) */ | ||
| 330 | |||
| 331 | sub eax, ecx ; /* len -= nbytes */ | ||
| 332 | rep movsb | ||
| 333 | mov rsi, rdi | ||
| 334 | sub rsi, r15 ; /* from = out - dist */ | ||
| 335 | jmp L_do_copy ; /* if (nbytes >= len) */ | ||
| 336 | |||
| 337 | ALIGN 4 | ||
| 338 | L_do_copy: | ||
| 339 | mov ecx, eax ; /* ecx = len */ | ||
| 340 | rep movsb | ||
| 341 | |||
| 342 | mov rsi, r8 ; /* move in back to %esi, toss from */ | ||
| 343 | jmp L_while_test | ||
| 344 | |||
| 345 | L_test_for_end_of_block: | ||
| 346 | test al, 32 | ||
| 347 | jz L_invalid_literal_length_code | ||
| 348 | mov dword ptr [rsp+116], 1 | ||
| 349 | jmp L_break_loop_with_status | ||
| 350 | |||
| 351 | L_invalid_literal_length_code: | ||
| 352 | mov dword ptr [rsp+116], 2 | ||
| 353 | jmp L_break_loop_with_status | ||
| 354 | |||
| 355 | L_invalid_distance_code: | ||
| 356 | mov dword ptr [rsp+116], 3 | ||
| 357 | jmp L_break_loop_with_status | ||
| 358 | |||
| 359 | L_invalid_distance_too_far: | ||
| 360 | mov dword ptr [rsp+116], 4 | ||
| 361 | jmp L_break_loop_with_status | ||
| 362 | |||
| 363 | L_break_loop: | ||
| 364 | mov dword ptr [rsp+116], 0 | ||
| 365 | |||
| 366 | L_break_loop_with_status: | ||
| 367 | ; /* put in, out, bits, and hold back into ar and pop esp */ | ||
| 368 | mov [rsp+16], rsi ; /* in */ | ||
| 369 | mov [rsp+32], rdi ; /* out */ | ||
| 370 | mov [rsp+88], ebx ; /* bits */ | ||
| 371 | mov [rsp+80], rdx ; /* hold */ | ||
| 372 | |||
| 373 | mov rax, [rsp] ; /* restore rbp and rsp */ | ||
| 374 | mov rbp, [rsp+8] | ||
| 375 | mov rsp, rax | ||
| 376 | |||
| 377 | |||
| 378 | |||
| 379 | mov rsi,[rsp-8] | ||
| 380 | mov rdi,[rsp-16] | ||
| 381 | mov r12,[rsp-24] | ||
| 382 | mov r13,[rsp-32] | ||
| 383 | mov r14,[rsp-40] | ||
| 384 | mov r15,[rsp-48] | ||
| 385 | mov rbx,[rsp-56] | ||
| 386 | |||
| 387 | ret 0 | ||
| 388 | ; : | ||
| 389 | ; : "m" (ar) | ||
| 390 | ; : "memory", "%rax", "%rbx", "%rcx", "%rdx", "%rsi", "%rdi", | ||
| 391 | ; "%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15" | ||
| 392 | ; ); | ||
| 393 | |||
| 394 | inffas8664fnc ENDP | ||
| 395 | ;_TEXT ENDS | ||
| 396 | END | ||
diff --git a/contrib/masmx64/readme.txt b/contrib/masmx64/readme.txt deleted file mode 100644 index 2da6733..0000000 --- a/contrib/masmx64/readme.txt +++ /dev/null | |||
| @@ -1,31 +0,0 @@ | |||
| 1 | Summary | ||
| 2 | ------- | ||
| 3 | This directory contains ASM implementations of the functions | ||
| 4 | longest_match() and inflate_fast(), for 64 bits x86 (both AMD64 and Intel EM64t), | ||
| 5 | for use with Microsoft Macro Assembler (x64) for AMD64 and Microsoft C++ 64 bits. | ||
| 6 | |||
| 7 | gvmat64.asm is written by Gilles Vollant (2005), by using Brian Raiter 686/32 bits | ||
| 8 | assembly optimized version from Jean-loup Gailly original longest_match function | ||
| 9 | |||
| 10 | inffasx64.asm and inffas8664.c were written by Chris Anderson, by optimizing | ||
| 11 | original function from Mark Adler | ||
| 12 | |||
| 13 | Use instructions | ||
| 14 | ---------------- | ||
| 15 | Assemble the .asm files using MASM and put the object files into the zlib source | ||
| 16 | directory. You can also get object files here: | ||
| 17 | |||
| 18 | http://www.winimage.com/zLibDll/zlib124_masm_obj.zip | ||
| 19 | |||
| 20 | define ASMV and ASMINF in your project. Include inffas8664.c in your source tree, | ||
| 21 | and inffasx64.obj and gvmat64.obj as object to link. | ||
| 22 | |||
| 23 | |||
| 24 | Build instructions | ||
| 25 | ------------------ | ||
| 26 | run bld_64.bat with Microsoft Macro Assembler (x64) for AMD64 (ml64.exe) | ||
| 27 | |||
| 28 | ml64.exe is given with Visual Studio 2005, Windows 2003 server DDK | ||
| 29 | |||
| 30 | You can get Windows 2003 server DDK with ml64 and cl for AMD64 from | ||
| 31 | http://www.microsoft.com/whdc/devtools/ddk/default.mspx for low price) | ||
diff --git a/contrib/masmx86/bld_ml32.bat b/contrib/masmx86/bld_ml32.bat deleted file mode 100644 index e1b86bf..0000000 --- a/contrib/masmx86/bld_ml32.bat +++ /dev/null | |||
| @@ -1,2 +0,0 @@ | |||
| 1 | ml /coff /Zi /c /Flmatch686.lst match686.asm | ||
| 2 | ml /coff /Zi /c /Flinffas32.lst inffas32.asm | ||
diff --git a/contrib/masmx86/inffas32.asm b/contrib/masmx86/inffas32.asm deleted file mode 100644 index 03d20f8..0000000 --- a/contrib/masmx86/inffas32.asm +++ /dev/null | |||
| @@ -1,1080 +0,0 @@ | |||
| 1 | ;/* inffas32.asm is a hand tuned assembler version of inffast.c -- fast decoding | ||
| 2 | ; * | ||
| 3 | ; * inffas32.asm is derivated from inffas86.c, with translation of assembly code | ||
| 4 | ; * | ||
| 5 | ; * Copyright (C) 1995-2003 Mark Adler | ||
| 6 | ; * For conditions of distribution and use, see copyright notice in zlib.h | ||
| 7 | ; * | ||
| 8 | ; * Copyright (C) 2003 Chris Anderson <christop@charm.net> | ||
| 9 | ; * Please use the copyright conditions above. | ||
| 10 | ; * | ||
| 11 | ; * Mar-13-2003 -- Most of this is derived from inffast.S which is derived from | ||
| 12 | ; * the gcc -S output of zlib-1.2.0/inffast.c. Zlib-1.2.0 is in beta release at | ||
| 13 | ; * the moment. I have successfully compiled and tested this code with gcc2.96, | ||
| 14 | ; * gcc3.2, icc5.0, msvc6.0. It is very close to the speed of inffast.S | ||
| 15 | ; * compiled with gcc -DNO_MMX, but inffast.S is still faster on the P3 with MMX | ||
| 16 | ; * enabled. I will attempt to merge the MMX code into this version. Newer | ||
| 17 | ; * versions of this and inffast.S can be found at | ||
| 18 | ; * http://www.eetbeetee.com/zlib/ and http://www.charm.net/~christop/zlib/ | ||
| 19 | ; * | ||
| 20 | ; * 2005 : modification by Gilles Vollant | ||
| 21 | ; */ | ||
| 22 | ; For Visual C++ 4.x and higher and ML 6.x and higher | ||
| 23 | ; ml.exe is in directory \MASM611C of Win95 DDK | ||
| 24 | ; ml.exe is also distributed in http://www.masm32.com/masmdl.htm | ||
| 25 | ; and in VC++2003 toolkit at http://msdn.microsoft.com/visualc/vctoolkit2003/ | ||
| 26 | ; | ||
| 27 | ; | ||
| 28 | ; compile with command line option | ||
| 29 | ; ml /coff /Zi /c /Flinffas32.lst inffas32.asm | ||
| 30 | |||
| 31 | ; if you define NO_GZIP (see inflate.h), compile with | ||
| 32 | ; ml /coff /Zi /c /Flinffas32.lst /DNO_GUNZIP inffas32.asm | ||
| 33 | |||
| 34 | |||
| 35 | ; zlib122sup is 0 fort zlib 1.2.2.1 and lower | ||
| 36 | ; zlib122sup is 8 fort zlib 1.2.2.2 and more (with addition of dmax and head | ||
| 37 | ; in inflate_state in inflate.h) | ||
| 38 | zlib1222sup equ 8 | ||
| 39 | |||
| 40 | |||
| 41 | IFDEF GUNZIP | ||
| 42 | INFLATE_MODE_TYPE equ 11 | ||
| 43 | INFLATE_MODE_BAD equ 26 | ||
| 44 | ELSE | ||
| 45 | IFNDEF NO_GUNZIP | ||
| 46 | INFLATE_MODE_TYPE equ 11 | ||
| 47 | INFLATE_MODE_BAD equ 26 | ||
| 48 | ELSE | ||
| 49 | INFLATE_MODE_TYPE equ 3 | ||
| 50 | INFLATE_MODE_BAD equ 17 | ||
| 51 | ENDIF | ||
| 52 | ENDIF | ||
| 53 | |||
| 54 | |||
| 55 | ; 75 "inffast.S" | ||
| 56 | ;FILE "inffast.S" | ||
| 57 | |||
| 58 | ;;;GLOBAL _inflate_fast | ||
| 59 | |||
| 60 | ;;;SECTION .text | ||
| 61 | |||
| 62 | |||
| 63 | |||
| 64 | .586p | ||
| 65 | .mmx | ||
| 66 | |||
| 67 | name inflate_fast_x86 | ||
| 68 | .MODEL FLAT | ||
| 69 | |||
| 70 | _DATA segment | ||
| 71 | inflate_fast_use_mmx: | ||
| 72 | dd 1 | ||
| 73 | |||
| 74 | |||
| 75 | _TEXT segment | ||
| 76 | |||
| 77 | |||
| 78 | |||
| 79 | ALIGN 4 | ||
| 80 | db 'Fast decoding Code from Chris Anderson' | ||
| 81 | db 0 | ||
| 82 | |||
| 83 | ALIGN 4 | ||
| 84 | invalid_literal_length_code_msg: | ||
| 85 | db 'invalid literal/length code' | ||
| 86 | db 0 | ||
| 87 | |||
| 88 | ALIGN 4 | ||
| 89 | invalid_distance_code_msg: | ||
| 90 | db 'invalid distance code' | ||
| 91 | db 0 | ||
| 92 | |||
| 93 | ALIGN 4 | ||
| 94 | invalid_distance_too_far_msg: | ||
| 95 | db 'invalid distance too far back' | ||
| 96 | db 0 | ||
| 97 | |||
| 98 | |||
| 99 | ALIGN 4 | ||
| 100 | inflate_fast_mask: | ||
| 101 | dd 0 | ||
| 102 | dd 1 | ||
| 103 | dd 3 | ||
| 104 | dd 7 | ||
| 105 | dd 15 | ||
| 106 | dd 31 | ||
| 107 | dd 63 | ||
| 108 | dd 127 | ||
| 109 | dd 255 | ||
| 110 | dd 511 | ||
| 111 | dd 1023 | ||
| 112 | dd 2047 | ||
| 113 | dd 4095 | ||
| 114 | dd 8191 | ||
| 115 | dd 16383 | ||
| 116 | dd 32767 | ||
| 117 | dd 65535 | ||
| 118 | dd 131071 | ||
| 119 | dd 262143 | ||
| 120 | dd 524287 | ||
| 121 | dd 1048575 | ||
| 122 | dd 2097151 | ||
| 123 | dd 4194303 | ||
| 124 | dd 8388607 | ||
| 125 | dd 16777215 | ||
| 126 | dd 33554431 | ||
| 127 | dd 67108863 | ||
| 128 | dd 134217727 | ||
| 129 | dd 268435455 | ||
| 130 | dd 536870911 | ||
| 131 | dd 1073741823 | ||
| 132 | dd 2147483647 | ||
| 133 | dd 4294967295 | ||
| 134 | |||
| 135 | |||
| 136 | mode_state equ 0 ;/* state->mode */ | ||
| 137 | wsize_state equ (32+zlib1222sup) ;/* state->wsize */ | ||
| 138 | write_state equ (36+4+zlib1222sup) ;/* state->write */ | ||
| 139 | window_state equ (40+4+zlib1222sup) ;/* state->window */ | ||
| 140 | hold_state equ (44+4+zlib1222sup) ;/* state->hold */ | ||
| 141 | bits_state equ (48+4+zlib1222sup) ;/* state->bits */ | ||
| 142 | lencode_state equ (64+4+zlib1222sup) ;/* state->lencode */ | ||
| 143 | distcode_state equ (68+4+zlib1222sup) ;/* state->distcode */ | ||
| 144 | lenbits_state equ (72+4+zlib1222sup) ;/* state->lenbits */ | ||
| 145 | distbits_state equ (76+4+zlib1222sup) ;/* state->distbits */ | ||
| 146 | |||
| 147 | |||
| 148 | ;;SECTION .text | ||
| 149 | ; 205 "inffast.S" | ||
| 150 | ;GLOBAL inflate_fast_use_mmx | ||
| 151 | |||
| 152 | ;SECTION .data | ||
| 153 | |||
| 154 | |||
| 155 | ; GLOBAL inflate_fast_use_mmx:object | ||
| 156 | ;.size inflate_fast_use_mmx, 4 | ||
| 157 | ; 226 "inffast.S" | ||
| 158 | ;SECTION .text | ||
| 159 | |||
| 160 | ALIGN 4 | ||
| 161 | _inflate_fast proc near | ||
| 162 | .FPO (16, 4, 0, 0, 1, 0) | ||
| 163 | push edi | ||
| 164 | push esi | ||
| 165 | push ebp | ||
| 166 | push ebx | ||
| 167 | pushfd | ||
| 168 | sub esp,64 | ||
| 169 | cld | ||
| 170 | |||
| 171 | |||
| 172 | |||
| 173 | |||
| 174 | mov esi, [esp+88] | ||
| 175 | mov edi, [esi+28] | ||
| 176 | |||
| 177 | |||
| 178 | |||
| 179 | |||
| 180 | |||
| 181 | |||
| 182 | |||
| 183 | mov edx, [esi+4] | ||
| 184 | mov eax, [esi+0] | ||
| 185 | |||
| 186 | add edx,eax | ||
| 187 | sub edx,11 | ||
| 188 | |||
| 189 | mov [esp+44],eax | ||
| 190 | mov [esp+20],edx | ||
| 191 | |||
| 192 | mov ebp, [esp+92] | ||
| 193 | mov ecx, [esi+16] | ||
| 194 | mov ebx, [esi+12] | ||
| 195 | |||
| 196 | sub ebp,ecx | ||
| 197 | neg ebp | ||
| 198 | add ebp,ebx | ||
| 199 | |||
| 200 | sub ecx,257 | ||
| 201 | add ecx,ebx | ||
| 202 | |||
| 203 | mov [esp+60],ebx | ||
| 204 | mov [esp+40],ebp | ||
| 205 | mov [esp+16],ecx | ||
| 206 | ; 285 "inffast.S" | ||
| 207 | mov eax, [edi+lencode_state] | ||
| 208 | mov ecx, [edi+distcode_state] | ||
| 209 | |||
| 210 | mov [esp+8],eax | ||
| 211 | mov [esp+12],ecx | ||
| 212 | |||
| 213 | mov eax,1 | ||
| 214 | mov ecx, [edi+lenbits_state] | ||
| 215 | shl eax,cl | ||
| 216 | dec eax | ||
| 217 | mov [esp+0],eax | ||
| 218 | |||
| 219 | mov eax,1 | ||
| 220 | mov ecx, [edi+distbits_state] | ||
| 221 | shl eax,cl | ||
| 222 | dec eax | ||
| 223 | mov [esp+4],eax | ||
| 224 | |||
| 225 | mov eax, [edi+wsize_state] | ||
| 226 | mov ecx, [edi+write_state] | ||
| 227 | mov edx, [edi+window_state] | ||
| 228 | |||
| 229 | mov [esp+52],eax | ||
| 230 | mov [esp+48],ecx | ||
| 231 | mov [esp+56],edx | ||
| 232 | |||
| 233 | mov ebp, [edi+hold_state] | ||
| 234 | mov ebx, [edi+bits_state] | ||
| 235 | ; 321 "inffast.S" | ||
| 236 | mov esi, [esp+44] | ||
| 237 | mov ecx, [esp+20] | ||
| 238 | cmp ecx,esi | ||
| 239 | ja L_align_long | ||
| 240 | |||
| 241 | add ecx,11 | ||
| 242 | sub ecx,esi | ||
| 243 | mov eax,12 | ||
| 244 | sub eax,ecx | ||
| 245 | lea edi, [esp+28] | ||
| 246 | rep movsb | ||
| 247 | mov ecx,eax | ||
| 248 | xor eax,eax | ||
| 249 | rep stosb | ||
| 250 | lea esi, [esp+28] | ||
| 251 | mov [esp+20],esi | ||
| 252 | jmp L_is_aligned | ||
| 253 | |||
| 254 | |||
| 255 | L_align_long: | ||
| 256 | test esi,3 | ||
| 257 | jz L_is_aligned | ||
| 258 | xor eax,eax | ||
| 259 | mov al, [esi] | ||
| 260 | inc esi | ||
| 261 | mov ecx,ebx | ||
| 262 | add ebx,8 | ||
| 263 | shl eax,cl | ||
| 264 | or ebp,eax | ||
| 265 | jmp L_align_long | ||
| 266 | |||
| 267 | L_is_aligned: | ||
| 268 | mov edi, [esp+60] | ||
| 269 | ; 366 "inffast.S" | ||
| 270 | L_check_mmx: | ||
| 271 | cmp dword ptr [inflate_fast_use_mmx],2 | ||
| 272 | je L_init_mmx | ||
| 273 | ja L_do_loop | ||
| 274 | |||
| 275 | push eax | ||
| 276 | push ebx | ||
| 277 | push ecx | ||
| 278 | push edx | ||
| 279 | pushfd | ||
| 280 | mov eax, [esp] | ||
| 281 | xor dword ptr [esp],0200000h | ||
| 282 | |||
| 283 | |||
| 284 | |||
| 285 | |||
| 286 | popfd | ||
| 287 | pushfd | ||
| 288 | pop edx | ||
| 289 | xor edx,eax | ||
| 290 | jz L_dont_use_mmx | ||
| 291 | xor eax,eax | ||
| 292 | cpuid | ||
| 293 | cmp ebx,0756e6547h | ||
| 294 | jne L_dont_use_mmx | ||
| 295 | cmp ecx,06c65746eh | ||
| 296 | jne L_dont_use_mmx | ||
| 297 | cmp edx,049656e69h | ||
| 298 | jne L_dont_use_mmx | ||
| 299 | mov eax,1 | ||
| 300 | cpuid | ||
| 301 | shr eax,8 | ||
| 302 | and eax,15 | ||
| 303 | cmp eax,6 | ||
| 304 | jne L_dont_use_mmx | ||
| 305 | test edx,0800000h | ||
| 306 | jnz L_use_mmx | ||
| 307 | jmp L_dont_use_mmx | ||
| 308 | L_use_mmx: | ||
| 309 | mov dword ptr [inflate_fast_use_mmx],2 | ||
| 310 | jmp L_check_mmx_pop | ||
| 311 | L_dont_use_mmx: | ||
| 312 | mov dword ptr [inflate_fast_use_mmx],3 | ||
| 313 | L_check_mmx_pop: | ||
| 314 | pop edx | ||
| 315 | pop ecx | ||
| 316 | pop ebx | ||
| 317 | pop eax | ||
| 318 | jmp L_check_mmx | ||
| 319 | ; 426 "inffast.S" | ||
| 320 | ALIGN 4 | ||
| 321 | L_do_loop: | ||
| 322 | ; 437 "inffast.S" | ||
| 323 | cmp bl,15 | ||
| 324 | ja L_get_length_code | ||
| 325 | |||
| 326 | xor eax,eax | ||
| 327 | lodsw | ||
| 328 | mov cl,bl | ||
| 329 | add bl,16 | ||
| 330 | shl eax,cl | ||
| 331 | or ebp,eax | ||
| 332 | |||
| 333 | L_get_length_code: | ||
| 334 | mov edx, [esp+0] | ||
| 335 | mov ecx, [esp+8] | ||
| 336 | and edx,ebp | ||
| 337 | mov eax, [ecx+edx*4] | ||
| 338 | |||
| 339 | L_dolen: | ||
| 340 | |||
| 341 | |||
| 342 | |||
| 343 | |||
| 344 | |||
| 345 | |||
| 346 | mov cl,ah | ||
| 347 | sub bl,ah | ||
| 348 | shr ebp,cl | ||
| 349 | |||
| 350 | |||
| 351 | |||
| 352 | |||
| 353 | |||
| 354 | |||
| 355 | test al,al | ||
| 356 | jnz L_test_for_length_base | ||
| 357 | |||
| 358 | shr eax,16 | ||
| 359 | stosb | ||
| 360 | |||
| 361 | L_while_test: | ||
| 362 | |||
| 363 | |||
| 364 | cmp [esp+16],edi | ||
| 365 | jbe L_break_loop | ||
| 366 | |||
| 367 | cmp [esp+20],esi | ||
| 368 | ja L_do_loop | ||
| 369 | jmp L_break_loop | ||
| 370 | |||
| 371 | L_test_for_length_base: | ||
| 372 | ; 502 "inffast.S" | ||
| 373 | mov edx,eax | ||
| 374 | shr edx,16 | ||
| 375 | mov cl,al | ||
| 376 | |||
| 377 | test al,16 | ||
| 378 | jz L_test_for_second_level_length | ||
| 379 | and cl,15 | ||
| 380 | jz L_save_len | ||
| 381 | cmp bl,cl | ||
| 382 | jae L_add_bits_to_len | ||
| 383 | |||
| 384 | mov ch,cl | ||
| 385 | xor eax,eax | ||
| 386 | lodsw | ||
| 387 | mov cl,bl | ||
| 388 | add bl,16 | ||
| 389 | shl eax,cl | ||
| 390 | or ebp,eax | ||
| 391 | mov cl,ch | ||
| 392 | |||
| 393 | L_add_bits_to_len: | ||
| 394 | mov eax,1 | ||
| 395 | shl eax,cl | ||
| 396 | dec eax | ||
| 397 | sub bl,cl | ||
| 398 | and eax,ebp | ||
| 399 | shr ebp,cl | ||
| 400 | add edx,eax | ||
| 401 | |||
| 402 | L_save_len: | ||
| 403 | mov [esp+24],edx | ||
| 404 | |||
| 405 | |||
| 406 | L_decode_distance: | ||
| 407 | ; 549 "inffast.S" | ||
| 408 | cmp bl,15 | ||
| 409 | ja L_get_distance_code | ||
| 410 | |||
| 411 | xor eax,eax | ||
| 412 | lodsw | ||
| 413 | mov cl,bl | ||
| 414 | add bl,16 | ||
| 415 | shl eax,cl | ||
| 416 | or ebp,eax | ||
| 417 | |||
| 418 | L_get_distance_code: | ||
| 419 | mov edx, [esp+4] | ||
| 420 | mov ecx, [esp+12] | ||
| 421 | and edx,ebp | ||
| 422 | mov eax, [ecx+edx*4] | ||
| 423 | |||
| 424 | |||
| 425 | L_dodist: | ||
| 426 | mov edx,eax | ||
| 427 | shr edx,16 | ||
| 428 | mov cl,ah | ||
| 429 | sub bl,ah | ||
| 430 | shr ebp,cl | ||
| 431 | ; 584 "inffast.S" | ||
| 432 | mov cl,al | ||
| 433 | |||
| 434 | test al,16 | ||
| 435 | jz L_test_for_second_level_dist | ||
| 436 | and cl,15 | ||
| 437 | jz L_check_dist_one | ||
| 438 | cmp bl,cl | ||
| 439 | jae L_add_bits_to_dist | ||
| 440 | |||
| 441 | mov ch,cl | ||
| 442 | xor eax,eax | ||
| 443 | lodsw | ||
| 444 | mov cl,bl | ||
| 445 | add bl,16 | ||
| 446 | shl eax,cl | ||
| 447 | or ebp,eax | ||
| 448 | mov cl,ch | ||
| 449 | |||
| 450 | L_add_bits_to_dist: | ||
| 451 | mov eax,1 | ||
| 452 | shl eax,cl | ||
| 453 | dec eax | ||
| 454 | sub bl,cl | ||
| 455 | and eax,ebp | ||
| 456 | shr ebp,cl | ||
| 457 | add edx,eax | ||
| 458 | jmp L_check_window | ||
| 459 | |||
| 460 | L_check_window: | ||
| 461 | ; 625 "inffast.S" | ||
| 462 | mov [esp+44],esi | ||
| 463 | mov eax,edi | ||
| 464 | sub eax, [esp+40] | ||
| 465 | |||
| 466 | cmp eax,edx | ||
| 467 | jb L_clip_window | ||
| 468 | |||
| 469 | mov ecx, [esp+24] | ||
| 470 | mov esi,edi | ||
| 471 | sub esi,edx | ||
| 472 | |||
| 473 | sub ecx,3 | ||
| 474 | mov al, [esi] | ||
| 475 | mov [edi],al | ||
| 476 | mov al, [esi+1] | ||
| 477 | mov dl, [esi+2] | ||
| 478 | add esi,3 | ||
| 479 | mov [edi+1],al | ||
| 480 | mov [edi+2],dl | ||
| 481 | add edi,3 | ||
| 482 | rep movsb | ||
| 483 | |||
| 484 | mov esi, [esp+44] | ||
| 485 | jmp L_while_test | ||
| 486 | |||
| 487 | ALIGN 4 | ||
| 488 | L_check_dist_one: | ||
| 489 | cmp edx,1 | ||
| 490 | jne L_check_window | ||
| 491 | cmp [esp+40],edi | ||
| 492 | je L_check_window | ||
| 493 | |||
| 494 | dec edi | ||
| 495 | mov ecx, [esp+24] | ||
| 496 | mov al, [edi] | ||
| 497 | sub ecx,3 | ||
| 498 | |||
| 499 | mov [edi+1],al | ||
| 500 | mov [edi+2],al | ||
| 501 | mov [edi+3],al | ||
| 502 | add edi,4 | ||
| 503 | rep stosb | ||
| 504 | |||
| 505 | jmp L_while_test | ||
| 506 | |||
| 507 | ALIGN 4 | ||
| 508 | L_test_for_second_level_length: | ||
| 509 | |||
| 510 | |||
| 511 | |||
| 512 | |||
| 513 | test al,64 | ||
| 514 | jnz L_test_for_end_of_block | ||
| 515 | |||
| 516 | mov eax,1 | ||
| 517 | shl eax,cl | ||
| 518 | dec eax | ||
| 519 | and eax,ebp | ||
| 520 | add eax,edx | ||
| 521 | mov edx, [esp+8] | ||
| 522 | mov eax, [edx+eax*4] | ||
| 523 | jmp L_dolen | ||
| 524 | |||
| 525 | ALIGN 4 | ||
| 526 | L_test_for_second_level_dist: | ||
| 527 | |||
| 528 | |||
| 529 | |||
| 530 | |||
| 531 | test al,64 | ||
| 532 | jnz L_invalid_distance_code | ||
| 533 | |||
| 534 | mov eax,1 | ||
| 535 | shl eax,cl | ||
| 536 | dec eax | ||
| 537 | and eax,ebp | ||
| 538 | add eax,edx | ||
| 539 | mov edx, [esp+12] | ||
| 540 | mov eax, [edx+eax*4] | ||
| 541 | jmp L_dodist | ||
| 542 | |||
| 543 | ALIGN 4 | ||
| 544 | L_clip_window: | ||
| 545 | ; 721 "inffast.S" | ||
| 546 | mov ecx,eax | ||
| 547 | mov eax, [esp+52] | ||
| 548 | neg ecx | ||
| 549 | mov esi, [esp+56] | ||
| 550 | |||
| 551 | cmp eax,edx | ||
| 552 | jb L_invalid_distance_too_far | ||
| 553 | |||
| 554 | add ecx,edx | ||
| 555 | cmp dword ptr [esp+48],0 | ||
| 556 | jne L_wrap_around_window | ||
| 557 | |||
| 558 | sub eax,ecx | ||
| 559 | add esi,eax | ||
| 560 | ; 749 "inffast.S" | ||
| 561 | mov eax, [esp+24] | ||
| 562 | cmp eax,ecx | ||
| 563 | jbe L_do_copy1 | ||
| 564 | |||
| 565 | sub eax,ecx | ||
| 566 | rep movsb | ||
| 567 | mov esi,edi | ||
| 568 | sub esi,edx | ||
| 569 | jmp L_do_copy1 | ||
| 570 | |||
| 571 | cmp eax,ecx | ||
| 572 | jbe L_do_copy1 | ||
| 573 | |||
| 574 | sub eax,ecx | ||
| 575 | rep movsb | ||
| 576 | mov esi,edi | ||
| 577 | sub esi,edx | ||
| 578 | jmp L_do_copy1 | ||
| 579 | |||
| 580 | L_wrap_around_window: | ||
| 581 | ; 793 "inffast.S" | ||
| 582 | mov eax, [esp+48] | ||
| 583 | cmp ecx,eax | ||
| 584 | jbe L_contiguous_in_window | ||
| 585 | |||
| 586 | add esi, [esp+52] | ||
| 587 | add esi,eax | ||
| 588 | sub esi,ecx | ||
| 589 | sub ecx,eax | ||
| 590 | |||
| 591 | |||
| 592 | mov eax, [esp+24] | ||
| 593 | cmp eax,ecx | ||
| 594 | jbe L_do_copy1 | ||
| 595 | |||
| 596 | sub eax,ecx | ||
| 597 | rep movsb | ||
| 598 | mov esi, [esp+56] | ||
| 599 | mov ecx, [esp+48] | ||
| 600 | cmp eax,ecx | ||
| 601 | jbe L_do_copy1 | ||
| 602 | |||
| 603 | sub eax,ecx | ||
| 604 | rep movsb | ||
| 605 | mov esi,edi | ||
| 606 | sub esi,edx | ||
| 607 | jmp L_do_copy1 | ||
| 608 | |||
| 609 | L_contiguous_in_window: | ||
| 610 | ; 836 "inffast.S" | ||
| 611 | add esi,eax | ||
| 612 | sub esi,ecx | ||
| 613 | |||
| 614 | |||
| 615 | mov eax, [esp+24] | ||
| 616 | cmp eax,ecx | ||
| 617 | jbe L_do_copy1 | ||
| 618 | |||
| 619 | sub eax,ecx | ||
| 620 | rep movsb | ||
| 621 | mov esi,edi | ||
| 622 | sub esi,edx | ||
| 623 | |||
| 624 | L_do_copy1: | ||
| 625 | ; 862 "inffast.S" | ||
| 626 | mov ecx,eax | ||
| 627 | rep movsb | ||
| 628 | |||
| 629 | mov esi, [esp+44] | ||
| 630 | jmp L_while_test | ||
| 631 | ; 878 "inffast.S" | ||
| 632 | ALIGN 4 | ||
| 633 | L_init_mmx: | ||
| 634 | emms | ||
| 635 | |||
| 636 | |||
| 637 | |||
| 638 | |||
| 639 | |||
| 640 | movd mm0,ebp | ||
| 641 | mov ebp,ebx | ||
| 642 | ; 896 "inffast.S" | ||
| 643 | movd mm4,dword ptr [esp+0] | ||
| 644 | movq mm3,mm4 | ||
| 645 | movd mm5,dword ptr [esp+4] | ||
| 646 | movq mm2,mm5 | ||
| 647 | pxor mm1,mm1 | ||
| 648 | mov ebx, [esp+8] | ||
| 649 | jmp L_do_loop_mmx | ||
| 650 | |||
| 651 | ALIGN 4 | ||
| 652 | L_do_loop_mmx: | ||
| 653 | psrlq mm0,mm1 | ||
| 654 | |||
| 655 | cmp ebp,32 | ||
| 656 | ja L_get_length_code_mmx | ||
| 657 | |||
| 658 | movd mm6,ebp | ||
| 659 | movd mm7,dword ptr [esi] | ||
| 660 | add esi,4 | ||
| 661 | psllq mm7,mm6 | ||
| 662 | add ebp,32 | ||
| 663 | por mm0,mm7 | ||
| 664 | |||
| 665 | L_get_length_code_mmx: | ||
| 666 | pand mm4,mm0 | ||
| 667 | movd eax,mm4 | ||
| 668 | movq mm4,mm3 | ||
| 669 | mov eax, [ebx+eax*4] | ||
| 670 | |||
| 671 | L_dolen_mmx: | ||
| 672 | movzx ecx,ah | ||
| 673 | movd mm1,ecx | ||
| 674 | sub ebp,ecx | ||
| 675 | |||
| 676 | test al,al | ||
| 677 | jnz L_test_for_length_base_mmx | ||
| 678 | |||
| 679 | shr eax,16 | ||
| 680 | stosb | ||
| 681 | |||
| 682 | L_while_test_mmx: | ||
| 683 | |||
| 684 | |||
| 685 | cmp [esp+16],edi | ||
| 686 | jbe L_break_loop | ||
| 687 | |||
| 688 | cmp [esp+20],esi | ||
| 689 | ja L_do_loop_mmx | ||
| 690 | jmp L_break_loop | ||
| 691 | |||
| 692 | L_test_for_length_base_mmx: | ||
| 693 | |||
| 694 | mov edx,eax | ||
| 695 | shr edx,16 | ||
| 696 | |||
| 697 | test al,16 | ||
| 698 | jz L_test_for_second_level_length_mmx | ||
| 699 | and eax,15 | ||
| 700 | jz L_decode_distance_mmx | ||
| 701 | |||
| 702 | psrlq mm0,mm1 | ||
| 703 | movd mm1,eax | ||
| 704 | movd ecx,mm0 | ||
| 705 | sub ebp,eax | ||
| 706 | and ecx, [inflate_fast_mask+eax*4] | ||
| 707 | add edx,ecx | ||
| 708 | |||
| 709 | L_decode_distance_mmx: | ||
| 710 | psrlq mm0,mm1 | ||
| 711 | |||
| 712 | cmp ebp,32 | ||
| 713 | ja L_get_dist_code_mmx | ||
| 714 | |||
| 715 | movd mm6,ebp | ||
| 716 | movd mm7,dword ptr [esi] | ||
| 717 | add esi,4 | ||
| 718 | psllq mm7,mm6 | ||
| 719 | add ebp,32 | ||
| 720 | por mm0,mm7 | ||
| 721 | |||
| 722 | L_get_dist_code_mmx: | ||
| 723 | mov ebx, [esp+12] | ||
| 724 | pand mm5,mm0 | ||
| 725 | movd eax,mm5 | ||
| 726 | movq mm5,mm2 | ||
| 727 | mov eax, [ebx+eax*4] | ||
| 728 | |||
| 729 | L_dodist_mmx: | ||
| 730 | |||
| 731 | movzx ecx,ah | ||
| 732 | mov ebx,eax | ||
| 733 | shr ebx,16 | ||
| 734 | sub ebp,ecx | ||
| 735 | movd mm1,ecx | ||
| 736 | |||
| 737 | test al,16 | ||
| 738 | jz L_test_for_second_level_dist_mmx | ||
| 739 | and eax,15 | ||
| 740 | jz L_check_dist_one_mmx | ||
| 741 | |||
| 742 | L_add_bits_to_dist_mmx: | ||
| 743 | psrlq mm0,mm1 | ||
| 744 | movd mm1,eax | ||
| 745 | movd ecx,mm0 | ||
| 746 | sub ebp,eax | ||
| 747 | and ecx, [inflate_fast_mask+eax*4] | ||
| 748 | add ebx,ecx | ||
| 749 | |||
| 750 | L_check_window_mmx: | ||
| 751 | mov [esp+44],esi | ||
| 752 | mov eax,edi | ||
| 753 | sub eax, [esp+40] | ||
| 754 | |||
| 755 | cmp eax,ebx | ||
| 756 | jb L_clip_window_mmx | ||
| 757 | |||
| 758 | mov ecx,edx | ||
| 759 | mov esi,edi | ||
| 760 | sub esi,ebx | ||
| 761 | |||
| 762 | sub ecx,3 | ||
| 763 | mov al, [esi] | ||
| 764 | mov [edi],al | ||
| 765 | mov al, [esi+1] | ||
| 766 | mov dl, [esi+2] | ||
| 767 | add esi,3 | ||
| 768 | mov [edi+1],al | ||
| 769 | mov [edi+2],dl | ||
| 770 | add edi,3 | ||
| 771 | rep movsb | ||
| 772 | |||
| 773 | mov esi, [esp+44] | ||
| 774 | mov ebx, [esp+8] | ||
| 775 | jmp L_while_test_mmx | ||
| 776 | |||
| 777 | ALIGN 4 | ||
| 778 | L_check_dist_one_mmx: | ||
| 779 | cmp ebx,1 | ||
| 780 | jne L_check_window_mmx | ||
| 781 | cmp [esp+40],edi | ||
| 782 | je L_check_window_mmx | ||
| 783 | |||
| 784 | dec edi | ||
| 785 | mov ecx,edx | ||
| 786 | mov al, [edi] | ||
| 787 | sub ecx,3 | ||
| 788 | |||
| 789 | mov [edi+1],al | ||
| 790 | mov [edi+2],al | ||
| 791 | mov [edi+3],al | ||
| 792 | add edi,4 | ||
| 793 | rep stosb | ||
| 794 | |||
| 795 | mov ebx, [esp+8] | ||
| 796 | jmp L_while_test_mmx | ||
| 797 | |||
| 798 | ALIGN 4 | ||
| 799 | L_test_for_second_level_length_mmx: | ||
| 800 | test al,64 | ||
| 801 | jnz L_test_for_end_of_block | ||
| 802 | |||
| 803 | and eax,15 | ||
| 804 | psrlq mm0,mm1 | ||
| 805 | movd ecx,mm0 | ||
| 806 | and ecx, [inflate_fast_mask+eax*4] | ||
| 807 | add ecx,edx | ||
| 808 | mov eax, [ebx+ecx*4] | ||
| 809 | jmp L_dolen_mmx | ||
| 810 | |||
| 811 | ALIGN 4 | ||
| 812 | L_test_for_second_level_dist_mmx: | ||
| 813 | test al,64 | ||
| 814 | jnz L_invalid_distance_code | ||
| 815 | |||
| 816 | and eax,15 | ||
| 817 | psrlq mm0,mm1 | ||
| 818 | movd ecx,mm0 | ||
| 819 | and ecx, [inflate_fast_mask+eax*4] | ||
| 820 | mov eax, [esp+12] | ||
| 821 | add ecx,ebx | ||
| 822 | mov eax, [eax+ecx*4] | ||
| 823 | jmp L_dodist_mmx | ||
| 824 | |||
| 825 | ALIGN 4 | ||
| 826 | L_clip_window_mmx: | ||
| 827 | |||
| 828 | mov ecx,eax | ||
| 829 | mov eax, [esp+52] | ||
| 830 | neg ecx | ||
| 831 | mov esi, [esp+56] | ||
| 832 | |||
| 833 | cmp eax,ebx | ||
| 834 | jb L_invalid_distance_too_far | ||
| 835 | |||
| 836 | add ecx,ebx | ||
| 837 | cmp dword ptr [esp+48],0 | ||
| 838 | jne L_wrap_around_window_mmx | ||
| 839 | |||
| 840 | sub eax,ecx | ||
| 841 | add esi,eax | ||
| 842 | |||
| 843 | cmp edx,ecx | ||
| 844 | jbe L_do_copy1_mmx | ||
| 845 | |||
| 846 | sub edx,ecx | ||
| 847 | rep movsb | ||
| 848 | mov esi,edi | ||
| 849 | sub esi,ebx | ||
| 850 | jmp L_do_copy1_mmx | ||
| 851 | |||
| 852 | cmp edx,ecx | ||
| 853 | jbe L_do_copy1_mmx | ||
| 854 | |||
| 855 | sub edx,ecx | ||
| 856 | rep movsb | ||
| 857 | mov esi,edi | ||
| 858 | sub esi,ebx | ||
| 859 | jmp L_do_copy1_mmx | ||
| 860 | |||
| 861 | L_wrap_around_window_mmx: | ||
| 862 | |||
| 863 | mov eax, [esp+48] | ||
| 864 | cmp ecx,eax | ||
| 865 | jbe L_contiguous_in_window_mmx | ||
| 866 | |||
| 867 | add esi, [esp+52] | ||
| 868 | add esi,eax | ||
| 869 | sub esi,ecx | ||
| 870 | sub ecx,eax | ||
| 871 | |||
| 872 | |||
| 873 | cmp edx,ecx | ||
| 874 | jbe L_do_copy1_mmx | ||
| 875 | |||
| 876 | sub edx,ecx | ||
| 877 | rep movsb | ||
| 878 | mov esi, [esp+56] | ||
| 879 | mov ecx, [esp+48] | ||
| 880 | cmp edx,ecx | ||
| 881 | jbe L_do_copy1_mmx | ||
| 882 | |||
| 883 | sub edx,ecx | ||
| 884 | rep movsb | ||
| 885 | mov esi,edi | ||
| 886 | sub esi,ebx | ||
| 887 | jmp L_do_copy1_mmx | ||
| 888 | |||
| 889 | L_contiguous_in_window_mmx: | ||
| 890 | |||
| 891 | add esi,eax | ||
| 892 | sub esi,ecx | ||
| 893 | |||
| 894 | |||
| 895 | cmp edx,ecx | ||
| 896 | jbe L_do_copy1_mmx | ||
| 897 | |||
| 898 | sub edx,ecx | ||
| 899 | rep movsb | ||
| 900 | mov esi,edi | ||
| 901 | sub esi,ebx | ||
| 902 | |||
| 903 | L_do_copy1_mmx: | ||
| 904 | |||
| 905 | |||
| 906 | mov ecx,edx | ||
| 907 | rep movsb | ||
| 908 | |||
| 909 | mov esi, [esp+44] | ||
| 910 | mov ebx, [esp+8] | ||
| 911 | jmp L_while_test_mmx | ||
| 912 | ; 1174 "inffast.S" | ||
| 913 | L_invalid_distance_code: | ||
| 914 | |||
| 915 | |||
| 916 | |||
| 917 | |||
| 918 | |||
| 919 | mov ecx, invalid_distance_code_msg | ||
| 920 | mov edx,INFLATE_MODE_BAD | ||
| 921 | jmp L_update_stream_state | ||
| 922 | |||
| 923 | L_test_for_end_of_block: | ||
| 924 | |||
| 925 | |||
| 926 | |||
| 927 | |||
| 928 | |||
| 929 | test al,32 | ||
| 930 | jz L_invalid_literal_length_code | ||
| 931 | |||
| 932 | mov ecx,0 | ||
| 933 | mov edx,INFLATE_MODE_TYPE | ||
| 934 | jmp L_update_stream_state | ||
| 935 | |||
| 936 | L_invalid_literal_length_code: | ||
| 937 | |||
| 938 | |||
| 939 | |||
| 940 | |||
| 941 | |||
| 942 | mov ecx, invalid_literal_length_code_msg | ||
| 943 | mov edx,INFLATE_MODE_BAD | ||
| 944 | jmp L_update_stream_state | ||
| 945 | |||
| 946 | L_invalid_distance_too_far: | ||
| 947 | |||
| 948 | |||
| 949 | |||
| 950 | mov esi, [esp+44] | ||
| 951 | mov ecx, invalid_distance_too_far_msg | ||
| 952 | mov edx,INFLATE_MODE_BAD | ||
| 953 | jmp L_update_stream_state | ||
| 954 | |||
| 955 | L_update_stream_state: | ||
| 956 | |||
| 957 | mov eax, [esp+88] | ||
| 958 | test ecx,ecx | ||
| 959 | jz L_skip_msg | ||
| 960 | mov [eax+24],ecx | ||
| 961 | L_skip_msg: | ||
| 962 | mov eax, [eax+28] | ||
| 963 | mov [eax+mode_state],edx | ||
| 964 | jmp L_break_loop | ||
| 965 | |||
| 966 | ALIGN 4 | ||
| 967 | L_break_loop: | ||
| 968 | ; 1243 "inffast.S" | ||
| 969 | cmp dword ptr [inflate_fast_use_mmx],2 | ||
| 970 | jne L_update_next_in | ||
| 971 | |||
| 972 | |||
| 973 | |||
| 974 | mov ebx,ebp | ||
| 975 | |||
| 976 | L_update_next_in: | ||
| 977 | ; 1266 "inffast.S" | ||
| 978 | mov eax, [esp+88] | ||
| 979 | mov ecx,ebx | ||
| 980 | mov edx, [eax+28] | ||
| 981 | shr ecx,3 | ||
| 982 | sub esi,ecx | ||
| 983 | shl ecx,3 | ||
| 984 | sub ebx,ecx | ||
| 985 | mov [eax+12],edi | ||
| 986 | mov [edx+bits_state],ebx | ||
| 987 | mov ecx,ebx | ||
| 988 | |||
| 989 | lea ebx, [esp+28] | ||
| 990 | cmp [esp+20],ebx | ||
| 991 | jne L_buf_not_used | ||
| 992 | |||
| 993 | sub esi,ebx | ||
| 994 | mov ebx, [eax+0] | ||
| 995 | mov [esp+20],ebx | ||
| 996 | add esi,ebx | ||
| 997 | mov ebx, [eax+4] | ||
| 998 | sub ebx,11 | ||
| 999 | add [esp+20],ebx | ||
| 1000 | |||
| 1001 | L_buf_not_used: | ||
| 1002 | mov [eax+0],esi | ||
| 1003 | |||
| 1004 | mov ebx,1 | ||
| 1005 | shl ebx,cl | ||
| 1006 | dec ebx | ||
| 1007 | |||
| 1008 | |||
| 1009 | |||
| 1010 | |||
| 1011 | |||
| 1012 | cmp dword ptr [inflate_fast_use_mmx],2 | ||
| 1013 | jne L_update_hold | ||
| 1014 | |||
| 1015 | |||
| 1016 | |||
| 1017 | psrlq mm0,mm1 | ||
| 1018 | movd ebp,mm0 | ||
| 1019 | |||
| 1020 | emms | ||
| 1021 | |||
| 1022 | L_update_hold: | ||
| 1023 | |||
| 1024 | |||
| 1025 | |||
| 1026 | and ebp,ebx | ||
| 1027 | mov [edx+hold_state],ebp | ||
| 1028 | |||
| 1029 | |||
| 1030 | |||
| 1031 | |||
| 1032 | mov ebx, [esp+20] | ||
| 1033 | cmp ebx,esi | ||
| 1034 | jbe L_last_is_smaller | ||
| 1035 | |||
| 1036 | sub ebx,esi | ||
| 1037 | add ebx,11 | ||
| 1038 | mov [eax+4],ebx | ||
| 1039 | jmp L_fixup_out | ||
| 1040 | L_last_is_smaller: | ||
| 1041 | sub esi,ebx | ||
| 1042 | neg esi | ||
| 1043 | add esi,11 | ||
| 1044 | mov [eax+4],esi | ||
| 1045 | |||
| 1046 | |||
| 1047 | |||
| 1048 | |||
| 1049 | L_fixup_out: | ||
| 1050 | |||
| 1051 | mov ebx, [esp+16] | ||
| 1052 | cmp ebx,edi | ||
| 1053 | jbe L_end_is_smaller | ||
| 1054 | |||
| 1055 | sub ebx,edi | ||
| 1056 | add ebx,257 | ||
| 1057 | mov [eax+16],ebx | ||
| 1058 | jmp L_done | ||
| 1059 | L_end_is_smaller: | ||
| 1060 | sub edi,ebx | ||
| 1061 | neg edi | ||
| 1062 | add edi,257 | ||
| 1063 | mov [eax+16],edi | ||
| 1064 | |||
| 1065 | |||
| 1066 | |||
| 1067 | |||
| 1068 | |||
| 1069 | L_done: | ||
| 1070 | add esp,64 | ||
| 1071 | popfd | ||
| 1072 | pop ebx | ||
| 1073 | pop ebp | ||
| 1074 | pop esi | ||
| 1075 | pop edi | ||
| 1076 | ret | ||
| 1077 | _inflate_fast endp | ||
| 1078 | |||
| 1079 | _TEXT ends | ||
| 1080 | end | ||
diff --git a/contrib/masmx86/match686.asm b/contrib/masmx86/match686.asm deleted file mode 100644 index 3b09212..0000000 --- a/contrib/masmx86/match686.asm +++ /dev/null | |||
| @@ -1,479 +0,0 @@ | |||
| 1 | ; match686.asm -- Asm portion of the optimized longest_match for 32 bits x86 | ||
| 2 | ; Copyright (C) 1995-1996 Jean-loup Gailly, Brian Raiter and Gilles Vollant. | ||
| 3 | ; File written by Gilles Vollant, by converting match686.S from Brian Raiter | ||
| 4 | ; for MASM. This is as assembly version of longest_match | ||
| 5 | ; from Jean-loup Gailly in deflate.c | ||
| 6 | ; | ||
| 7 | ; http://www.zlib.net | ||
| 8 | ; http://www.winimage.com/zLibDll | ||
| 9 | ; http://www.muppetlabs.com/~breadbox/software/assembly.html | ||
| 10 | ; | ||
| 11 | ; For Visual C++ 4.x and higher and ML 6.x and higher | ||
| 12 | ; ml.exe is distributed in | ||
| 13 | ; http://www.microsoft.com/downloads/details.aspx?FamilyID=7a1c9da0-0510-44a2-b042-7ef370530c64 | ||
| 14 | ; | ||
| 15 | ; this file contain two implementation of longest_match | ||
| 16 | ; | ||
| 17 | ; this longest_match was written by Brian raiter (1998), optimized for Pentium Pro | ||
| 18 | ; (and the faster known version of match_init on modern Core 2 Duo and AMD Phenom) | ||
| 19 | ; | ||
| 20 | ; for using an assembly version of longest_match, you need define ASMV in project | ||
| 21 | ; | ||
| 22 | ; compile the asm file running | ||
| 23 | ; ml /coff /Zi /c /Flmatch686.lst match686.asm | ||
| 24 | ; and do not include match686.obj in your project | ||
| 25 | ; | ||
| 26 | ; note: contrib of zLib 1.2.3 and earlier contained both a deprecated version for | ||
| 27 | ; Pentium (prior Pentium Pro) and this version for Pentium Pro and modern processor | ||
| 28 | ; with autoselect (with cpu detection code) | ||
| 29 | ; if you want support the old pentium optimization, you can still use these version | ||
| 30 | ; | ||
| 31 | ; this file is not optimized for old pentium, but it compatible with all x86 32 bits | ||
| 32 | ; processor (starting 80386) | ||
| 33 | ; | ||
| 34 | ; | ||
| 35 | ; see below : zlib1222add must be adjuster if you use a zlib version < 1.2.2.2 | ||
| 36 | |||
| 37 | ;uInt longest_match(s, cur_match) | ||
| 38 | ; deflate_state *s; | ||
| 39 | ; IPos cur_match; /* current match */ | ||
| 40 | |||
| 41 | NbStack equ 76 | ||
| 42 | cur_match equ dword ptr[esp+NbStack-0] | ||
| 43 | str_s equ dword ptr[esp+NbStack-4] | ||
| 44 | ; 5 dword on top (ret,ebp,esi,edi,ebx) | ||
| 45 | adrret equ dword ptr[esp+NbStack-8] | ||
| 46 | pushebp equ dword ptr[esp+NbStack-12] | ||
| 47 | pushedi equ dword ptr[esp+NbStack-16] | ||
| 48 | pushesi equ dword ptr[esp+NbStack-20] | ||
| 49 | pushebx equ dword ptr[esp+NbStack-24] | ||
| 50 | |||
| 51 | chain_length equ dword ptr [esp+NbStack-28] | ||
| 52 | limit equ dword ptr [esp+NbStack-32] | ||
| 53 | best_len equ dword ptr [esp+NbStack-36] | ||
| 54 | window equ dword ptr [esp+NbStack-40] | ||
| 55 | prev equ dword ptr [esp+NbStack-44] | ||
| 56 | scan_start equ word ptr [esp+NbStack-48] | ||
| 57 | wmask equ dword ptr [esp+NbStack-52] | ||
| 58 | match_start_ptr equ dword ptr [esp+NbStack-56] | ||
| 59 | nice_match equ dword ptr [esp+NbStack-60] | ||
| 60 | scan equ dword ptr [esp+NbStack-64] | ||
| 61 | |||
| 62 | windowlen equ dword ptr [esp+NbStack-68] | ||
| 63 | match_start equ dword ptr [esp+NbStack-72] | ||
| 64 | strend equ dword ptr [esp+NbStack-76] | ||
| 65 | NbStackAdd equ (NbStack-24) | ||
| 66 | |||
| 67 | .386p | ||
| 68 | |||
| 69 | name gvmatch | ||
| 70 | .MODEL FLAT | ||
| 71 | |||
| 72 | |||
| 73 | |||
| 74 | ; all the +zlib1222add offsets are due to the addition of fields | ||
| 75 | ; in zlib in the deflate_state structure since the asm code was first written | ||
| 76 | ; (if you compile with zlib 1.0.4 or older, use "zlib1222add equ (-4)"). | ||
| 77 | ; (if you compile with zlib between 1.0.5 and 1.2.2.1, use "zlib1222add equ 0"). | ||
| 78 | ; if you compile with zlib 1.2.2.2 or later , use "zlib1222add equ 8"). | ||
| 79 | |||
| 80 | zlib1222add equ 8 | ||
| 81 | |||
| 82 | ; Note : these value are good with a 8 bytes boundary pack structure | ||
| 83 | dep_chain_length equ 74h+zlib1222add | ||
| 84 | dep_window equ 30h+zlib1222add | ||
| 85 | dep_strstart equ 64h+zlib1222add | ||
| 86 | dep_prev_length equ 70h+zlib1222add | ||
| 87 | dep_nice_match equ 88h+zlib1222add | ||
| 88 | dep_w_size equ 24h+zlib1222add | ||
| 89 | dep_prev equ 38h+zlib1222add | ||
| 90 | dep_w_mask equ 2ch+zlib1222add | ||
| 91 | dep_good_match equ 84h+zlib1222add | ||
| 92 | dep_match_start equ 68h+zlib1222add | ||
| 93 | dep_lookahead equ 6ch+zlib1222add | ||
| 94 | |||
| 95 | |||
| 96 | _TEXT segment | ||
| 97 | |||
| 98 | IFDEF NOUNDERLINE | ||
| 99 | public longest_match | ||
| 100 | public match_init | ||
| 101 | ELSE | ||
| 102 | public _longest_match | ||
| 103 | public _match_init | ||
| 104 | ENDIF | ||
| 105 | |||
| 106 | MAX_MATCH equ 258 | ||
| 107 | MIN_MATCH equ 3 | ||
| 108 | MIN_LOOKAHEAD equ (MAX_MATCH+MIN_MATCH+1) | ||
| 109 | |||
| 110 | |||
| 111 | |||
| 112 | MAX_MATCH equ 258 | ||
| 113 | MIN_MATCH equ 3 | ||
| 114 | MIN_LOOKAHEAD equ (MAX_MATCH + MIN_MATCH + 1) | ||
| 115 | MAX_MATCH_8_ equ ((MAX_MATCH + 7) AND 0FFF0h) | ||
| 116 | |||
| 117 | |||
| 118 | ;;; stack frame offsets | ||
| 119 | |||
| 120 | chainlenwmask equ esp + 0 ; high word: current chain len | ||
| 121 | ; low word: s->wmask | ||
| 122 | window equ esp + 4 ; local copy of s->window | ||
| 123 | windowbestlen equ esp + 8 ; s->window + bestlen | ||
| 124 | scanstart equ esp + 16 ; first two bytes of string | ||
| 125 | scanend equ esp + 12 ; last two bytes of string | ||
| 126 | scanalign equ esp + 20 ; dword-misalignment of string | ||
| 127 | nicematch equ esp + 24 ; a good enough match size | ||
| 128 | bestlen equ esp + 28 ; size of best match so far | ||
| 129 | scan equ esp + 32 ; ptr to string wanting match | ||
| 130 | |||
| 131 | LocalVarsSize equ 36 | ||
| 132 | ; saved ebx byte esp + 36 | ||
| 133 | ; saved edi byte esp + 40 | ||
| 134 | ; saved esi byte esp + 44 | ||
| 135 | ; saved ebp byte esp + 48 | ||
| 136 | ; return address byte esp + 52 | ||
| 137 | deflatestate equ esp + 56 ; the function arguments | ||
| 138 | curmatch equ esp + 60 | ||
| 139 | |||
| 140 | ;;; Offsets for fields in the deflate_state structure. These numbers | ||
| 141 | ;;; are calculated from the definition of deflate_state, with the | ||
| 142 | ;;; assumption that the compiler will dword-align the fields. (Thus, | ||
| 143 | ;;; changing the definition of deflate_state could easily cause this | ||
| 144 | ;;; program to crash horribly, without so much as a warning at | ||
| 145 | ;;; compile time. Sigh.) | ||
| 146 | |||
| 147 | dsWSize equ 36+zlib1222add | ||
| 148 | dsWMask equ 44+zlib1222add | ||
| 149 | dsWindow equ 48+zlib1222add | ||
| 150 | dsPrev equ 56+zlib1222add | ||
| 151 | dsMatchLen equ 88+zlib1222add | ||
| 152 | dsPrevMatch equ 92+zlib1222add | ||
| 153 | dsStrStart equ 100+zlib1222add | ||
| 154 | dsMatchStart equ 104+zlib1222add | ||
| 155 | dsLookahead equ 108+zlib1222add | ||
| 156 | dsPrevLen equ 112+zlib1222add | ||
| 157 | dsMaxChainLen equ 116+zlib1222add | ||
| 158 | dsGoodMatch equ 132+zlib1222add | ||
| 159 | dsNiceMatch equ 136+zlib1222add | ||
| 160 | |||
| 161 | |||
| 162 | ;;; match686.asm -- Pentium-Pro-optimized version of longest_match() | ||
| 163 | ;;; Written for zlib 1.1.2 | ||
| 164 | ;;; Copyright (C) 1998 Brian Raiter <breadbox@muppetlabs.com> | ||
| 165 | ;;; You can look at http://www.muppetlabs.com/~breadbox/software/assembly.html | ||
| 166 | ;;; | ||
| 167 | ;; | ||
| 168 | ;; This software is provided 'as-is', without any express or implied | ||
| 169 | ;; warranty. In no event will the authors be held liable for any damages | ||
| 170 | ;; arising from the use of this software. | ||
| 171 | ;; | ||
| 172 | ;; Permission is granted to anyone to use this software for any purpose, | ||
| 173 | ;; including commercial applications, and to alter it and redistribute it | ||
| 174 | ;; freely, subject to the following restrictions: | ||
| 175 | ;; | ||
| 176 | ;; 1. The origin of this software must not be misrepresented; you must not | ||
| 177 | ;; claim that you wrote the original software. If you use this software | ||
| 178 | ;; in a product, an acknowledgment in the product documentation would be | ||
| 179 | ;; appreciated but is not required. | ||
| 180 | ;; 2. Altered source versions must be plainly marked as such, and must not be | ||
| 181 | ;; misrepresented as being the original software | ||
| 182 | ;; 3. This notice may not be removed or altered from any source distribution. | ||
| 183 | ;; | ||
| 184 | |||
| 185 | ;GLOBAL _longest_match, _match_init | ||
| 186 | |||
| 187 | |||
| 188 | ;SECTION .text | ||
| 189 | |||
| 190 | ;;; uInt longest_match(deflate_state *deflatestate, IPos curmatch) | ||
| 191 | |||
| 192 | ;_longest_match: | ||
| 193 | IFDEF NOUNDERLINE | ||
| 194 | longest_match proc near | ||
| 195 | ELSE | ||
| 196 | _longest_match proc near | ||
| 197 | ENDIF | ||
| 198 | .FPO (9, 4, 0, 0, 1, 0) | ||
| 199 | |||
| 200 | ;;; Save registers that the compiler may be using, and adjust esp to | ||
| 201 | ;;; make room for our stack frame. | ||
| 202 | |||
| 203 | push ebp | ||
| 204 | push edi | ||
| 205 | push esi | ||
| 206 | push ebx | ||
| 207 | sub esp, LocalVarsSize | ||
| 208 | |||
| 209 | ;;; Retrieve the function arguments. ecx will hold cur_match | ||
| 210 | ;;; throughout the entire function. edx will hold the pointer to the | ||
| 211 | ;;; deflate_state structure during the function's setup (before | ||
| 212 | ;;; entering the main loop. | ||
| 213 | |||
| 214 | mov edx, [deflatestate] | ||
| 215 | mov ecx, [curmatch] | ||
| 216 | |||
| 217 | ;;; uInt wmask = s->w_mask; | ||
| 218 | ;;; unsigned chain_length = s->max_chain_length; | ||
| 219 | ;;; if (s->prev_length >= s->good_match) { | ||
| 220 | ;;; chain_length >>= 2; | ||
| 221 | ;;; } | ||
| 222 | |||
| 223 | mov eax, [edx + dsPrevLen] | ||
| 224 | mov ebx, [edx + dsGoodMatch] | ||
| 225 | cmp eax, ebx | ||
| 226 | mov eax, [edx + dsWMask] | ||
| 227 | mov ebx, [edx + dsMaxChainLen] | ||
| 228 | jl LastMatchGood | ||
| 229 | shr ebx, 2 | ||
| 230 | LastMatchGood: | ||
| 231 | |||
| 232 | ;;; chainlen is decremented once beforehand so that the function can | ||
| 233 | ;;; use the sign flag instead of the zero flag for the exit test. | ||
| 234 | ;;; It is then shifted into the high word, to make room for the wmask | ||
| 235 | ;;; value, which it will always accompany. | ||
| 236 | |||
| 237 | dec ebx | ||
| 238 | shl ebx, 16 | ||
| 239 | or ebx, eax | ||
| 240 | mov [chainlenwmask], ebx | ||
| 241 | |||
| 242 | ;;; if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; | ||
| 243 | |||
| 244 | mov eax, [edx + dsNiceMatch] | ||
| 245 | mov ebx, [edx + dsLookahead] | ||
| 246 | cmp ebx, eax | ||
| 247 | jl LookaheadLess | ||
| 248 | mov ebx, eax | ||
| 249 | LookaheadLess: mov [nicematch], ebx | ||
| 250 | |||
| 251 | ;;; register Bytef *scan = s->window + s->strstart; | ||
| 252 | |||
| 253 | mov esi, [edx + dsWindow] | ||
| 254 | mov [window], esi | ||
| 255 | mov ebp, [edx + dsStrStart] | ||
| 256 | lea edi, [esi + ebp] | ||
| 257 | mov [scan], edi | ||
| 258 | |||
| 259 | ;;; Determine how many bytes the scan ptr is off from being | ||
| 260 | ;;; dword-aligned. | ||
| 261 | |||
| 262 | mov eax, edi | ||
| 263 | neg eax | ||
| 264 | and eax, 3 | ||
| 265 | mov [scanalign], eax | ||
| 266 | |||
| 267 | ;;; IPos limit = s->strstart > (IPos)MAX_DIST(s) ? | ||
| 268 | ;;; s->strstart - (IPos)MAX_DIST(s) : NIL; | ||
| 269 | |||
| 270 | mov eax, [edx + dsWSize] | ||
| 271 | sub eax, MIN_LOOKAHEAD | ||
| 272 | sub ebp, eax | ||
| 273 | jg LimitPositive | ||
| 274 | xor ebp, ebp | ||
| 275 | LimitPositive: | ||
| 276 | |||
| 277 | ;;; int best_len = s->prev_length; | ||
| 278 | |||
| 279 | mov eax, [edx + dsPrevLen] | ||
| 280 | mov [bestlen], eax | ||
| 281 | |||
| 282 | ;;; Store the sum of s->window + best_len in esi locally, and in esi. | ||
| 283 | |||
| 284 | add esi, eax | ||
| 285 | mov [windowbestlen], esi | ||
| 286 | |||
| 287 | ;;; register ush scan_start = *(ushf*)scan; | ||
| 288 | ;;; register ush scan_end = *(ushf*)(scan+best_len-1); | ||
| 289 | ;;; Posf *prev = s->prev; | ||
| 290 | |||
| 291 | movzx ebx, word ptr [edi] | ||
| 292 | mov [scanstart], ebx | ||
| 293 | movzx ebx, word ptr [edi + eax - 1] | ||
| 294 | mov [scanend], ebx | ||
| 295 | mov edi, [edx + dsPrev] | ||
| 296 | |||
| 297 | ;;; Jump into the main loop. | ||
| 298 | |||
| 299 | mov edx, [chainlenwmask] | ||
| 300 | jmp short LoopEntry | ||
| 301 | |||
| 302 | align 4 | ||
| 303 | |||
| 304 | ;;; do { | ||
| 305 | ;;; match = s->window + cur_match; | ||
| 306 | ;;; if (*(ushf*)(match+best_len-1) != scan_end || | ||
| 307 | ;;; *(ushf*)match != scan_start) continue; | ||
| 308 | ;;; [...] | ||
| 309 | ;;; } while ((cur_match = prev[cur_match & wmask]) > limit | ||
| 310 | ;;; && --chain_length != 0); | ||
| 311 | ;;; | ||
| 312 | ;;; Here is the inner loop of the function. The function will spend the | ||
| 313 | ;;; majority of its time in this loop, and majority of that time will | ||
| 314 | ;;; be spent in the first ten instructions. | ||
| 315 | ;;; | ||
| 316 | ;;; Within this loop: | ||
| 317 | ;;; ebx = scanend | ||
| 318 | ;;; ecx = curmatch | ||
| 319 | ;;; edx = chainlenwmask - i.e., ((chainlen << 16) | wmask) | ||
| 320 | ;;; esi = windowbestlen - i.e., (window + bestlen) | ||
| 321 | ;;; edi = prev | ||
| 322 | ;;; ebp = limit | ||
| 323 | |||
| 324 | LookupLoop: | ||
| 325 | and ecx, edx | ||
| 326 | movzx ecx, word ptr [edi + ecx*2] | ||
| 327 | cmp ecx, ebp | ||
| 328 | jbe LeaveNow | ||
| 329 | sub edx, 00010000h | ||
| 330 | js LeaveNow | ||
| 331 | LoopEntry: movzx eax, word ptr [esi + ecx - 1] | ||
| 332 | cmp eax, ebx | ||
| 333 | jnz LookupLoop | ||
| 334 | mov eax, [window] | ||
| 335 | movzx eax, word ptr [eax + ecx] | ||
| 336 | cmp eax, [scanstart] | ||
| 337 | jnz LookupLoop | ||
| 338 | |||
| 339 | ;;; Store the current value of chainlen. | ||
| 340 | |||
| 341 | mov [chainlenwmask], edx | ||
| 342 | |||
| 343 | ;;; Point edi to the string under scrutiny, and esi to the string we | ||
| 344 | ;;; are hoping to match it up with. In actuality, esi and edi are | ||
| 345 | ;;; both pointed (MAX_MATCH_8 - scanalign) bytes ahead, and edx is | ||
| 346 | ;;; initialized to -(MAX_MATCH_8 - scanalign). | ||
| 347 | |||
| 348 | mov esi, [window] | ||
| 349 | mov edi, [scan] | ||
| 350 | add esi, ecx | ||
| 351 | mov eax, [scanalign] | ||
| 352 | mov edx, 0fffffef8h; -(MAX_MATCH_8) | ||
| 353 | lea edi, [edi + eax + 0108h] ;MAX_MATCH_8] | ||
| 354 | lea esi, [esi + eax + 0108h] ;MAX_MATCH_8] | ||
| 355 | |||
| 356 | ;;; Test the strings for equality, 8 bytes at a time. At the end, | ||
| 357 | ;;; adjust edx so that it is offset to the exact byte that mismatched. | ||
| 358 | ;;; | ||
| 359 | ;;; We already know at this point that the first three bytes of the | ||
| 360 | ;;; strings match each other, and they can be safely passed over before | ||
| 361 | ;;; starting the compare loop. So what this code does is skip over 0-3 | ||
| 362 | ;;; bytes, as much as necessary in order to dword-align the edi | ||
| 363 | ;;; pointer. (esi will still be misaligned three times out of four.) | ||
| 364 | ;;; | ||
| 365 | ;;; It should be confessed that this loop usually does not represent | ||
| 366 | ;;; much of the total running time. Replacing it with a more | ||
| 367 | ;;; straightforward "rep cmpsb" would not drastically degrade | ||
| 368 | ;;; performance. | ||
| 369 | |||
| 370 | LoopCmps: | ||
| 371 | mov eax, [esi + edx] | ||
| 372 | xor eax, [edi + edx] | ||
| 373 | jnz LeaveLoopCmps | ||
| 374 | mov eax, [esi + edx + 4] | ||
| 375 | xor eax, [edi + edx + 4] | ||
| 376 | jnz LeaveLoopCmps4 | ||
| 377 | add edx, 8 | ||
| 378 | jnz LoopCmps | ||
| 379 | jmp short LenMaximum | ||
| 380 | LeaveLoopCmps4: add edx, 4 | ||
| 381 | LeaveLoopCmps: test eax, 0000FFFFh | ||
| 382 | jnz LenLower | ||
| 383 | add edx, 2 | ||
| 384 | shr eax, 16 | ||
| 385 | LenLower: sub al, 1 | ||
| 386 | adc edx, 0 | ||
| 387 | |||
| 388 | ;;; Calculate the length of the match. If it is longer than MAX_MATCH, | ||
| 389 | ;;; then automatically accept it as the best possible match and leave. | ||
| 390 | |||
| 391 | lea eax, [edi + edx] | ||
| 392 | mov edi, [scan] | ||
| 393 | sub eax, edi | ||
| 394 | cmp eax, MAX_MATCH | ||
| 395 | jge LenMaximum | ||
| 396 | |||
| 397 | ;;; If the length of the match is not longer than the best match we | ||
| 398 | ;;; have so far, then forget it and return to the lookup loop. | ||
| 399 | |||
| 400 | mov edx, [deflatestate] | ||
| 401 | mov ebx, [bestlen] | ||
| 402 | cmp eax, ebx | ||
| 403 | jg LongerMatch | ||
| 404 | mov esi, [windowbestlen] | ||
| 405 | mov edi, [edx + dsPrev] | ||
| 406 | mov ebx, [scanend] | ||
| 407 | mov edx, [chainlenwmask] | ||
| 408 | jmp LookupLoop | ||
| 409 | |||
| 410 | ;;; s->match_start = cur_match; | ||
| 411 | ;;; best_len = len; | ||
| 412 | ;;; if (len >= nice_match) break; | ||
| 413 | ;;; scan_end = *(ushf*)(scan+best_len-1); | ||
| 414 | |||
| 415 | LongerMatch: mov ebx, [nicematch] | ||
| 416 | mov [bestlen], eax | ||
| 417 | mov [edx + dsMatchStart], ecx | ||
| 418 | cmp eax, ebx | ||
| 419 | jge LeaveNow | ||
| 420 | mov esi, [window] | ||
| 421 | add esi, eax | ||
| 422 | mov [windowbestlen], esi | ||
| 423 | movzx ebx, word ptr [edi + eax - 1] | ||
| 424 | mov edi, [edx + dsPrev] | ||
| 425 | mov [scanend], ebx | ||
| 426 | mov edx, [chainlenwmask] | ||
| 427 | jmp LookupLoop | ||
| 428 | |||
| 429 | ;;; Accept the current string, with the maximum possible length. | ||
| 430 | |||
| 431 | LenMaximum: mov edx, [deflatestate] | ||
| 432 | mov dword ptr [bestlen], MAX_MATCH | ||
| 433 | mov [edx + dsMatchStart], ecx | ||
| 434 | |||
| 435 | ;;; if ((uInt)best_len <= s->lookahead) return (uInt)best_len; | ||
| 436 | ;;; return s->lookahead; | ||
| 437 | |||
| 438 | LeaveNow: | ||
| 439 | mov edx, [deflatestate] | ||
| 440 | mov ebx, [bestlen] | ||
| 441 | mov eax, [edx + dsLookahead] | ||
| 442 | cmp ebx, eax | ||
| 443 | jg LookaheadRet | ||
| 444 | mov eax, ebx | ||
| 445 | LookaheadRet: | ||
| 446 | |||
| 447 | ;;; Restore the stack and return from whence we came. | ||
| 448 | |||
| 449 | add esp, LocalVarsSize | ||
| 450 | pop ebx | ||
| 451 | pop esi | ||
| 452 | pop edi | ||
| 453 | pop ebp | ||
| 454 | |||
| 455 | ret | ||
| 456 | ; please don't remove this string ! | ||
| 457 | ; Your can freely use match686 in any free or commercial app if you don't remove the string in the binary! | ||
| 458 | db 0dh,0ah,"asm686 with masm, optimised assembly code from Brian Raiter, written 1998",0dh,0ah | ||
| 459 | |||
| 460 | |||
| 461 | IFDEF NOUNDERLINE | ||
| 462 | longest_match endp | ||
| 463 | ELSE | ||
| 464 | _longest_match endp | ||
| 465 | ENDIF | ||
| 466 | |||
| 467 | IFDEF NOUNDERLINE | ||
| 468 | match_init proc near | ||
| 469 | ret | ||
| 470 | match_init endp | ||
| 471 | ELSE | ||
| 472 | _match_init proc near | ||
| 473 | ret | ||
| 474 | _match_init endp | ||
| 475 | ENDIF | ||
| 476 | |||
| 477 | |||
| 478 | _TEXT ends | ||
| 479 | end | ||
diff --git a/contrib/masmx86/readme.txt b/contrib/masmx86/readme.txt deleted file mode 100644 index 3271f72..0000000 --- a/contrib/masmx86/readme.txt +++ /dev/null | |||
| @@ -1,27 +0,0 @@ | |||
| 1 | |||
| 2 | Summary | ||
| 3 | ------- | ||
| 4 | This directory contains ASM implementations of the functions | ||
| 5 | longest_match() and inflate_fast(). | ||
| 6 | |||
| 7 | |||
| 8 | Use instructions | ||
| 9 | ---------------- | ||
| 10 | Assemble using MASM, and copy the object files into the zlib source | ||
| 11 | directory, then run the appropriate makefile, as suggested below. You can | ||
| 12 | donwload MASM from here: | ||
| 13 | |||
| 14 | http://www.microsoft.com/downloads/details.aspx?displaylang=en&FamilyID=7a1c9da0-0510-44a2-b042-7ef370530c64 | ||
| 15 | |||
| 16 | You can also get objects files here: | ||
| 17 | |||
| 18 | http://www.winimage.com/zLibDll/zlib124_masm_obj.zip | ||
| 19 | |||
| 20 | Build instructions | ||
| 21 | ------------------ | ||
| 22 | * With Microsoft C and MASM: | ||
| 23 | nmake -f win32/Makefile.msc LOC="-DASMV -DASMINF" OBJA="match686.obj inffas32.obj" | ||
| 24 | |||
| 25 | * With Borland C and TASM: | ||
| 26 | make -f win32/Makefile.bor LOCAL_ZLIB="-DASMV -DASMINF" OBJA="match686.obj inffas32.obj" OBJPA="+match686c.obj+match686.obj+inffas32.obj" | ||
| 27 | |||
diff --git a/win32/Makefile.bor b/win32/Makefile.bor index d152bbb..4495353 100644 --- a/win32/Makefile.bor +++ b/win32/Makefile.bor | |||
| @@ -3,7 +3,6 @@ | |||
| 3 | # | 3 | # |
| 4 | # Usage: | 4 | # Usage: |
| 5 | # make -f win32/Makefile.bor | 5 | # make -f win32/Makefile.bor |
| 6 | # make -f win32/Makefile.bor LOCAL_ZLIB=-DASMV OBJA=match.obj OBJPA=+match.obj | ||
| 7 | 6 | ||
| 8 | # ------------ Borland C++ ------------ | 7 | # ------------ Borland C++ ------------ |
| 9 | 8 | ||
diff --git a/win32/Makefile.gcc b/win32/Makefile.gcc index 305be50..081e391 100644 --- a/win32/Makefile.gcc +++ b/win32/Makefile.gcc | |||
| @@ -11,10 +11,6 @@ | |||
| 11 | # | 11 | # |
| 12 | # make -fwin32/Makefile.gcc; make test testdll -fwin32/Makefile.gcc | 12 | # make -fwin32/Makefile.gcc; make test testdll -fwin32/Makefile.gcc |
| 13 | # | 13 | # |
| 14 | # To use the asm code, type: | ||
| 15 | # cp contrib/asm?86/match.S ./match.S | ||
| 16 | # make LOC=-DASMV OBJA=match.o -fwin32/Makefile.gcc | ||
| 17 | # | ||
| 18 | # To install libz.a, zconf.h and zlib.h in the system directories, type: | 14 | # To install libz.a, zconf.h and zlib.h in the system directories, type: |
| 19 | # | 15 | # |
| 20 | # make install -fwin32/Makefile.gcc | 16 | # make install -fwin32/Makefile.gcc |
| @@ -38,7 +34,6 @@ IMPLIB = libz.dll.a | |||
| 38 | # | 34 | # |
| 39 | SHARED_MODE=0 | 35 | SHARED_MODE=0 |
| 40 | 36 | ||
| 41 | #LOC = -DASMV | ||
| 42 | #LOC = -DZLIB_DEBUG -g | 37 | #LOC = -DZLIB_DEBUG -g |
| 43 | 38 | ||
| 44 | PREFIX = | 39 | PREFIX = |
diff --git a/win32/Makefile.msc b/win32/Makefile.msc index 6831882..9c65153 100644 --- a/win32/Makefile.msc +++ b/win32/Makefile.msc | |||
| @@ -4,10 +4,6 @@ | |||
| 4 | # Usage: | 4 | # Usage: |
| 5 | # nmake -f win32/Makefile.msc (standard build) | 5 | # nmake -f win32/Makefile.msc (standard build) |
| 6 | # nmake -f win32/Makefile.msc LOC=-DFOO (nonstandard build) | 6 | # nmake -f win32/Makefile.msc LOC=-DFOO (nonstandard build) |
| 7 | # nmake -f win32/Makefile.msc LOC="-DASMV -DASMINF" \ | ||
| 8 | # OBJA="inffas32.obj match686.obj" (use ASM code, x86) | ||
| 9 | # nmake -f win32/Makefile.msc AS=ml64 LOC="-DASMV -DASMINF -I." \ | ||
| 10 | # OBJA="inffasx64.obj gvmat64.obj inffas8664.obj" (use ASM code, x64) | ||
| 11 | 7 | ||
| 12 | # The toplevel directory of the source tree. | 8 | # The toplevel directory of the source tree. |
| 13 | # | 9 | # |
