diff options
Diffstat (limited to 'contrib/asm586')
| -rw-r--r-- | contrib/asm586/README.586 | 43 | ||||
| -rw-r--r-- | contrib/asm586/match.S | 354 |
2 files changed, 397 insertions, 0 deletions
diff --git a/contrib/asm586/README.586 b/contrib/asm586/README.586 new file mode 100644 index 0000000..6bb78f3 --- /dev/null +++ b/contrib/asm586/README.586 | |||
| @@ -0,0 +1,43 @@ | |||
| 1 | This is a patched version of zlib modified to use | ||
| 2 | Pentium-optimized assembly code in the deflation algorithm. The files | ||
| 3 | changed/added by this patch are: | ||
| 4 | |||
| 5 | README.586 | ||
| 6 | match.S | ||
| 7 | |||
| 8 | The effectiveness of these modifications is a bit marginal, as the the | ||
| 9 | program's bottleneck seems to be mostly L1-cache contention, for which | ||
| 10 | there is no real way to work around without rewriting the basic | ||
| 11 | algorithm. The speedup on average is around 5-10% (which is generally | ||
| 12 | less than the amount of variance between subsequent executions). | ||
| 13 | However, when used at level 9 compression, the cache contention can | ||
| 14 | drop enough for the assembly version to achieve 10-20% speedup (and | ||
| 15 | sometimes more, depending on the amount of overall redundancy in the | ||
| 16 | files). Even here, though, cache contention can still be the limiting | ||
| 17 | factor, depending on the nature of the program using the zlib library. | ||
| 18 | This may also mean that better improvements will be seen on a Pentium | ||
| 19 | with MMX, which suffers much less from L1-cache contention, but I have | ||
| 20 | not yet verified this. | ||
| 21 | |||
| 22 | Note that this code has been tailored for the Pentium in particular, | ||
| 23 | and will not perform well on the Pentium Pro (due to the use of a | ||
| 24 | partial register in the inner loop). | ||
| 25 | |||
| 26 | If you are using an assembler other than GNU as, you will have to | ||
| 27 | translate match.S to use your assembler's syntax. (Have fun.) | ||
| 28 | |||
| 29 | Brian Raiter | ||
| 30 | breadbox@muppetlabs.com | ||
| 31 | April, 1998 | ||
| 32 | |||
| 33 | |||
| 34 | Added for zlib 1.1.3: | ||
| 35 | |||
| 36 | The patches come from | ||
| 37 | http://www.muppetlabs.com/~breadbox/software/assembly.html | ||
| 38 | |||
| 39 | To compile zlib with this asm file, copy match.S to the zlib directory | ||
| 40 | then do: | ||
| 41 | |||
| 42 | CFLAGS="-O3 -DASMV" ./configure | ||
| 43 | make OBJA=match.o | ||
diff --git a/contrib/asm586/match.S b/contrib/asm586/match.S new file mode 100644 index 0000000..8f16140 --- /dev/null +++ b/contrib/asm586/match.S | |||
| @@ -0,0 +1,354 @@ | |||
| 1 | /* match.s -- Pentium-optimized version of longest_match() | ||
| 2 | * Written for zlib 1.1.2 | ||
| 3 | * Copyright (C) 1998 Brian Raiter <breadbox@muppetlabs.com> | ||
| 4 | * | ||
| 5 | * This is free software; you can redistribute it and/or modify it | ||
| 6 | * under the terms of the GNU General Public License. | ||
| 7 | */ | ||
| 8 | |||
| 9 | #ifndef NO_UNDERLINE | ||
| 10 | #define match_init _match_init | ||
| 11 | #define longest_match _longest_match | ||
| 12 | #endif | ||
| 13 | |||
| 14 | #define MAX_MATCH (258) | ||
| 15 | #define MIN_MATCH (3) | ||
| 16 | #define MIN_LOOKAHEAD (MAX_MATCH + MIN_MATCH + 1) | ||
| 17 | #define MAX_MATCH_8 ((MAX_MATCH + 7) & ~7) | ||
| 18 | |||
| 19 | /* stack frame offsets */ | ||
| 20 | |||
| 21 | #define wmask 0 /* local copy of s->wmask */ | ||
| 22 | #define window 4 /* local copy of s->window */ | ||
| 23 | #define windowbestlen 8 /* s->window + bestlen */ | ||
| 24 | #define chainlenscanend 12 /* high word: current chain len */ | ||
| 25 | /* low word: last bytes sought */ | ||
| 26 | #define scanstart 16 /* first two bytes of string */ | ||
| 27 | #define scanalign 20 /* dword-misalignment of string */ | ||
| 28 | #define nicematch 24 /* a good enough match size */ | ||
| 29 | #define bestlen 28 /* size of best match so far */ | ||
| 30 | #define scan 32 /* ptr to string wanting match */ | ||
| 31 | |||
| 32 | #define LocalVarsSize (36) | ||
| 33 | /* saved ebx 36 */ | ||
| 34 | /* saved edi 40 */ | ||
| 35 | /* saved esi 44 */ | ||
| 36 | /* saved ebp 48 */ | ||
| 37 | /* return address 52 */ | ||
| 38 | #define deflatestate 56 /* the function arguments */ | ||
| 39 | #define curmatch 60 | ||
| 40 | |||
| 41 | /* Offsets for fields in the deflate_state structure. These numbers | ||
| 42 | * are calculated from the definition of deflate_state, with the | ||
| 43 | * assumption that the compiler will dword-align the fields. (Thus, | ||
| 44 | * changing the definition of deflate_state could easily cause this | ||
| 45 | * program to crash horribly, without so much as a warning at | ||
| 46 | * compile time. Sigh.) | ||
| 47 | */ | ||
| 48 | #define dsWSize 36 | ||
| 49 | #define dsWMask 44 | ||
| 50 | #define dsWindow 48 | ||
| 51 | #define dsPrev 56 | ||
| 52 | #define dsMatchLen 88 | ||
| 53 | #define dsPrevMatch 92 | ||
| 54 | #define dsStrStart 100 | ||
| 55 | #define dsMatchStart 104 | ||
| 56 | #define dsLookahead 108 | ||
| 57 | #define dsPrevLen 112 | ||
| 58 | #define dsMaxChainLen 116 | ||
| 59 | #define dsGoodMatch 132 | ||
| 60 | #define dsNiceMatch 136 | ||
| 61 | |||
| 62 | |||
| 63 | .file "match.S" | ||
| 64 | |||
| 65 | .globl match_init, longest_match | ||
| 66 | |||
| 67 | .text | ||
| 68 | |||
| 69 | /* uInt longest_match(deflate_state *deflatestate, IPos curmatch) */ | ||
| 70 | |||
| 71 | longest_match: | ||
| 72 | |||
| 73 | /* Save registers that the compiler may be using, and adjust %esp to */ | ||
| 74 | /* make room for our stack frame. */ | ||
| 75 | |||
| 76 | pushl %ebp | ||
| 77 | pushl %edi | ||
| 78 | pushl %esi | ||
| 79 | pushl %ebx | ||
| 80 | subl $LocalVarsSize, %esp | ||
| 81 | |||
| 82 | /* Retrieve the function arguments. %ecx will hold cur_match */ | ||
| 83 | /* throughout the entire function. %edx will hold the pointer to the */ | ||
| 84 | /* deflate_state structure during the function's setup (before */ | ||
| 85 | /* entering the main loop). */ | ||
| 86 | |||
| 87 | movl deflatestate(%esp), %edx | ||
| 88 | movl curmatch(%esp), %ecx | ||
| 89 | |||
| 90 | /* if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; */ | ||
| 91 | |||
| 92 | movl dsNiceMatch(%edx), %eax | ||
| 93 | movl dsLookahead(%edx), %ebx | ||
| 94 | cmpl %eax, %ebx | ||
| 95 | jl LookaheadLess | ||
| 96 | movl %eax, %ebx | ||
| 97 | LookaheadLess: movl %ebx, nicematch(%esp) | ||
| 98 | |||
| 99 | /* register Bytef *scan = s->window + s->strstart; */ | ||
| 100 | |||
| 101 | movl dsWindow(%edx), %esi | ||
| 102 | movl %esi, window(%esp) | ||
| 103 | movl dsStrStart(%edx), %ebp | ||
| 104 | lea (%esi,%ebp), %edi | ||
| 105 | movl %edi, scan(%esp) | ||
| 106 | |||
| 107 | /* Determine how many bytes the scan ptr is off from being */ | ||
| 108 | /* dword-aligned. */ | ||
| 109 | |||
| 110 | movl %edi, %eax | ||
| 111 | negl %eax | ||
| 112 | andl $3, %eax | ||
| 113 | movl %eax, scanalign(%esp) | ||
| 114 | |||
| 115 | /* IPos limit = s->strstart > (IPos)MAX_DIST(s) ? */ | ||
| 116 | /* s->strstart - (IPos)MAX_DIST(s) : NIL; */ | ||
| 117 | |||
| 118 | movl dsWSize(%edx), %eax | ||
| 119 | subl $MIN_LOOKAHEAD, %eax | ||
| 120 | subl %eax, %ebp | ||
| 121 | jg LimitPositive | ||
| 122 | xorl %ebp, %ebp | ||
| 123 | LimitPositive: | ||
| 124 | |||
| 125 | /* unsigned chain_length = s->max_chain_length; */ | ||
| 126 | /* if (s->prev_length >= s->good_match) { */ | ||
| 127 | /* chain_length >>= 2; */ | ||
| 128 | /* } */ | ||
| 129 | |||
| 130 | movl dsPrevLen(%edx), %eax | ||
| 131 | movl dsGoodMatch(%edx), %ebx | ||
| 132 | cmpl %ebx, %eax | ||
| 133 | movl dsMaxChainLen(%edx), %ebx | ||
| 134 | jl LastMatchGood | ||
| 135 | shrl $2, %ebx | ||
| 136 | LastMatchGood: | ||
| 137 | |||
| 138 | /* chainlen is decremented once beforehand so that the function can */ | ||
| 139 | /* use the sign flag instead of the zero flag for the exit test. */ | ||
| 140 | /* It is then shifted into the high word, to make room for the scanend */ | ||
| 141 | /* scanend value, which it will always accompany. */ | ||
| 142 | |||
| 143 | decl %ebx | ||
| 144 | shll $16, %ebx | ||
| 145 | |||
| 146 | /* int best_len = s->prev_length; */ | ||
| 147 | |||
| 148 | movl dsPrevLen(%edx), %eax | ||
| 149 | movl %eax, bestlen(%esp) | ||
| 150 | |||
| 151 | /* Store the sum of s->window + best_len in %esi locally, and in %esi. */ | ||
| 152 | |||
| 153 | addl %eax, %esi | ||
| 154 | movl %esi, windowbestlen(%esp) | ||
| 155 | |||
| 156 | /* register ush scan_start = *(ushf*)scan; */ | ||
| 157 | /* register ush scan_end = *(ushf*)(scan+best_len-1); */ | ||
| 158 | |||
| 159 | movw (%edi), %bx | ||
| 160 | movw %bx, scanstart(%esp) | ||
| 161 | movw -1(%edi,%eax), %bx | ||
| 162 | movl %ebx, chainlenscanend(%esp) | ||
| 163 | |||
| 164 | /* Posf *prev = s->prev; */ | ||
| 165 | /* uInt wmask = s->w_mask; */ | ||
| 166 | |||
| 167 | movl dsPrev(%edx), %edi | ||
| 168 | movl dsWMask(%edx), %edx | ||
| 169 | mov %edx, wmask(%esp) | ||
| 170 | |||
| 171 | /* Jump into the main loop. */ | ||
| 172 | |||
| 173 | jmp LoopEntry | ||
| 174 | |||
| 175 | .balign 16 | ||
| 176 | |||
| 177 | /* do { | ||
| 178 | * match = s->window + cur_match; | ||
| 179 | * if (*(ushf*)(match+best_len-1) != scan_end || | ||
| 180 | * *(ushf*)match != scan_start) continue; | ||
| 181 | * [...] | ||
| 182 | * } while ((cur_match = prev[cur_match & wmask]) > limit | ||
| 183 | * && --chain_length != 0); | ||
| 184 | * | ||
| 185 | * Here is the inner loop of the function. The function will spend the | ||
| 186 | * majority of its time in this loop, and majority of that time will | ||
| 187 | * be spent in the first ten instructions. | ||
| 188 | * | ||
| 189 | * Within this loop: | ||
| 190 | * %ebx = chainlenscanend - i.e., ((chainlen << 16) | scanend) | ||
| 191 | * %ecx = curmatch | ||
| 192 | * %edx = curmatch & wmask | ||
| 193 | * %esi = windowbestlen - i.e., (window + bestlen) | ||
| 194 | * %edi = prev | ||
| 195 | * %ebp = limit | ||
| 196 | * | ||
| 197 | * Two optimization notes on the choice of instructions: | ||
| 198 | * | ||
| 199 | * The first instruction uses a 16-bit address, which costs an extra, | ||
| 200 | * unpairable cycle. This is cheaper than doing a 32-bit access and | ||
| 201 | * zeroing the high word, due to the 3-cycle misalignment penalty which | ||
| 202 | * would occur half the time. This also turns out to be cheaper than | ||
| 203 | * doing two separate 8-bit accesses, as the memory is so rarely in the | ||
| 204 | * L1 cache. | ||
| 205 | * | ||
| 206 | * The window buffer, however, apparently spends a lot of time in the | ||
| 207 | * cache, and so it is faster to retrieve the word at the end of the | ||
| 208 | * match string with two 8-bit loads. The instructions that test the | ||
| 209 | * word at the beginning of the match string, however, are executed | ||
| 210 | * much less frequently, and there it was cheaper to use 16-bit | ||
| 211 | * instructions, which avoided the necessity of saving off and | ||
| 212 | * subsequently reloading one of the other registers. | ||
| 213 | */ | ||
| 214 | LookupLoop: | ||
| 215 | /* 1 U & V */ | ||
| 216 | movw (%edi,%edx,2), %cx /* 2 U pipe */ | ||
| 217 | movl wmask(%esp), %edx /* 2 V pipe */ | ||
| 218 | cmpl %ebp, %ecx /* 3 U pipe */ | ||
| 219 | jbe LeaveNow /* 3 V pipe */ | ||
| 220 | subl $0x00010000, %ebx /* 4 U pipe */ | ||
| 221 | js LeaveNow /* 4 V pipe */ | ||
| 222 | LoopEntry: movb -1(%esi,%ecx), %al /* 5 U pipe */ | ||
| 223 | andl %ecx, %edx /* 5 V pipe */ | ||
| 224 | cmpb %bl, %al /* 6 U pipe */ | ||
| 225 | jnz LookupLoop /* 6 V pipe */ | ||
| 226 | movb (%esi,%ecx), %ah | ||
| 227 | cmpb %bh, %ah | ||
| 228 | jnz LookupLoop | ||
| 229 | movl window(%esp), %eax | ||
| 230 | movw (%eax,%ecx), %ax | ||
| 231 | cmpw scanstart(%esp), %ax | ||
| 232 | jnz LookupLoop | ||
| 233 | |||
| 234 | /* Store the current value of chainlen. */ | ||
| 235 | |||
| 236 | movl %ebx, chainlenscanend(%esp) | ||
| 237 | |||
| 238 | /* Point %edi to the string under scrutiny, and %esi to the string we */ | ||
| 239 | /* are hoping to match it up with. In actuality, %esi and %edi are */ | ||
| 240 | /* both pointed (MAX_MATCH_8 - scanalign) bytes ahead, and %edx is */ | ||
| 241 | /* initialized to -(MAX_MATCH_8 - scanalign). */ | ||
| 242 | |||
| 243 | movl window(%esp), %esi | ||
| 244 | movl scan(%esp), %edi | ||
| 245 | addl %ecx, %esi | ||
| 246 | movl scanalign(%esp), %eax | ||
| 247 | movl $(-MAX_MATCH_8), %edx | ||
| 248 | lea MAX_MATCH_8(%edi,%eax), %edi | ||
| 249 | lea MAX_MATCH_8(%esi,%eax), %esi | ||
| 250 | |||
| 251 | /* Test the strings for equality, 8 bytes at a time. At the end, | ||
| 252 | * adjust %edx so that it is offset to the exact byte that mismatched. | ||
| 253 | * | ||
| 254 | * We already know at this point that the first three bytes of the | ||
| 255 | * strings match each other, and they can be safely passed over before | ||
| 256 | * starting the compare loop. So what this code does is skip over 0-3 | ||
| 257 | * bytes, as much as necessary in order to dword-align the %edi | ||
| 258 | * pointer. (%esi will still be misaligned three times out of four.) | ||
| 259 | * | ||
| 260 | * It should be confessed that this loop usually does not represent | ||
| 261 | * much of the total running time. Replacing it with a more | ||
| 262 | * straightforward "rep cmpsb" would not drastically degrade | ||
| 263 | * performance. | ||
| 264 | */ | ||
| 265 | LoopCmps: | ||
| 266 | movl (%esi,%edx), %eax | ||
| 267 | movl (%edi,%edx), %ebx | ||
| 268 | xorl %ebx, %eax | ||
| 269 | jnz LeaveLoopCmps | ||
| 270 | movl 4(%esi,%edx), %eax | ||
| 271 | movl 4(%edi,%edx), %ebx | ||
| 272 | xorl %ebx, %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 chainlenscanend(%esp), %ebx | ||
| 302 | movl windowbestlen(%esp), %esi | ||
| 303 | movl dsPrev(%edx), %edi | ||
| 304 | movl wmask(%esp), %edx | ||
| 305 | andl %ecx, %edx | ||
| 306 | jmp LookupLoop | ||
| 307 | |||
| 308 | /* s->match_start = cur_match; */ | ||
| 309 | /* best_len = len; */ | ||
| 310 | /* if (len >= nice_match) break; */ | ||
| 311 | /* scan_end = *(ushf*)(scan+best_len-1); */ | ||
| 312 | |||
| 313 | LongerMatch: movl nicematch(%esp), %ebx | ||
| 314 | movl %eax, bestlen(%esp) | ||
| 315 | movl %ecx, dsMatchStart(%edx) | ||
| 316 | cmpl %ebx, %eax | ||
| 317 | jge LeaveNow | ||
| 318 | movl window(%esp), %esi | ||
| 319 | addl %eax, %esi | ||
| 320 | movl %esi, windowbestlen(%esp) | ||
| 321 | movl chainlenscanend(%esp), %ebx | ||
| 322 | movw -1(%edi,%eax), %bx | ||
| 323 | movl dsPrev(%edx), %edi | ||
| 324 | movl %ebx, chainlenscanend(%esp) | ||
| 325 | movl wmask(%esp), %edx | ||
| 326 | andl %ecx, %edx | ||
| 327 | jmp LookupLoop | ||
| 328 | |||
| 329 | /* Accept the current string, with the maximum possible length. */ | ||
| 330 | |||
| 331 | LenMaximum: movl deflatestate(%esp), %edx | ||
| 332 | movl $MAX_MATCH, bestlen(%esp) | ||
| 333 | movl %ecx, dsMatchStart(%edx) | ||
| 334 | |||
| 335 | /* if ((uInt)best_len <= s->lookahead) return (uInt)best_len; */ | ||
| 336 | /* return s->lookahead; */ | ||
| 337 | |||
| 338 | LeaveNow: | ||
| 339 | movl deflatestate(%esp), %edx | ||
| 340 | movl bestlen(%esp), %ebx | ||
| 341 | movl dsLookahead(%edx), %eax | ||
| 342 | cmpl %eax, %ebx | ||
| 343 | jg LookaheadRet | ||
| 344 | movl %ebx, %eax | ||
| 345 | LookaheadRet: | ||
| 346 | |||
| 347 | /* Restore the stack and return from whence we came. */ | ||
| 348 | |||
| 349 | addl $LocalVarsSize, %esp | ||
| 350 | popl %ebx | ||
| 351 | popl %esi | ||
| 352 | popl %edi | ||
| 353 | popl %ebp | ||
| 354 | match_init: ret | ||
