aboutsummaryrefslogtreecommitdiff
path: root/contrib/amd64
diff options
context:
space:
mode:
authorMark Adler <zlib@madler.net>2017-10-12 20:08:53 -0700
committerMark Adler <zlib@madler.net>2017-10-12 20:27:14 -0700
commit288f1080317b954b6bdca33708631c011549c008 (patch)
tree9629f01104722ba8e490f04a0790c56513ba989a /contrib/amd64
parenta5773513942b1c57d0eff51fcb2ebac72796ed95 (diff)
downloadzlib-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.
Diffstat (limited to 'contrib/amd64')
-rw-r--r--contrib/amd64/amd64-match.S452
1 files changed, 0 insertions, 452 deletions
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
70void print_depl()
71{
72deflate_state ds;
73deflate_state *s=&ds;
74printf("size pointer=%u\n",(int)sizeof(void*));
75
76printf("#define dsWSize (%3u)(%%rdi)\n",(int)(((char*)&(s->w_size))-((char*)s)));
77printf("#define dsWMask (%3u)(%%rdi)\n",(int)(((char*)&(s->w_mask))-((char*)s)));
78printf("#define dsWindow (%3u)(%%rdi)\n",(int)(((char*)&(s->window))-((char*)s)));
79printf("#define dsPrev (%3u)(%%rdi)\n",(int)(((char*)&(s->prev))-((char*)s)));
80printf("#define dsMatchLen (%3u)(%%rdi)\n",(int)(((char*)&(s->match_length))-((char*)s)));
81printf("#define dsPrevMatch (%3u)(%%rdi)\n",(int)(((char*)&(s->prev_match))-((char*)s)));
82printf("#define dsStrStart (%3u)(%%rdi)\n",(int)(((char*)&(s->strstart))-((char*)s)));
83printf("#define dsMatchStart (%3u)(%%rdi)\n",(int)(((char*)&(s->match_start))-((char*)s)));
84printf("#define dsLookahead (%3u)(%%rdi)\n",(int)(((char*)&(s->lookahead))-((char*)s)));
85printf("#define dsPrevLen (%3u)(%%rdi)\n",(int)(((char*)&(s->prev_length))-((char*)s)));
86printf("#define dsMaxChainLen (%3u)(%%rdi)\n",(int)(((char*)&(s->max_chain_length))-((char*)s)));
87printf("#define dsGoodMatch (%3u)(%%rdi)\n",(int)(((char*)&(s->good_match))-((char*)s)));
88printf("#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
145longest_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
171LastMatchGood:
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
189LookaheadLess: 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 */
249LookupLoop:
250 andl %chainlenwmask, %curmatchd
251 movzwl (%prev, %curmatch, 2), %curmatchd
252 cmpl %limitd, %curmatchd
253 jbe LeaveNow
254 subl $0x00010000, %chainlenwmask
255 js LeaveNow
256LoopEntry: 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
290LoopCmps:
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
345LeaveLoopCmps: 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
368LeaveLoopCmps: 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
372LeaveLoopCmps16:
373 add $8, %rdx
374LeaveLoopCmps8:
375 add $8, %rdx
376LeaveLoopCmps: testl $0xFFFFFFFF, %eax /* Check the first 4 bytes */
377 jnz Check16
378 add $4, %rdx
379 shr $32, %rax
380Check16: testw $0xFFFF, %ax
381 jnz LenLower
382 add $2, %rdx
383 shrl $16, %eax
384LenLower: 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
412LongerMatch:
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
428LenMaximum:
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
435LeaveNow:
436 movl dsLookahead, %eax
437 cmpl %eax, %bestlend
438 cmovngl %bestlend, %eax
439LookaheadRet:
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
452match_init: ret
generated by cgit v1.2.3-55-g6feb (git 2.39.1) at 2025-03-25 10:06:25 +0000