Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(1184)

Side by Side Diff: third_party/zlib/fill_window_sse.c

Issue 552123005: Integrate SIMD optimisations for zlib (Closed) Base URL: https://chromium.googlesource.com/chromium/src.git@master
Patch Set: Fix issues in fallback (non-SIMD) code Created 6 years, 3 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch
OLDNEW
(Empty)
1 /*
2 * Fill Window with SSE2-optimized hash shifting
3 *
4 * Copyright (C) 2013 Intel Corporation
5 * Authors:
6 * Arjan van de Ven <arjan@linux.intel.com>
7 * Jim Kukunas <james.t.kukunas@linux.intel.com>
8 *
9 * For conditions of distribution and use, see copyright notice in zlib.h
10 */
11 #ifdef HAVE_SSE2
12
13 #include <immintrin.h>
14 #include "deflate.h"
15
16 extern int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
17
18 void fill_window_sse(deflate_state *s)
19 {
20 const __m128i xmm_wsize = _mm_set1_epi16(s->w_size);
21
22 register unsigned n;
23 register Posf *p;
24 unsigned more; /* Amount of free space at the end of the window. */
25 uInt wsize = s->w_size;
26
27 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
28
29 do {
30 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
31
32 /* Deal with !@#$% 64K limit: */
33 if (sizeof(int) <= 2) {
34 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
35 more = wsize;
36
37 } else if (more == (unsigned)(-1)) {
38 /* Very unlikely, but possible on 16 bit machine if
39 * strstart == 0 && lookahead == 1 (input done a byte at time)
40 */
41 more--;
42 }
43 }
44
45 /* If the window is almost full and there is insufficient lookahead,
46 * move the upper half to the lower one to make room in the upper half.
47 */
48 if (s->strstart >= wsize+MAX_DIST(s)) {
49
50 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
51 s->match_start -= wsize;
52 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
53 s->block_start -= (long) wsize;
54
55 /* Slide the hash table (could be avoided with 32 bit values
56 at the expense of memory usage). We slide even when level == 0
57 to keep the hash table consistent if we switch back to level > 0
58 later. (Using level 0 permanently is not an optimal usage of
59 zlib, so we don't care about this pathological case.)
60 */
61 n = s->hash_size;
62 p = &s->head[n];
63 p -= 8;
64 do {
65 __m128i value, result;
66
67 value = _mm_loadu_si128((__m128i *)p);
68 result = _mm_subs_epu16(value, xmm_wsize);
69 _mm_storeu_si128((__m128i *)p, result);
70
71 p -= 8;
72 n -= 8;
73 } while (n > 0);
74
75 n = wsize;
76 #ifndef FASTEST
77 p = &s->prev[n];
78 p -= 8;
79 do {
80 __m128i value, result;
81
82 value = _mm_loadu_si128((__m128i *)p);
83 result = _mm_subs_epu16(value, xmm_wsize);
84 _mm_storeu_si128((__m128i *)p, result);
85
86 p -= 8;
87 n -= 8;
88 } while (n > 0);
89 #endif
90 more += wsize;
91 }
92 if (s->strm->avail_in == 0) break;
93
94 /* If there was no sliding:
95 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
96 * more == window_size - lookahead - strstart
97 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
98 * => more >= window_size - 2*WSIZE + 2
99 * In the BIG_MEM or MMAP case (not yet supported),
100 * window_size == input_size + MIN_LOOKAHEAD &&
101 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
102 * Otherwise, window_size == 2*WSIZE so more >= 2.
103 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
104 */
105 Assert(more >= 2, "more < 2");
106
107 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
108 s->lookahead += n;
109
110 /* Initialize the hash value now that we have some input: */
111 if (s->lookahead >= MIN_MATCH) {
112 uInt str = s->strstart;
113 s->ins_h = s->window[str];
114 if (str >= 1)
115 UPDATE_HASH(s, s->ins_h, str + 1 - (MIN_MATCH-1));
116 #if MIN_MATCH != 3
117 Call UPDATE_HASH() MIN_MATCH-3 more times
118 #endif
119 }
120 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
121 * but this is not important since only literal bytes will be emitted.
122 */
123
124 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
125
126 /* If the WIN_INIT bytes after the end of the current data have never been
127 * written, then zero those bytes in order to avoid memory check reports of
128 * the use of uninitialized (or uninitialised as Julian writes) bytes by
129 * the longest match routines. Update the high water mark for the next
130 * time through here. WIN_INIT is set to MAX_MATCH since the longest match
131 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
132 */
133 if (s->high_water < s->window_size) {
134 ulg curr = s->strstart + (ulg)(s->lookahead);
135 ulg init;
136
137 if (s->high_water < curr) {
138 /* Previous high water mark below current data -- zero WIN_INIT
139 * bytes or up to end of window, whichever is less.
140 */
141 init = s->window_size - curr;
142 if (init > WIN_INIT)
143 init = WIN_INIT;
144 zmemzero(s->window + curr, (unsigned)init);
145 s->high_water = curr + init;
146 }
147 else if (s->high_water < (ulg)curr + WIN_INIT) {
148 /* High water mark at or above current data, but below current data
149 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
150 * to end of window, whichever is less.
151 */
152 init = (ulg)curr + WIN_INIT - s->high_water;
153 if (init > s->window_size - s->high_water)
154 init = s->window_size - s->high_water;
155 zmemzero(s->window + s->high_water, (unsigned)init);
156 s->high_water += init;
157 }
158 }
159
160 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
161 "not enough room for search");
162 }
163 #endif
OLDNEW

Powered by Google App Engine
This is Rietveld 408576698