/src/zlib-ng/deflate_quick.c
Line | Count | Source |
1 | | /* |
2 | | * The deflate_quick deflate strategy, designed to be used when cycles are |
3 | | * at a premium. |
4 | | * |
5 | | * Copyright (C) 2013 Intel Corporation. All rights reserved. |
6 | | * Authors: |
7 | | * Wajdi Feghali <wajdi.k.feghali@intel.com> |
8 | | * Jim Guilford <james.guilford@intel.com> |
9 | | * Vinodh Gopal <vinodh.gopal@intel.com> |
10 | | * Erdinc Ozturk <erdinc.ozturk@intel.com> |
11 | | * Jim Kukunas <james.t.kukunas@linux.intel.com> |
12 | | * |
13 | | * Portions are Copyright (C) 2016 12Sided Technology, LLC. |
14 | | * Author: |
15 | | * Phil Vachon <pvachon@12sidedtech.com> |
16 | | * |
17 | | * For conditions of distribution and use, see copyright notice in zlib.h |
18 | | */ |
19 | | |
20 | | #include "zbuild.h" |
21 | | #include "zmemory.h" |
22 | | #include "deflate.h" |
23 | | #include "deflate_p.h" |
24 | | #include "functable.h" |
25 | | #include "trees_emit.h" |
26 | | #include "insert_string_p.h" |
27 | | |
28 | | extern const ct_data static_ltree[L_CODES+2]; |
29 | | extern const ct_data static_dtree[D_CODES]; |
30 | | |
31 | 1.29k | Z_FORCEINLINE static void quick_start_block(deflate_state *s, uint32_t strstart, int last) { |
32 | 1.29k | zng_tr_emit_tree(s, STATIC_TREES, last); |
33 | 1.29k | s->block_open = 1 + last; |
34 | 1.29k | s->block_start = (int)strstart; |
35 | 1.29k | } |
36 | | |
37 | 2.15k | Z_FORCEINLINE static int quick_end_block(deflate_state *s, uint32_t strstart, int last) { |
38 | 2.15k | if (s->block_open) { |
39 | 1.29k | zng_tr_emit_end_block(s, static_ltree, last); |
40 | 1.29k | s->block_open = 0; |
41 | 1.29k | s->block_start = (int)strstart; |
42 | 1.29k | PREFIX(flush_pending)(s->strm); |
43 | 1.29k | return (s->strm->avail_out == 0); |
44 | 1.29k | } |
45 | 853 | return 0; |
46 | 2.15k | } |
47 | | |
48 | | Z_FORCEINLINE static block_state deflate_quick_impl(deflate_state *s, int flush, |
49 | 2.07k | uint32_t strstart, uint32_t lookahead) { |
50 | 2.07k | unsigned char *window; |
51 | 2.07k | unsigned last = (flush == Z_FINISH) ? 1 : 0; |
52 | | |
53 | 2.07k | if (UNLIKELY(last && s->block_open != 2)) { |
54 | | /* Emit end of previous block */ |
55 | 1.07k | if (quick_end_block(s, strstart, 0)) |
56 | 8 | return need_more; |
57 | | /* Emit start of last block */ |
58 | 1.07k | quick_start_block(s, strstart, last); |
59 | 1.07k | } else if (UNLIKELY(s->block_open == 0 && lookahead > 0)) { |
60 | | /* Start new block only when we have lookahead data, so that if no |
61 | | input data is given an empty block will not be written */ |
62 | 0 | quick_start_block(s, strstart, last); |
63 | 0 | } |
64 | | |
65 | 2.06k | window = s->window; |
66 | | |
67 | 74.9M | for (;;) { |
68 | 74.9M | if (UNLIKELY(s->pending + ((BIT_BUF_SIZE + 7) >> 3) >= s->pending_buf_size)) { |
69 | 957 | PREFIX(flush_pending)(s->strm); |
70 | 957 | if (s->strm->avail_out == 0) { |
71 | 193 | s->lookahead = lookahead; |
72 | 193 | s->strstart = strstart; |
73 | 193 | return (last && s->strm->avail_in == 0 && s->bi_valid == 0 && s->block_open == 0) ? finish_started : need_more; |
74 | 193 | } |
75 | 957 | } |
76 | | |
77 | 74.9M | if (UNLIKELY(lookahead < MIN_LOOKAHEAD)) { |
78 | 88.9k | s->lookahead = lookahead; |
79 | 88.9k | s->strstart = strstart; |
80 | 88.9k | PREFIX(fill_window)(s); |
81 | 88.9k | lookahead = s->lookahead; |
82 | 88.9k | strstart = s->strstart; |
83 | 88.9k | if (UNLIKELY(lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH)) |
84 | 800 | return need_more; |
85 | 88.1k | if (UNLIKELY(lookahead == 0)) |
86 | 1.07k | break; |
87 | | |
88 | 87.0k | if (UNLIKELY(s->block_open == 0)) { |
89 | | /* Start new block when we have lookahead data, so that if no |
90 | | input data is given an empty block will not be written */ |
91 | 226 | quick_start_block(s, strstart, last); |
92 | 226 | } |
93 | 87.0k | } |
94 | | |
95 | 74.9M | uint32_t str_val = Z_U32_FROM_LE(zng_memread_4(window + strstart)); |
96 | | |
97 | 74.9M | if (LIKELY(lookahead >= WANT_MIN_MATCH)) { |
98 | 74.9M | uint32_t hash_head = insert_knuth_val(s, strstart, str_val); |
99 | 74.9M | int64_t dist = (int64_t)strstart - hash_head; |
100 | | |
101 | 74.9M | if (dist <= MAX_DIST(s) && dist > 0) { |
102 | 35.7M | const uint8_t *match_start = window + hash_head; |
103 | 35.7M | uint32_t match_val = Z_U32_FROM_LE(zng_memread_4(match_start)); |
104 | | |
105 | 35.7M | if (str_val == match_val) { |
106 | 345k | const uint8_t *scan_start = window + strstart; |
107 | 345k | uint32_t match_len = FUNCTABLE_CALL(compare256)(scan_start+2, match_start+2) + 2; |
108 | | |
109 | 345k | if (match_len >= WANT_MIN_MATCH) { |
110 | 345k | if (UNLIKELY(match_len > lookahead)) |
111 | 96 | match_len = lookahead; |
112 | | |
113 | 345k | Assert(match_len <= STD_MAX_MATCH, "match too long"); |
114 | 345k | Assert(strstart <= UINT16_MAX, "strstart should fit in uint16_t"); |
115 | 345k | check_match(s, strstart, hash_head, match_len); |
116 | | |
117 | 345k | zng_tr_emit_dist(s, static_ltree, static_dtree, match_len - STD_MIN_MATCH, (uint32_t)dist); |
118 | 345k | lookahead -= match_len; |
119 | 345k | strstart += match_len; |
120 | 345k | continue; |
121 | 345k | } |
122 | 345k | } |
123 | 35.7M | } |
124 | 74.9M | } |
125 | | |
126 | 74.6M | zng_tr_emit_lit(s, static_ltree, (uint8_t)str_val); |
127 | 74.6M | strstart++; |
128 | 74.6M | lookahead--; |
129 | 74.6M | } |
130 | | |
131 | 1.07k | s->lookahead = lookahead; |
132 | 1.07k | s->strstart = strstart; |
133 | 1.07k | s->insert = strstart < (STD_MIN_MATCH - 1) ? strstart : (STD_MIN_MATCH - 1); |
134 | 1.07k | if (UNLIKELY(last)) { |
135 | 1.07k | if (quick_end_block(s, strstart, 1)) |
136 | 209 | return finish_started; |
137 | 862 | return finish_done; |
138 | 1.07k | } |
139 | | |
140 | 0 | if (quick_end_block(s, strstart, 0)) |
141 | 0 | return need_more; |
142 | 0 | return block_done; |
143 | 0 | } |
144 | | |
145 | 2.07k | Z_INTERNAL block_state deflate_quick(deflate_state *s, int flush) { |
146 | 2.07k | return deflate_quick_impl(s, flush, s->strstart, s->lookahead); |
147 | 2.07k | } |