1
/* inffas8664.c is a hand tuned assembler version of inffast.c - fast decoding
2
* version for AMD64 on Windows using Microsoft C compiler
4
* Copyright (C) 1995-2003 Mark Adler
5
* For conditions of distribution and use, see copyright notice in zlib.h
7
* Copyright (C) 2003 Chris Anderson <christop@charm.net>
8
* Please use the copyright conditions above.
10
* 2005 - Adaptation to Microsoft C Compiler for AMD64 by Gilles Vollant
12
* inffas8664.c call function inffas8664fnc in inffasx64.asm
13
* inffasx64.asm is automatically convert from AMD64 portion of inffas86.c
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.
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/
41
/* Mark Adler's comments from inffast.c: */
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.
55
strm->avail_out >= 258
56
start >= strm->avail_out
59
On return, state->mode is one of:
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
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.
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
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*/
106
void inflate_fast(strm, start)
108
unsigned start; /* inflate()'s starting value for strm->avail_out */
110
struct inflate_state FAR *state;
112
void inffas8664fnc(struct inffast_ar * par);
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
120
#define PAD_AVAIL_IN 5
121
#define PAD_AVAIL_OUT 257
124
/* copy state to local variables */
125
state = (struct inflate_state FAR *)strm->state;
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;
142
/* decode literals and length/distances until end-of-block or not enough
143
input data or output space */
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;
155
strm->msg = "invalid literal/length code";
156
else if (ar.status == 3)
157
strm->msg = "invalid distance code";
159
strm->msg = "invalid distance too far back";
162
else if ( ar.status == 1 ) {
166
/* return unused bytes (on entry, bits < 8, so in won't go too far back) */
167
ar.len = ar.bits >> 3;
169
ar.bits -= ar.len << 3;
170
ar.hold &= (1U << ar.bits) - 1;
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;