forked from KolibriOS/kolibrios
5f948ae4c6
git-svn-id: svn://kolibrios.org@6854 a494cfbc-eb01-0410-851d-a64ba20cac60
380 lines
12 KiB
PHP
380 lines
12 KiB
PHP
; deflate.inc -- internal compression state
|
|
; Copyright (C) 1995-2012 Jean-loup Gailly
|
|
; For conditions of distribution and use, see copyright notice in zlib.inc
|
|
|
|
; WARNING: this file should *not* be used by applications. It is
|
|
; part of the implementation of the compression library and is
|
|
; subject to change. Applications should only use zlib.inc.
|
|
|
|
include 'zutil.inc'
|
|
|
|
; ===========================================================================
|
|
; Internal compression state.
|
|
|
|
|
|
LENGTH_CODES equ 29
|
|
; number of length codes, not counting the special END_BLOCK code
|
|
|
|
LITERALS equ 256
|
|
; number of literal bytes 0..255
|
|
|
|
L_CODES equ (LITERALS+1+LENGTH_CODES)
|
|
; number of Literal or Length codes, including the END_BLOCK code
|
|
|
|
D_CODES equ 30
|
|
; number of distance codes
|
|
|
|
BL_CODES equ 19
|
|
; number of codes used to transfer the bit lengths
|
|
|
|
HEAP_SIZE equ (2*L_CODES+1)
|
|
; maximum heap size
|
|
|
|
MAX_BITS equ 15
|
|
; All codes must not exceed MAX_BITS bits
|
|
|
|
Buf_size equ 16
|
|
; size of bit buffer in bi_buf
|
|
|
|
INIT_STATE equ 42
|
|
EXTRA_STATE equ 69
|
|
NAME_STATE equ 73
|
|
COMMENT_STATE equ 91
|
|
HCRC_STATE equ 103
|
|
BUSY_STATE equ 113
|
|
FINISH_STATE equ 800
|
|
; Stream status
|
|
|
|
; Data structure describing a single value and its code string.
|
|
struct ct_data
|
|
fc dw ? ;union
|
|
;uint_16 freq ;frequency count
|
|
;uint_16 code ;bit string
|
|
dale dw ? ;union
|
|
;uint_16 dad ;father node in Huffman tree
|
|
;uint_16 len ;length of bit string
|
|
ends
|
|
|
|
Freq equ ct_data.fc ;.freq
|
|
Code equ ct_data.fc ;.code
|
|
Dad equ ct_data.dale ;.dad
|
|
Len equ ct_data.dale ;.len
|
|
|
|
struct tree_desc
|
|
dyn_tree dd ? ;ct_data * ;the dynamic tree
|
|
max_code dd ? ;int ;largest code with non zero frequency
|
|
stat_desc dd ? ;static_tree_desc * ;the corresponding static tree
|
|
ends
|
|
|
|
; A Pos is an index in the character window. We use short instead of int to
|
|
; save space in the various tables. IPos is used only for parameter passing.
|
|
|
|
struct deflate_state ;internal_state
|
|
strm dd ? ;z_streamp ;pointer back to this zlib stream
|
|
status dd ? ;int ;as the name implies
|
|
pending_buf dd ? ;Bytef *;output still pending
|
|
pending_buf_size dd ? ;ulg ;size of pending_buf
|
|
pending_out dd ? ;Bytef * ;next pending byte to output to the stream
|
|
pending dd ? ;uInt ;nb of bytes in the pending buffer
|
|
wrap dd ? ;int ;bit 0 true for zlib, bit 1 true for gzip
|
|
gzhead dd ? ;gz_headerp ;gzip header information to write
|
|
gzindex dd ? ;uInt ;where in extra, name, or comment
|
|
method db ? ;Byte ;can only be DEFLATED
|
|
rb 3 ;for align
|
|
last_flush dd ? ;int ;value of flush param for previous deflate call
|
|
|
|
; used by deflate.asm:
|
|
|
|
w_size dd ? ;uInt ;LZ77 window size (32K by default)
|
|
w_bits dd ? ;uInt ;log2(w_size) (8..16)
|
|
w_mask dd ? ;uInt ;w_size - 1
|
|
|
|
window dd ? ;Bytef *
|
|
; Sliding window. Input bytes are read into the second half of the window,
|
|
; and move to the first half later to keep a dictionary of at least wSize
|
|
; bytes. With this organization, matches are limited to a distance of
|
|
; wSize-MAX_MATCH bytes, but this ensures that IO is always
|
|
; performed with a length multiple of the block size. Also, it limits
|
|
; the window size to 64K, which is quite useful on MSDOS.
|
|
; To do: use the user input buffer as sliding window.
|
|
|
|
window_size dd ? ;ulg
|
|
; Actual size of window: 2*wSize, except when the user input buffer
|
|
; is directly used as sliding window.
|
|
|
|
prev dd ? ;Posf *
|
|
; Link to older string with same hash index. To limit the size of this
|
|
; array to 64K, this link is maintained only for the last 32K strings.
|
|
; An index in this array is thus a window index modulo 32K.
|
|
|
|
head dd ? ;Posf * ;Heads of the hash chains or NIL.
|
|
|
|
ins_h dd ? ;uInt ;hash index of string to be inserted
|
|
hash_size dd ? ;uInt ;number of elements in hash table
|
|
hash_bits dd ? ;uInt ;log2(hash_size)
|
|
hash_mask dd ? ;uInt ;hash_size-1
|
|
|
|
hash_shift dd ? ;uInt
|
|
; Number of bits by which ins_h must be shifted at each input
|
|
; step. It must be such that after MIN_MATCH steps, the oldest
|
|
; byte no longer takes part in the hash key, that is:
|
|
; hash_shift * MIN_MATCH >= hash_bits
|
|
|
|
block_start dd ? ;long
|
|
; Window position at the beginning of the current output block. Gets
|
|
; negative when the window is moved backwards.
|
|
|
|
match_length dd ? ;uInt ;length of best match
|
|
prev_match dd ? ;IPos ;previous match
|
|
match_available dd ? ;int ;set if previous match exists
|
|
strstart dd ? ;uInt ;start of string to insert
|
|
match_start dd ? ;uInt ;start of matching string
|
|
lookahead dd ? ;uInt ;number of valid bytes ahead in window
|
|
|
|
prev_length dd ? ;uInt
|
|
; Length of the best match at previous step. Matches not greater than this
|
|
; are discarded. This is used in the lazy match evaluation.
|
|
|
|
max_chain_length dd ? ;uInt
|
|
; To speed up deflation, hash chains are never searched beyond this
|
|
; length. A higher limit improves compression ratio but degrades the
|
|
; speed.
|
|
|
|
max_lazy_match dd ? ;uInt
|
|
; Attempt to find a better match only when the current match is strictly
|
|
; smaller than this value. This mechanism is used only for compression
|
|
; levels >= 4.
|
|
|
|
level dw ? ;int ;compression level (1..9)
|
|
rb 2 ;for align
|
|
strategy dw ? ;int ;favor or force Huffman coding
|
|
rb 2 ;for align
|
|
|
|
good_match dd ? ;uInt
|
|
; Use a faster search when the previous match is longer than this
|
|
|
|
nice_match dd ? ;int ;Stop searching when current match exceeds this
|
|
|
|
; used by trees.asm:
|
|
; Didn't use ct_data typedef below to suppress compiler warning
|
|
dyn_ltree rb sizeof.ct_data * HEAP_SIZE ;literal and length tree
|
|
dyn_dtree rb sizeof.ct_data * (2*D_CODES+1) ;distance tree
|
|
bl_tree rb sizeof.ct_data * (2*BL_CODES+1) ;Huffman tree for bit lengths
|
|
|
|
l_desc tree_desc ;desc. for literal tree
|
|
d_desc tree_desc ;desc. for distance tree
|
|
bl_desc tree_desc ;desc. for bit length tree
|
|
|
|
bl_count rw MAX_BITS+1 ;uint_16[]
|
|
; number of codes at each bit length for an optimal tree
|
|
|
|
heap rd 2*L_CODES+1 ;int[] ;heap used to build the Huffman trees
|
|
heap_len dd ? ;int ;number of elements in the heap
|
|
heap_max dd ? ;int ;element of largest frequency
|
|
; The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
|
|
; The same heap array is used to build all trees.
|
|
|
|
depth rb ((2*L_CODES+1)+3) and (not 3) ;uch[]
|
|
; Depth of each subtree used as tie breaker for trees of equal frequency
|
|
|
|
l_buf dd ? ;uchf * ;buffer for literals or lengths
|
|
|
|
lit_bufsize dd ? ;uInt
|
|
; Size of match buffer for literals/lengths. There are 4 reasons for
|
|
; limiting lit_bufsize to 64K:
|
|
; - frequencies can be kept in 16 bit counters
|
|
; - if compression is not successful for the first block, all input
|
|
; data is still in the window so we can still emit a stored block even
|
|
; when input comes from standard input. (This can also be done for
|
|
; all blocks if lit_bufsize is not greater than 32K.)
|
|
; - if compression is not successful for a file smaller than 64K, we can
|
|
; even emit a stored file instead of a stored block (saving 5 bytes).
|
|
; This is applicable only for zip (not gzip or zlib).
|
|
; - creating new Huffman trees less frequently may not provide fast
|
|
; adaptation to changes in the input data statistics. (Take for
|
|
; example a binary file with poorly compressible code followed by
|
|
; a highly compressible string table.) Smaller buffer sizes give
|
|
; fast adaptation but have of course the overhead of transmitting
|
|
; trees more frequently.
|
|
; - I can't count above 4
|
|
|
|
last_lit dd ? ;uInt ;running index in l_buf
|
|
|
|
d_buf dd ? ;uint_16p
|
|
; Buffer for distances. To simplify the code, d_buf and l_buf have
|
|
; the same number of elements. To use different lengths, an extra flag
|
|
; array would be necessary.
|
|
|
|
opt_len dd ? ;ulg ;bit length of current block with optimal trees
|
|
static_len dd ? ;ulg ;bit length of current block with static trees
|
|
matches dd ? ;uInt ;number of string matches in current block
|
|
insert dd ? ;uInt ;bytes at end of window left to insert
|
|
|
|
if DEBUG eq 1
|
|
;compressed_len dd ? ;ulg ;total bit length of compressed file mod 2^32
|
|
;bits_sent dd ? ;ulg ;bit length of compressed data sent mod 2^32
|
|
end if
|
|
|
|
bi_buf dw ? ;uint_16
|
|
rb 2 ;for align
|
|
; Output buffer. bits are inserted starting at the bottom (least
|
|
; significant bits).
|
|
|
|
bi_valid dd ? ;int
|
|
; Number of valid bits in bi_buf. All bits above the last valid bit
|
|
; are always zero.
|
|
|
|
high_water dd ? ;ulg
|
|
; High water mark offset in window for initialized bytes -- bytes above
|
|
; this are set to zero in order to avoid memory check warnings when
|
|
; longest match routines access bytes past the input. This is then
|
|
; updated to the new high water mark.
|
|
ends
|
|
|
|
deflate_state.max_insert_length equ deflate_state.max_lazy_match
|
|
; Insert new strings in the hash table only if the match length is not
|
|
; greater than this length. This saves time but degrades compression.
|
|
; max_insert_length is used only for compression levels <= 3.
|
|
|
|
; Output a byte on the stream.
|
|
; IN assertion: there is enough room in pending_buf.
|
|
|
|
macro put_byte s, c
|
|
{
|
|
mov eax,[s+deflate_state.pending]
|
|
add eax,[s+deflate_state.pending_buf]
|
|
mov byte[eax],c
|
|
inc dword[s+deflate_state.pending]
|
|
}
|
|
macro put_dword s, d
|
|
{
|
|
mov eax,[s+deflate_state.pending]
|
|
add eax,[s+deflate_state.pending_buf]
|
|
mov dword[eax],d
|
|
add dword[s+deflate_state.pending],4
|
|
}
|
|
|
|
MIN_LOOKAHEAD equ (MAX_MATCH+MIN_MATCH+1)
|
|
; Minimum amount of lookahead, except at the end of the input file.
|
|
; See deflate.asm for comments about the MIN_MATCH+1.
|
|
|
|
macro MAX_DIST s
|
|
{
|
|
mov eax,[s+deflate_state.w_size]
|
|
sub eax,MIN_LOOKAHEAD
|
|
}
|
|
; In order to simplify the code, particularly on 16 bit machines, match
|
|
; distances are limited to MAX_DIST instead of WSIZE.
|
|
|
|
WIN_INIT equ MAX_MATCH
|
|
; Number of bytes after end of data in window to initialize in order to avoid
|
|
; memory checker errors from longest match routines
|
|
|
|
macro d_code dist
|
|
{
|
|
;if (dist < 256) _dist_code[dist]
|
|
;else _dist_code[ 256+(dist>>7) ]
|
|
local .end0
|
|
if dist eq eax
|
|
else
|
|
mov eax,dist
|
|
end if
|
|
cmp eax,256
|
|
jb .end0
|
|
shr eax,7
|
|
add eax,256
|
|
.end0:
|
|
movzx eax,byte[eax+_dist_code]
|
|
}
|
|
; Mapping from a distance to a distance code. dist is the distance - 1 and
|
|
; must not have side effects. _dist_code[256] and _dist_code[257] are never
|
|
; used.
|
|
|
|
macro _tr_tally_lit s, c, flush
|
|
{
|
|
if DEBUG eq 0
|
|
; Inline versions of _tr_tally for speed:
|
|
if c eq eax
|
|
else
|
|
mov eax,c
|
|
end if
|
|
push ecx
|
|
mov ecx,[s+deflate_state.last_lit]
|
|
shl ecx,1
|
|
add ecx,[s+deflate_state.d_buf]
|
|
mov word[ecx],0
|
|
mov ecx,[s+deflate_state.last_lit]
|
|
add ecx,[s+deflate_state.l_buf]
|
|
mov byte[ecx],al
|
|
inc dword[s+deflate_state.last_lit]
|
|
and eax,0xff
|
|
inc word[s+sizeof.ct_data*eax+deflate_state.dyn_ltree+Freq]
|
|
xor eax,eax
|
|
mov ecx,[s+deflate_state.lit_bufsize]
|
|
dec ecx
|
|
cmp [s+deflate_state.last_lit],ecx
|
|
sete al ;flush = (..==..)
|
|
mov flush, eax
|
|
pop ecx
|
|
else
|
|
stdcall _tr_tally, s, 0, c
|
|
mov flush, eax
|
|
end if
|
|
}
|
|
macro _tr_tally_dist s, distance, length, flush
|
|
{
|
|
if DEBUG eq 0
|
|
push ecx
|
|
|
|
;s.d_buf[s.last_lit] = dist
|
|
mov ecx,[s+deflate_state.last_lit]
|
|
shl ecx,1
|
|
add ecx,[s+deflate_state.d_buf]
|
|
if distance eq eax
|
|
mov [ecx],ax
|
|
else
|
|
mov word[ecx],distance
|
|
end if
|
|
|
|
;s.l_buf[s.last_lit++] = len
|
|
mov ecx,[s+deflate_state.last_lit]
|
|
add ecx,[s+deflate_state.l_buf]
|
|
if length eq eax
|
|
mov [ecx],al
|
|
else if length eq ebx
|
|
mov [ecx],bl
|
|
else
|
|
... ;mov byte[ecx],length
|
|
end if
|
|
inc dword[s+deflate_state.last_lit]
|
|
|
|
;dist--
|
|
if distance eq eax
|
|
else
|
|
mov eax,distance
|
|
end if
|
|
dec eax
|
|
|
|
;s.dyn_ltree[_length_code[len]+LITERALS+1].Freq++
|
|
movzx ecx,byte[ecx]
|
|
movzx ecx,byte[ecx+_length_code]
|
|
inc word[s+sizeof.ct_data*ecx+deflate_state.dyn_ltree+sizeof.ct_data*(LITERALS+1)+Freq]
|
|
|
|
;s.dyn_dtree[d_code(dist)].Freq++
|
|
d_code eax
|
|
inc word[s+sizeof.ct_data*eax+deflate_state.dyn_dtree+Freq]
|
|
|
|
;flush = (s.last_lit == s.lit_bufsize-1)
|
|
mov ecx,[s+deflate_state.lit_bufsize]
|
|
dec ecx
|
|
xor eax,eax
|
|
cmp [s+deflate_state.last_lit],ecx
|
|
sete al
|
|
mov flush,eax
|
|
pop ecx
|
|
else
|
|
stdcall _tr_tally, s, distance, length
|
|
mov flush, eax
|
|
end if
|
|
}
|