URL https://opencores.org/ocsvn/openrisc/openrisc/trunk

Subversion Repositories openrisc

[/] [openrisc/] [trunk/] [gnu-dev/] [or1k-gcc/] [zlib/] [contrib/] [masmx86/] [inffas32.asm] - Blame information for rev 745

Details | Compare with Previous | View Log


;/* inffas32.asm is a hand tuned assembler version of inffast.c -- fast decoding
; *
; * inffas32.asm is derivated from inffas86.c, with translation of assembly code
; *
; * Copyright (C) 1995-2003 Mark Adler
; * For conditions of distribution and use, see copyright notice in zlib.h
; *
; * Copyright (C) 2003 Chris Anderson 
; * Please use the copyright conditions above.
; *
; * Mar-13-2003 -- Most of this is derived from inffast.S which is derived from
; * the gcc -S output of zlib-1.2.0/inffast.c.  Zlib-1.2.0 is in beta release at
; * the moment.  I have successfully compiled and tested this code with gcc2.96,
; * gcc3.2, icc5.0, msvc6.0.  It is very close to the speed of inffast.S
; * compiled with gcc -DNO_MMX, but inffast.S is still faster on the P3 with MMX
; * enabled.  I will attempt to merge the MMX code into this version.  Newer
; * versions of this and inffast.S can be found at
; * http://www.eetbeetee.com/zlib/ and http://www.charm.net/~christop/zlib/
; *
; * 2005 : modification by Gilles Vollant
; */
; For Visual C++ 4.x and higher and ML 6.x and higher
;   ml.exe is in directory \MASM611C of Win95 DDK
;   ml.exe is also distributed in http://www.masm32.com/masmdl.htm
;    and in VC++2003 toolkit at http://msdn.microsoft.com/visualc/vctoolkit2003/
;
;
;   compile with command line option
;   ml  /coff /Zi /c /Flinffas32.lst inffas32.asm
 
;   if you define NO_GZIP (see inflate.h), compile with
;   ml  /coff /Zi /c /Flinffas32.lst /DNO_GUNZIP inffas32.asm
 
 
; zlib122sup is 0 fort zlib 1.2.2.1 and lower
; zlib122sup is 8 fort zlib 1.2.2.2 and more (with addition of dmax and head
;        in inflate_state in inflate.h)
zlib1222sup      equ    8
 
 
IFDEF GUNZIP
  INFLATE_MODE_TYPE    equ 11
  INFLATE_MODE_BAD     equ 26
ELSE
  IFNDEF NO_GUNZIP
    INFLATE_MODE_TYPE    equ 11
    INFLATE_MODE_BAD     equ 26
  ELSE
    INFLATE_MODE_TYPE    equ 3
    INFLATE_MODE_BAD     equ 17
  ENDIF
ENDIF
 
 
; 75 "inffast.S"
;FILE "inffast.S"
 
;;;GLOBAL _inflate_fast
 
;;;SECTION .text
 
 
 
        .586p
        .mmx
 
        name    inflate_fast_x86
        .MODEL  FLAT
 
_DATA                   segment
inflate_fast_use_mmx:
        dd      1
 
 
_TEXT                   segment
PUBLIC _inflate_fast
 
ALIGN 4
_inflate_fast:
        jmp inflate_fast_entry
 
 
 
ALIGN 4
        db      'Fast decoding Code from Chris Anderson'
        db      0
 
ALIGN 4
invalid_literal_length_code_msg:
        db      'invalid literal/length code'
        db      0
 
ALIGN 4
invalid_distance_code_msg:
        db      'invalid distance code'
        db      0
 
ALIGN 4
invalid_distance_too_far_msg:
        db      'invalid distance too far back'
        db      0
 
 
ALIGN 4
inflate_fast_mask:
dd      0
dd      1
dd      3
dd      7
dd      15
dd      31
dd      63
dd      127
dd      255
dd      511
dd      1023
dd      2047
dd      4095
dd      8191
dd      16383
dd      32767
dd      65535
dd      131071
dd      262143
dd      524287
dd      1048575
dd      2097151
dd      4194303
dd      8388607
dd      16777215
dd      33554431
dd      67108863
dd      134217727
dd      268435455
dd      536870911
dd      1073741823
dd      2147483647
dd      4294967295
 
 
mode_state       equ    0        ;/* state->mode */
wsize_state      equ    (32+zlib1222sup)        ;/* state->wsize */
write_state      equ    (36+4+zlib1222sup)      ;/* state->write */
window_state     equ    (40+4+zlib1222sup)      ;/* state->window */
hold_state       equ    (44+4+zlib1222sup)      ;/* state->hold */
bits_state       equ    (48+4+zlib1222sup)      ;/* state->bits */
lencode_state    equ    (64+4+zlib1222sup)      ;/* state->lencode */
distcode_state   equ    (68+4+zlib1222sup)      ;/* state->distcode */
lenbits_state    equ    (72+4+zlib1222sup)      ;/* state->lenbits */
distbits_state   equ    (76+4+zlib1222sup)      ;/* state->distbits */
 
 
;;SECTION .text
; 205 "inffast.S"
;GLOBAL inflate_fast_use_mmx
 
;SECTION .data
 
 
; GLOBAL inflate_fast_use_mmx:object
;.size inflate_fast_use_mmx, 4
; 226 "inffast.S"
;SECTION .text
 
ALIGN 4
inflate_fast_entry:
        push  edi
        push  esi
        push  ebp
        push  ebx
        pushfd
        sub  esp,64
        cld
 
 
 
 
        mov  esi, [esp+88]
        mov  edi, [esi+28]
 
 
 
 
 
 
 
        mov  edx, [esi+4]
        mov  eax, [esi+0]
 
        add  edx,eax
        sub  edx,11
 
        mov  [esp+44],eax
        mov  [esp+20],edx
 
        mov  ebp, [esp+92]
        mov  ecx, [esi+16]
        mov  ebx, [esi+12]
 
        sub  ebp,ecx
        neg  ebp
        add  ebp,ebx
 
        sub  ecx,257
        add  ecx,ebx
 
        mov  [esp+60],ebx
        mov  [esp+40],ebp
        mov  [esp+16],ecx
; 285 "inffast.S"
        mov  eax, [edi+lencode_state]
        mov  ecx, [edi+distcode_state]
 
        mov  [esp+8],eax
        mov  [esp+12],ecx
 
        mov  eax,1
        mov  ecx, [edi+lenbits_state]
        shl  eax,cl
        dec  eax
        mov  [esp+0],eax
 
        mov  eax,1
        mov  ecx, [edi+distbits_state]
        shl  eax,cl
        dec  eax
        mov  [esp+4],eax
 
        mov  eax, [edi+wsize_state]
        mov  ecx, [edi+write_state]
        mov  edx, [edi+window_state]
 
        mov  [esp+52],eax
        mov  [esp+48],ecx
        mov  [esp+56],edx
 
        mov  ebp, [edi+hold_state]
        mov  ebx, [edi+bits_state]
; 321 "inffast.S"
        mov  esi, [esp+44]
        mov  ecx, [esp+20]
        cmp  ecx,esi
        ja   L_align_long
 
        add  ecx,11
        sub  ecx,esi
        mov  eax,12
        sub  eax,ecx
        lea  edi, [esp+28]
        rep movsb
        mov  ecx,eax
        xor  eax,eax
        rep stosb
        lea  esi, [esp+28]
        mov  [esp+20],esi
        jmp  L_is_aligned
 
 
L_align_long:
        test  esi,3
        jz   L_is_aligned
        xor  eax,eax
        mov  al, [esi]
        inc  esi
        mov  ecx,ebx
        add  ebx,8
        shl  eax,cl
        or  ebp,eax
        jmp L_align_long
 
L_is_aligned:
        mov  edi, [esp+60]
; 366 "inffast.S"
L_check_mmx:
        cmp  dword ptr [inflate_fast_use_mmx],2
        je   L_init_mmx
        ja   L_do_loop
 
        push  eax
        push  ebx
        push  ecx
        push  edx
        pushfd
        mov  eax, [esp]
        xor  dword ptr [esp],0200000h
 
 
 
 
        popfd
        pushfd
        pop  edx
        xor  edx,eax
        jz   L_dont_use_mmx
        xor  eax,eax
        cpuid
        cmp  ebx,0756e6547h
        jne  L_dont_use_mmx
        cmp  ecx,06c65746eh
        jne  L_dont_use_mmx
        cmp  edx,049656e69h
        jne  L_dont_use_mmx
        mov  eax,1
        cpuid
        shr  eax,8
        and  eax,15
        cmp  eax,6
        jne  L_dont_use_mmx
        test  edx,0800000h
        jnz  L_use_mmx
        jmp  L_dont_use_mmx
L_use_mmx:
        mov  dword ptr [inflate_fast_use_mmx],2
        jmp  L_check_mmx_pop
L_dont_use_mmx:
        mov  dword ptr [inflate_fast_use_mmx],3
L_check_mmx_pop:
        pop  edx
        pop  ecx
        pop  ebx
        pop  eax
        jmp  L_check_mmx
; 426 "inffast.S"
ALIGN 4
L_do_loop:
; 437 "inffast.S"
        cmp  bl,15
        ja   L_get_length_code
 
        xor  eax,eax
        lodsw
        mov  cl,bl
        add  bl,16
        shl  eax,cl
        or  ebp,eax
 
L_get_length_code:
        mov  edx, [esp+0]
        mov  ecx, [esp+8]
        and  edx,ebp
        mov  eax, [ecx+edx*4]
 
L_dolen:
 
 
 
 
 
 
        mov  cl,ah
        sub  bl,ah
        shr  ebp,cl
 
 
 
 
 
 
        test  al,al
        jnz   L_test_for_length_base
 
        shr  eax,16
        stosb
 
L_while_test:
 
 
        cmp  [esp+16],edi
        jbe  L_break_loop
 
        cmp  [esp+20],esi
        ja   L_do_loop
        jmp  L_break_loop
 
L_test_for_length_base:
; 502 "inffast.S"
        mov  edx,eax
        shr  edx,16
        mov  cl,al
 
        test  al,16
        jz   L_test_for_second_level_length
        and  cl,15
        jz   L_save_len
        cmp  bl,cl
        jae  L_add_bits_to_len
 
        mov  ch,cl
        xor  eax,eax
        lodsw
        mov  cl,bl
        add  bl,16
        shl  eax,cl
        or  ebp,eax
        mov  cl,ch
 
L_add_bits_to_len:
        mov  eax,1
        shl  eax,cl
        dec  eax
        sub  bl,cl
        and  eax,ebp
        shr  ebp,cl
        add  edx,eax
 
L_save_len:
        mov  [esp+24],edx
 
 
L_decode_distance:
; 549 "inffast.S"
        cmp  bl,15
        ja   L_get_distance_code
 
        xor  eax,eax
        lodsw
        mov  cl,bl
        add  bl,16
        shl  eax,cl
        or  ebp,eax
 
L_get_distance_code:
        mov  edx, [esp+4]
        mov  ecx, [esp+12]
        and  edx,ebp
        mov  eax, [ecx+edx*4]
 
 
L_dodist:
        mov  edx,eax
        shr  edx,16
        mov  cl,ah
        sub  bl,ah
        shr  ebp,cl
; 584 "inffast.S"
        mov  cl,al
 
        test  al,16
        jz  L_test_for_second_level_dist
        and  cl,15
        jz  L_check_dist_one
        cmp  bl,cl
        jae  L_add_bits_to_dist
 
        mov  ch,cl
        xor  eax,eax
        lodsw
        mov  cl,bl
        add  bl,16
        shl  eax,cl
        or  ebp,eax
        mov  cl,ch
 
L_add_bits_to_dist:
        mov  eax,1
        shl  eax,cl
        dec  eax
        sub  bl,cl
        and  eax,ebp
        shr  ebp,cl
        add  edx,eax
        jmp  L_check_window
 
L_check_window:
; 625 "inffast.S"
        mov  [esp+44],esi
        mov  eax,edi
        sub  eax, [esp+40]
 
        cmp  eax,edx
        jb   L_clip_window
 
        mov  ecx, [esp+24]
        mov  esi,edi
        sub  esi,edx
 
        sub  ecx,3
        mov  al, [esi]
        mov  [edi],al
        mov  al, [esi+1]
        mov  dl, [esi+2]
        add  esi,3
        mov  [edi+1],al
        mov  [edi+2],dl
        add  edi,3
        rep movsb
 
        mov  esi, [esp+44]
        jmp  L_while_test
 
ALIGN 4
L_check_dist_one:
        cmp  edx,1
        jne  L_check_window
        cmp  [esp+40],edi
        je  L_check_window
 
        dec  edi
        mov  ecx, [esp+24]
        mov  al, [edi]
        sub  ecx,3
 
        mov  [edi+1],al
        mov  [edi+2],al
        mov  [edi+3],al
        add  edi,4
        rep stosb
 
        jmp  L_while_test
 
ALIGN 4
L_test_for_second_level_length:
 
 
 
 
        test  al,64
        jnz   L_test_for_end_of_block
 
        mov  eax,1
        shl  eax,cl
        dec  eax
        and  eax,ebp
        add  eax,edx
        mov  edx, [esp+8]
        mov  eax, [edx+eax*4]
        jmp  L_dolen
 
ALIGN 4
L_test_for_second_level_dist:
 
 
 
 
        test  al,64
        jnz   L_invalid_distance_code
 
        mov  eax,1
        shl  eax,cl
        dec  eax
        and  eax,ebp
        add  eax,edx
        mov  edx, [esp+12]
        mov  eax, [edx+eax*4]
        jmp  L_dodist
 
ALIGN 4
L_clip_window:
; 721 "inffast.S"
        mov  ecx,eax
        mov  eax, [esp+52]
        neg  ecx
        mov  esi, [esp+56]
 
        cmp  eax,edx
        jb   L_invalid_distance_too_far
 
        add  ecx,edx
        cmp  dword ptr [esp+48],0
        jne  L_wrap_around_window
 
        sub  eax,ecx
        add  esi,eax
; 749 "inffast.S"
        mov  eax, [esp+24]
        cmp  eax,ecx
        jbe  L_do_copy1
 
        sub  eax,ecx
        rep movsb
        mov  esi,edi
        sub  esi,edx
        jmp  L_do_copy1
 
        cmp  eax,ecx
        jbe  L_do_copy1
 
        sub  eax,ecx
        rep movsb
        mov  esi,edi
        sub  esi,edx
        jmp  L_do_copy1
 
L_wrap_around_window:
; 793 "inffast.S"
        mov  eax, [esp+48]
        cmp  ecx,eax
        jbe  L_contiguous_in_window
 
        add  esi, [esp+52]
        add  esi,eax
        sub  esi,ecx
        sub  ecx,eax
 
 
        mov  eax, [esp+24]
        cmp  eax,ecx
        jbe  L_do_copy1
 
        sub  eax,ecx
        rep movsb
        mov  esi, [esp+56]
        mov  ecx, [esp+48]
        cmp  eax,ecx
        jbe  L_do_copy1
 
        sub  eax,ecx
        rep movsb
        mov  esi,edi
        sub  esi,edx
        jmp  L_do_copy1
 
L_contiguous_in_window:
; 836 "inffast.S"
        add  esi,eax
        sub  esi,ecx
 
 
        mov  eax, [esp+24]
        cmp  eax,ecx
        jbe  L_do_copy1
 
        sub  eax,ecx
        rep movsb
        mov  esi,edi
        sub  esi,edx
 
L_do_copy1:
; 862 "inffast.S"
        mov  ecx,eax
        rep movsb
 
        mov  esi, [esp+44]
        jmp  L_while_test
; 878 "inffast.S"
ALIGN 4
L_init_mmx:
        emms
 
 
 
 
 
        movd mm0,ebp
        mov  ebp,ebx
; 896 "inffast.S"
        movd mm4,[esp+0]
        movq mm3,mm4
        movd mm5,[esp+4]
        movq mm2,mm5
        pxor mm1,mm1
        mov  ebx, [esp+8]
        jmp  L_do_loop_mmx
 
ALIGN 4
L_do_loop_mmx:
        psrlq mm0,mm1
 
        cmp  ebp,32
        ja  L_get_length_code_mmx
 
        movd mm6,ebp
        movd mm7,[esi]
        add  esi,4
        psllq mm7,mm6
        add  ebp,32
        por mm0,mm7
 
L_get_length_code_mmx:
        pand mm4,mm0
        movd eax,mm4
        movq mm4,mm3
        mov  eax, [ebx+eax*4]
 
L_dolen_mmx:
        movzx  ecx,ah
        movd mm1,ecx
        sub  ebp,ecx
 
        test  al,al
        jnz L_test_for_length_base_mmx
 
        shr  eax,16
        stosb
 
L_while_test_mmx:
 
 
        cmp  [esp+16],edi
        jbe L_break_loop
 
        cmp  [esp+20],esi
        ja L_do_loop_mmx
        jmp L_break_loop
 
L_test_for_length_base_mmx:
 
        mov  edx,eax
        shr  edx,16
 
        test  al,16
        jz  L_test_for_second_level_length_mmx
        and  eax,15
        jz L_decode_distance_mmx
 
        psrlq mm0,mm1
        movd mm1,eax
        movd ecx,mm0
        sub  ebp,eax
        and  ecx, [inflate_fast_mask+eax*4]
        add  edx,ecx
 
L_decode_distance_mmx:
        psrlq mm0,mm1
 
        cmp  ebp,32
        ja L_get_dist_code_mmx
 
        movd mm6,ebp
        movd mm7,[esi]
        add  esi,4
        psllq mm7,mm6
        add  ebp,32
        por mm0,mm7
 
L_get_dist_code_mmx:
        mov  ebx, [esp+12]
        pand mm5,mm0
        movd eax,mm5
        movq mm5,mm2
        mov  eax, [ebx+eax*4]
 
L_dodist_mmx:
 
        movzx  ecx,ah
        mov  ebx,eax
        shr  ebx,16
        sub  ebp,ecx
        movd mm1,ecx
 
        test  al,16
        jz L_test_for_second_level_dist_mmx
        and  eax,15
        jz L_check_dist_one_mmx
 
L_add_bits_to_dist_mmx:
        psrlq mm0,mm1
        movd mm1,eax
        movd ecx,mm0
        sub  ebp,eax
        and  ecx, [inflate_fast_mask+eax*4]
        add  ebx,ecx
 
L_check_window_mmx:
        mov  [esp+44],esi
        mov  eax,edi
        sub  eax, [esp+40]
 
        cmp  eax,ebx
        jb L_clip_window_mmx
 
        mov  ecx,edx
        mov  esi,edi
        sub  esi,ebx
 
        sub  ecx,3
        mov  al, [esi]
        mov  [edi],al
        mov  al, [esi+1]
        mov  dl, [esi+2]
        add  esi,3
        mov  [edi+1],al
        mov  [edi+2],dl
        add  edi,3
        rep movsb
 
        mov  esi, [esp+44]
        mov  ebx, [esp+8]
        jmp  L_while_test_mmx
 
ALIGN 4
L_check_dist_one_mmx:
        cmp  ebx,1
        jne  L_check_window_mmx
        cmp  [esp+40],edi
        je   L_check_window_mmx
 
        dec  edi
        mov  ecx,edx
        mov  al, [edi]
        sub  ecx,3
 
        mov  [edi+1],al
        mov  [edi+2],al
        mov  [edi+3],al
        add  edi,4
        rep stosb
 
        mov  ebx, [esp+8]
        jmp  L_while_test_mmx
 
ALIGN 4
L_test_for_second_level_length_mmx:
        test  al,64
        jnz L_test_for_end_of_block
 
        and  eax,15
        psrlq mm0,mm1
        movd ecx,mm0
        and  ecx, [inflate_fast_mask+eax*4]
        add  ecx,edx
        mov  eax, [ebx+ecx*4]
        jmp L_dolen_mmx
 
ALIGN 4
L_test_for_second_level_dist_mmx:
        test  al,64
        jnz L_invalid_distance_code
 
        and  eax,15
        psrlq mm0,mm1
        movd ecx,mm0
        and  ecx, [inflate_fast_mask+eax*4]
        mov  eax, [esp+12]
        add  ecx,ebx
        mov  eax, [eax+ecx*4]
        jmp  L_dodist_mmx
 
ALIGN 4
L_clip_window_mmx:
 
        mov  ecx,eax
        mov  eax, [esp+52]
        neg  ecx
        mov  esi, [esp+56]
 
        cmp  eax,ebx
        jb  L_invalid_distance_too_far
 
        add  ecx,ebx
        cmp  dword ptr [esp+48],0
        jne  L_wrap_around_window_mmx
 
        sub  eax,ecx
        add  esi,eax
 
        cmp  edx,ecx
        jbe  L_do_copy1_mmx
 
        sub  edx,ecx
        rep movsb
        mov  esi,edi
        sub  esi,ebx
        jmp  L_do_copy1_mmx
 
        cmp  edx,ecx
        jbe  L_do_copy1_mmx
 
        sub  edx,ecx
        rep movsb
        mov  esi,edi
        sub  esi,ebx
        jmp  L_do_copy1_mmx
 
L_wrap_around_window_mmx:
 
        mov  eax, [esp+48]
        cmp  ecx,eax
        jbe  L_contiguous_in_window_mmx
 
        add  esi, [esp+52]
        add  esi,eax
        sub  esi,ecx
        sub  ecx,eax
 
 
        cmp  edx,ecx
        jbe  L_do_copy1_mmx
 
        sub  edx,ecx
        rep movsb
        mov  esi, [esp+56]
        mov  ecx, [esp+48]
        cmp  edx,ecx
        jbe  L_do_copy1_mmx
 
        sub  edx,ecx
        rep movsb
        mov  esi,edi
        sub  esi,ebx
        jmp  L_do_copy1_mmx
 
L_contiguous_in_window_mmx:
 
        add  esi,eax
        sub  esi,ecx
 
 
        cmp  edx,ecx
        jbe  L_do_copy1_mmx
 
        sub  edx,ecx
        rep movsb
        mov  esi,edi
        sub  esi,ebx
 
L_do_copy1_mmx:
 
 
        mov  ecx,edx
        rep movsb
 
        mov  esi, [esp+44]
        mov  ebx, [esp+8]
        jmp  L_while_test_mmx
; 1174 "inffast.S"
L_invalid_distance_code:
 
 
 
 
 
        mov  ecx, invalid_distance_code_msg
        mov  edx,INFLATE_MODE_BAD
        jmp  L_update_stream_state
 
L_test_for_end_of_block:
 
 
 
 
 
        test  al,32
        jz  L_invalid_literal_length_code
 
        mov  ecx,0
        mov  edx,INFLATE_MODE_TYPE
        jmp  L_update_stream_state
 
L_invalid_literal_length_code:
 
 
 
 
 
        mov  ecx, invalid_literal_length_code_msg
        mov  edx,INFLATE_MODE_BAD
        jmp  L_update_stream_state
 
L_invalid_distance_too_far:
 
 
 
        mov  esi, [esp+44]
        mov  ecx, invalid_distance_too_far_msg
        mov  edx,INFLATE_MODE_BAD
        jmp  L_update_stream_state
 
L_update_stream_state:
 
        mov  eax, [esp+88]
        test  ecx,ecx
        jz  L_skip_msg
        mov  [eax+24],ecx
L_skip_msg:
        mov  eax, [eax+28]
        mov  [eax+mode_state],edx
        jmp  L_break_loop
 
ALIGN 4
L_break_loop:
; 1243 "inffast.S"
        cmp  dword ptr [inflate_fast_use_mmx],2
        jne  L_update_next_in
 
 
 
        mov  ebx,ebp
 
L_update_next_in:
; 1266 "inffast.S"
        mov  eax, [esp+88]
        mov  ecx,ebx
        mov  edx, [eax+28]
        shr  ecx,3
        sub  esi,ecx
        shl  ecx,3
        sub  ebx,ecx
        mov  [eax+12],edi
        mov  [edx+bits_state],ebx
        mov  ecx,ebx
 
        lea  ebx, [esp+28]
        cmp  [esp+20],ebx
        jne  L_buf_not_used
 
        sub  esi,ebx
        mov  ebx, [eax+0]
        mov  [esp+20],ebx
        add  esi,ebx
        mov  ebx, [eax+4]
        sub  ebx,11
        add  [esp+20],ebx
 
L_buf_not_used:
        mov  [eax+0],esi
 
        mov  ebx,1
        shl  ebx,cl
        dec  ebx
 
 
 
 
 
        cmp  dword ptr [inflate_fast_use_mmx],2
        jne  L_update_hold
 
 
 
        psrlq mm0,mm1
        movd ebp,mm0
 
        emms
 
L_update_hold:
 
 
 
        and  ebp,ebx
        mov  [edx+hold_state],ebp
 
 
 
 
        mov  ebx, [esp+20]
        cmp  ebx,esi
        jbe  L_last_is_smaller
 
        sub  ebx,esi
        add  ebx,11
        mov  [eax+4],ebx
        jmp  L_fixup_out
L_last_is_smaller:
        sub  esi,ebx
        neg  esi
        add  esi,11
        mov  [eax+4],esi
 
 
 
 
L_fixup_out:
 
        mov  ebx, [esp+16]
        cmp  ebx,edi
        jbe  L_end_is_smaller
 
        sub  ebx,edi
        add  ebx,257
        mov  [eax+16],ebx
        jmp  L_done
L_end_is_smaller:
        sub  edi,ebx
        neg  edi
        add  edi,257
        mov  [eax+16],edi
 
 
 
 
 
L_done:
        add  esp,64
        popfd
        pop  ebx
        pop  ebp
        pop  esi
        pop  edi
        ret
 
_TEXT   ends
end

Browse

Tools

Subversion Repositories openrisc

[/] [openrisc/] [trunk/] [gnu-dev/] [or1k-gcc/] [zlib/] [contrib/] [masmx86/] [inffas32.asm] - Blame information for rev 745

Line No.	Rev	Author	Line
1	745	jeremybenn	`;/* inffas32.asm is a hand tuned assembler version of inffast.c -- fast decoding`
2			`; *`
3			`; * inffas32.asm is derivated from inffas86.c, with translation of assembly code`
4			`; *`
5			`; * Copyright (C) 1995-2003 Mark Adler`
6			`; * For conditions of distribution and use, see copyright notice in zlib.h`
7			`; *`
8			`; * Copyright (C) 2003 Chris Anderson`
9			`; * Please use the copyright conditions above.`
10			`; *`
11			`; * Mar-13-2003 -- Most of this is derived from inffast.S which is derived from`
12			`; * the gcc -S output of zlib-1.2.0/inffast.c. Zlib-1.2.0 is in beta release at`
13			`; * the moment. I have successfully compiled and tested this code with gcc2.96,`
14			`; * gcc3.2, icc5.0, msvc6.0. It is very close to the speed of inffast.S`
15			`; * compiled with gcc -DNO_MMX, but inffast.S is still faster on the P3 with MMX`
16			`; * enabled. I will attempt to merge the MMX code into this version. Newer`
17			`; * versions of this and inffast.S can be found at`
18			`; * http://www.eetbeetee.com/zlib/ and http://www.charm.net/~christop/zlib/`
19			`; *`
20			`; * 2005 : modification by Gilles Vollant`
21			`; */`
22			`; For Visual C++ 4.x and higher and ML 6.x and higher`
23			`; ml.exe is in directory \MASM611C of Win95 DDK`
24			`; ml.exe is also distributed in http://www.masm32.com/masmdl.htm`
25			`; and in VC++2003 toolkit at http://msdn.microsoft.com/visualc/vctoolkit2003/`
26			`;`
27			`;`
28			`; compile with command line option`
29			`; ml /coff /Zi /c /Flinffas32.lst inffas32.asm`
30
31			`; if you define NO_GZIP (see inflate.h), compile with`
32			`; ml /coff /Zi /c /Flinffas32.lst /DNO_GUNZIP inffas32.asm`
33
34
35			`; zlib122sup is 0 fort zlib 1.2.2.1 and lower`
36			`; zlib122sup is 8 fort zlib 1.2.2.2 and more (with addition of dmax and head`
37			`; in inflate_state in inflate.h)`
38			`zlib1222sup equ 8`
39
40
41			`IFDEF GUNZIP`
42			`INFLATE_MODE_TYPE equ 11`
43			`INFLATE_MODE_BAD equ 26`
44			`ELSE`
45			`IFNDEF NO_GUNZIP`
46			`INFLATE_MODE_TYPE equ 11`
47			`INFLATE_MODE_BAD equ 26`
48			`ELSE`
49			`INFLATE_MODE_TYPE equ 3`
50			`INFLATE_MODE_BAD equ 17`
51			`ENDIF`
52			`ENDIF`
53
54
55			`; 75 "inffast.S"`
56			`;FILE "inffast.S"`
57
58			`;;;GLOBAL _inflate_fast`
59
60			`;;;SECTION .text`
61
62
63
64			`.586p`
65			`.mmx`
66
67			`name inflate_fast_x86`
68			`.MODEL FLAT`
69
70			`_DATA segment`
71			`inflate_fast_use_mmx:`
72			`dd 1`
73
74
75			`_TEXT segment`
76			`PUBLIC _inflate_fast`
77
78			`ALIGN 4`
79			`_inflate_fast:`
80			`jmp inflate_fast_entry`
81
82
83
84			`ALIGN 4`
85			`db 'Fast decoding Code from Chris Anderson'`
86			`db 0`
87
88			`ALIGN 4`
89			`invalid_literal_length_code_msg:`
90			`db 'invalid literal/length code'`
91			`db 0`
92
93			`ALIGN 4`
94			`invalid_distance_code_msg:`
95			`db 'invalid distance code'`
96			`db 0`
97
98			`ALIGN 4`
99			`invalid_distance_too_far_msg:`
100			`db 'invalid distance too far back'`
101			`db 0`
102
103
104			`ALIGN 4`
105			`inflate_fast_mask:`
106			`dd 0`
107			`dd 1`
108			`dd 3`
109			`dd 7`
110			`dd 15`
111			`dd 31`
112			`dd 63`
113			`dd 127`
114			`dd 255`
115			`dd 511`
116			`dd 1023`
117			`dd 2047`
118			`dd 4095`
119			`dd 8191`
120			`dd 16383`
121			`dd 32767`
122			`dd 65535`
123			`dd 131071`
124			`dd 262143`
125			`dd 524287`
126			`dd 1048575`
127			`dd 2097151`
128			`dd 4194303`
129			`dd 8388607`
130			`dd 16777215`
131			`dd 33554431`
132			`dd 67108863`
133			`dd 134217727`
134			`dd 268435455`
135			`dd 536870911`
136			`dd 1073741823`
137			`dd 2147483647`
138			`dd 4294967295`
139
140
141			`mode_state equ 0 ;/* state->mode */`
142			`wsize_state equ (32+zlib1222sup) ;/* state->wsize */`
143			`write_state equ (36+4+zlib1222sup) ;/* state->write */`
144			`window_state equ (40+4+zlib1222sup) ;/* state->window */`
145			`hold_state equ (44+4+zlib1222sup) ;/* state->hold */`
146			`bits_state equ (48+4+zlib1222sup) ;/* state->bits */`
147			`lencode_state equ (64+4+zlib1222sup) ;/* state->lencode */`
148			`distcode_state equ (68+4+zlib1222sup) ;/* state->distcode */`
149			`lenbits_state equ (72+4+zlib1222sup) ;/* state->lenbits */`
150			`distbits_state equ (76+4+zlib1222sup) ;/* state->distbits */`
151
152
153			`;;SECTION .text`
154			`; 205 "inffast.S"`
155			`;GLOBAL inflate_fast_use_mmx`
156
157			`;SECTION .data`
158
159
160			`; GLOBAL inflate_fast_use_mmx:object`
161			`;.size inflate_fast_use_mmx, 4`
162			`; 226 "inffast.S"`
163			`;SECTION .text`
164
165			`ALIGN 4`
166			`inflate_fast_entry:`
167			`push edi`
168			`push esi`
169			`push ebp`
170			`push ebx`
171			`pushfd`
172			`sub esp,64`
173			`cld`
174
175
176
177
178			`mov esi, [esp+88]`
179			`mov edi, [esi+28]`
180
181
182
183
184
185
186
187			`mov edx, [esi+4]`
188			`mov eax, [esi+0]`
189
190			`add edx,eax`
191			`sub edx,11`
192
193			`mov [esp+44],eax`
194			`mov [esp+20],edx`
195
196			`mov ebp, [esp+92]`
197			`mov ecx, [esi+16]`
198			`mov ebx, [esi+12]`
199
200			`sub ebp,ecx`
201			`neg ebp`
202			`add ebp,ebx`
203
204			`sub ecx,257`
205			`add ecx,ebx`
206
207			`mov [esp+60],ebx`
208			`mov [esp+40],ebp`
209			`mov [esp+16],ecx`
210			`; 285 "inffast.S"`
211			`mov eax, [edi+lencode_state]`
212			`mov ecx, [edi+distcode_state]`
213
214			`mov [esp+8],eax`
215			`mov [esp+12],ecx`
216
217			`mov eax,1`
218			`mov ecx, [edi+lenbits_state]`
219			`shl eax,cl`
220			`dec eax`
221			`mov [esp+0],eax`
222
223			`mov eax,1`
224			`mov ecx, [edi+distbits_state]`
225			`shl eax,cl`
226			`dec eax`
227			`mov [esp+4],eax`
228
229			`mov eax, [edi+wsize_state]`
230			`mov ecx, [edi+write_state]`
231			`mov edx, [edi+window_state]`
232
233			`mov [esp+52],eax`
234			`mov [esp+48],ecx`
235			`mov [esp+56],edx`
236
237			`mov ebp, [edi+hold_state]`
238			`mov ebx, [edi+bits_state]`
239			`; 321 "inffast.S"`
240			`mov esi, [esp+44]`
241			`mov ecx, [esp+20]`
242			`cmp ecx,esi`
243			`ja L_align_long`
244
245			`add ecx,11`
246			`sub ecx,esi`
247			`mov eax,12`
248			`sub eax,ecx`
249			`lea edi, [esp+28]`
250			`rep movsb`
251			`mov ecx,eax`
252			`xor eax,eax`
253			`rep stosb`
254			`lea esi, [esp+28]`
255			`mov [esp+20],esi`
256			`jmp L_is_aligned`
257
258
259			`L_align_long:`
260			`test esi,3`
261			`jz L_is_aligned`
262			`xor eax,eax`
263			`mov al, [esi]`
264			`inc esi`
265			`mov ecx,ebx`
266			`add ebx,8`
267			`shl eax,cl`
268			`or ebp,eax`
269			`jmp L_align_long`
270
271			`L_is_aligned:`
272			`mov edi, [esp+60]`
273			`; 366 "inffast.S"`
274			`L_check_mmx:`
275			`cmp dword ptr [inflate_fast_use_mmx],2`
276			`je L_init_mmx`
277			`ja L_do_loop`
278
279			`push eax`
280			`push ebx`
281			`push ecx`
282			`push edx`
283			`pushfd`
284			`mov eax, [esp]`
285			`xor dword ptr [esp],0200000h`
286
287
288
289
290			`popfd`
291			`pushfd`
292			`pop edx`
293			`xor edx,eax`
294			`jz L_dont_use_mmx`
295			`xor eax,eax`
296			`cpuid`
297			`cmp ebx,0756e6547h`
298			`jne L_dont_use_mmx`
299			`cmp ecx,06c65746eh`
300			`jne L_dont_use_mmx`
301			`cmp edx,049656e69h`
302			`jne L_dont_use_mmx`
303			`mov eax,1`
304			`cpuid`
305			`shr eax,8`
306			`and eax,15`
307			`cmp eax,6`
308			`jne L_dont_use_mmx`
309			`test edx,0800000h`
310			`jnz L_use_mmx`
311			`jmp L_dont_use_mmx`
312			`L_use_mmx:`
313			`mov dword ptr [inflate_fast_use_mmx],2`
314			`jmp L_check_mmx_pop`
315			`L_dont_use_mmx:`
316			`mov dword ptr [inflate_fast_use_mmx],3`
317			`L_check_mmx_pop:`
318			`pop edx`
319			`pop ecx`
320			`pop ebx`
321			`pop eax`
322			`jmp L_check_mmx`
323			`; 426 "inffast.S"`
324			`ALIGN 4`
325			`L_do_loop:`
326			`; 437 "inffast.S"`
327			`cmp bl,15`
328			`ja L_get_length_code`
329
330			`xor eax,eax`
331			`lodsw`
332			`mov cl,bl`
333			`add bl,16`
334			`shl eax,cl`
335			`or ebp,eax`
336
337			`L_get_length_code:`
338			`mov edx, [esp+0]`
339			`mov ecx, [esp+8]`
340			`and edx,ebp`
341			`mov eax, [ecx+edx*4]`
342
343			`L_dolen:`
344
345
346
347
348
349
350			`mov cl,ah`
351			`sub bl,ah`
352			`shr ebp,cl`
353
354
355
356
357
358
359			`test al,al`
360			`jnz L_test_for_length_base`
361
362			`shr eax,16`
363			`stosb`
364
365			`L_while_test:`
366
367
368			`cmp [esp+16],edi`
369			`jbe L_break_loop`
370
371			`cmp [esp+20],esi`
372			`ja L_do_loop`
373			`jmp L_break_loop`
374
375			`L_test_for_length_base:`
376			`; 502 "inffast.S"`
377			`mov edx,eax`
378			`shr edx,16`
379			`mov cl,al`
380
381			`test al,16`
382			`jz L_test_for_second_level_length`
383			`and cl,15`
384			`jz L_save_len`
385			`cmp bl,cl`
386			`jae L_add_bits_to_len`
387
388			`mov ch,cl`
389			`xor eax,eax`
390			`lodsw`
391			`mov cl,bl`
392			`add bl,16`
393			`shl eax,cl`
394			`or ebp,eax`
395			`mov cl,ch`
396
397			`L_add_bits_to_len:`
398			`mov eax,1`
399			`shl eax,cl`
400			`dec eax`
401			`sub bl,cl`
402			`and eax,ebp`
403			`shr ebp,cl`
404			`add edx,eax`
405
406			`L_save_len:`
407			`mov [esp+24],edx`
408
409
410			`L_decode_distance:`
411			`; 549 "inffast.S"`
412			`cmp bl,15`
413			`ja L_get_distance_code`
414
415			`xor eax,eax`
416			`lodsw`
417			`mov cl,bl`
418			`add bl,16`
419			`shl eax,cl`
420			`or ebp,eax`
421
422			`L_get_distance_code:`
423			`mov edx, [esp+4]`
424			`mov ecx, [esp+12]`
425			`and edx,ebp`
426			`mov eax, [ecx+edx*4]`
427
428
429			`L_dodist:`
430			`mov edx,eax`
431			`shr edx,16`
432			`mov cl,ah`
433			`sub bl,ah`
434			`shr ebp,cl`
435			`; 584 "inffast.S"`
436			`mov cl,al`
437
438			`test al,16`
439			`jz L_test_for_second_level_dist`
440			`and cl,15`
441			`jz L_check_dist_one`
442			`cmp bl,cl`
443			`jae L_add_bits_to_dist`
444
445			`mov ch,cl`
446			`xor eax,eax`
447			`lodsw`
448			`mov cl,bl`
449			`add bl,16`
450			`shl eax,cl`
451			`or ebp,eax`
452			`mov cl,ch`
453
454			`L_add_bits_to_dist:`
455			`mov eax,1`
456			`shl eax,cl`
457			`dec eax`
458			`sub bl,cl`
459			`and eax,ebp`
460			`shr ebp,cl`
461			`add edx,eax`
462			`jmp L_check_window`
463
464			`L_check_window:`
465			`; 625 "inffast.S"`
466			`mov [esp+44],esi`
467			`mov eax,edi`
468			`sub eax, [esp+40]`
469
470			`cmp eax,edx`
471			`jb L_clip_window`
472
473			`mov ecx, [esp+24]`
474			`mov esi,edi`
475			`sub esi,edx`
476
477			`sub ecx,3`
478			`mov al, [esi]`
479			`mov [edi],al`
480			`mov al, [esi+1]`
481			`mov dl, [esi+2]`
482			`add esi,3`
483			`mov [edi+1],al`
484			`mov [edi+2],dl`
485			`add edi,3`
486			`rep movsb`
487
488			`mov esi, [esp+44]`
489			`jmp L_while_test`
490
491			`ALIGN 4`
492			`L_check_dist_one:`
493			`cmp edx,1`
494			`jne L_check_window`
495			`cmp [esp+40],edi`
496			`je L_check_window`
497
498			`dec edi`
499			`mov ecx, [esp+24]`
500			`mov al, [edi]`
501			`sub ecx,3`
502
503			`mov [edi+1],al`
504			`mov [edi+2],al`
505			`mov [edi+3],al`
506			`add edi,4`
507			`rep stosb`
508
509			`jmp L_while_test`
510
511			`ALIGN 4`
512			`L_test_for_second_level_length:`
513
514
515
516
517			`test al,64`
518			`jnz L_test_for_end_of_block`
519
520			`mov eax,1`
521			`shl eax,cl`
522			`dec eax`
523			`and eax,ebp`
524			`add eax,edx`
525			`mov edx, [esp+8]`
526			`mov eax, [edx+eax*4]`
527			`jmp L_dolen`
528
529			`ALIGN 4`
530			`L_test_for_second_level_dist:`
531
532
533
534
535			`test al,64`
536			`jnz L_invalid_distance_code`
537
538			`mov eax,1`
539			`shl eax,cl`
540			`dec eax`
541			`and eax,ebp`
542			`add eax,edx`
543			`mov edx, [esp+12]`
544			`mov eax, [edx+eax*4]`
545			`jmp L_dodist`
546
547			`ALIGN 4`
548			`L_clip_window:`
549			`; 721 "inffast.S"`
550			`mov ecx,eax`
551			`mov eax, [esp+52]`
552			`neg ecx`
553			`mov esi, [esp+56]`
554
555			`cmp eax,edx`
556			`jb L_invalid_distance_too_far`
557
558			`add ecx,edx`
559			`cmp dword ptr [esp+48],0`
560			`jne L_wrap_around_window`
561
562			`sub eax,ecx`
563			`add esi,eax`
564			`; 749 "inffast.S"`
565			`mov eax, [esp+24]`
566			`cmp eax,ecx`
567			`jbe L_do_copy1`
568
569			`sub eax,ecx`
570			`rep movsb`
571			`mov esi,edi`
572			`sub esi,edx`
573			`jmp L_do_copy1`
574
575			`cmp eax,ecx`
576			`jbe L_do_copy1`
577
578			`sub eax,ecx`
579			`rep movsb`
580			`mov esi,edi`
581			`sub esi,edx`
582			`jmp L_do_copy1`
583
584			`L_wrap_around_window:`
585			`; 793 "inffast.S"`
586			`mov eax, [esp+48]`
587			`cmp ecx,eax`
588			`jbe L_contiguous_in_window`
589
590			`add esi, [esp+52]`
591			`add esi,eax`
592			`sub esi,ecx`
593			`sub ecx,eax`
594
595
596			`mov eax, [esp+24]`
597			`cmp eax,ecx`
598			`jbe L_do_copy1`
599
600			`sub eax,ecx`
601			`rep movsb`
602			`mov esi, [esp+56]`
603			`mov ecx, [esp+48]`
604			`cmp eax,ecx`
605			`jbe L_do_copy1`
606
607			`sub eax,ecx`
608			`rep movsb`
609			`mov esi,edi`
610			`sub esi,edx`
611			`jmp L_do_copy1`
612
613			`L_contiguous_in_window:`
614			`; 836 "inffast.S"`
615			`add esi,eax`
616			`sub esi,ecx`
617
618
619			`mov eax, [esp+24]`
620			`cmp eax,ecx`
621			`jbe L_do_copy1`
622
623			`sub eax,ecx`
624			`rep movsb`
625			`mov esi,edi`
626			`sub esi,edx`
627
628			`L_do_copy1:`
629			`; 862 "inffast.S"`
630			`mov ecx,eax`
631			`rep movsb`
632
633			`mov esi, [esp+44]`
634			`jmp L_while_test`
635			`; 878 "inffast.S"`
636			`ALIGN 4`
637			`L_init_mmx:`
638			`emms`
639
640
641
642
643
644			`movd mm0,ebp`
645			`mov ebp,ebx`
646			`; 896 "inffast.S"`
647			`movd mm4,[esp+0]`
648			`movq mm3,mm4`
649			`movd mm5,[esp+4]`
650			`movq mm2,mm5`
651			`pxor mm1,mm1`
652			`mov ebx, [esp+8]`
653			`jmp L_do_loop_mmx`
654
655			`ALIGN 4`
656			`L_do_loop_mmx:`
657			`psrlq mm0,mm1`
658
659			`cmp ebp,32`
660			`ja L_get_length_code_mmx`
661
662			`movd mm6,ebp`
663			`movd mm7,[esi]`
664			`add esi,4`
665			`psllq mm7,mm6`
666			`add ebp,32`
667			`por mm0,mm7`
668
669			`L_get_length_code_mmx:`
670			`pand mm4,mm0`
671			`movd eax,mm4`
672			`movq mm4,mm3`
673			`mov eax, [ebx+eax*4]`
674
675			`L_dolen_mmx:`
676			`movzx ecx,ah`
677			`movd mm1,ecx`
678			`sub ebp,ecx`
679
680			`test al,al`
681			`jnz L_test_for_length_base_mmx`
682
683			`shr eax,16`
684			`stosb`
685
686			`L_while_test_mmx:`
687
688
689			`cmp [esp+16],edi`
690			`jbe L_break_loop`
691
692			`cmp [esp+20],esi`
693			`ja L_do_loop_mmx`
694			`jmp L_break_loop`
695
696			`L_test_for_length_base_mmx:`
697
698			`mov edx,eax`
699			`shr edx,16`
700
701			`test al,16`
702			`jz L_test_for_second_level_length_mmx`
703			`and eax,15`
704			`jz L_decode_distance_mmx`
705
706			`psrlq mm0,mm1`
707			`movd mm1,eax`
708			`movd ecx,mm0`
709			`sub ebp,eax`
710			`and ecx, [inflate_fast_mask+eax*4]`
711			`add edx,ecx`
712
713			`L_decode_distance_mmx:`
714			`psrlq mm0,mm1`
715
716			`cmp ebp,32`
717			`ja L_get_dist_code_mmx`
718
719			`movd mm6,ebp`
720			`movd mm7,[esi]`
721			`add esi,4`
722			`psllq mm7,mm6`
723			`add ebp,32`
724			`por mm0,mm7`
725
726			`L_get_dist_code_mmx:`
727			`mov ebx, [esp+12]`
728			`pand mm5,mm0`
729			`movd eax,mm5`
730			`movq mm5,mm2`
731			`mov eax, [ebx+eax*4]`
732
733			`L_dodist_mmx:`
734
735			`movzx ecx,ah`
736			`mov ebx,eax`
737			`shr ebx,16`
738			`sub ebp,ecx`
739			`movd mm1,ecx`
740
741			`test al,16`
742			`jz L_test_for_second_level_dist_mmx`
743			`and eax,15`
744			`jz L_check_dist_one_mmx`
745
746			`L_add_bits_to_dist_mmx:`
747			`psrlq mm0,mm1`
748			`movd mm1,eax`
749			`movd ecx,mm0`
750			`sub ebp,eax`
751			`and ecx, [inflate_fast_mask+eax*4]`
752			`add ebx,ecx`
753
754			`L_check_window_mmx:`
755			`mov [esp+44],esi`
756			`mov eax,edi`
757			`sub eax, [esp+40]`
758
759			`cmp eax,ebx`
760			`jb L_clip_window_mmx`
761
762			`mov ecx,edx`
763			`mov esi,edi`
764			`sub esi,ebx`
765
766			`sub ecx,3`
767			`mov al, [esi]`
768			`mov [edi],al`
769			`mov al, [esi+1]`
770			`mov dl, [esi+2]`
771			`add esi,3`
772			`mov [edi+1],al`
773			`mov [edi+2],dl`
774			`add edi,3`
775			`rep movsb`
776
777			`mov esi, [esp+44]`
778			`mov ebx, [esp+8]`
779			`jmp L_while_test_mmx`
780
781			`ALIGN 4`
782			`L_check_dist_one_mmx:`
783			`cmp ebx,1`
784			`jne L_check_window_mmx`
785			`cmp [esp+40],edi`
786			`je L_check_window_mmx`
787
788			`dec edi`
789			`mov ecx,edx`
790			`mov al, [edi]`
791			`sub ecx,3`
792
793			`mov [edi+1],al`
794			`mov [edi+2],al`
795			`mov [edi+3],al`
796			`add edi,4`
797			`rep stosb`
798
799			`mov ebx, [esp+8]`
800			`jmp L_while_test_mmx`
801
802			`ALIGN 4`
803			`L_test_for_second_level_length_mmx:`
804			`test al,64`
805			`jnz L_test_for_end_of_block`
806
807			`and eax,15`
808			`psrlq mm0,mm1`
809			`movd ecx,mm0`
810			`and ecx, [inflate_fast_mask+eax*4]`
811			`add ecx,edx`
812			`mov eax, [ebx+ecx*4]`
813			`jmp L_dolen_mmx`
814
815			`ALIGN 4`
816			`L_test_for_second_level_dist_mmx:`
817			`test al,64`
818			`jnz L_invalid_distance_code`
819
820			`and eax,15`
821			`psrlq mm0,mm1`
822			`movd ecx,mm0`
823			`and ecx, [inflate_fast_mask+eax*4]`
824			`mov eax, [esp+12]`
825			`add ecx,ebx`
826			`mov eax, [eax+ecx*4]`
827			`jmp L_dodist_mmx`
828
829			`ALIGN 4`
830			`L_clip_window_mmx:`
831
832			`mov ecx,eax`
833			`mov eax, [esp+52]`
834			`neg ecx`
835			`mov esi, [esp+56]`
836
837			`cmp eax,ebx`
838			`jb L_invalid_distance_too_far`
839
840			`add ecx,ebx`
841			`cmp dword ptr [esp+48],0`
842			`jne L_wrap_around_window_mmx`
843
844			`sub eax,ecx`
845			`add esi,eax`
846
847			`cmp edx,ecx`
848			`jbe L_do_copy1_mmx`
849
850			`sub edx,ecx`
851			`rep movsb`
852			`mov esi,edi`
853			`sub esi,ebx`
854			`jmp L_do_copy1_mmx`
855
856			`cmp edx,ecx`
857			`jbe L_do_copy1_mmx`
858
859			`sub edx,ecx`
860			`rep movsb`
861			`mov esi,edi`
862			`sub esi,ebx`
863			`jmp L_do_copy1_mmx`
864
865			`L_wrap_around_window_mmx:`
866
867			`mov eax, [esp+48]`
868			`cmp ecx,eax`
869			`jbe L_contiguous_in_window_mmx`
870
871			`add esi, [esp+52]`
872			`add esi,eax`
873			`sub esi,ecx`
874			`sub ecx,eax`
875
876
877			`cmp edx,ecx`
878			`jbe L_do_copy1_mmx`
879
880			`sub edx,ecx`
881			`rep movsb`
882			`mov esi, [esp+56]`
883			`mov ecx, [esp+48]`
884			`cmp edx,ecx`
885			`jbe L_do_copy1_mmx`
886
887			`sub edx,ecx`
888			`rep movsb`
889			`mov esi,edi`
890			`sub esi,ebx`
891			`jmp L_do_copy1_mmx`
892
893			`L_contiguous_in_window_mmx:`
894
895			`add esi,eax`
896			`sub esi,ecx`
897
898
899			`cmp edx,ecx`
900			`jbe L_do_copy1_mmx`
901
902			`sub edx,ecx`
903			`rep movsb`
904			`mov esi,edi`
905			`sub esi,ebx`
906
907			`L_do_copy1_mmx:`
908
909
910			`mov ecx,edx`
911			`rep movsb`
912
913			`mov esi, [esp+44]`
914			`mov ebx, [esp+8]`
915			`jmp L_while_test_mmx`
916			`; 1174 "inffast.S"`
917			`L_invalid_distance_code:`
918
919
920
921
922
923			`mov ecx, invalid_distance_code_msg`
924			`mov edx,INFLATE_MODE_BAD`
925			`jmp L_update_stream_state`
926
927			`L_test_for_end_of_block:`
928
929
930
931
932
933			`test al,32`
934			`jz L_invalid_literal_length_code`
935
936			`mov ecx,0`
937			`mov edx,INFLATE_MODE_TYPE`
938			`jmp L_update_stream_state`
939
940			`L_invalid_literal_length_code:`
941
942
943
944
945
946			`mov ecx, invalid_literal_length_code_msg`
947			`mov edx,INFLATE_MODE_BAD`
948			`jmp L_update_stream_state`
949
950			`L_invalid_distance_too_far:`
951
952
953
954			`mov esi, [esp+44]`
955			`mov ecx, invalid_distance_too_far_msg`
956			`mov edx,INFLATE_MODE_BAD`
957			`jmp L_update_stream_state`
958
959			`L_update_stream_state:`
960
961			`mov eax, [esp+88]`
962			`test ecx,ecx`
963			`jz L_skip_msg`
964			`mov [eax+24],ecx`
965			`L_skip_msg:`
966			`mov eax, [eax+28]`
967			`mov [eax+mode_state],edx`
968			`jmp L_break_loop`
969
970			`ALIGN 4`
971			`L_break_loop:`
972			`; 1243 "inffast.S"`
973			`cmp dword ptr [inflate_fast_use_mmx],2`
974			`jne L_update_next_in`
975
976
977
978			`mov ebx,ebp`
979
980			`L_update_next_in:`
981			`; 1266 "inffast.S"`
982			`mov eax, [esp+88]`
983			`mov ecx,ebx`
984			`mov edx, [eax+28]`
985			`shr ecx,3`
986			`sub esi,ecx`
987			`shl ecx,3`
988			`sub ebx,ecx`
989			`mov [eax+12],edi`
990			`mov [edx+bits_state],ebx`
991			`mov ecx,ebx`
992
993			`lea ebx, [esp+28]`
994			`cmp [esp+20],ebx`
995			`jne L_buf_not_used`
996
997			`sub esi,ebx`
998			`mov ebx, [eax+0]`
999			`mov [esp+20],ebx`
1000			`add esi,ebx`
1001			`mov ebx, [eax+4]`
1002			`sub ebx,11`
1003			`add [esp+20],ebx`
1004
1005			`L_buf_not_used:`
1006			`mov [eax+0],esi`
1007
1008			`mov ebx,1`
1009			`shl ebx,cl`
1010			`dec ebx`
1011
1012
1013
1014
1015
1016			`cmp dword ptr [inflate_fast_use_mmx],2`
1017			`jne L_update_hold`
1018
1019
1020
1021			`psrlq mm0,mm1`
1022			`movd ebp,mm0`
1023
1024			`emms`
1025
1026			`L_update_hold:`
1027
1028
1029
1030			`and ebp,ebx`
1031			`mov [edx+hold_state],ebp`
1032
1033
1034
1035
1036			`mov ebx, [esp+20]`
1037			`cmp ebx,esi`
1038			`jbe L_last_is_smaller`
1039
1040			`sub ebx,esi`
1041			`add ebx,11`
1042			`mov [eax+4],ebx`
1043			`jmp L_fixup_out`
1044			`L_last_is_smaller:`
1045			`sub esi,ebx`
1046			`neg esi`
1047			`add esi,11`
1048			`mov [eax+4],esi`
1049
1050
1051
1052
1053			`L_fixup_out:`
1054
1055			`mov ebx, [esp+16]`
1056			`cmp ebx,edi`
1057			`jbe L_end_is_smaller`
1058
1059			`sub ebx,edi`
1060			`add ebx,257`
1061			`mov [eax+16],ebx`
1062			`jmp L_done`
1063			`L_end_is_smaller:`
1064			`sub edi,ebx`
1065			`neg edi`
1066			`add edi,257`
1067			`mov [eax+16],edi`
1068
1069
1070
1071
1072
1073			`L_done:`
1074			`add esp,64`
1075			`popfd`
1076			`pop ebx`
1077			`pop ebp`
1078			`pop esi`
1079			`pop edi`
1080			`ret`
1081
1082			`_TEXT ends`
1083			`end`