common/dist/zlib/contrib/masmx64/inffas8664.c

   1 /*      $NetBSD: inffas8664.c,v 1.1.1.1 2006/01/14 20:10:55 christos Exp $      */
   2
   3 /* inffas8664.c is a hand tuned assembler version of inffast.c - fast decoding
   4  * version for AMD64 on Windows using Microsoft C compiler
   5  *
   6  * Copyright (C) 1995-2003 Mark Adler
   7  * For conditions of distribution and use, see copyright notice in zlib.h
   8  *
   9  * Copyright (C) 2003 Chris Anderson <christop@charm.net>
  10  * Please use the copyright conditions above.
  11  *
  12  * 2005 - Adaptation to Microsoft C Compiler for AMD64 by Gilles Vollant
  13  *
  14  * inffas8664.c call function inffas8664fnc in inffasx64.asm
  15  *  inffasx64.asm is automatically convert from AMD64 portion of inffas86.c
  16  *
  17  * Dec-29-2003 -- I added AMD64 inflate asm support.  This version is also
  18  * slightly quicker on x86 systems because, instead of using rep movsb to copy
  19  * data, it uses rep movsw, which moves data in 2-byte chunks instead of single
  20  * bytes.  I've tested the AMD64 code on a Fedora Core 1 + the x86_64 updates
  21  * from http://fedora.linux.duke.edu/fc1_x86_64
  22  * which is running on an Athlon 64 3000+ / Gigabyte GA-K8VT800M system with
  23  * 1GB ram.  The 64-bit version is about 4% faster than the 32-bit version,
  24  * when decompressing mozilla-source-1.3.tar.gz.
  25  *
  26  * Mar-13-2003 -- Most of this is derived from inffast.S which is derived from
  27  * the gcc -S output of zlib-1.2.0/inffast.c.  Zlib-1.2.0 is in beta release at
  28  * the moment.  I have successfully compiled and tested this code with gcc2.96,
  29  * gcc3.2, icc5.0, msvc6.0.  It is very close to the speed of inffast.S
  30  * compiled with gcc -DNO_MMX, but inffast.S is still faster on the P3 with MMX
  31  * enabled.  I will attempt to merge the MMX code into this version.  Newer
  32  * versions of this and inffast.S can be found at
  33  * http://www.eetbeetee.com/zlib/ and http://www.charm.net/~christop/zlib/
  34  *
  35  */
  36
  37 #include <stdio.h>
  38 #include "zutil.h"
  39 #include "inftrees.h"
  40 #include "inflate.h"
  41 #include "inffast.h"
  42
  43 /* Mark Adler's comments from inffast.c: */
  44
  45 /*
  46    Decode literal, length, and distance codes and write out the resulting
  47    literal and match bytes until either not enough input or output is
  48    available, an end-of-block is encountered, or a data error is encountered.
  49    When large enough input and output buffers are supplied to inflate(), for
  50    example, a 16K input buffer and a 64K output buffer, more than 95% of the
  51    inflate execution time is spent in this routine.
  52
  53    Entry assumptions:
  54
  55         state->mode == LEN
  56         strm->avail_in >= 6
  57         strm->avail_out >= 258
  58         start >= strm->avail_out
  59         state->bits < 8
  60
  61    On return, state->mode is one of:
  62
  63         LEN -- ran out of enough output space or enough available input
  64         TYPE -- reached end of block code, inflate() to interpret next block
  65         BAD -- error in block data
  66
  67    Notes:
  68
  69     - The maximum input bits used by a length/distance pair is 15 bits for the
  70       length code, 5 bits for the length extra, 15 bits for the distance code,
  71       and 13 bits for the distance extra.  This totals 48 bits, or six bytes.
  72       Therefore if strm->avail_in >= 6, then there is enough input to avoid
  73       checking for available input while decoding.
  74
  75     - The maximum bytes that a single length/distance pair can output is 258
  76       bytes, which is the maximum length that can be coded.  inflate_fast()
  77       requires strm->avail_out >= 258 for each loop to avoid checking for
  78       output space.
  79  */
  80
  81
  82
  83     typedef struct inffast_ar {
  84 /* 64   32                               x86  x86_64 */
  85 /* ar offset                              register */
  86 /*  0    0 */ void *esp;                /* esp save */
  87 /*  8    4 */ void *ebp;                /* ebp save */
  88 /* 16    8 */ unsigned char FAR *in;    /* esi rsi  local strm->next_in */
  89 /* 24   12 */ unsigned char FAR *last;  /*     r9   while in < last */
  90 /* 32   16 */ unsigned char FAR *out;   /* edi rdi  local strm->next_out */
  91 /* 40   20 */ unsigned char FAR *beg;   /*          inflate()'s init next_out */
  92 /* 48   24 */ unsigned char FAR *end;   /*     r10  while out < end */
  93 /* 56   28 */ unsigned char FAR *window;/*          size of window, wsize!=0 */
  94 /* 64   32 */ code const FAR *lcode;    /* ebp rbp  local strm->lencode */
  95 /* 72   36 */ code const FAR *dcode;    /*     r11  local strm->distcode */
  96 /* 80   40 */ size_t /*unsigned long */hold;       /* edx rdx  local strm->hold */
  97 /* 88   44 */ unsigned bits;            /* ebx rbx  local strm->bits */
  98 /* 92   48 */ unsigned wsize;           /*          window size */
  99 /* 96   52 */ unsigned write;           /*          window write index */
 100 /*100   56 */ unsigned lmask;           /*     r12  mask for lcode */
 101 /*104   60 */ unsigned dmask;           /*     r13  mask for dcode */
 102 /*108   64 */ unsigned len;             /*     r14  match length */
 103 /*112   68 */ unsigned dist;            /*     r15  match distance */
 104 /*116   72 */ unsigned status;          /*          set when state chng*/
 105     } type_ar;
 106 #ifdef ASMINF
 107
 108 void inflate_fast(strm, start)
 109 z_streamp strm;
 110 unsigned start;         /* inflate()'s starting value for strm->avail_out */
 111 {
 112     struct inflate_state FAR *state;
 113     type_ar ar;
 114     void inffas8664fnc(struct inffast_ar * par);
 115
 116
 117
 118 #if (defined( __GNUC__ ) && defined( __amd64__ ) && ! defined( __i386 )) || (defined(_MSC_VER) && defined(_M_AMD64))
 119 #define PAD_AVAIL_IN 6
 120 #define PAD_AVAIL_OUT 258
 121 #else
 122 #define PAD_AVAIL_IN 5
 123 #define PAD_AVAIL_OUT 257
 124 #endif
 125
 126     /* copy state to local variables */
 127     state = (struct inflate_state FAR *)strm->state;
 128
 129     ar.in = strm->next_in;
 130     ar.last = ar.in + (strm->avail_in - PAD_AVAIL_IN);
 131     ar.out = strm->next_out;
 132     ar.beg = ar.out - (start - strm->avail_out);
 133     ar.end = ar.out + (strm->avail_out - PAD_AVAIL_OUT);
 134     ar.wsize = state->wsize;
 135     ar.write = state->write;
 136     ar.window = state->window;
 137     ar.hold = state->hold;
 138     ar.bits = state->bits;
 139     ar.lcode = state->lencode;
 140     ar.dcode = state->distcode;
 141     ar.lmask = (1U << state->lenbits) - 1;
 142     ar.dmask = (1U << state->distbits) - 1;
 143
 144     /* decode literals and length/distances until end-of-block or not enough
 145        input data or output space */
 146
 147     /* align in on 1/2 hold size boundary */
 148     while (((size_t)(void *)ar.in & (sizeof(ar.hold) / 2 - 1)) != 0) {
 149         ar.hold += (unsigned long)*ar.in++ << ar.bits;
 150         ar.bits += 8;
 151     }
 152
 153     inffas8664fnc(&ar);
 154
 155     if (ar.status > 1) {
 156         if (ar.status == 2)
 157             strm->msg = "invalid literal/length code";
 158         else if (ar.status == 3)
 159             strm->msg = "invalid distance code";
 160         else
 161             strm->msg = "invalid distance too far back";
 162         state->mode = BAD;
 163     }
 164     else if ( ar.status == 1 ) {
 165         state->mode = TYPE;
 166     }
 167
 168     /* return unused bytes (on entry, bits < 8, so in won't go too far back) */
 169     ar.len = ar.bits >> 3;
 170     ar.in -= ar.len;
 171     ar.bits -= ar.len << 3;
 172     ar.hold &= (1U << ar.bits) - 1;
 173
 174     /* update state and return */
 175     strm->next_in = ar.in;
 176     strm->next_out = ar.out;
 177     strm->avail_in = (unsigned)(ar.in < ar.last ?
 178                                 PAD_AVAIL_IN + (ar.last - ar.in) :
 179                                 PAD_AVAIL_IN - (ar.in - ar.last));
 180     strm->avail_out = (unsigned)(ar.out < ar.end ?
 181                                  PAD_AVAIL_OUT + (ar.end - ar.out) :
 182                                  PAD_AVAIL_OUT - (ar.out - ar.end));
 183     state->hold = (unsigned long)ar.hold;
 184     state->bits = ar.bits;
 185     return;
 186 }
 187
 188 #endif