Chris@4: /* inffas8664.c is a hand tuned assembler version of inffast.c - fast decoding Chris@4: * version for AMD64 on Windows using Microsoft C compiler Chris@4: * Chris@4: * Copyright (C) 1995-2003 Mark Adler Chris@4: * For conditions of distribution and use, see copyright notice in zlib.h Chris@4: * Chris@4: * Copyright (C) 2003 Chris Anderson Chris@4: * Please use the copyright conditions above. Chris@4: * Chris@4: * 2005 - Adaptation to Microsoft C Compiler for AMD64 by Gilles Vollant Chris@4: * Chris@4: * inffas8664.c call function inffas8664fnc in inffasx64.asm Chris@4: * inffasx64.asm is automatically convert from AMD64 portion of inffas86.c Chris@4: * Chris@4: * Dec-29-2003 -- I added AMD64 inflate asm support. This version is also Chris@4: * slightly quicker on x86 systems because, instead of using rep movsb to copy Chris@4: * data, it uses rep movsw, which moves data in 2-byte chunks instead of single Chris@4: * bytes. I've tested the AMD64 code on a Fedora Core 1 + the x86_64 updates Chris@4: * from http://fedora.linux.duke.edu/fc1_x86_64 Chris@4: * which is running on an Athlon 64 3000+ / Gigabyte GA-K8VT800M system with Chris@4: * 1GB ram. The 64-bit version is about 4% faster than the 32-bit version, Chris@4: * when decompressing mozilla-source-1.3.tar.gz. Chris@4: * Chris@4: * Mar-13-2003 -- Most of this is derived from inffast.S which is derived from Chris@4: * the gcc -S output of zlib-1.2.0/inffast.c. Zlib-1.2.0 is in beta release at Chris@4: * the moment. I have successfully compiled and tested this code with gcc2.96, Chris@4: * gcc3.2, icc5.0, msvc6.0. It is very close to the speed of inffast.S Chris@4: * compiled with gcc -DNO_MMX, but inffast.S is still faster on the P3 with MMX Chris@4: * enabled. I will attempt to merge the MMX code into this version. Newer Chris@4: * versions of this and inffast.S can be found at Chris@4: * http://www.eetbeetee.com/zlib/ and http://www.charm.net/~christop/zlib/ Chris@4: * Chris@4: */ Chris@4: Chris@4: #include Chris@4: #include "zutil.h" Chris@4: #include "inftrees.h" Chris@4: #include "inflate.h" Chris@4: #include "inffast.h" Chris@4: Chris@4: /* Mark Adler's comments from inffast.c: */ Chris@4: Chris@4: /* Chris@4: Decode literal, length, and distance codes and write out the resulting Chris@4: literal and match bytes until either not enough input or output is Chris@4: available, an end-of-block is encountered, or a data error is encountered. Chris@4: When large enough input and output buffers are supplied to inflate(), for Chris@4: example, a 16K input buffer and a 64K output buffer, more than 95% of the Chris@4: inflate execution time is spent in this routine. Chris@4: Chris@4: Entry assumptions: Chris@4: Chris@4: state->mode == LEN Chris@4: strm->avail_in >= 6 Chris@4: strm->avail_out >= 258 Chris@4: start >= strm->avail_out Chris@4: state->bits < 8 Chris@4: Chris@4: On return, state->mode is one of: Chris@4: Chris@4: LEN -- ran out of enough output space or enough available input Chris@4: TYPE -- reached end of block code, inflate() to interpret next block Chris@4: BAD -- error in block data Chris@4: Chris@4: Notes: Chris@4: Chris@4: - The maximum input bits used by a length/distance pair is 15 bits for the Chris@4: length code, 5 bits for the length extra, 15 bits for the distance code, Chris@4: and 13 bits for the distance extra. This totals 48 bits, or six bytes. Chris@4: Therefore if strm->avail_in >= 6, then there is enough input to avoid Chris@4: checking for available input while decoding. Chris@4: Chris@4: - The maximum bytes that a single length/distance pair can output is 258 Chris@4: bytes, which is the maximum length that can be coded. inflate_fast() Chris@4: requires strm->avail_out >= 258 for each loop to avoid checking for Chris@4: output space. Chris@4: */ Chris@4: Chris@4: Chris@4: Chris@4: typedef struct inffast_ar { Chris@4: /* 64 32 x86 x86_64 */ Chris@4: /* ar offset register */ Chris@4: /* 0 0 */ void *esp; /* esp save */ Chris@4: /* 8 4 */ void *ebp; /* ebp save */ Chris@4: /* 16 8 */ unsigned char FAR *in; /* esi rsi local strm->next_in */ Chris@4: /* 24 12 */ unsigned char FAR *last; /* r9 while in < last */ Chris@4: /* 32 16 */ unsigned char FAR *out; /* edi rdi local strm->next_out */ Chris@4: /* 40 20 */ unsigned char FAR *beg; /* inflate()'s init next_out */ Chris@4: /* 48 24 */ unsigned char FAR *end; /* r10 while out < end */ Chris@4: /* 56 28 */ unsigned char FAR *window;/* size of window, wsize!=0 */ Chris@4: /* 64 32 */ code const FAR *lcode; /* ebp rbp local strm->lencode */ Chris@4: /* 72 36 */ code const FAR *dcode; /* r11 local strm->distcode */ Chris@4: /* 80 40 */ size_t /*unsigned long */hold; /* edx rdx local strm->hold */ Chris@4: /* 88 44 */ unsigned bits; /* ebx rbx local strm->bits */ Chris@4: /* 92 48 */ unsigned wsize; /* window size */ Chris@4: /* 96 52 */ unsigned write; /* window write index */ Chris@4: /*100 56 */ unsigned lmask; /* r12 mask for lcode */ Chris@4: /*104 60 */ unsigned dmask; /* r13 mask for dcode */ Chris@4: /*108 64 */ unsigned len; /* r14 match length */ Chris@4: /*112 68 */ unsigned dist; /* r15 match distance */ Chris@4: /*116 72 */ unsigned status; /* set when state chng*/ Chris@4: } type_ar; Chris@4: #ifdef ASMINF Chris@4: Chris@4: void inflate_fast(strm, start) Chris@4: z_streamp strm; Chris@4: unsigned start; /* inflate()'s starting value for strm->avail_out */ Chris@4: { Chris@4: struct inflate_state FAR *state; Chris@4: type_ar ar; Chris@4: void inffas8664fnc(struct inffast_ar * par); Chris@4: Chris@4: Chris@4: Chris@4: #if (defined( __GNUC__ ) && defined( __amd64__ ) && ! defined( __i386 )) || (defined(_MSC_VER) && defined(_M_AMD64)) Chris@4: #define PAD_AVAIL_IN 6 Chris@4: #define PAD_AVAIL_OUT 258 Chris@4: #else Chris@4: #define PAD_AVAIL_IN 5 Chris@4: #define PAD_AVAIL_OUT 257 Chris@4: #endif Chris@4: Chris@4: /* copy state to local variables */ Chris@4: state = (struct inflate_state FAR *)strm->state; Chris@4: Chris@4: ar.in = strm->next_in; Chris@4: ar.last = ar.in + (strm->avail_in - PAD_AVAIL_IN); Chris@4: ar.out = strm->next_out; Chris@4: ar.beg = ar.out - (start - strm->avail_out); Chris@4: ar.end = ar.out + (strm->avail_out - PAD_AVAIL_OUT); Chris@4: ar.wsize = state->wsize; Chris@4: ar.write = state->wnext; Chris@4: ar.window = state->window; Chris@4: ar.hold = state->hold; Chris@4: ar.bits = state->bits; Chris@4: ar.lcode = state->lencode; Chris@4: ar.dcode = state->distcode; Chris@4: ar.lmask = (1U << state->lenbits) - 1; Chris@4: ar.dmask = (1U << state->distbits) - 1; Chris@4: Chris@4: /* decode literals and length/distances until end-of-block or not enough Chris@4: input data or output space */ Chris@4: Chris@4: /* align in on 1/2 hold size boundary */ Chris@4: while (((size_t)(void *)ar.in & (sizeof(ar.hold) / 2 - 1)) != 0) { Chris@4: ar.hold += (unsigned long)*ar.in++ << ar.bits; Chris@4: ar.bits += 8; Chris@4: } Chris@4: Chris@4: inffas8664fnc(&ar); Chris@4: Chris@4: if (ar.status > 1) { Chris@4: if (ar.status == 2) Chris@4: strm->msg = "invalid literal/length code"; Chris@4: else if (ar.status == 3) Chris@4: strm->msg = "invalid distance code"; Chris@4: else Chris@4: strm->msg = "invalid distance too far back"; Chris@4: state->mode = BAD; Chris@4: } Chris@4: else if ( ar.status == 1 ) { Chris@4: state->mode = TYPE; Chris@4: } Chris@4: Chris@4: /* return unused bytes (on entry, bits < 8, so in won't go too far back) */ Chris@4: ar.len = ar.bits >> 3; Chris@4: ar.in -= ar.len; Chris@4: ar.bits -= ar.len << 3; Chris@4: ar.hold &= (1U << ar.bits) - 1; Chris@4: Chris@4: /* update state and return */ Chris@4: strm->next_in = ar.in; Chris@4: strm->next_out = ar.out; Chris@4: strm->avail_in = (unsigned)(ar.in < ar.last ? Chris@4: PAD_AVAIL_IN + (ar.last - ar.in) : Chris@4: PAD_AVAIL_IN - (ar.in - ar.last)); Chris@4: strm->avail_out = (unsigned)(ar.out < ar.end ? Chris@4: PAD_AVAIL_OUT + (ar.end - ar.out) : Chris@4: PAD_AVAIL_OUT - (ar.out - ar.end)); Chris@4: state->hold = (unsigned long)ar.hold; Chris@4: state->bits = ar.bits; Chris@4: return; Chris@4: } Chris@4: Chris@4: #endif