X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=include%2Fwimlib%2Fdecompress_common.h;h=a06ede1457359efe7a16de0573969f0418c6ffcc;hp=6762043c8b0ab30081c51b83454fd76820d27ecb;hb=32158cb5b4df58eb71a1986762e5aaf12bce9d30;hpb=0ecb0529b5fcacc1abafa1f3f02a40c44783ada8 diff --git a/include/wimlib/decompress_common.h b/include/wimlib/decompress_common.h index 6762043c..a06ede14 100644 --- a/include/wimlib/decompress_common.h +++ b/include/wimlib/decompress_common.h @@ -3,16 +3,29 @@ * * Header for decompression code shared by multiple compression formats. * - * The author dedicates this file to the public domain. - * You can do whatever you want with this file. + * The following copying information applies to this specific source code file: + * + * Written in 2012-2016 by Eric Biggers + * + * To the extent possible under law, the author(s) have dedicated all copyright + * and related and neighboring rights to this software to the public domain + * worldwide via the Creative Commons Zero 1.0 Universal Public Domain + * Dedication (the "CC0"). + * + * This software is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS + * FOR A PARTICULAR PURPOSE. See the CC0 for more details. + * + * You should have received a copy of the CC0 along with this software; if not + * see . */ #ifndef _WIMLIB_DECOMPRESS_COMMON_H #define _WIMLIB_DECOMPRESS_COMMON_H -#include "wimlib/assert.h" +#include + #include "wimlib/compiler.h" -#include "wimlib/endianness.h" #include "wimlib/types.h" #include "wimlib/unaligned.h" @@ -60,7 +73,6 @@ static inline void bitstream_ensure_bits(struct input_bitstream *is, const unsigned num_bits) { /* This currently works for at most 17 bits. */ - wimlib_assert2(num_bits <= 17); if (is->bitsleft >= num_bits) return; @@ -68,7 +80,7 @@ bitstream_ensure_bits(struct input_bitstream *is, const unsigned num_bits) if (unlikely(is->end - is->next < 2)) goto overflow; - is->bitbuf |= (u32)get_unaligned_u16_le(is->next) << (16 - is->bitsleft); + is->bitbuf |= (u32)get_unaligned_le16(is->next) << (16 - is->bitsleft); is->next += 2; is->bitsleft += 16; @@ -76,7 +88,7 @@ bitstream_ensure_bits(struct input_bitstream *is, const unsigned num_bits) if (unlikely(is->end - is->next < 2)) goto overflow; - is->bitbuf |= (u32)get_unaligned_u16_le(is->next); + is->bitbuf |= (u32)get_unaligned_le16(is->next); is->next += 2; is->bitsleft = 32; } @@ -93,9 +105,7 @@ overflow: static inline u32 bitstream_peek_bits(const struct input_bitstream *is, const unsigned num_bits) { - if (unlikely(num_bits == 0)) - return 0; - return is->bitbuf >> (32 - num_bits); + return (is->bitbuf >> 1) >> (sizeof(is->bitbuf) * 8 - num_bits - 1); } /* Remove @num_bits from the bitstream. There must be at least @num_bits @@ -144,7 +154,7 @@ bitstream_read_u16(struct input_bitstream *is) if (unlikely(is->end - is->next < 2)) return 0; - v = get_unaligned_u16_le(is->next); + v = get_unaligned_le16(is->next); is->next += 2; return v; } @@ -157,23 +167,22 @@ bitstream_read_u32(struct input_bitstream *is) if (unlikely(is->end - is->next < 4)) return 0; - v = get_unaligned_u32_le(is->next); + v = get_unaligned_le32(is->next); is->next += 4; return v; } -/* Read an array of literal bytes embedded in the bitstream. Return a pointer - * to the resulting array, or NULL if the read overflows the input buffer. */ -static inline const u8 * -bitstream_read_bytes(struct input_bitstream *is, size_t count) +/* Read into @dst_buffer an array of literal bytes embedded in the bitstream. + * Return either a pointer to the byte past the last written, or NULL if the + * read overflows the input buffer. */ +static inline void * +bitstream_read_bytes(struct input_bitstream *is, void *dst_buffer, size_t count) { - const u8 *p; - if (unlikely(is->end - is->next < count)) return NULL; - p = is->next; + memcpy(dst_buffer, is->next, count); is->next += count; - return p; + return (u8 *)dst_buffer + count; } /* Align the input bitstream on a coding-unit boundary. */ @@ -215,9 +224,9 @@ bitstream_align(struct input_bitstream *is) * input data is exhausted, the Huffman symbol is decoded as if the missing bits * are all zeroes. * - * XXX: This is mostly duplicated in lzms_huffman_decode_symbol() in - * lzms-decompress.c. */ -static inline u16 + * XXX: This is mostly duplicated in lzms_decode_huffman_symbol() in + * lzms_decompress.c. */ +static inline unsigned read_huffsym(struct input_bitstream *istream, const u16 decode_table[], unsigned table_bits, unsigned max_codeword_len) { @@ -254,6 +263,25 @@ make_huffman_decode_table(u16 decode_table[], unsigned num_syms, unsigned num_bits, const u8 lens[], unsigned max_codeword_len); +static inline void +copy_word_unaligned(const void *src, void *dst) +{ + store_word_unaligned(load_word_unaligned(src), dst); +} + +static inline machine_word_t +repeat_byte(u8 b) +{ + machine_word_t v; + + STATIC_ASSERT(WORDBITS == 32 || WORDBITS == 64); + + v = b; + v |= v << 8; + v |= v << 16; + v |= v << ((WORDBITS == 64) ? 32 : 0); + return v; +} /* * Copy an LZ77 match at (dst - offset) to dst. @@ -282,13 +310,11 @@ lz_copy(u8 *dst, u32 length, u32 offset, const u8 *winend, u32 min_length) * example, if a word is 8 bytes and the match is of length 5, then * we'll simply copy 8 bytes. This is okay as long as we don't write * beyond the end of the output buffer, hence the check for (winend - - * end >= WORDSIZE - 1). + * end >= WORDBYTES - 1). */ - if (UNALIGNED_ACCESS_IS_VERY_FAST && - likely(winend - end >= WORDSIZE - 1)) - { + if (UNALIGNED_ACCESS_IS_FAST && likely(winend - end >= WORDBYTES - 1)) { - if (offset >= WORDSIZE) { + if (offset >= WORDBYTES) { /* The source and destination words don't overlap. */ /* To improve branch prediction, one iteration of this @@ -298,14 +324,14 @@ lz_copy(u8 *dst, u32 length, u32 offset, const u8 *winend, u32 min_length) * and we'll need to continue copying. */ copy_word_unaligned(src, dst); - src += WORDSIZE; - dst += WORDSIZE; + src += WORDBYTES; + dst += WORDBYTES; if (dst < end) { do { copy_word_unaligned(src, dst); - src += WORDSIZE; - dst += WORDSIZE; + src += WORDBYTES; + dst += WORDBYTES; } while (dst < end); } return; @@ -318,19 +344,19 @@ lz_copy(u8 *dst, u32 length, u32 offset, const u8 *winend, u32 min_length) machine_word_t v = repeat_byte(*(dst - 1)); do { store_word_unaligned(v, dst); - src += WORDSIZE; - dst += WORDSIZE; + src += WORDBYTES; + dst += WORDBYTES; } while (dst < end); return; } /* * We don't bother with special cases for other 'offset < - * WORDSIZE', which are usually rarer than 'offset == 1'. Extra - * checks will just slow things down. Actually, it's possible - * to handle all the 'offset < WORDSIZE' cases using the same - * code, but it still becomes more complicated doesn't seem any - * faster overall; it definitely slows down the more common - * 'offset == 1' case. + * WORDBYTES', which are usually rarer than 'offset == 1'. + * Extra checks will just slow things down. Actually, it's + * possible to handle all the 'offset < WORDBYTES' cases using + * the same code, but it still becomes more complicated doesn't + * seem any faster overall; it definitely slows down the more + * common 'offset == 1' case. */ }