From: Kostya Shishkov Date: Fri, 23 Dec 2022 07:39:09 +0000 (+0100) Subject: core/compr: make a faster version of Inflate::uncompress() X-Git-Url: https://git.nihav.org/?a=commitdiff_plain;h=7bf82b3370b7a5579ebc62a1b003f4bf2e3515f8;p=nihav.git core/compr: make a faster version of Inflate::uncompress() --- diff --git a/nihav-core/src/compr/deflate.rs b/nihav-core/src/compr/deflate.rs index 71e0dfe..8fc6a13 100644 --- a/nihav-core/src/compr/deflate.rs +++ b/nihav-core/src/compr/deflate.rs @@ -172,6 +172,23 @@ impl<'a> CurrentSource<'a> { } Ok(()) } + fn skip_bytes(&mut self, nbytes: usize) -> BitReaderResult<()> { + self.align(); + let cached = usize::from(self.br.bits / 8); + if nbytes <= cached { + self.skip((nbytes as u32) * 8)?; + } else { + self.skip((cached as u32) * 8)?; + self.br.bits = 0; + self.br.bitbuf = 0; + self.br.pos += nbytes - cached; + if self.br.pos > self.src.len() { + return Err(BitReaderError::BitstreamEnd); + } + self.refill(); + } + Ok(()) + } fn align(&mut self) { let b = self.br.bits & 7; if b != 0 { @@ -181,6 +198,9 @@ impl<'a> CurrentSource<'a> { fn left(&self) -> isize { ((self.src.len() as isize) - (self.br.pos as isize)) * 8 + (self.br.bits as isize) } + fn tell(&self) -> usize { + self.br.pos - usize::from(self.br.bits / 8) + } } impl<'a, S: Copy> CodebookReader for CurrentSource<'a> { @@ -772,8 +792,210 @@ impl Inflate { ///! Decompresses input data into output returning the uncompressed data length. pub fn uncompress(src: &[u8], dst: &mut [u8]) -> DecompressResult { - let mut inflate = Self::new(); - inflate.decompress_data(src, dst, false) + let mut csrc = CurrentSource::new(src, BitReaderState::default()); + if src.len() > 2 { + let cm = src[0] & 0xF; + let cinfo = src[0] >> 4; + let hdr = (u16::from(src[0]) << 8) | u16::from(src[1]); + if cm == 8 && cinfo <= 7 && (hdr % 31) == 0 { + csrc.skip(16).unwrap(); + } + } + + let mut fix_len_cb = None; + + let mut dst_idx = 0; + let mut final_block = false; + while !final_block { + final_block = csrc.read_bool()?; + + let bmode = csrc.read(2)?; + match bmode { + 0 => { + csrc.align(); + let len = csrc.read(16)? as usize; + let inv_len = csrc.read(16)? as usize; + if (len ^ inv_len) != 0xFFFF { + return Err(DecompressError::InvalidHeader); + } + let src_pos = csrc.tell(); + if src_pos + len > src.len() { + return Err(DecompressError::ShortData); + } + if dst_idx + len > dst.len() { + return Err(DecompressError::OutputFull); + } + dst[dst_idx..][..len].copy_from_slice(&src[src_pos..][..len]); + dst_idx += len; + csrc.skip_bytes(len)?; + }, + 1 => { + if fix_len_cb.is_none() { + let mut cr = FixedLenCodeReader {}; + fix_len_cb = Some(Codebook::new(&mut cr, CodebookMode::LSB).unwrap()); + } + if let Some(ref len_cb) = &fix_len_cb { + loop { + let val = csrc.read_cb(len_cb)?; + if val < 256 { + if dst_idx >= dst.len() { + return Err(DecompressError::OutputFull); + } + dst[dst_idx] = val as u8; + dst_idx += 1; + } else if val == 256 { + break; + } else { + let len_idx = (val - 257) as usize; + if len_idx >= LENGTH_BASE.len() { + return Err(DecompressError::InvalidData); + } + let len_bits = LENGTH_ADD_BITS[len_idx]; + let mut length = LENGTH_BASE[len_idx] as usize; + if len_bits > 0 { + length += csrc.read(len_bits)? as usize; + } + let dist_idx = reverse_bits(csrc.read(5)?, 5) as usize; + if dist_idx >= DIST_BASE.len() { + return Err(DecompressError::InvalidData); + } + let dist_bits = DIST_ADD_BITS[dist_idx]; + let mut dist = DIST_BASE[dist_idx] as usize; + if dist_bits > 0 { + dist += csrc.read(dist_bits)? as usize; + } + + if dst_idx + length > dst.len() { + return Err(DecompressError::OutputFull); + } + if dist > dst_idx { + return Err(DecompressError::InvalidData); + } + lz_copy(dst, dst_idx, dist, length); + dst_idx += length; + } + } + } else { + unreachable!(); + } + }, + 2 => { + let hlit = csrc.read(5)? as usize + 257; + if hlit >= 287 { + return Err(DecompressError::InvalidHeader); + } + let hdist = csrc.read(5)? as usize + 1; + let hclen = csrc.read(4)? as usize + 4; + let mut cur_len_idx = 0; + let mut len_lengths = [0; 19]; + let mut all_lengths = [0; NUM_LITERALS + NUM_DISTS]; + + for _ in 0..hclen { + len_lengths[LEN_RECODE[cur_len_idx]] = csrc.read(3)? as u8; + cur_len_idx += 1; + } + let mut len_codes = [ShortCodebookDesc { code: 0, bits: 0 }; 19]; + lengths_to_codes(&len_lengths, &mut len_codes)?; + let mut cr = ShortCodebookDescReader::new(len_codes.to_vec()); + let ret = Codebook::new(&mut cr, CodebookMode::LSB); + if ret.is_err() { + return Err(DecompressError::InvalidHeader); + } + let dyn_len_cb = ret.unwrap(); + + let mut cur_len_idx = 0; + while cur_len_idx < hlit + hdist { + let val = csrc.read_cb(&dyn_len_cb)?; + if val < 16 { + all_lengths[cur_len_idx] = val as u8; + cur_len_idx += 1; + } else { + let mode = (val as usize) - 16; + if mode > 2 { + return Err(DecompressError::InvalidHeader); + } + let base = REPEAT_BASE[mode] as usize; + let bits = REPEAT_BITS[mode]; + let len = base + (csrc.read(bits)? as usize); + if cur_len_idx + len > hlit + hdist { + return Err(DecompressError::InvalidHeader); + } + let rpt = if mode == 0 { + if cur_len_idx == 0 { + return Err(DecompressError::InvalidHeader); + } + all_lengths[cur_len_idx - 1] + } else { + 0 + }; + for _ in 0..len { + all_lengths[cur_len_idx] = rpt; + cur_len_idx += 1; + } + } + } + let (lit_lengths, dist_lengths) = all_lengths.split_at(hlit); + + let mut lit_codes = [ShortCodebookDesc { code: 0, bits: 0 }; NUM_LITERALS]; + lengths_to_codes(lit_lengths, &mut lit_codes)?; + let mut cr = ShortCodebookDescReader::new(lit_codes.to_vec()); + let ret = Codebook::new(&mut cr, CodebookMode::LSB); + if ret.is_err() { return Err(DecompressError::InvalidHeader); } + let dyn_lit_cb = ret.unwrap(); + + let mut dist_codes = [ShortCodebookDesc { code: 0, bits: 0 }; NUM_DISTS]; + lengths_to_codes(&dist_lengths[..hdist], &mut dist_codes)?; + let mut cr = ShortCodebookDescReader::new(dist_codes.to_vec()); + let ret = Codebook::new(&mut cr, CodebookMode::LSB); + if ret.is_err() { return Err(DecompressError::InvalidHeader); } + let dyn_dist_cb = ret.unwrap(); + + loop { + let val = csrc.read_cb(&dyn_lit_cb)?; + if val < 256 { + if dst_idx >= dst.len() { + return Err(DecompressError::OutputFull); + } + dst[dst_idx] = val as u8; + dst_idx += 1; + } else if val == 256 { + break; + } else { + let len_idx = (val - 257) as usize; + if len_idx >= LENGTH_BASE.len() { + return Err(DecompressError::InvalidData); + } + let len_bits = LENGTH_ADD_BITS[len_idx]; + let mut length = LENGTH_BASE[len_idx] as usize; + if len_bits > 0 { + length += csrc.read(len_bits)? as usize; + } + + let dist_idx = csrc.read_cb(&dyn_dist_cb)? as usize; + if dist_idx >= DIST_BASE.len() { + return Err(DecompressError::InvalidData); + } + let dist_bits = DIST_ADD_BITS[dist_idx]; + let mut dist = DIST_BASE[dist_idx] as usize; + if dist_bits > 0 { + dist += csrc.read(dist_bits)? as usize; + } + + if dst_idx + length > dst.len() { + return Err(DecompressError::OutputFull); + } + if dist > dst_idx { + return Err(DecompressError::InvalidData); + } + lz_copy(dst, dst_idx, dist, length); + dst_idx += length; + } + } + }, + _ => return Err(DecompressError::InvalidHeader), + }; + } + Ok(dst_idx) } }