From f2f1afb2cec901bf936d742c208aca2322955cc2 Mon Sep 17 00:00:00 2001 From: Kostya Shishkov Date: Fri, 9 Jun 2017 20:25:01 +0200 Subject: [PATCH] deblocking for I.263 --- src/codecs/blockdec.rs | 79 +++++++++++++++++++---- src/codecs/intel263.rs | 138 +++++++++++++++++++++++++++++++++++++++-- 2 files changed, 201 insertions(+), 16 deletions(-) diff --git a/src/codecs/blockdec.rs b/src/codecs/blockdec.rs index 23e22d4..64eb6cb 100644 --- a/src/codecs/blockdec.rs +++ b/src/codecs/blockdec.rs @@ -36,12 +36,13 @@ pub struct PicInfo { umv: bool, pb: Option, ts: u8, + deblock: bool, } #[allow(dead_code)] impl PicInfo { - pub fn new(w: usize, h: usize, mode: Type, quant: u8, apm: bool, umv: bool, ts: u8, pb: Option) -> Self { - PicInfo{ w: w, h: h, mode: mode, quant: quant, apm: apm, umv: umv, ts: ts, pb: pb } + pub fn new(w: usize, h: usize, mode: Type, quant: u8, apm: bool, umv: bool, ts: u8, pb: Option, deblock: bool) -> Self { + PicInfo{ w: w, h: h, mode: mode, quant: quant, apm: apm, umv: umv, ts: ts, pb: pb, deblock: deblock } } pub fn get_width(&self) -> usize { self.w } pub fn get_height(&self) -> usize { self.h } @@ -246,8 +247,9 @@ pub trait BlockDecoder { fn decode_block_header(&mut self, pinfo: &PicInfo, sinfo: &Slice) -> DecoderResult; fn decode_block_intra(&mut self, info: &BlockInfo, quant: u8, no: usize, coded: bool, blk: &mut [i16; 64]) -> DecoderResult<()>; fn decode_block_inter(&mut self, info: &BlockInfo, quant: u8, no: usize, coded: bool, blk: &mut [i16; 64]) -> DecoderResult<()>; - fn calc_mv(&mut self, vec: MV); fn is_slice_end(&mut self) -> bool; + + fn filter_row(&mut self, buf: &mut NAVideoBuffer, mb_y: usize, mb_w: usize, cbpi: &CBPInfo); } #[allow(dead_code)] @@ -339,6 +341,52 @@ impl MVInfo { } } +#[allow(dead_code)] +pub struct CBPInfo { + cbp: Vec, + q: Vec, + mb_w: usize, +} + +impl CBPInfo { + fn new() -> Self { CBPInfo{ cbp: Vec::new(), q: Vec::new(), mb_w: 0 } } + fn reset(&mut self, mb_w: usize) { + self.mb_w = mb_w; + self.cbp.truncate(0); + self.cbp.resize(self.mb_w * 2, 0); + self.q.truncate(0); + self.q.resize(self.mb_w * 2, 0); + } + fn update_row(&mut self) { + for i in 0..self.mb_w { + self.cbp[i] = self.cbp[self.mb_w + i]; + self.q[i] = self.q[self.mb_w + i]; + } + } + fn set_cbp(&mut self, mb_x: usize, cbp: u8) { + self.cbp[self.mb_w + mb_x] = cbp; + } + fn set_q(&mut self, mb_x: usize, q: u8) { + self.q[self.mb_w + mb_x] = q; + } + pub fn get_q(&self, mb_x: usize) -> u8 { self.q[mb_x] } + pub fn is_coded(&self, mb_x: usize, blk_no: usize) -> bool { + (self.cbp[self.mb_w + mb_x] & (1 << (5 - blk_no))) != 0 + } + pub fn is_coded_top(&self, mb_x: usize, blk_no: usize) -> bool { + let cbp = self.cbp[self.mb_w + mb_x]; + let cbp_top = self.cbp[mb_x]; + match blk_no { + 0 => { (cbp_top & 0b001000) != 0 }, + 1 => { (cbp_top & 0b000100) != 0 }, + 2 => { (cbp & 0b100000) != 0 }, + 3 => { (cbp & 0b010000) != 0 }, + 4 => { (cbp_top & 0b000010) != 0 }, + _ => { (cbp_top & 0b000001) != 0 }, + } + } +} + fn copy_blocks(dst: &mut NAVideoBuffer, src: &NAVideoBuffer, xpos: usize, ypos: usize, w: usize, h: usize, mv: MV) { let srcx = ((mv.x >> 1) as isize) + (xpos as isize); let srcy = ((mv.y >> 1) as isize) + (ypos as isize); @@ -402,6 +450,7 @@ impl DCT8x8VideoDecoder { pub fn parse_frame(&mut self, bd: &mut BlockDecoder) -> DecoderResult { let pinfo = bd.decode_pichdr()?; let mut mvi = MVInfo::new(); + let mut cbpi = CBPInfo::new(); //todo handle res change self.w = pinfo.w; @@ -417,7 +466,6 @@ impl DCT8x8VideoDecoder { } mem::swap(&mut self.cur_frm, &mut self.prev_frm); -// if self.ftype == Type::I && !pinfo.is_pb() { self.prev_frm = None; } let tsdiff = pinfo.ts.wrapping_sub(self.last_ts); let bsdiff = if pinfo.is_pb() { pinfo.get_pbinfo().get_trb() } else { 0 }; @@ -431,6 +479,7 @@ impl DCT8x8VideoDecoder { let mut slice = Slice::get_default_slice(&pinfo); mvi.reset(self.mb_w, 0, pinfo.get_umv()); + cbpi.reset(self.mb_w); let mut blk: [[i16; 64]; 6] = [[0; 64]; 6]; for mb_y in 0..self.mb_h { @@ -441,11 +490,14 @@ impl DCT8x8VideoDecoder { //println!("new slice @{}.{}!",mb_x,mb_y); slice = bd.decode_slice_header(&pinfo)?; mvi.reset(self.mb_w, mb_x, pinfo.get_umv()); + //cbpi.reset(self.mb_w); } let binfo = bd.decode_block_header(&pinfo, &slice)?; let cbp = binfo.get_cbp(); //println!("mb {}.{} CBP {:X} type {:?}, {} mvs skip {}", mb_x,mb_y, cbp, binfo.get_mode(), binfo.get_num_mvs(),binfo.is_skipped()); + cbpi.set_cbp(mb_x, cbp); + cbpi.set_q(mb_x, binfo.get_q()); if binfo.is_intra() { for i in 0..6 { bd.decode_block_intra(&binfo, binfo.get_q(), i, (cbp & (1 << (5 - i))) != 0, &mut blk[i])?; @@ -523,8 +575,13 @@ impl DCT8x8VideoDecoder { self.b_data.push(b_mb); } } + if pinfo.deblock { + bd.filter_row(&mut buf, mb_y, self.mb_w, &cbpi); + } mvi.update_row(); + cbpi.update_row(); } + self.cur_frm = Some(buf); self.last_ts = pinfo.ts; //println!("unpacked all"); @@ -557,17 +614,16 @@ impl DCT8x8VideoDecoder { fn recon_b_frame(b_buf: &mut NAVideoBuffer, bck_buf: &NAVideoBuffer, fwd_buf: &NAVideoBuffer, mb_w: usize, mb_h: usize, b_data: &Vec) { + let mut cbpi = CBPInfo::new(); let mut cur_mb = 0; + cbpi.reset(mb_w); for mb_y in 0..mb_h { for mb_x in 0..mb_w { let num_mv = b_data[cur_mb].num_mv; let is_fwd = b_data[cur_mb].fwd; - if num_mv == 0 { - copy_blocks(b_buf, bck_buf, mb_x * 16, mb_y * 16, 16, 16, ZERO_MV); - if !is_fwd { - avg_blocks(b_buf, fwd_buf, mb_x * 16, mb_y * 16, 16, 16, ZERO_MV); - } - } else if num_mv == 1 { + let cbp = b_data[cur_mb].cbp; + cbpi.set_cbp(mb_x, cbp); + if num_mv == 1 { copy_blocks(b_buf, bck_buf, mb_x * 16, mb_y * 16, 16, 16, b_data[cur_mb].mv_f[0]); if !is_fwd { avg_blocks(b_buf, fwd_buf, mb_x * 16, mb_y * 16, 16, 16, b_data[cur_mb].mv_b[0]); @@ -582,10 +638,11 @@ fn recon_b_frame(b_buf: &mut NAVideoBuffer, bck_buf: &NAVideoBuffer, fwd } } } - if num_mv != 0 && b_data[cur_mb].cbp != 0 { + if cbp != 0 { blockdsp::add_blocks(b_buf, mb_x, mb_y, &b_data[cur_mb].blk); } cur_mb += 1; } + cbpi.update_row(); } } diff --git a/src/codecs/intel263.rs b/src/codecs/intel263.rs index 61e1cc4..99f0d4c 100644 --- a/src/codecs/intel263.rs +++ b/src/codecs/intel263.rs @@ -151,16 +151,19 @@ impl<'a> BlockDecoder for Intel263BR<'a> { br.read(1)?; // syntax arithmetic coding let apm = br.read_bool()?; self.is_pb = br.read_bool()?; + let deblock; if sfmt == 0b111 { sfmt = br.read(3)?; validate!((sfmt != 0b000) && (sfmt != 0b111)); br.read(2)?; // unknown flags - let deblock = br.read_bool()?; + deblock = br.read_bool()?; br.read(1)?; // unknown flag let pbplus = br.read_bool()?; br.read(5)?; // unknown flags let marker = br.read(5)?; validate!(marker == 1); + } else { + deblock = false; } let w; let h; if sfmt == 0b110 { @@ -198,7 +201,7 @@ impl<'a> BlockDecoder for Intel263BR<'a> { self.mb_w = (w + 15) >> 4; let ftype = if is_intra { Type::I } else { Type::P }; - let picinfo = PicInfo::new(w, h, ftype, quant as u8, apm, umv, tr, pbinfo); + let picinfo = PicInfo::new(w, h, ftype, quant as u8, apm, umv, tr, pbinfo, deblock); Ok(picinfo) } @@ -307,10 +310,135 @@ impl<'a> BlockDecoder for Intel263BR<'a> { self.decode_block(quant, false, coded, blk) } -#[allow(unused_variables)] - fn calc_mv(&mut self, vec: MV) {} - fn is_slice_end(&mut self) -> bool { self.br.peek(16) == 0 } + + fn filter_row(&mut self, buf: &mut NAVideoBuffer, mb_y: usize, mb_w: usize, cbpi: &CBPInfo) { + let stride = buf.get_stride(0); + let mut off = buf.get_offset(0) + mb_y * 16 * stride; + for mb_x in 0..mb_w { + let coff = off; + let coded0 = cbpi.is_coded(mb_x, 0); + let coded1 = cbpi.is_coded(mb_x, 1); + let q = cbpi.get_q(mb_w + mb_x); + if mb_y != 0 { + if coded0 && cbpi.is_coded_top(mb_x, 0) { deblock_hor(buf, 0, q, coff); } + if coded1 && cbpi.is_coded_top(mb_x, 1) { deblock_hor(buf, 0, q, coff + 8); } + } + let coff = off + 8 * stride; + if cbpi.is_coded(mb_x, 2) && coded0 { deblock_hor(buf, 0, q, coff); } + if cbpi.is_coded(mb_x, 3) && coded1 { deblock_hor(buf, 0, q, coff + 8); } + off += 16; + } + let mut leftt = false; + let mut leftc = false; + let mut off = buf.get_offset(0) + mb_y * 16 * stride; + for mb_x in 0..mb_w { + let ctop0 = cbpi.is_coded_top(mb_x, 0); + let ctop1 = cbpi.is_coded_top(mb_x, 0); + let ccur0 = cbpi.is_coded(mb_x, 0); + let ccur1 = cbpi.is_coded(mb_x, 1); + let q = cbpi.get_q(mb_w + mb_x); + if mb_y != 0 { + let coff = off - 8 * stride; + let qtop = cbpi.get_q(mb_x); + if leftt && ctop0 { deblock_ver(buf, 0, qtop, coff); } + if ctop0 && ctop1 { deblock_ver(buf, 0, qtop, coff + 8); } + } + if leftc && ccur0 { deblock_ver(buf, 0, q, off); } + if ccur0 && ccur1 { deblock_ver(buf, 0, q, off + 8); } + leftt = ctop1; + leftc = ccur1; + off += 16; + } + let strideu = buf.get_stride(1); + let stridev = buf.get_stride(2); + let offu = buf.get_offset(1) + mb_y * 8 * strideu; + let offv = buf.get_offset(2) + mb_y * 8 * stridev; + if mb_y != 0 { + for mb_x in 0..mb_w { + let ctu = cbpi.is_coded_top(mb_x, 4); + let ccu = cbpi.is_coded(mb_x, 4); + let ctv = cbpi.is_coded_top(mb_x, 5); + let ccv = cbpi.is_coded(mb_x, 5); + let q = cbpi.get_q(mb_w + mb_x); + if ctu && ccu { deblock_hor(buf, 1, q, offu + mb_x * 8); } + if ctv && ccv { deblock_hor(buf, 2, q, offv + mb_x * 8); } + } + let mut leftu = false; + let mut leftv = false; + let offu = buf.get_offset(1) + (mb_y - 1) * 8 * strideu; + let offv = buf.get_offset(2) + (mb_y - 1) * 8 * stridev; + for mb_x in 0..mb_w { + let ctu = cbpi.is_coded_top(mb_x, 4); + let ctv = cbpi.is_coded_top(mb_x, 5); + let qt = cbpi.get_q(mb_x); + if leftu && ctu { deblock_ver(buf, 1, qt, offu + mb_x * 8); } + if leftv && ctv { deblock_ver(buf, 2, qt, offv + mb_x * 8); } + leftu = ctu; + leftv = ctv; + } + } + } +} + +fn deblock_hor(buf: &mut NAVideoBuffer, comp: usize, q: u8, off: usize) { + let stride = buf.get_stride(comp); + let mut dptr = buf.get_data_mut(); + let mut buf = dptr.as_mut_slice(); + for x in 0..8 { + let a = buf[off - 2 * stride + x] as i16; + let b = buf[off - 1 * stride + x] as i16; + let c = buf[off + 0 * stride + x] as i16; + let d = buf[off + 1 * stride + x] as i16; + let diff = ((a - d) * 3 + (c - b) * 8) >> 4; + if (diff != 0) && (diff >= -32) && (diff < 32) { + let d0 = diff.abs() * 2 - (q as i16); + let d1 = if d0 < 0 { 0 } else { d0 }; + let d2 = diff.abs() - d1; + let d3 = if d2 < 0 { 0 } else { d2 }; + + let delta = if diff < 0 { -d3 } else { d3 }; + + let b1 = b + delta; + if b1 < 0 { buf[off - 1 * stride + x] = 0; } + else if b1 > 255 { buf[off - 1 * stride + x] = 0xFF; } + else { buf[off - 1 * stride + x] = b1 as u8; } + let c1 = c - delta; + if c1 < 0 { buf[off + x] = 0; } + else if c1 > 255 { buf[off + x] = 0xFF; } + else { buf[off + x] = c1 as u8; } + } + } +} + +fn deblock_ver(buf: &mut NAVideoBuffer, comp: usize, q: u8, off: usize) { + let stride = buf.get_stride(comp); + let mut dptr = buf.get_data_mut(); + let mut buf = dptr.as_mut_slice(); + for y in 0..8 { + let a = buf[off - 2 + y * stride] as i16; + let b = buf[off - 1 + y * stride] as i16; + let c = buf[off + 0 + y * stride] as i16; + let d = buf[off + 1 + y * stride] as i16; + let diff = ((a - d) * 3 + (c - b) * 8) >> 4; + if (diff != 0) && (diff >= -32) && (diff < 32) { + let d0 = diff.abs() * 2 - (q as i16); + let d1 = if d0 < 0 { 0 } else { d0 }; + let d2 = diff.abs() - d1; + let d3 = if d2 < 0 { 0 } else { d2 }; + + let delta = if diff < 0 { -d3 } else { d3 }; + + let b1 = b + delta; + if b1 < 0 { buf[off - 1 + y * stride] = 0; } + else if b1 > 255 { buf[off - 1 + y * stride] = 0xFF; } + else { buf[off - 1 + y * stride] = b1 as u8; } + let c1 = c - delta; + if c1 < 0 { buf[off + y * stride] = 0; } + else if c1 > 255 { buf[off + y * stride] = 0xFF; } + else { buf[off + y * stride] = c1 as u8; } + } + } } impl Intel263Decoder { -- 2.30.2