X-Git-Url: https://git.nihav.org/?a=blobdiff_plain;f=nihav-codec-support%2Fsrc%2Fcodecs%2Fh263%2Fcode.rs;h=cbeb12042fad4946a79d04ce14f238ec6da0de1c;hb=b7c882c1ce6f86c07c2340751200e3a060942826;hp=1f00beb77881d0b978f8413233d1c67d27efa617;hpb=5131a66c0d27dfd69c0599df16e35be79970ff68;p=nihav.git diff --git a/nihav-codec-support/src/codecs/h263/code.rs b/nihav-codec-support/src/codecs/h263/code.rs index 1f00beb..cbeb120 100644 --- a/nihav-codec-support/src/codecs/h263/code.rs +++ b/nihav-codec-support/src/codecs/h263/code.rs @@ -221,6 +221,119 @@ pub fn h263_idct(blk: &mut [i16; 64]) { for i in 0..8 { idct_col(blk, i); } } +struct IDCTAnnexW {} + +impl IDCTAnnexW { + const CPO8: i32 = 0x539f; + const SPO8: i32 = 0x4546; + const CPO16: i32 = 0x7d8a; + const SPO16: i32 = 0x18f9; + const C3PO16: i32 = 0x6a6e; + const S3PO16: i32 = 0x471d; + const OOR2: i32 = 0x5a82; + + fn rotate(a: i32, b: i32, c: i32, s: i32, cs: i8, ss: i8) -> (i32, i32) { + let (t00, t10) = if cs > 0 { + ((a * c) >> cs, (b * c) >> cs) + } else { + ((a * c) << -cs, (b * c) << -cs) + }; + let (t01, t11) = if ss > 0 { + ((a * s) >> ss, (b * s) >> ss) + } else { + ((a * s) << -ss, (b * s) << -ss) + }; + ((t01 - t10 + 0x7FFF) >> 16, (t00 + t11 + 0x7FFF) >> 16) + } + + fn bfly(a: i32, b: i32) -> (i32, i32) { (a + b, a - b) } + + fn idct_row(dst: &mut [i32; 64], src: &[i16; 64]) { + for (drow, srow) in dst.chunks_mut(8).zip(src.chunks(8)) { + let s0 = i32::from(srow[0]) << 4; + let s1 = i32::from(srow[1]) << 4; + let s2 = i32::from(srow[2]) << 4; + let s3 = i32::from(srow[3]) << 4; + let s4 = i32::from(srow[4]) << 4; + let s5 = i32::from(srow[5]) << 4; + let s6 = i32::from(srow[6]) << 4; + let s7 = i32::from(srow[7]) << 4; + + let (s2, s6) = Self::rotate(s2, s6, Self::CPO8, Self::SPO8, -2, -1); + let (s1, s7) = Self::rotate(s1, s7, Self::CPO16, Self::SPO16, -1, -1); + let (s3, s5) = Self::rotate(s3, s5, Self::C3PO16, Self::S3PO16, -1, -1); + let (s0, s4) = Self::bfly(s0, s4); + + let (s3, s1) = Self::bfly(s1, s3); + let (s5, s7) = Self::bfly(s7, s5); + let (s0, s6) = Self::bfly(s0, s6); + let (s4, s2) = Self::bfly(s4, s2); + + let (s3, s7) = Self::bfly(s7, s3); + let s1 = (s1 * Self::OOR2 * 4).saturating_add(0x7FFF) >> 16; + let s5 = (s5 * Self::OOR2 * 4).saturating_add(0x7FFF) >> 16; + + drow[1] = s4 + s3; + drow[6] = s4 - s3; + drow[2] = s2 + s7; + drow[5] = s2 - s7; + drow[0] = s0 + s5; + drow[7] = s0 - s5; + drow[3] = s6 + s1; + drow[4] = s6 - s1; + } + } + + #[allow(clippy::erasing_op)] + #[allow(clippy::identity_op)] + fn idct_col(dst: &mut [i16; 64], src: &[i32; 64]) { + for i in 0..8 { + let s0 = src[i + 8 * 0]; + let s1 = src[i + 8 * 1]; + let s2 = src[i + 8 * 2]; + let s3 = src[i + 8 * 3]; + let s4 = src[i + 8 * 4]; + let s5 = src[i + 8 * 5]; + let s6 = src[i + 8 * 6]; + let s7 = src[i + 8 * 7]; + + let (s2, s6) = Self::rotate(s2, s6, Self::CPO8, Self::SPO8, -1, 0); + let (s1, s7) = Self::rotate(s1, s7, Self::CPO16, Self::SPO16, 0, 0); + let (s3, s5) = Self::rotate(s3, s5, Self::C3PO16, Self::S3PO16, 0, 0); + let (a, b) = Self::bfly(s0, s4); + let (s0, s4) = if s4 >= 0 { + (a >> 1, b >> 1) + } else { + ((a + 1) >> 1, (b + 1) >> 1) + }; + + let (s3, s1) = Self::bfly(s1, s3); + let (s5, s7) = Self::bfly(s7, s5); + let (s0, s6) = Self::bfly(s0, s6); + let (s4, s2) = Self::bfly(s4, s2); + + let (s3, s7) = Self::bfly(s7, s3); + let s1 = (s1 * Self::OOR2 * 4).saturating_add(0x7FFF) >> 16; + let s5 = (s5 * Self::OOR2 * 4).saturating_add(0x7FFF) >> 16; + + dst[i + 8 * 1] = ((s4 + s3) >> 6) as i16; + dst[i + 8 * 6] = ((s4 - s3) >> 6) as i16; + dst[i + 8 * 2] = ((s2 + s7) >> 6) as i16; + dst[i + 8 * 5] = ((s2 - s7) >> 6) as i16; + dst[i + 8 * 0] = ((s0 + s5) >> 6) as i16; + dst[i + 8 * 7] = ((s0 - s5) >> 6) as i16; + dst[i + 8 * 3] = ((s6 + s1) >> 6) as i16; + dst[i + 8 * 4] = ((s6 - s1) >> 6) as i16; + } + } +} +#[allow(dead_code)] +pub fn h263_annex_w_idct(blk: &mut [i16; 64]) { + let mut tmp = [0i32; 64]; + IDCTAnnexW::idct_row(&mut tmp, blk); + IDCTAnnexW::idct_col(blk, &tmp); +} + fn h263_interp00(dst: &mut [u8], dstride: usize, src: &[u8], sstride: usize, bw: usize, bh: usize) { let mut didx = 0; @@ -339,6 +452,7 @@ fn h263_interp11_avg(dst: &mut [u8], dstride: usize, src: &[u8], sstride: usize, pub const H263_INTERP_AVG_FUNCS: &[blockdsp::BlkInterpFunc] = &[ h263_interp00_avg, h263_interp01_avg, h263_interp10_avg, h263_interp11_avg ]; +#[derive(Default)] pub struct H263BlockDSP { } impl H263BlockDSP { @@ -360,7 +474,7 @@ fn deblock_hor(buf: &mut NAVideoBuffer, comp: usize, strength: u8, off: usiz let diff = ((a - d) + (c - b) * 4) / 8; if (diff != 0) && (diff > -24) && (diff < 24) { let d1a = (diff.abs() - 2 * (diff.abs() - (strength as i16)).max(0)).max(0); - let d1 = if d1a < 0 { 0 } else { d1a }; + let d1 = if diff < 0 { -d1a } else { d1a }; let hd1 = d1a / 2; let d2 = ((a - d) / 4).max(-hd1).min(hd1); @@ -384,7 +498,7 @@ fn deblock_ver(buf: &mut NAVideoBuffer, comp: usize, strength: u8, off: usiz let diff = (a - d + (c - b) * 4) / 8; if (diff != 0) && (diff > -24) && (diff < 24) { let d1a = (diff.abs() - 2 * (diff.abs() - (strength as i16)).max(0)).max(0); - let d1 = if d1a < 0 { 0 } else { d1a }; + let d1 = if diff < 0 { -d1a } else { d1a }; let hd1 = d1a / 2; let d2 = ((a - d) / 4).max(-hd1).min(hd1); @@ -486,7 +600,7 @@ impl BlockDSP for H263BlockDSP { blockdsp::copy_block(&mut dst, src.clone(), 0, xpos, ypos, mv.x >> 1, mv.y >> 1, 16, 16, 0, 1, mode, H263_INTERP_FUNCS); blockdsp::copy_block(&mut dst, src.clone(), 1, xpos >> 1, ypos >> 1, mv.x >> 2, mv.y >> 2, 8, 8, 0, 1, cmode, H263_INTERP_FUNCS); - blockdsp::copy_block(&mut dst, src.clone(), 2, xpos >> 1, ypos >> 1, mv.x >> 2, mv.y >> 2, 8, 8, 0, 1, cmode, H263_INTERP_FUNCS); + blockdsp::copy_block(&mut dst, src, 2, xpos >> 1, ypos >> 1, mv.x >> 2, mv.y >> 2, 8, 8, 0, 1, cmode, H263_INTERP_FUNCS); } fn copy_blocks8x8(&self, dst: &mut NAVideoBuffer, src: NAVideoBufferRef, xpos: usize, ypos: usize, mvs: &[MV; 4]) { let mut dst = NASimpleVideoFrame::from_video_buf(dst).unwrap(); @@ -515,7 +629,7 @@ impl BlockDSP for H263BlockDSP { blockdsp::copy_block(&mut dst, src.clone(), 0, xpos, ypos, mv.x >> 1, mv.y >> 1, 16, 16, 0, 1, mode, H263_INTERP_AVG_FUNCS); blockdsp::copy_block(&mut dst, src.clone(), 1, xpos >> 1, ypos >> 1, mv.x >> 2, mv.y >> 2, 8, 8, 0, 1, cmode, H263_INTERP_AVG_FUNCS); - blockdsp::copy_block(&mut dst, src.clone(), 2, xpos >> 1, ypos >> 1, mv.x >> 2, mv.y >> 2, 8, 8, 0, 1, cmode, H263_INTERP_AVG_FUNCS); + blockdsp::copy_block(&mut dst, src, 2, xpos >> 1, ypos >> 1, mv.x >> 2, mv.y >> 2, 8, 8, 0, 1, cmode, H263_INTERP_AVG_FUNCS); } fn avg_blocks8x8(&self, dst: &mut NAVideoBuffer, src: NAVideoBufferRef, xpos: usize, ypos: usize, mvs: &[MV; 4]) { let mut dst = NASimpleVideoFrame::from_video_buf(dst).unwrap();