tb_den: u32,
}
+fn copy_luma_default(dst: &mut [u8], dstride: usize, src: &[u8], sstride: usize, w: usize, h: usize) {
+ for (dline, sline) in dst.chunks_mut(dstride)
+ .zip(src.chunks(sstride))
+ .take(h) {
+ dline[..w].copy_from_slice(&sline[..w]);
+ }
+}
+#[cfg(not(target_arch="x86_64"))]
+fn copy_luma(dst: &mut [u8], dstride: usize, src: &[u8], sstride: usize, w: usize, h: usize) {
+ copy_luma_default(dst, dstride, src, sstride, w, h);
+}
+#[cfg(not(target_arch="x86_64"))]
+fn deint_chroma(frm: NASimpleVideoFrame<u8>, src: &[u8], sstride: usize) {
+ let mut uoff = frm.offset[1];
+ let mut voff = frm.offset[2];
+ for cline in src.chunks(sstride).take(frm.height[1]) {
+ for (x, pair) in cline.chunks_exact(2).take(frm.width[1]).enumerate() {
+ frm.data[uoff + x] = pair[0];
+ frm.data[voff + x] = pair[1];
+ }
+ uoff += frm.stride[1];
+ voff += frm.stride[2];
+ }
+}
+
+#[cfg(target_arch="x86_64")]
+use std::arch::asm;
+#[cfg(target_arch="x86_64")]
+fn copy_luma(dst: &mut [u8], dstride: usize, src: &[u8], sstride: usize, w: usize, h: usize) {
+ if !is_x86_feature_detected!("avx") {
+ copy_luma_default(dst, dstride, src, sstride, w, h);
+ return;
+ }
+ if dst.as_ptr().align_offset(32) == 0 && src.as_ptr().align_offset(32) == 0 &&
+ (w % 64) == 0 && ((dstride | sstride) % 32) == 0 {
+ unsafe {
+ asm!(
+ "2:",
+ " mov {x}, {w}",
+ " 3:",
+ " vmovdqa ymm0, [{src}]",
+ " vmovdqa ymm1, [{src}+32]",
+ " vmovdqa [{dst}], ymm0",
+ " vmovdqa [{dst}+32], ymm1",
+ " add {src}, 64",
+ " add {dst}, 64",
+ " sub {x}, 64",
+ " jnz 3b",
+ " add {src}, {sstep}",
+ " add {dst}, {dstep}",
+ " dec {h}",
+ " jnz 2b",
+ dst = inout(reg) dst.as_mut_ptr() => _,
+ src = inout(reg) src.as_ptr() => _,
+ sstep = in(reg) sstride - w,
+ dstep = in(reg) dstride - w,
+ w = in(reg) w,
+ h = in(reg) h,
+ x = out(reg) _,
+ out("ymm0") _,
+ out("ymm1") _,
+ );
+ }
+ } else {
+ for (dline, sline) in dst.chunks_mut(dstride)
+ .zip(src.chunks(sstride))
+ .take(h) {
+ dline[..w].copy_from_slice(&sline[..w]);
+ }
+ }
+}
+#[cfg(target_arch="x86_64")]
+fn deint_chroma(frm: NASimpleVideoFrame<u8>, src: &[u8], sstride: usize) {
+ unsafe {
+ let width = frm.width[1];
+ let height = frm.height[1];
+ let dst = frm.data.as_mut_ptr();
+ let udst = dst.add(frm.offset[1]);
+ let vdst = dst.add(frm.offset[2]);
+ let dstep = frm.stride[1] - width;
+ let sstep = sstride - width * 2;
+ asm!(
+ "2:",
+ " mov {tmp}, {width}",
+ " test {width}, 8",
+ " jz 3f",
+ " movaps xmm0, [{src}]",
+ " movaps xmm1, xmm0",
+ " psllw xmm0, 8",
+ " psrlw xmm1, 8",
+ " psrlw xmm0, 8",
+ " packuswb xmm1, xmm1",
+ " packuswb xmm0, xmm0",
+ " movq [{vdst}], xmm1",
+ " movq [{udst}], xmm0",
+ " add {src}, 16",
+ " add {vdst}, 8",
+ " add {udst}, 8",
+ " sub {tmp}, 8",
+ " 3:",
+ " movaps xmm0, [{src}]",
+ " movaps xmm1, [{src} + 16]",
+ " movaps xmm2, xmm0",
+ " movaps xmm3, xmm1",
+ " psllw xmm0, 8",
+ " psllw xmm1, 8",
+ " psrlw xmm2, 8",
+ " psrlw xmm3, 8",
+ " psrlw xmm0, 8",
+ " psrlw xmm1, 8",
+ " packuswb xmm2, xmm3",
+ " packuswb xmm0, xmm1",
+ " movups [{vdst}], xmm2",
+ " movups [{udst}], xmm0",
+ " add {src}, 32",
+ " add {vdst}, 16",
+ " add {udst}, 16",
+ " sub {tmp}, 16",
+ " jnz 3b",
+ " add {udst}, {dstep}",
+ " add {vdst}, {dstep}",
+ " add {src}, {sstep}",
+ " dec {height}",
+ " jnz 2b",
+ src = inout(reg) src.as_ptr() => _,
+ udst = inout(reg) udst => _,
+ vdst = inout(reg) vdst => _,
+ width = in(reg) width,
+ height = inout(reg) height => _,
+ dstep = in(reg) dstep,
+ sstep = in(reg) sstep,
+ tmp = out(reg) _,
+ out("xmm0") _,
+ out("xmm1") _,
+ out("xmm2") _,
+ out("xmm3") _,
+ );
+ }
+}
+
fn fill_frame(ifmt: VAImageFormat, pic: &Picture<PictureSync>, frm: &mut NABufferType) -> DecoderResult<()> {
let mut vbuf = frm.get_vbuf().unwrap();
let (w, h) = pic.surface_size();
validate!(iimg.width == (frm.width[0] as u16));
validate!(iimg.height == (frm.height[0] as u16));
- for (dline, sline) in frm.data[frm.offset[0]..].chunks_mut(frm.stride[0])
- .zip(imgdata[iimg.offsets[0] as usize..].chunks(iimg.pitches[0] as usize))
- .take(frm.height[0]) {
- dline[..frm.width[0]].copy_from_slice(&sline[..frm.width[0]]);
- }
+ copy_luma(&mut frm.data[frm.offset[0]..], frm.stride[0], &imgdata[iimg.offsets[0] as usize..], iimg.pitches[0] as usize, frm.width[0], frm.height[0]);
- let mut uoff = frm.offset[1];
- let mut voff = frm.offset[2];
- for cline in imgdata[iimg.offsets[1] as usize..].chunks(iimg.pitches[1] as usize).take(frm.height[1]) {
- for (x, pair) in cline.chunks_exact(2).take(frm.width[1]).enumerate() {
- frm.data[uoff + x] = pair[0];
- frm.data[voff + x] = pair[1];
- }
- uoff += frm.stride[1];
- voff += frm.stride[2];
- }
+ deint_chroma(frm, &imgdata[iimg.offsets[1] as usize..], iimg.pitches[1] as usize);
},
_ => unimplemented!(),
};
match profile {
100 | 110 | 122 | 144 => {
let b = br.read_byte()?;
- validate!((b & 0xFC) == 0xFC);
+ // some encoders put something different here
+ if (b & 0xFC) != 0xFC {
+ return Ok(());
+ }
// b & 3 -> chroma format
let b = br.read_byte()?;
validate!((b & 0xF8) == 0xF8);
projects / nihav-player.git / blobdiff