pub struct VaapiH264Decoder {
info: NACodecInfoRef,
vaapi: Option<VaapiInternals>,
+ needs_derive: bool,
spses: Vec<SeqParameterSet>,
ppses: Vec<PicParameterSet>,
frame_refs: FrameRefs,
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];
#[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 {
+ let copy_len = dstride.min(w);
+ for (dline, sline) in dst.chunks_mut(dstride)
+ .zip(src.chunks(sstride))
+ .take(h) {
+ dline[..copy_len].copy_from_slice(&sline[..copy_len]);
+ }
+ }
+}
+#[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 width = (frm.width[1] + 7) & !7;
+ let height = (frm.height[1] + 7) & !7;
let dst = frm.data.as_mut_ptr();
let udst = dst.add(frm.offset[1]);
let vdst = dst.add(frm.offset[2]);
}
}
-fn fill_frame(ifmt: VAImageFormat, pic: &Picture<PictureSync>, frm: &mut NABufferType) -> DecoderResult<()> {
+fn fill_frame(ifmt: VAImageFormat, pic: &Picture<PictureSync>, frm: &mut NABufferType, needs_derive: bool) -> DecoderResult<()> {
let mut vbuf = frm.get_vbuf().unwrap();
let (w, h) = pic.surface_size();
//let cur_ts = pic.timestamp();
- let img = Image::new(pic, ifmt, w, h, true).expect("get image");
+ let img = Image::new(pic, ifmt, w, h, !needs_derive).expect("get image");
let iimg = img.image();
let imgdata: &[u8] = img.as_ref();
match iimg.format.fourcc().map_err(|_| DecoderError::InvalidData)? {
VAFourcc::NV12 => {
let frm = NASimpleVideoFrame::from_video_buf(&mut vbuf).unwrap();
- validate!(iimg.width == (frm.width[0] as u16));
- validate!(iimg.height == (frm.height[0] as u16));
+ validate!(iimg.width == (((frm.width[0] + 15) & !15) as u16));
+ validate!(iimg.height == (((frm.height[0] + 15) & !15) 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] + 15) & !15, (frm.height[0] + 15) & !15);
deint_chroma(frm, &imgdata[iimg.offsets[1] as usize..], iimg.pitches[1] as usize);
},
+ VAFourcc::YV12 => {
+ let frm = NASimpleVideoFrame::from_video_buf(&mut vbuf).unwrap();
+ validate!(iimg.width == (((frm.width[0] + 15) & !15) as u16));
+ validate!(iimg.height == (((frm.height[0] + 15) & !15) as u16));
+
+ 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] + 15) & !15, (frm.height[0] + 15) & !15);
+ copy_luma(&mut frm.data[frm.offset[2]..], frm.stride[2], &imgdata[iimg.offsets[1] as usize..], iimg.pitches[1] as usize, (frm.width[1] + 15) & !15, (frm.height[1] + 15) & !15);
+ copy_luma(&mut frm.data[frm.offset[1]..], frm.stride[1], &imgdata[iimg.offsets[2] as usize..], iimg.pitches[2] as usize, (frm.width[2] + 15) & !15, (frm.height[2] + 15) & !15);
+ },
_ => unimplemented!(),
};
Ok(())
Self {
info: NACodecInfoRef::default(),
vaapi: None,
+ needs_derive: false,
spses: Vec::with_capacity(1),
ppses: Vec::with_capacity(4),
frame_refs: FrameRefs::new(),
let _compatibility = br.read_byte()?;
let _level = br.read_byte()?;
let b = br.read_byte()?;
- validate!((b & 0xFC) == 0xFC);
+ //validate!((b & 0xFC) == 0xFC);
self.nal_len = (b & 3) + 1;
let b = br.read_byte()?;
- validate!((b & 0xE0) == 0xE0);
+ //validate!((b & 0xE0) == 0xE0);
let num_sps = (b & 0x1F) as usize;
for _ in 0..num_sps {
let len = br.read_u16be()? as usize;
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);
return Err(DecoderError::Bug);
}
+ let needs_derive= if let Ok(vendor) = display.query_vendor_string() {
+ vendor.contains("Kaby Lake")
+ } else { false };
+
let config = display.create_config(vec![
VAConfigAttrib { type_: VAConfigAttribType::VAConfigAttribRTFormat, value: RTFormat::YUV420.into() },
], va_profile, VAEntrypoint::VAEntrypointVLD).map_err(|_| {
}
self.vaapi = Some(VaapiInternals { display, context, ref_pics, surfaces, ifmt });
+ self.needs_derive = needs_derive;
- let vinfo = NAVideoInfo::new(width, height, false, YUV420_FORMAT);
+ let vinfo = NAVideoInfo::new(vinfo.get_width(), vinfo.get_height(), false, YUV420_FORMAT);
self.info = NACodecInfo::new_ref(info.get_name(), NACodecTypeInfo::Video(vinfo), info.get_extradata()).into_ref();
self.out_frm = alloc_video_buffer(vinfo, 4)?;
let is_ref = frm.is_ref;
let ftype = frm.ftype;
if let Ok(pic) = frm.pic.sync() {
- let _ = fill_frame(vactx.ifmt, &pic, &mut self.out_frm);
+ let _ = fill_frame(vactx.ifmt, &pic, &mut self.out_frm, self.needs_derive);
if !is_ref {
if let Ok(surf) = pic.take_surface() {
let is_ref = frm.is_ref;
let ftype = frm.ftype;
if let Ok(pic) = frm.pic.sync() {
- let _ = fill_frame(vactx.ifmt, &pic, &mut self.out_frm);
+ let _ = fill_frame(vactx.ifmt, &pic, &mut self.out_frm, self.needs_derive);
if !is_ref {
if let Ok(surf) = pic.take_surface() {
projects / nihav-player.git / blobdiff