From: Kostya Shishkov Date: Thu, 14 Feb 2019 15:14:06 +0000 (+0100) Subject: scaler initial work X-Git-Url: https://git.nihav.org/?a=commitdiff_plain;h=03accf760cf79bc56d7c6dc6e82cd885fb7e1e13;p=nihav.git scaler initial work --- diff --git a/nihav-core/src/lib.rs b/nihav-core/src/lib.rs index 615a0cf..d36297f 100644 --- a/nihav-core/src/lib.rs +++ b/nihav-core/src/lib.rs @@ -10,6 +10,7 @@ pub mod io; pub mod refs; pub mod register; pub mod detect; +pub mod scale; #[cfg(feature="dsp")] pub mod dsp; diff --git a/nihav-core/src/scale/colorcvt.rs b/nihav-core/src/scale/colorcvt.rs new file mode 100644 index 0000000..2e2f299 --- /dev/null +++ b/nihav-core/src/scale/colorcvt.rs @@ -0,0 +1,323 @@ +use super::*; +use super::kernel::Kernel; + +const YUV_PARAMS: &[[f32; 2]] = &[ + [ 0.333, 0.333 ], // RGB + [ 0.2126, 0.0722 ], // ITU-R BT709 + [ 0.333, 0.333 ], // unspecified + [ 0.333, 0.333 ], // reserved + [ 0.299, 0.114 ], // ITU-R BT601 + [ 0.299, 0.114 ], // ITU-R BT470 + [ 0.299, 0.114 ], // SMPTE 170M + [ 0.212, 0.087 ], // SMPTE 240M + [ 0.333, 0.333 ], // YCoCg + [ 0.2627, 0.0593 ], // ITU-R BT2020 + [ 0.2627, 0.0593 ], // ITU-R BT2020 +]; + +const BT_PAL_COEFFS: [f32; 2] = [ 0.493, 0.877 ]; + +const SMPTE_NTSC_COEFFS: &[f32; 4] = &[ -0.268, 0.7358, 0.4127, 0.4778 ]; + +/*const RGB2YCOCG: [[f32; 3]; 3] = [ + [ 0.25, 0.5, 0.25 ], + [ -0.25, 0.5, -0.25 ], + [ 0.5, 0.0, -0.5 ] +]; +const YCOCG2RGB: [[f32; 3]; 3] = [ + [ 1.0, -1.0, 1.0 ], + [ 1.0, 1.0, 0.0 ], + [ 1.0, -1.0, -1.0 ] +]; + +const XYZ2RGB: [[f32; 3]; 3] = [ + [ 0.49, 0.31, 0.2 ], + [ 0.17697, 0.8124, 0.01063 ], + [ 0.0, 0.01, 0.99 ] +]; +const RGB2XYZ: [[f32; 3]; 3] = [ + [ 2.364613, -0.89654, -0.46807 ], + [ -0.515167, 1.42641, 0.08876 ], + [ 0.0052, -0.01441, 1.00920 ] +];*/ + +fn make_rgb2yuv(kr: f32, kb: f32, mat: &mut [[f32; 3]; 3]) { + // Y + mat[0][0] = kr; + mat[0][1] = 1.0 - kr - kb; + mat[0][2] = kb; + // Cb + mat[1][0] = -mat[0][0] * 0.5 / (1.0 - kb); + mat[1][1] = -mat[0][1] * 0.5 / (1.0 - kb); + mat[1][2] = 0.5; + // Cr + mat[2][0] = 0.5; + mat[2][1] = -mat[0][1] * 0.5 / (1.0 - kr); + mat[2][2] = -mat[0][2] * 0.5 / (1.0 - kr); +} + +fn make_yuv2rgb(kr: f32, kb: f32, mat: &mut [[f32; 3]; 3]) { + let kg = 1.0 - kr - kb; + + // R + mat[0][0] = 1.0; + mat[0][1] = 0.0; + mat[0][2] = 2.0 * (1.0 - kr); + // G + mat[1][0] = 1.0; + mat[1][1] = -kb * 2.0 * (1.0 - kb) / kg; + mat[1][2] = -kr * 2.0 * (1.0 - kr) / kg; + // B + mat[2][0] = 1.0; + mat[2][1] = 2.0 * (1.0 - kb); + mat[2][2] = 0.0; +} + +fn apply_pal_rgb2yuv(eu: f32, ev: f32, mat: &mut [[f32; 3]; 3]) { + let ufac = 2.0 * (1.0 - mat[0][2]) * eu; + let vfac = 2.0 * (1.0 - mat[0][0]) * ev; + + // U + mat[1][0] *= ufac; + mat[1][1] *= ufac; + mat[1][2] = eu * (1.0 - mat[0][2]); + // V + mat[2][0] = ev * (1.0 - mat[0][0]); + mat[2][1] *= vfac; + mat[2][2] *= vfac; +} + +fn apply_pal_yuv2rgb(eu: f32, ev: f32, mat: &mut [[f32; 3]; 3]) { + let ufac = 1.0 / (mat[2][1] * eu); + let vfac = 1.0 / (mat[0][2] * ev); + + // R + mat[0][2] *= vfac; + // G + mat[1][1] *= ufac; + mat[1][2] *= vfac; + // B + mat[2][1] *= ufac; +} + +fn apply_ntsc_rgb2yiq(params: &[f32; 4], mat: &mut [[f32; 3]; 3]) { + let ufac = 2.0 * (1.0 - mat[0][2]); + let vfac = 2.0 * (1.0 - mat[0][0]); + let mut tmp: [[f32; 3]; 2] = [[0.0; 3]; 2]; + + for i in 0..3 { + tmp[0][i] = mat[1][i] * ufac; + tmp[1][i] = mat[2][i] * vfac; + } + for i in 0..3 { + mat[1][i] = params[0] * tmp[0][i] + params[1] * tmp[1][i]; + mat[2][i] = params[2] * tmp[0][i] + params[3] * tmp[1][i]; + } +} + +fn subm_det(mat: &[[f32; 3]; 3], col: usize, row: usize) -> f32 { + let row0 = if row == 0 { 1 } else { 0 }; + let row1 = if (row == 1) || (row0 == 1) { 2 } else { 1 }; + let col0 = if col == 0 { 1 } else { 0 }; + let col1 = if (col == 1) || (col0 == 1) { 2 } else { 1 }; + + let det = mat[row0][col0] * mat[row1][col1] - mat[row0][col1] * mat[row1][col0]; + if ((col ^ row) & 1) == 0 { + det + } else { + -det + } +} + +fn invert_matrix(mat: &mut [[f32; 3]; 3]) { + let d00 = subm_det(mat, 0, 0); + let d01 = subm_det(mat, 0, 1); + let d02 = subm_det(mat, 0, 2); + let d10 = subm_det(mat, 1, 0); + let d11 = subm_det(mat, 1, 1); + let d12 = subm_det(mat, 1, 2); + let d20 = subm_det(mat, 2, 0); + let d21 = subm_det(mat, 2, 1); + let d22 = subm_det(mat, 2, 2); + let det = 1.0 / (mat[0][0] * d00 + mat[0][1] * d10 + mat[0][2] * d20).abs(); + + mat[0][0] = det * d00; + mat[0][1] = det * d01; + mat[0][2] = det * d02; + mat[1][0] = det * d10; + mat[1][1] = det * d11; + mat[1][2] = det * d12; + mat[2][0] = det * d20; + mat[2][1] = det * d21; + mat[2][2] = det * d22; +} + +fn matrix_mul(mat: &[[f32; 3]; 3], a: f32, b: f32, c: f32) -> (f32, f32, f32) { + (a * mat[0][0] + b * mat[0][1] + c * mat[0][2], + a * mat[1][0] + b * mat[1][1] + c * mat[1][2], + a * mat[2][0] + b * mat[2][1] + c * mat[2][2] ) +} + +#[derive(Default)] +struct RgbToYuv { + matrix: [[f32; 3]; 3], +} + +impl RgbToYuv { + fn new() -> Self { Self::default() } +} + +impl Kernel for RgbToYuv { + fn init(&mut self, in_fmt: &ScaleInfo, dest_fmt: &ScaleInfo) -> ScaleResult { + let mut df = dest_fmt.fmt; +//todo coeff selection + make_rgb2yuv(YUV_PARAMS[2][0], YUV_PARAMS[2][1], &mut self.matrix); + if let ColorModel::YUV(yuvsm) = df.get_model() { + match yuvsm { + YUVSubmodel::YCbCr => {}, + YUVSubmodel::YIQ => { apply_ntsc_rgb2yiq(SMPTE_NTSC_COEFFS, &mut self.matrix); }, + YUVSubmodel::YUVJ => { apply_pal_rgb2yuv(BT_PAL_COEFFS[0], BT_PAL_COEFFS[1], &mut self.matrix); }, + }; + } else { + return Err(ScaleError::InvalidArgument); + } + for i in 0..MAX_CHROMATONS { + if let Some(ref mut chr) = df.comp_info[i] { + chr.packed = false; + chr.comp_offs = i as u8; + chr.h_ss = 0; + chr.v_ss = 0; + } + } +println!(" [intermediate format {}]", df); + let res = alloc_video_buffer(NAVideoInfo::new(in_fmt.width, in_fmt.height, false, df), 3); + if res.is_err() { return Err(ScaleError::AllocError); } + Ok(res.unwrap()) + } + fn process(&mut self, pic_in: &NABufferType, pic_out: &mut NABufferType) { + if let (Some(ref sbuf), Some(ref mut dbuf)) = (pic_in.get_vbuf(), pic_out.get_vbuf()) { + let istrides = [sbuf.get_stride(0), sbuf.get_stride(1), sbuf.get_stride(2)]; + let dstrides = [dbuf.get_stride(0), dbuf.get_stride(1), dbuf.get_stride(2)]; + let (w, h) = sbuf.get_dimensions(0); + + let mut roff = sbuf.get_offset(0); + let mut goff = sbuf.get_offset(1); + let mut boff = sbuf.get_offset(2); + let mut yoff = dbuf.get_offset(0); + let mut uoff = dbuf.get_offset(1); + let mut voff = dbuf.get_offset(2); + let src = sbuf.get_data(); + let dst = dbuf.get_data_mut().unwrap(); + for _y in 0..h { + for x in 0..w { + let r = src[roff + x] as f32; + let g = src[goff + x] as f32; + let b = src[boff + x] as f32; + let (y, u, v) = matrix_mul(&self.matrix, r, g, b); + + dst[yoff + x] = (y as i16).max(0).min(255) as u8; + dst[uoff + x] = ((u as i16).max(-128).min(128) + 128) as u8; + dst[voff + x] = ((v as i16).max(-128).min(128) + 128) as u8; + } + roff += istrides[0]; + goff += istrides[1]; + boff += istrides[2]; + yoff += dstrides[0]; + uoff += dstrides[1]; + voff += dstrides[2]; + } + } + } +} + +pub fn create_rgb2yuv() -> Box { + Box::new(RgbToYuv::new()) +} + +#[derive(Default)] +struct YuvToRgb { + matrix: [[f32; 3]; 3], +} + +impl YuvToRgb { + fn new() -> Self { Self::default() } +} + +impl Kernel for YuvToRgb { + fn init(&mut self, in_fmt: &ScaleInfo, dest_fmt: &ScaleInfo) -> ScaleResult { + let mut df = dest_fmt.fmt; +//todo coeff selection + make_yuv2rgb(YUV_PARAMS[2][0], YUV_PARAMS[2][1], &mut self.matrix); + if let ColorModel::YUV(yuvsm) = in_fmt.fmt.get_model() { + match yuvsm { + YUVSubmodel::YCbCr => {}, + YUVSubmodel::YIQ => { + make_rgb2yuv(YUV_PARAMS[2][0], YUV_PARAMS[2][1], &mut self.matrix); + apply_ntsc_rgb2yiq(SMPTE_NTSC_COEFFS, &mut self.matrix); + invert_matrix(&mut self.matrix); + }, + YUVSubmodel::YUVJ => { + apply_pal_yuv2rgb(BT_PAL_COEFFS[0], BT_PAL_COEFFS[1], &mut self.matrix); + }, + }; + } else { + return Err(ScaleError::InvalidArgument); + } + for i in 0..MAX_CHROMATONS { + if let Some(ref mut chr) = df.comp_info[i] { + chr.packed = false; + chr.comp_offs = i as u8; + } + } +println!(" [intermediate format {}]", df); + let res = alloc_video_buffer(NAVideoInfo::new(in_fmt.width, in_fmt.height, false, df), 3); + if res.is_err() { return Err(ScaleError::AllocError); } + Ok(res.unwrap()) + } + fn process(&mut self, pic_in: &NABufferType, pic_out: &mut NABufferType) { + if let (Some(ref sbuf), Some(ref mut dbuf)) = (pic_in.get_vbuf(), pic_out.get_vbuf()) { + let istrides = [sbuf.get_stride(0), sbuf.get_stride(1), sbuf.get_stride(2)]; + let dstrides = [dbuf.get_stride(0), dbuf.get_stride(1), dbuf.get_stride(2)]; + let (w, h) = sbuf.get_dimensions(0); + let (sv0, sh0) = sbuf.get_info().get_format().get_chromaton(1).unwrap().get_subsampling(); + let (sv1, sh1) = sbuf.get_info().get_format().get_chromaton(2).unwrap().get_subsampling(); + + let uhmask = (1 << sh0) - 1; + let vhmask = (1 << sh1) - 1; + let mut roff = dbuf.get_offset(0); + let mut goff = dbuf.get_offset(1); + let mut boff = dbuf.get_offset(2); + let mut yoff = sbuf.get_offset(0); + let mut uoff = sbuf.get_offset(1); + let mut voff = sbuf.get_offset(2); + let src = sbuf.get_data(); + let dst = dbuf.get_data_mut().unwrap(); + for y in 0..h { + for x in 0..w { + let y = src[yoff + x] as f32; + let u = ((src[uoff + (x >> sv0)] as i16) - 128) as f32; + let v = ((src[voff + (x >> sv1)] as i16) - 128) as f32; + + let (r, g, b) = matrix_mul(&self.matrix, y, u, v); + dst[roff + x] = (r as i16).max(0).min(255) as u8; + dst[goff + x] = (g as i16).max(0).min(255) as u8; + dst[boff + x] = (b as i16).max(0).min(255) as u8; + } + roff += dstrides[0]; + goff += dstrides[1]; + boff += dstrides[2]; + yoff += istrides[0]; + if (y & uhmask) == uhmask { + uoff += istrides[1]; + } + if (y & vhmask) == vhmask { + voff += istrides[2]; + } + } + } + } +} + +pub fn create_yuv2rgb() -> Box { + Box::new(YuvToRgb::new()) +} diff --git a/nihav-core/src/scale/kernel.rs b/nihav-core/src/scale/kernel.rs new file mode 100644 index 0000000..84a69b4 --- /dev/null +++ b/nihav-core/src/scale/kernel.rs @@ -0,0 +1,7 @@ +use crate::frame::*; +use super::{ScaleInfo, ScaleResult}; + +pub trait Kernel { + fn init(&mut self, in_fmt: &ScaleInfo, dest_fmt: &ScaleInfo) -> ScaleResult; + fn process(&mut self, pic_in: &NABufferType, pic_out: &mut NABufferType); +} diff --git a/nihav-core/src/scale/mod.rs b/nihav-core/src/scale/mod.rs new file mode 100644 index 0000000..c569b2f --- /dev/null +++ b/nihav-core/src/scale/mod.rs @@ -0,0 +1,348 @@ +use crate::frame::*; + +mod kernel; + +mod colorcvt; +mod repack; +mod scale; + +#[derive(Clone,Copy,PartialEq)] +pub struct ScaleInfo { + pub fmt: NAPixelFormaton, + pub width: usize, + pub height: usize, +} + +impl std::fmt::Display for ScaleInfo { + fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result { + write!(f, "({}x{}, {})", self.width, self.height, self.fmt) + } +} + +#[derive(Debug,Clone,Copy,PartialEq)] +#[allow(dead_code)] +pub enum ScaleError { + NoFrame, + AllocError, + InvalidArgument, + NotImplemented, + Bug, +} + +pub type ScaleResult = Result; + +/*trait Kernel { + fn init(&mut self, in_fmt: &ScaleInfo, dest_fmt: &ScaleInfo) -> ScaleResult; + fn process(&mut self, pic_in: &NABufferType, pic_out: &mut NABufferType); +}*/ + +struct KernelDesc { + name: &'static str, + create: fn () -> Box, +} + +impl KernelDesc { + fn find(name: &str) -> ScaleResult> { + for kern in KERNELS.iter() { + if kern.name == name { + return Ok((kern.create)()); + } + } + Err(ScaleError::InvalidArgument) + } +} + +const KERNELS: &[KernelDesc] = &[ + KernelDesc { name: "pack", create: repack::create_pack }, + KernelDesc { name: "unpack", create: repack::create_unpack }, + KernelDesc { name: "depal", create: repack::create_depal }, + KernelDesc { name: "scale", create: scale::create_scale }, + KernelDesc { name: "rgb_to_yuv", create: colorcvt::create_rgb2yuv }, + KernelDesc { name: "yuv_to_rgb", create: colorcvt::create_yuv2rgb }, +]; + +struct Stage { + fmt_out: ScaleInfo, + tmp_pic: NABufferType, + next: Option>, + worker: Box, +} + +pub fn get_scale_fmt_from_pic(pic: &NABufferType) -> ScaleInfo { + let info = pic.get_video_info().unwrap(); + ScaleInfo { fmt: info.get_format(), width: info.get_width(), height: info.get_height() } +} + +impl Stage { + fn new(name: &str, in_fmt: &ScaleInfo, dest_fmt: &ScaleInfo) -> ScaleResult { + let mut worker = KernelDesc::find(name)?; + let tmp_pic = worker.init(in_fmt, dest_fmt)?; + let fmt_out = get_scale_fmt_from_pic(&tmp_pic); + Ok(Self { fmt_out, tmp_pic, next: None, worker }) + } + fn add(&mut self, new: Stage) { + if let Some(ref mut next) = self.next { + next.add(new); + } else { + self.next = Some(Box::new(new)); + } + } + fn process(&mut self, pic_in: &NABufferType, pic_out: &mut NABufferType) -> ScaleResult<()> { + if let Some(ref mut nextstage) = self.next { + self.worker.process(pic_in, &mut self.tmp_pic); + nextstage.process(&self.tmp_pic, pic_out)?; + } else { + self.worker.process(pic_in, pic_out); + } + Ok(()) + } + fn drop_last_tmp(&mut self) { + if let Some(ref mut nextstage) = self.next { + nextstage.drop_last_tmp(); + } else { + self.tmp_pic = NABufferType::None; + } + } +} + +pub struct NAScale { + fmt_in: ScaleInfo, + fmt_out: ScaleInfo, + just_convert: bool, + pipeline: Option, +} + +fn check_format(in_fmt: NAVideoInfo, ref_fmt: &ScaleInfo, just_convert: bool) -> ScaleResult<()> { + if in_fmt.get_format() != ref_fmt.fmt { return Err(ScaleError::InvalidArgument); } + if !just_convert && (in_fmt.get_width() != ref_fmt.width || in_fmt.get_height() != ref_fmt.height) { + return Err(ScaleError::InvalidArgument); + } + Ok(()) +} + +fn copy(pic_in: &NABufferType, pic_out: &mut NABufferType) +{ + if let (Some(ref sbuf), Some(ref mut dbuf)) = (pic_in.get_vbuf(), pic_out.get_vbuf()) { + let sdata = sbuf.get_data(); + let ddata = dbuf.get_data_mut().unwrap(); + ddata.copy_from_slice(&sdata[0..]); + } else { + unimplemented!(); + } +} + +macro_rules! add_stage { + ($head:expr, $new:expr) => { + if let Some(ref mut h) = $head { + h.add($new); + } else { + $head = Some($new); + } + }; +} +fn is_better_fmt(a: &ScaleInfo, b: &ScaleInfo) -> bool { + if (a.width >= b.width) && (a.height >= b.height) { + return true; + } + if a.fmt.get_max_depth() > b.fmt.get_max_depth() { + return true; + } + if a.fmt.get_max_subsampling() < b.fmt.get_max_subsampling() { + return true; + } + false +} +fn build_pipeline(ifmt: &ScaleInfo, ofmt: &ScaleInfo, just_convert: bool) -> ScaleResult> { + let inname = ifmt.fmt.get_model().get_short_name(); + let outname = ofmt.fmt.get_model().get_short_name(); + +println!("convert {} -> {}", ifmt, ofmt); + let mut needs_scale = !just_convert; + if (ofmt.fmt.get_max_subsampling() > 0) && + (ofmt.fmt.get_max_subsampling() != ifmt.fmt.get_max_subsampling()) { + needs_scale = true; + } + let needs_unpack = needs_scale || !ifmt.fmt.is_unpacked(); + let needs_pack = !ofmt.fmt.is_unpacked(); + let mut needs_convert = false; + if inname != outname { + needs_convert = true; + } + let scale_before_cvt = is_better_fmt(&ifmt, &ofmt) && needs_convert + && (ofmt.fmt.get_max_subsampling() == 0); +//todo stages for model and gamma conversion + + let mut stages: Option = None; + let mut cur_fmt = *ifmt; + + if needs_unpack { +println!("[adding unpack]"); + let new_stage; + if !cur_fmt.fmt.is_paletted() { + new_stage = Stage::new("unpack", &cur_fmt, &ofmt)?; + } else { + new_stage = Stage::new("depal", &cur_fmt, &ofmt)?; + } + cur_fmt = new_stage.fmt_out; + add_stage!(stages, new_stage); + } + if needs_scale && scale_before_cvt { +println!("[adding scale]"); + let new_stage = Stage::new("scale", &cur_fmt, &ofmt)?; + cur_fmt = new_stage.fmt_out; + add_stage!(stages, new_stage); + } + if needs_convert { +println!("[adding convert]"); + let cvtname = format!("{}_to_{}", inname, outname); +println!("[{}]", cvtname); + let new_stage = Stage::new(&cvtname, &cur_fmt, &ofmt)?; +//todo if fails try converting via RGB or YUV + cur_fmt = new_stage.fmt_out; + add_stage!(stages, new_stage); +//todo alpha plane copy/add + } + if needs_scale && !scale_before_cvt { +println!("[adding scale]"); + let new_stage = Stage::new("scale", &cur_fmt, &ofmt)?; + cur_fmt = new_stage.fmt_out; + add_stage!(stages, new_stage); + } +//todo flip if needed + if needs_pack { +println!("[adding pack]"); + let new_stage = Stage::new("pack", &cur_fmt, &ofmt)?; + //cur_fmt = new_stage.fmt_out; + add_stage!(stages, new_stage); + } + + if let Some(ref mut head) = stages { + head.drop_last_tmp(); + } + + Ok(stages) +} + +impl NAScale { + pub fn new(fmt_in: ScaleInfo, fmt_out: ScaleInfo) -> ScaleResult { + let pipeline; + let just_convert = (fmt_in.width == fmt_out.width) && (fmt_in.height == fmt_out.height); + if fmt_in != fmt_out { + pipeline = build_pipeline(&fmt_in, &fmt_out, just_convert)?; + } else { + pipeline = None; + } + Ok(Self { fmt_in, fmt_out, just_convert, pipeline }) + } + pub fn needs_processing(&self) -> bool { self.pipeline.is_some() } + pub fn get_in_fmt(&self) -> ScaleInfo { self.fmt_in } + pub fn get_out_fmt(&self) -> ScaleInfo { self.fmt_out } + pub fn convert(&mut self, pic_in: &NABufferType, pic_out: &mut NABufferType) -> ScaleResult<()> { + let in_info = pic_in.get_video_info(); + let out_info = pic_out.get_video_info(); + if in_info.is_none() || out_info.is_none() { return Err(ScaleError::InvalidArgument); } + let in_info = in_info.unwrap(); + let out_info = out_info.unwrap(); + if self.just_convert && + (in_info.get_width() != out_info.get_width() || in_info.get_height() != out_info.get_height()) { + return Err(ScaleError::InvalidArgument); + } + check_format(in_info, &self.fmt_in, self.just_convert)?; + check_format(out_info, &self.fmt_out, self.just_convert)?; + if let Some(ref mut pipe) = self.pipeline { + pipe.process(pic_in, pic_out) + } else { + copy(pic_in, pic_out); + Ok(()) + } + } +} + +#[cfg(test)] +mod test { + use super::*; + + fn fill_pic(pic: &mut NABufferType, val: u8) { + if let Some(ref mut buf) = pic.get_vbuf() { + let data = buf.get_data_mut().unwrap(); + for el in data.iter_mut() { *el = val; } + } else if let Some(ref mut buf) = pic.get_vbuf16() { + let data = buf.get_data_mut().unwrap(); + for el in data.iter_mut() { *el = val as u16; } + } else if let Some(ref mut buf) = pic.get_vbuf32() { + let data = buf.get_data_mut().unwrap(); + for el in data.iter_mut() { *el = (val as u32) * 0x01010101; } + } + } + #[test] + fn test_convert() { + let mut in_pic = alloc_video_buffer(NAVideoInfo::new(1, 1, false, RGB565_FORMAT), 3).unwrap(); + fill_pic(&mut in_pic, 42); + let mut out_pic = alloc_video_buffer(NAVideoInfo::new(1, 1, false, RGB24_FORMAT), 3).unwrap(); + fill_pic(&mut out_pic, 0); + let ifmt = get_scale_fmt_from_pic(&in_pic); + let ofmt = get_scale_fmt_from_pic(&out_pic); + let mut scaler = NAScale::new(ifmt, ofmt).unwrap(); + scaler.convert(&in_pic, &mut out_pic).unwrap(); + let obuf = out_pic.get_vbuf().unwrap(); + let odata = obuf.get_data(); + assert_eq!(odata[0], 0x0); + assert_eq!(odata[1], 0x4); + assert_eq!(odata[2], 0x52); + + let mut in_pic = alloc_video_buffer(NAVideoInfo::new(4, 4, false, RGB24_FORMAT), 3).unwrap(); + fill_pic(&mut in_pic, 42); + let mut out_pic = alloc_video_buffer(NAVideoInfo::new(4, 4, false, YUV420_FORMAT), 3).unwrap(); + fill_pic(&mut out_pic, 0); + let ifmt = get_scale_fmt_from_pic(&in_pic); + let ofmt = get_scale_fmt_from_pic(&out_pic); + let mut scaler = NAScale::new(ifmt, ofmt).unwrap(); + scaler.convert(&in_pic, &mut out_pic).unwrap(); + let obuf = out_pic.get_vbuf().unwrap(); + let yoff = obuf.get_offset(0); + let uoff = obuf.get_offset(1); + let voff = obuf.get_offset(2); + let odata = obuf.get_data(); + assert_eq!(odata[yoff], 42); + assert!(((odata[uoff] ^ 0x80) as i8).abs() <= 1); + assert!(((odata[voff] ^ 0x80) as i8).abs() <= 1); + let mut scaler = NAScale::new(ofmt, ifmt).unwrap(); + scaler.convert(&out_pic, &mut in_pic).unwrap(); + let obuf = in_pic.get_vbuf().unwrap(); + let odata = obuf.get_data(); + assert_eq!(odata[0], 42); + } + #[test] + fn test_scale() { + let mut in_pic = alloc_video_buffer(NAVideoInfo::new(2, 2, false, RGB565_FORMAT), 3).unwrap(); + fill_pic(&mut in_pic, 42); + let mut out_pic = alloc_video_buffer(NAVideoInfo::new(3, 3, false, RGB565_FORMAT), 3).unwrap(); + fill_pic(&mut out_pic, 0); + let ifmt = get_scale_fmt_from_pic(&in_pic); + let ofmt = get_scale_fmt_from_pic(&out_pic); + let mut scaler = NAScale::new(ifmt, ofmt).unwrap(); + scaler.convert(&in_pic, &mut out_pic).unwrap(); + let obuf = out_pic.get_vbuf16().unwrap(); + let odata = obuf.get_data(); + assert_eq!(odata[0], 42); + } + #[test] + fn test_scale_and_convert() { + let mut in_pic = alloc_video_buffer(NAVideoInfo::new(7, 3, false, RGB565_FORMAT), 3).unwrap(); + fill_pic(&mut in_pic, 42); + let mut out_pic = alloc_video_buffer(NAVideoInfo::new(4, 4, false, YUV420_FORMAT), 3).unwrap(); + fill_pic(&mut out_pic, 0); + let ifmt = get_scale_fmt_from_pic(&in_pic); + let ofmt = get_scale_fmt_from_pic(&out_pic); + let mut scaler = NAScale::new(ifmt, ofmt).unwrap(); + scaler.convert(&in_pic, &mut out_pic).unwrap(); + let obuf = out_pic.get_vbuf().unwrap(); + let yoff = obuf.get_offset(0); + let uoff = obuf.get_offset(1); + let voff = obuf.get_offset(2); + let odata = obuf.get_data(); + assert_eq!(odata[yoff], 28); + assert_eq!(odata[uoff], 154); + assert_eq!(odata[voff], 103); + } +} diff --git a/nihav-core/src/scale/repack.rs b/nihav-core/src/scale/repack.rs new file mode 100644 index 0000000..4f2ce99 --- /dev/null +++ b/nihav-core/src/scale/repack.rs @@ -0,0 +1,298 @@ +use crate::formats::*; +use super::*; +use super::kernel::Kernel; + +fn convert_depth(val: u32, indepth: u8, outdepth: u8) -> u32 { + if indepth >= outdepth { + val >> (indepth - outdepth) + } else { + let mut val2 = val << (outdepth - indepth); + let mut shift = outdepth - indepth; + while shift >= indepth { + shift -= indepth; + val2 |= val << shift; + } + if shift > 0 { + val2 |= val >> (indepth - shift); + } + val2 + } +} + +#[derive(Default)] +struct PackKernel { + shifts: [u8; MAX_CHROMATONS], + depths: [u8; MAX_CHROMATONS], + ncomps: usize, + osize: [u8; MAX_CHROMATONS], + ooff: [usize; MAX_CHROMATONS], +} + +impl PackKernel { + fn new() -> Self { Self::default() } +} + +impl Kernel for PackKernel { + fn init(&mut self, in_fmt: &ScaleInfo, dest_fmt: &ScaleInfo) -> ScaleResult { + self.ncomps = in_fmt.fmt.components as usize; + for i in 0..self.ncomps { + let ichr = in_fmt.fmt.comp_info[i].unwrap(); + let ochr = dest_fmt.fmt.comp_info[i].unwrap(); + self.shifts[i] = ochr.shift; + self.osize[i] = ochr.depth; + self.depths[i] = ichr.depth; + self.ooff[i] = ochr.comp_offs as usize; + } + let res = alloc_video_buffer(NAVideoInfo::new(in_fmt.width, in_fmt.height, false, dest_fmt.fmt), 3); + if res.is_err() { return Err(ScaleError::AllocError); } + Ok(res.unwrap()) + } + fn process(&mut self, pic_in: &NABufferType, pic_out: &mut NABufferType) { + if let Some(ref buf) = pic_in.get_vbuf() { + if let Some(ref mut dbuf) = pic_out.get_vbuf() { + let dstride = dbuf.get_stride(0); + for comp in 0..self.ncomps { + let ioff = buf.get_offset(comp); + let istride = buf.get_stride(comp); + let step = dbuf.get_info().get_format().get_chromaton(comp).unwrap().get_step() as usize; + let (w, h) = dbuf.get_dimensions(comp); + let sdata = buf.get_data(); + let sdata = &sdata[ioff..]; + let ddata = dbuf.get_data_mut().unwrap(); + for (src, dst) in sdata.chunks(istride).zip(ddata.chunks_mut(dstride)).take(h) { + for x in 0..w { + dst[x * step + self.ooff[comp]] = convert_depth(src[x] as u32, self.depths[comp], self.osize[comp]) as u8; + } + } + } + } else if let Some(ref mut dbuf) = pic_out.get_vbuf16() { + let (w, h) = dbuf.get_dimensions(0); + let dstride = dbuf.get_stride(0); + let ddata = dbuf.get_data_mut().unwrap(); + let src = buf.get_data(); + let mut ioff: [usize; MAX_CHROMATONS] = [0; MAX_CHROMATONS]; + let mut istride: [usize; MAX_CHROMATONS] = [0; MAX_CHROMATONS]; + for comp in 0..self.ncomps { + ioff[comp] = buf.get_offset(comp); + istride[comp] = buf.get_stride(comp); + } + for dst in ddata.chunks_mut(dstride).take(h) { + for x in 0..w { + let mut elem: u32 = 0; + for comp in 0..self.ncomps { + let c = src[ioff[comp] + x] as u32; + elem |= convert_depth(c, self.depths[comp], self.osize[comp]) << self.shifts[comp]; + } + dst[x] = elem as u16; + } + for comp in 0..self.ncomps { + ioff[comp] += istride[comp]; + } + } + } else { +unimplemented!(); + } + } else if let Some(ref _buf) = pic_in.get_vbuf16() { +unimplemented!(); + } else if let Some(ref _buf) = pic_in.get_vbuf32() { +unimplemented!(); + } else { + unreachable!(); + } + } +} + +pub fn create_pack() -> Box { + Box::new(PackKernel::new()) +} + +#[derive(Default)] +struct UnpackKernel { + shifts: [u8; MAX_CHROMATONS], + masks: [u32; MAX_CHROMATONS], + depths: [u8; MAX_CHROMATONS], + ncomps: usize, + osize: [u8; MAX_CHROMATONS], +} + +impl UnpackKernel { + fn new() -> Self { Self::default() } +} + +impl Kernel for UnpackKernel { + fn init(&mut self, in_fmt: &ScaleInfo, dest_fmt: &ScaleInfo) -> ScaleResult { + self.ncomps = in_fmt.fmt.components as usize; + let mut chr: Vec> = Vec::with_capacity(MAX_CHROMATONS); + for i in 0..self.ncomps { + let ichr = in_fmt.fmt.comp_info[i].unwrap(); + let ochr = dest_fmt.fmt.comp_info[i].unwrap(); + self.shifts[i] = ichr.shift; + self.masks[i] = (1 << ichr.depth) - 1; + if ochr.depth > ichr.depth { + self.osize[i] = ochr.depth; + } else { + self.osize[i] = (ichr.depth + 7) & !7; + } + self.depths[i] = ichr.depth; + let mut dchr = ichr; + dchr.packed = false; + dchr.depth = self.osize[i]; + dchr.shift = 0; + dchr.comp_offs = 0; + dchr.next_elem = 0; + chr.push(Some(dchr)); + } + let mut df = in_fmt.fmt; + for i in 0..self.ncomps { + df.comp_info[i] = chr[i]; + } +println!(" [intermediate format {}]", df); + let res = alloc_video_buffer(NAVideoInfo::new(in_fmt.width, in_fmt.height, false, df), 3); + if res.is_err() { return Err(ScaleError::AllocError); } + Ok(res.unwrap()) + } + fn process(&mut self, pic_in: &NABufferType, pic_out: &mut NABufferType) { + if let Some(ref buf) = pic_in.get_vbuf() { + let step = buf.get_info().get_format().get_chromaton(0).unwrap().get_step() as usize; + let mut soff: [usize; MAX_CHROMATONS] = [0; MAX_CHROMATONS]; + for i in 0..self.ncomps { + soff[i] = buf.get_info().get_format().get_chromaton(i).unwrap().get_offset() as usize; + } + let (w, h) = buf.get_dimensions(0); + let ioff = buf.get_offset(0); + let istride = buf.get_stride(0); + let sdata1 = buf.get_data(); + let sdata = &sdata1[ioff..]; + if let Some(ref mut dbuf) = pic_out.get_vbuf() { + let mut ostride: [usize; MAX_CHROMATONS] = [0; MAX_CHROMATONS]; + let mut offs: [usize; MAX_CHROMATONS] = [0; MAX_CHROMATONS]; + for i in 0..self.ncomps { + ostride[i] = dbuf.get_stride(i); + offs[i] = dbuf.get_offset(i); + } + let dst = dbuf.get_data_mut().unwrap(); + for src in sdata.chunks(istride).take(h) { + for x in 0..w { + for i in 0..self.ncomps { + dst[offs[i] + x] = src[x * step + soff[i]]; + } + } + for i in 0..self.ncomps { offs[i] += ostride[i]; } + } + } else { +unimplemented!(); + } + } else if let Some(ref buf) = pic_in.get_vbuf16() { + let (w, h) = buf.get_dimensions(0); + let ioff = buf.get_offset(0); + let istride = buf.get_stride(0); + let sdata1 = buf.get_data(); + let sdata = &sdata1[ioff..]; + if let Some(ref mut dbuf) = pic_out.get_vbuf() { + let mut ostride: [usize; MAX_CHROMATONS] = [0; MAX_CHROMATONS]; + let mut offs: [usize; MAX_CHROMATONS] = [0; MAX_CHROMATONS]; + for i in 0..self.ncomps { + ostride[i] = dbuf.get_stride(i); + offs[i] = dbuf.get_offset(i); + } + let dst = dbuf.get_data_mut().unwrap(); + for src in sdata.chunks(istride).take(h) { + for x in 0..w { + let elem = src[x] as u32; + for i in 0..self.ncomps { + dst[offs[i] + x] = convert_depth((elem >> self.shifts[i]) & self.masks[i], self.depths[i], self.osize[i]) as u8; + } + } + for i in 0..self.ncomps { offs[i] += ostride[i]; } + } + } else { +unimplemented!(); + } + } else if let Some(ref _buf) = pic_in.get_vbuf32() { +unimplemented!(); + } else { + unreachable!(); + } + } +} + +pub fn create_unpack() -> Box { + Box::new(UnpackKernel::new()) +} + +#[derive(Default)] +struct DepalKernel { + depths: [u8; MAX_CHROMATONS], + coffs: [usize; MAX_CHROMATONS], + ncomps: usize, + palstep: usize, +} + +impl DepalKernel { + fn new() -> Self { Self::default() } +} + +impl Kernel for DepalKernel { + fn init(&mut self, in_fmt: &ScaleInfo, _dest_fmt: &ScaleInfo) -> ScaleResult { +//todo select output more fitting for dest_fmt if possible + self.ncomps = in_fmt.fmt.components as usize; + let mut chr: Vec> = Vec::with_capacity(MAX_CHROMATONS); + self.palstep = in_fmt.fmt.elem_size as usize; + for i in 0..self.ncomps { + let ichr = in_fmt.fmt.comp_info[i].unwrap(); + self.coffs[i] = ichr.comp_offs as usize; + self.depths[i] = ichr.depth; + let mut dchr = ichr; + dchr.packed = false; + dchr.depth = 8; + dchr.shift = 0; + dchr.comp_offs = i as u8; + dchr.next_elem = 0; + chr.push(Some(dchr)); + } + let mut df = in_fmt.fmt; + df.palette = false; + for i in 0..self.ncomps { + df.comp_info[i] = chr[i]; + } +println!(" [intermediate format {}]", df); + let res = alloc_video_buffer(NAVideoInfo::new(in_fmt.width, in_fmt.height, false, df), 3); + if res.is_err() { return Err(ScaleError::AllocError); } + Ok(res.unwrap()) + } + fn process(&mut self, pic_in: &NABufferType, pic_out: &mut NABufferType) { + if let (Some(ref sbuf), Some(ref mut dbuf)) = (pic_in.get_vbuf(), pic_out.get_vbuf()) { + let ioff = sbuf.get_offset(0); + let paloff = sbuf.get_offset(1); + let (w, h) = sbuf.get_dimensions(0); + let istride = sbuf.get_stride(0); + let sdata1 = sbuf.get_data(); + let sdata = &sdata1[ioff..]; + let pal = &sdata1[paloff..]; + + let mut ostride: [usize; MAX_CHROMATONS] = [0; MAX_CHROMATONS]; + let mut offs: [usize; MAX_CHROMATONS] = [0; MAX_CHROMATONS]; + for i in 0..self.ncomps { + ostride[i] = dbuf.get_stride(i); + offs[i] = dbuf.get_offset(i); + } + let dst = dbuf.get_data_mut().unwrap(); + for src in sdata.chunks(istride).take(h) { + for x in 0..w { + let palidx = src[x] as usize; + for i in 0..self.ncomps { + let elem = pal[palidx * self.palstep + self.coffs[i]] as u32; + dst[offs[i] + x] = convert_depth(elem, self.depths[i], 8) as u8; + } + } + for i in 0..self.ncomps { offs[i] += ostride[i]; } + } + } else { + unreachable!(); + } + } +} + +pub fn create_depal() -> Box { + Box::new(DepalKernel::new()) +} diff --git a/nihav-core/src/scale/scale.rs b/nihav-core/src/scale/scale.rs new file mode 100644 index 0000000..197b7c4 --- /dev/null +++ b/nihav-core/src/scale/scale.rs @@ -0,0 +1,58 @@ +use super::*; +use super::kernel::Kernel; + +struct NNResampler {} + +impl NNResampler { + fn new() -> Self { Self{} } +} + +macro_rules! scale_loop { + ($sbuf:expr, $dbuf:expr) => { + let fmt = $sbuf.get_info().get_format(); + let ncomp = fmt.get_num_comp(); + for comp in 0..ncomp { + let istride = $sbuf.get_stride(comp); + let dstride = $dbuf.get_stride(comp); + let (sw, sh) = $sbuf.get_dimensions(comp); + let (dw, dh) = $dbuf.get_dimensions(comp); + let ioff = $sbuf.get_offset(comp); + let mut doff = $dbuf.get_offset(comp); + let src = $sbuf.get_data(); + let dst = $dbuf.get_data_mut().unwrap(); + for y in 0..dh { + let sy = y * sh / dh; + let soff = ioff + sy * istride; + for x in 0..dw { + let sx = x * sw / dw; + dst[doff + x] = src[soff + sx]; + } + doff += dstride; + } + } + }; +} + +impl Kernel for NNResampler { + fn init(&mut self, in_fmt: &ScaleInfo, dest_fmt: &ScaleInfo) -> ScaleResult { + let res = alloc_video_buffer(NAVideoInfo::new(dest_fmt.width, dest_fmt.height, false, in_fmt.fmt), 3); + if res.is_err() { return Err(ScaleError::AllocError); } + Ok(res.unwrap()) + } + fn process(&mut self, pic_in: &NABufferType, pic_out: &mut NABufferType) { + if let (Some(ref sbuf), Some(ref mut dbuf)) = (pic_in.get_vbuf(), pic_out.get_vbuf()) { + scale_loop!(sbuf, dbuf); + } else if let (Some(ref sbuf), Some(ref mut dbuf)) = (pic_in.get_vbuf16(), pic_out.get_vbuf16()) { + scale_loop!(sbuf, dbuf); + } else if let (Some(ref sbuf), Some(ref mut dbuf)) = (pic_in.get_vbuf32(), pic_out.get_vbuf32()) { + scale_loop!(sbuf, dbuf); + } else { + unreachable!(); + } + } +} + +pub fn create_scale() -> Box { + Box::new(NNResampler::new()) +} +