small optimizations and info binary

12 files changed, 63 insertions, 35 deletions

diff --git a/evc/scripts/bench_modes b/evc/scripts/bench_modes
index 7cc6abd..9fa17f6 100755
--- a/evc/scripts/bench_modes
+++ b/evc/scripts/bench_modes
@@ -4,14 +4,15 @@ set h $argv[2]
 set t $argv[3]
 
 ffmpeg -hide_banner -i $argv[4] -to {$t} -vf scale={$w}x{$h},fps=30,format=rgb24 -f rawvideo pipe:1 >samples/raw
-ffmpeg -hide_banner -y -i $argv[4] -to {$t} -vf scale={$w}x{$h},fps=30,format=rgb24 -c:v vp9 samples/reference.webm
+# ffmpeg -hide_banner -y -i $argv[4] -to {$t} -vf scale={$w}x{$h},fps=30,format=rgb24 -c:v vp9 samples/reference.webm
 
 echo
 echo "file: "$argv[4]
 echo "resolution: "{$w}x{$h}
 echo "frames: "(math $t \* 30)
 echo "reference (raw): "(du -h samples/raw | cut -f 1)
-echo "reference (vp8): "(du -h samples/reference.webm | cut -f 1)
+echo "reference (input): "(du -h $argv[4] | cut -f 1)
+# echo "reference (vp8): "(du -h samples/reference.webm | cut -f 1)
 for mode in trivial simple-exhaustive simple-fast advanced advanced-partial
     echo -----------
     echo "mode: $mode"
diff --git a/evc/scripts/bench_out b/evc/scripts/bench_out
new file mode 100755
index 0000000..5bb0639
--- /dev/null
+++ b/evc/scripts/bench_out
@@ -0,0 +1,10 @@
+#!/bin/fish
+set w $argv[1]
+set h $argv[2]
+
+for mode in trivial simple-exhaustive simple-fast advanced advanced-partial
+    cargo run --release --bin decode -- --debug < samples/encoded-$mode | 
+        ffmpeg -y -hide_banner -framerate 25 -video_size {$w}x{$h} -pixel_format rgb24 -f rawvideo -i pipe:0 samples/decoded-$mode-debug.mp4
+    cargo run --release --bin decode -- < samples/encoded-$mode | 
+        ffmpeg -y -hide_banner -framerate 25 -video_size {$w}x{$h} -pixel_format rgb24 -f rawvideo -i pipe:0 samples/decoded-$mode.mp4
+end
diff --git a/evc/src/bin/encode.rs b/evc/src/bin/encode.rs
index c9d8792..32af242 100644
--- a/evc/src/bin/encode.rs
+++ b/evc/src/bin/encode.rs
@@ -26,7 +26,7 @@ pub struct EncodeArgs {
     #[arg(short, long, default_value = "8")]
     jobs: usize,
 
-    #[arg(short, long, default_value = "8")]
+    #[arg(short, long, default_value = "4")]
     min_block_size: isize,
 
     #[arg(short = 't', long, default_value = "200")]
diff --git a/evc/src/bin/info.rs b/evc/src/bin/info.rs
new file mode 100644
index 0000000..1e1472a
--- /dev/null
+++ b/evc/src/bin/info.rs
@@ -0,0 +1,11 @@
+use std::io::BufReader;
+
+use anyhow::Context;
+use evc::format::{header::Header, ser::Source};
+
+fn main() {
+    env_logger::init_from_env("LOG");
+    let mut input = BufReader::new(std::io::stdin());
+    let header = input.get::<Header>().context("reading header").unwrap();
+    eprintln!("{header:#?}")
+}
diff --git a/evc/src/block.rs b/evc/src/block.rs
index 8d7686f..cbf69bf 100644
--- a/evc/src/block.rs
+++ b/evc/src/block.rs
@@ -5,15 +5,16 @@ use crate::{
 };
 use anyhow::bail;
 
-#[derive(Clone, Debug)]
+#[derive(Clone, Debug, PartialEq)]
 pub enum Block {
     Literal(Vec<Pixel>),
+    CompressedLiteral(Vec<Pixel>),
     Split(Box<[Block; 2]>),
     Reference { translation: Vec2<isize> },
     AdvancedReference(AdvancedReference),
 }
 
-#[derive(Clone, Debug)]
+#[derive(Clone, Debug, PartialEq)]
 pub struct AdvancedReference {
     pub translation: Vec2<i8>,
     pub transform: Mat2<i8>,
@@ -31,14 +32,15 @@ impl Block {
                 sink.put(0u8)?;
                 pixels.write_const_size(sink, size.area() as usize)?;
             }
+            Block::CompressedLiteral(_) => bail!("compressed literal is not supported"),
             Block::Split(box [a, b]) => {
-                sink.put(1u8)?;
+                sink.put(2u8)?;
                 let [asize, bsize] = split_size(size);
                 a.write(sink, asize)?;
                 b.write(sink, bsize)?;
             }
             Block::Reference { translation } => {
-                sink.put(2u8)?;
+                sink.put(3u8)?;
                 sink.put(Small(*translation))?;
             }
             Block::AdvancedReference(AdvancedReference {
@@ -46,7 +48,7 @@ impl Block {
                 transform,
                 value_scale,
             }) => {
-                sink.put(3u8)?;
+                sink.put(4u8)?;
                 sink.put((*translation, *transform, *value_scale))?;
             }
         }
@@ -56,16 +58,17 @@ impl Block {
     pub fn read(source: &mut impl std::io::Read, size: Vec2<isize>) -> anyhow::Result<Self> {
         Ok(match source.get::<u8>()? {
             0 => Block::Literal(Vec::read_const_size(source, size.area() as usize)?),
-            1 => Block::Split(Box::new({
+            1 => bail!("compressed literal is not supported"),
+            2 => Block::Split(Box::new({
                 let [asize, bsize] = split_size(size);
                 let a = Block::read(source, asize)?;
                 let b = Block::read(source, bsize)?;
                 [a, b]
             })),
-            2 => Block::Reference {
+            3 => Block::Reference {
                 translation: source.get::<Small<Vec2<isize>>>()?.0,
             },
-            3 => Block::AdvancedReference(AdvancedReference {
+            4 => Block::AdvancedReference(AdvancedReference {
                 translation: source.get()?,
                 transform: source.get()?,
                 value_scale: source.get()?,
@@ -95,13 +98,8 @@ impl Block {
     pub fn is_literal(&self) -> bool {
         matches!(self, Block::Literal(..))
     }
-    pub fn is_ref_without_translation(&self) -> bool {
-        matches!(
-            self,
-            Block::Reference {
-                translation: Vec2::<isize>::ZERO
-            }
-        )
+    pub fn identical_ref(a: &Self, b: &Self) -> bool {
+        matches!(a, Block::Reference { .. }) && a == b
     }
 }
 
diff --git a/evc/src/codec/decode.rs b/evc/src/codec/decode.rs
index 197028c..61234da 100644
--- a/evc/src/codec/decode.rs
+++ b/evc/src/codec/decode.rs
@@ -29,6 +29,7 @@ pub fn decode_block(
                 config.max_threads,
             );
         }
+        Block::CompressedLiteral(_) => todo!(),
         Block::Reference { translation } => target.copy_from(&prev.offset(*translation)),
         Block::AdvancedReference(r) => target.copy_from_sampler(&Sampler::from_refblock(prev, r)),
     }
diff --git a/evc/src/codec/encode/mod.rs b/evc/src/codec/encode/mod.rs
index 2ccac64..336f298 100644
--- a/evc/src/codec/encode/mod.rs
+++ b/evc/src/codec/encode/mod.rs
@@ -69,15 +69,19 @@ pub fn encode_block(view: View<&Frame>, prev: View<&Frame>, config: &EncodeConfi
             let config = unsafe { std::mem::transmute::<_, &'static EncodeConfig>(config) };
 
             // only bother to do multithreading, when the block is big.
-            let (a, b) = both_par(
-                || encode_block(av, ap, config),
-                || encode_block(bv, bp, config),
-                config.max_threads,
-            );
+            let (a, b) = if view.area() > 100 {
+                both_par(
+                    || encode_block(av, ap, config),
+                    || encode_block(bv, bp, config),
+                    config.max_threads,
+                )
+            } else {
+                (encode_block(av, ap, config), encode_block(bv, bp, config))
+            };
 
             if a.is_literal() && b.is_literal() {
                 Block::Literal(view.pixels())
-            } else if a.is_ref_without_translation() && b.is_ref_without_translation() {
+            } else if Block::identical_ref(&a, &b) {
                 Block::Reference {
                     translation: Vec2::<isize>::ZERO,
                 }
diff --git a/evc/src/codec/encode/simple.rs b/evc/src/codec/encode/simple.rs
index cabbc15..2a971af 100644
--- a/evc/src/codec/encode/simple.rs
+++ b/evc/src/codec/encode/simple.rs
@@ -45,6 +45,11 @@ pub fn fast(
                 }
             };
 
+            probe(offset + Vec2 { x: 8, y: 0 });
+            probe(offset + Vec2 { x: -8, y: 0 });
+            probe(offset + Vec2 { x: 0, y: 8 });
+            probe(offset + Vec2 { x: 0, y: -8 });
+
             probe(offset + Vec2 { x: 1, y: 0 });
             probe(offset + Vec2 { x: -1, y: 0 });
             probe(offset + Vec2 { x: 0, y: 1 });
diff --git a/evc/src/debug.rs b/evc/src/debug.rs
index 99723b8..c3ea7d0 100644
--- a/evc/src/debug.rs
+++ b/evc/src/debug.rs
@@ -64,7 +64,7 @@ impl Pixel {
 
 pub fn draw_debug(block: &Block, mut target: View<&mut Frame>) {
     match &block {
-        Block::Literal(_) => {
+        Block::Literal(_) | Block::CompressedLiteral(_) => {
             target.draw_box(Pixel::GREEN);
         }
         Block::Split(box [a, b]) => {
diff --git a/evc/src/frame.rs b/evc/src/frame.rs
index 81fdf3b..78d0e73 100644
--- a/evc/src/frame.rs
+++ b/evc/src/frame.rs
@@ -63,11 +63,8 @@ impl Index<Vec2<isize>> for Frame {
     type Output = Pixel;
     #[inline]
     fn index(&self, Vec2 { x, y }: Vec2<isize>) -> &Self::Output {
-        if x >= 0 && y >= 0 && x < self.size.x && y < self.size.y {
-            &self.buffer[(x + y * self.size.x) as usize]
-        } else {
-            &Pixel::BLACK
-        }
+        &self.buffer
+            [(x.clamp(0, self.size.x - 1) + y.clamp(0, self.size.y - 1) * self.size.x) as usize]
     }
 }
 impl IndexMut<Vec2<isize>> for Frame {
diff --git a/evc/src/helpers/matrix.rs b/evc/src/helpers/matrix.rs
index 87c0e7d..c3c120b 100644
--- a/evc/src/helpers/matrix.rs
+++ b/evc/src/helpers/matrix.rs
@@ -1,6 +1,6 @@
 use crate::helpers::vector::Vec2;
 
-#[derive(Debug, Clone, Copy)]
+#[derive(Debug, Clone, Copy, PartialEq)]
 pub struct Mat2<T> {
     pub a: T,
     pub b: T,
diff --git a/evc/src/helpers/pixel.rs b/evc/src/helpers/pixel.rs
index 816d7dc..39fe98c 100644
--- a/evc/src/helpers/pixel.rs
+++ b/evc/src/helpers/pixel.rs
@@ -1,6 +1,6 @@
 use crate::format::ser::{Ser, Sink, Source};
 
-#[derive(Copy, Clone, Debug, Default)]
+#[derive(Copy, Clone, Debug, Default, PartialEq)]
 pub struct Pixel {
     pub r: u8,
     pub g: u8,
@@ -24,11 +24,12 @@ impl Pixel {
     pub fn distance(a: Pixel, b: Pixel) -> usize {
         let (rd, gd, bd) = (
             a.r.abs_diff(b.r) as usize,
-            a.r.abs_diff(b.r) as usize,
-            a.r.abs_diff(b.r) as usize,
+            a.g.abs_diff(b.g) as usize,
+            a.b.abs_diff(b.b) as usize,
         );
         // fast_sqrt(rd * rd + gd * gd + bd * bd)
-        SQRT[rd + gd + bd]
+        // SQRT[rd + gd + bd]
+        rd + gd + bd
     }
 
     #[inline]