infra for compressed literal

4 files changed, 173 insertions, 14 deletions

diff --git a/evc/src/codec/compress.rs b/evc/src/codec/compress.rs
new file mode 100644
index 0000000..e011669
--- /dev/null
+++ b/evc/src/codec/compress.rs
@@ -0,0 +1,151 @@
+use crate::{
+    block::Block,
+    frame::Frame,
+    helpers::{pixel::Pixel, vector::Vec2},
+    view::View,
+};
+use rustdct::DctPlanner;
+
+pub fn compress_block(block: &mut Block, size: Vec2<usize>) {
+    match block {
+        Block::Literal(p) => {
+            let comp = lit_compress(size.x, size.y, &p);
+            *block = Block::CompressedLiteral(comp)
+        }
+        Block::Split(box [a, b]) => {
+            let vert = size.x > size.y;
+            compress_block(
+                a,
+                if vert {
+                    Vec2 {
+                        x: size.x / 2,
+                        y: size.y,
+                    }
+                } else {
+                    Vec2 {
+                        x: size.x,
+                        y: size.y / 2,
+                    }
+                },
+            );
+            compress_block(
+                b,
+                if vert {
+                    Vec2 {
+                        x: size.x - size.x / 2,
+                        y: size.y,
+                    }
+                } else {
+                    Vec2 {
+                        x: size.x,
+                        y: size.y - size.y / 2,
+                    }
+                },
+            );
+        }
+        _ => (),
+    }
+}
+
+pub fn lit_compress(w: usize, h: usize, pixels: &[Pixel]) -> Vec<u8> {
+    let mut out = vec![];
+    for ci in 0..3 {
+        let mut ch = vec![0.; w * h];
+        for y in 0..h {
+            for x in 0..w {
+                let p = pixels[x + y * w];
+                ch[x + y * w] = match ci {
+                    0 => p.r,
+                    1 => p.g,
+                    2 => p.b,
+                    _ => unreachable!(),
+                } as f32
+            }
+        }
+
+        norm_dct_channel(w, h, &mut ch);
+        for i in 0..w * h {
+            out.push(ch[i] as u8)
+        }
+    }
+    out
+}
+pub fn lit_decompress(compressed: &[u8], mut target: View<&mut Frame>) {
+    let (w, h) = (target.size.x as usize, target.size.y as usize);
+    for ci in 0..3 {
+        let mut ch = compressed[ci * w * h..(ci + 1) * w * h]
+            .iter()
+            .map(|v| *v as f32)
+            .collect::<Vec<_>>();
+        norm_idct_channel(w, h, &mut ch);
+        for y in 0..h {
+            for x in 0..w {
+                let p = &mut target[Vec2 {
+                    x: x as isize,
+                    y: y as isize,
+                }];
+                *(match ci {
+                    0 => &mut p.r,
+                    1 => &mut p.g,
+                    2 => &mut p.b,
+                    _ => unreachable!(),
+                }) = ch[x + y * w] as u8
+            }
+        }
+    }
+}
+
+fn norm_dct_channel(w: usize, h: usize, values: &mut [f32]) {
+    let d = DctPlanner::new().plan_dct2(w);
+    let mut temp = vec![0.; d.get_scratch_len()];
+    for c in values.chunks_exact_mut(w) {
+        d.process_dct2_with_scratch(c, &mut temp)
+    }
+
+    let mut values_trans = transpose(w, h, values);
+    let d = DctPlanner::new().plan_dct2(h);
+    let mut temp = vec![0.; d.get_scratch_len()];
+    for c in values_trans.chunks_exact_mut(h) {
+        d.process_dct2_with_scratch(c, &mut temp)
+    }
+
+    let scale = 2. / (w as f32 * h as f32).sqrt();
+    for (i, v) in values_trans.iter().enumerate() {
+        values[i] = *v * scale
+    }
+}
+
+fn norm_idct_channel(w: usize, h: usize, values: &mut [f32]) {
+    let d = DctPlanner::new().plan_dct2(w);
+    let mut temp = vec![0.; d.get_scratch_len()];
+    for c in values.chunks_exact_mut(w) {
+        d.process_dct3_with_scratch(c, &mut temp)
+    }
+
+    let mut values_trans = transpose(w, h, values);
+    let d = DctPlanner::new().plan_dct2(h);
+    let mut temp = vec![0.; d.get_scratch_len()];
+    for c in values_trans.chunks_exact_mut(h) {
+        d.process_dct3_with_scratch(c, &mut temp)
+    }
+
+    let scale = 2. / (w as f32 * h as f32).sqrt();
+    for (i, v) in values_trans.iter().enumerate() {
+        values[i] = *v * scale
+    }
+}
+
+fn transpose(w: usize, h: usize, values: &[f32]) -> Vec<f32> {
+    let mut i = 0;
+    let mut it;
+    let mut transposed = vec![0.; values.len()];
+    for row in 0..h {
+        it = row;
+        for _ in 0..w {
+            transposed[it] = values[i];
+            i += 1;
+            it += h;
+        }
+    }
+    transposed
+}
diff --git a/evc/src/codec/decode.rs b/evc/src/codec/decode.rs
index 61234da..087483c 100644
--- a/evc/src/codec/decode.rs
+++ b/evc/src/codec/decode.rs
@@ -2,6 +2,8 @@ use crate::{
     block::Block, frame::Frame, helpers::threading::both_par, refsampler::Sampler, view::View,
 };
 
+use super::compress::lit_decompress;
+
 pub struct DecodeConfig {
     pub max_threads: usize,
 }
@@ -29,7 +31,9 @@ pub fn decode_block(
                 config.max_threads,
             );
         }
-        Block::CompressedLiteral(_) => todo!(),
+        Block::CompressedLiteral(data) => {
+            lit_decompress(&data, target);
+        }
         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 336f298..76fb481 100644
--- a/evc/src/codec/encode/mod.rs
+++ b/evc/src/codec/encode/mod.rs
@@ -30,7 +30,7 @@ pub enum EncodeMode {
     AdvancedPartial,
 }
 
-pub fn encode_block(view: View<&Frame>, prev: View<&Frame>, config: &EncodeConfig) -> Block {
+pub fn encode_block(view: View<&Frame>, prev: View<&Frame>, config: &EncodeConfig) -> (f64, Block) {
     let (diff, refblock) = if view.area() > config.max_diff_area {
         (
             f64::INFINITY,
@@ -57,10 +57,10 @@ pub fn encode_block(view: View<&Frame>, prev: View<&Frame>, config: &EncodeConfi
         (diff - irrelevance, refblock)
     };
     if diff < config.ref_thres {
-        refblock
+        (diff, refblock)
     } else {
         if view.size.x < config.min_block_size || view.size.y < config.min_block_size {
-            Block::Literal(view.pixels())
+            (0.0, Block::Literal(view.pixels()))
         } else {
             let [av, bv] =
                 unsafe { std::mem::transmute::<_, [View<&'static Frame>; 2]>(view.split()) };
@@ -69,7 +69,7 @@ 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) = if view.area() > 100 {
+            let ((ad, a), (bd, b)) = if view.area() > 100 {
                 both_par(
                     || encode_block(av, ap, config),
                     || encode_block(bv, bp, config),
@@ -79,15 +79,18 @@ pub fn encode_block(view: View<&Frame>, prev: View<&Frame>, config: &EncodeConfi
                 (encode_block(av, ap, config), encode_block(bv, bp, config))
             };
 
-            if a.is_literal() && b.is_literal() {
-                Block::Literal(view.pixels())
-            } else if Block::identical_ref(&a, &b) {
-                Block::Reference {
-                    translation: Vec2::<isize>::ZERO,
-                }
-            } else {
-                Block::Split(Box::new([a, b]))
-            }
+            (
+                ad + bd,
+                if a.is_literal() && b.is_literal() {
+                    Block::Literal(view.pixels())
+                } else if Block::identical_ref(&a, &b) {
+                    Block::Reference {
+                        translation: Vec2::<isize>::ZERO,
+                    }
+                } else {
+                    Block::Split(Box::new([a, b]))
+                },
+            )
         }
     }
 }
diff --git a/evc/src/codec/mod.rs b/evc/src/codec/mod.rs
index 1c80d24..3203e6e 100644
--- a/evc/src/codec/mod.rs
+++ b/evc/src/codec/mod.rs
@@ -1,2 +1,3 @@
 pub mod decode;
 pub mod encode;
+pub mod compress;