From a713143ef9c1187c37004043b1d3322d773f9ea0 Mon Sep 17 00:00:00 2001
From: metamuffin <metamuffin@disroot.org>
Date: Tue, 6 Dec 2022 22:16:50 +0100
Subject: webm + variable amount of threads

---
 evc/scripts/gen         |  4 ++--
 evc/src/bin/encode.rs   |  3 ++-
 evc/src/codec/encode.rs | 19 ++++++++++++++-----
 evc/src/threading.rs    |  4 ++--
 4 files changed, 20 insertions(+), 10 deletions(-)

diff --git a/evc/scripts/gen b/evc/scripts/gen
index 1cf4493..620c69f 100755
--- a/evc/scripts/gen
+++ b/evc/scripts/gen
@@ -5,6 +5,6 @@ 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 |
     LOG=info cargo run --release --bin encode -- -W {$w} -H {$h} > samples/encoded
 LOG=info cargo run --release --bin decode -- < samples/encoded |
-    ffmpeg -hide_banner -y -framerate 30 -video_size {$w}x{$h} -pixel_format rgb24 -f rawvideo -i pipe:0 samples/decoded.mp4
+    ffmpeg -hide_banner -y -framerate 30 -video_size {$w}x{$h} -pixel_format rgb24 -f rawvideo -i pipe:0 samples/decoded.webm
 LOG=info cargo run --release --bin decode -- --debug < samples/encoded |
-    ffmpeg -hide_banner -y -framerate 30 -video_size {$w}x{$h} -pixel_format rgb24 -f rawvideo -i pipe:0 samples/decoded-debug.mp4
+    ffmpeg -hide_banner -y -framerate 30 -video_size {$w}x{$h} -pixel_format rgb24 -f rawvideo -i pipe:0 samples/decoded-debug.webm
diff --git a/evc/src/bin/encode.rs b/evc/src/bin/encode.rs
index fb5a18b..4857a80 100644
--- a/evc/src/bin/encode.rs
+++ b/evc/src/bin/encode.rs
@@ -38,7 +38,8 @@ fn main() -> anyhow::Result<()> {
         translate: !args.no_translation,
         ref_thres: 2.0,
         max_diff_size: 10_000,
-        min_block_size: 12,
+        min_block_size: 4,
+        max_threads: 12,
     };
 
     let size = Vec2 {
diff --git a/evc/src/codec/encode.rs b/evc/src/codec/encode.rs
index 13df17d..dbefdb4 100644
--- a/evc/src/codec/encode.rs
+++ b/evc/src/codec/encode.rs
@@ -8,6 +8,7 @@ pub struct EncodeConfig {
     pub ref_thres: f64,
     pub max_diff_size: isize,
     pub min_block_size: isize,
+    pub max_threads: usize,
     // pub importance_k: f64,
     // pub importance_scale: f64,
 }
@@ -40,7 +41,7 @@ pub fn encode_block(view: View<&Frame>, prev: View<&Frame>, config: &EncodeConfi
     if diff < (config.ref_thres) {
         Block::Reference { translation }
     } else {
-        if view.size.x < config.min_block_size {
+        if view.size.x < config.min_block_size || view.size.y < config.min_block_size {
             Block::Literal(view.pixels())
         } else {
             let [av, bv] =
@@ -48,10 +49,18 @@ pub fn encode_block(view: View<&Frame>, prev: View<&Frame>, config: &EncodeConfi
             let [ap, bp] =
                 unsafe { std::mem::transmute::<_, [View<&'static Frame>; 2]>(prev.split()) };
             let config = unsafe { std::mem::transmute::<_, &'static EncodeConfig>(config) };
-            let (a, b) = both_par(
-                || encode_block(av, ap, config),
-                || encode_block(bv, bp, config),
-            );
+
+            // only bother to do multithreading, when the block is big.
+            let (a, b) = if view.size.x > 64 {
+                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 {
diff --git a/evc/src/threading.rs b/evc/src/threading.rs
index 012b5bf..3291172 100644
--- a/evc/src/threading.rs
+++ b/evc/src/threading.rs
@@ -5,13 +5,13 @@ use std::{
 
 static THREADS_RUNNING: AtomicUsize = AtomicUsize::new(0);
 
-pub fn both_par<F1, F2, O1, O2>(f1: F1, f2: F2) -> (O1, O2)
+pub fn both_par<F1, F2, O1, O2>(f1: F1, f2: F2, max_threads: usize) -> (O1, O2)
 where
     F1: FnOnce() -> O1 + Send + 'static,
     O1: Send + 'static,
     F2: FnOnce() -> O2,
 {
-    if THREADS_RUNNING.load(Ordering::Relaxed) < 12 {
+    if THREADS_RUNNING.load(Ordering::Relaxed) < max_threads {
         THREADS_RUNNING.fetch_add(1, Ordering::Relaxed);
 
         let o1h = thread::spawn(move || f1());
-- 
cgit v1.2.3-70-g09d2