alloc: Reduce unnecessary Vec allocations and indirections

* Changed literal_probs array from a Vec<Vec<u16>> to a Vec2D backed by a contiguous Vec<u16>.
* BitTrees in LenDecoder and DecoderState are now stored inline

Author: chyyran

src/decode/lzma.rs

@@ -1,6 +1,7 @@
 use crate::decode::lzbuffer::{LzBuffer, LzCircularBuffer};
 use crate::decode::rangecoder;
 use crate::decode::rangecoder::RangeDecoder;
+use crate::decode::vec2d::Vec2D;
 use crate::decompress::Options;
 use crate::decompress::UnpackedSize;
 use crate::error;
@@ -167,8 +168,8 @@ pub(crate) struct DecoderState {
     partial_input_buf: std::io::Cursor<[u8; MAX_REQUIRED_INPUT]>,
     pub(crate) lzma_props: LzmaProperties,
     unpacked_size: Option<u64>,
-    literal_probs: Vec<Vec<u16>>,
-    pos_slot_decoder: Vec<rangecoder::BitTree>,
+    literal_probs: Vec2D<u16>,
+    pos_slot_decoder: [rangecoder::BitTree; 4],
     align_decoder: rangecoder::BitTree,
     pos_decoders: [u16; 115],
     is_match: [u16; 192], // true = LZ, false = literal
@@ -190,8 +191,13 @@ impl DecoderState {
             partial_input_buf: std::io::Cursor::new([0; MAX_REQUIRED_INPUT]),
             lzma_props,
             unpacked_size,
-            literal_probs: vec![vec![0x400; 0x300]; 1 << (lzma_props.lc + lzma_props.lp)],
-            pos_slot_decoder: vec![rangecoder::BitTree::new(6); 4],
+            literal_probs: Vec2D::init(1 << (lzma_props.lc + lzma_props.lp), 0x300, 0x400),
+            pos_slot_decoder: [
+                rangecoder::BitTree::new(6),
+                rangecoder::BitTree::new(6),
+                rangecoder::BitTree::new(6),
+                rangecoder::BitTree::new(6),
+            ],
             align_decoder: rangecoder::BitTree::new(4),
             pos_decoders: [0x400; 115],
             is_match: [0x400; 192],
@@ -211,10 +217,10 @@ impl DecoderState {
         new_props.validate();
         if self.lzma_props.lc + self.lzma_props.lp == new_props.lc + new_props.lp {
             // We can reset here by filling the existing buffer with 0x400.
-            self.literal_probs.iter_mut().for_each(|v| v.fill(0x400))
+            self.literal_probs.fill(0x400);
         } else {
             // We need to reallocate because of the new size of `lc+lp`.
-            self.literal_probs = vec![vec![0x400; 0x300]; 1 << (new_props.lc + new_props.lp)];
+            self.literal_probs = Vec2D::init(1 << (new_props.lc + new_props.lp), 0x300, 0x400);
         }
 
         self.lzma_props = new_props;

src/decode/mod.rs

@@ -6,6 +6,7 @@ pub mod lzma2;
 pub mod options;
 pub mod rangecoder;
 pub mod util;
+pub mod vec2d;
 pub mod xz;
 
 #[cfg(feature = "stream")]

src/decode/rangecoder.rs

@@ -190,8 +190,8 @@ impl BitTree {
 pub struct LenDecoder {
     choice: u16,
     choice2: u16,
-    low_coder: Vec<BitTree>,
-    mid_coder: Vec<BitTree>,
+    low_coder: [BitTree; 16],
+    mid_coder: [BitTree; 16],
     high_coder: BitTree,
 }
 
@@ -200,8 +200,42 @@ impl LenDecoder {
         LenDecoder {
             choice: 0x400,
             choice2: 0x400,
-            low_coder: vec![BitTree::new(3); 16],
-            mid_coder: vec![BitTree::new(3); 16],
+            low_coder: [
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+            ],
+            mid_coder: [
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+                BitTree::new(3),
+            ],
             high_coder: BitTree::new(8),
         }
     }

src/decode/vec2d.rs

@@ -0,0 +1,74 @@
+use std::ops::{Index, IndexMut};
+
+/// A 2 dimensional matrix in row-major order backed by a contiguous `Vec`
+#[derive(Debug)]
+pub struct Vec2D<T> {
+    data: Vec<T>,
+    cols: usize,
+    rows: usize,
+}
+
+impl<T> Vec2D<T> {
+    /// Initialize a grid of size `rows * cols` with the given data element.
+    pub fn init(rows: usize, cols: usize, data: T) -> Vec2D<T>
+    where
+        T: Clone,
+    {
+        Vec2D {
+            data: vec![data; rows * cols],
+            cols,
+            rows,
+        }
+    }
+
+    /// Fills the grid with elements by cloning `value`.
+    pub fn fill(&mut self, value: T)
+    where
+        T: Clone,
+    {
+        self.data.fill(value)
+    }
+}
+
+impl<T> Index<usize> for Vec2D<T> {
+    type Output = [T];
+
+    fn index(&self, idx: usize) -> &Self::Output {
+        if idx < self.rows {
+            let start_idx = idx * self.cols;
+            &self.data[start_idx..start_idx + self.cols]
+        } else {
+            panic!("row index {:?} out of bounds.", idx);
+        }
+    }
+}
+
+impl<T> IndexMut<usize> for Vec2D<T> {
+    fn index_mut(&mut self, idx: usize) -> &mut Self::Output {
+        &mut self.data[(idx * self.cols)..]
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn vec2d_fill() {
+        let mut vec2d = Vec2D::init(3, 3, 0);
+        vec2d.fill(7);
+        assert_eq!(vec2d[0], [7, 7, 7]);
+        assert_eq!(vec2d[1], [7, 7, 7]);
+    }
+
+    #[test]
+    fn vec2d_index() {
+        let vec2d = Vec2D {
+            data: vec![0, 1, 2, 3, 4, 5],
+            rows: 3,
+            cols: 3,
+        };
+        assert_eq!(vec2d[0], [0, 1, 2]);
+        assert_eq!(vec2d[1], [3, 4, 5]);
+    }
+}