From b4baed337a70c317c5d6a2fa245bda21f461fc6b Mon Sep 17 00:00:00 2001
From: oyvindln <oyvindln@users.noreply.github.com>
Date: Wed, 29 May 2024 02:33:44 +0200
Subject: [PATCH] feat(deflate): evade a bounds check in deflate for a small
 perf improvement

---
 benches/bench.rs                | 6 ++++++
 miniz_oxide/src/deflate/core.rs | 6 ++++++
 2 files changed, 12 insertions(+)

diff --git a/benches/bench.rs b/benches/bench.rs
index 94bcb23..d3b3ab8 100755
--- a/benches/bench.rs
+++ b/benches/bench.rs
@@ -107,6 +107,12 @@ mod oxide {
         1,
         "benches/data/bin"
     );
+    compress_bench!(
+        compress_bin_lvl_2,
+        tdefl_compress_mem_to_heap,
+        2,
+        "benches/data/bin"
+    );
     compress_bench!(
         compress_bin_lvl_6,
         tdefl_compress_mem_to_heap,
diff --git a/miniz_oxide/src/deflate/core.rs b/miniz_oxide/src/deflate/core.rs
index 30762b3..6deddd0 100644
--- a/miniz_oxide/src/deflate/core.rs
+++ b/miniz_oxide/src/deflate/core.rs
@@ -1242,6 +1242,8 @@ impl DictOxide {
         let pos = pos & LZ_DICT_SIZE_MASK;
         let end = pos + 4;
         // Somehow this assertion makes things faster.
+        // TODO: as of may 2024 this does not seem to make any difference
+        // so consider removing.
         assert!(end < LZ_DICT_FULL_SIZE);
 
         let bytes: [u8; 4] = self.b.dict[pos..end].try_into().unwrap();
@@ -1252,6 +1254,10 @@ impl DictOxide {
     /// type T.
     #[inline]
     fn read_unaligned_u64(&self, pos: usize) -> u64 {
+        // Help evade bounds/panic code check by masking the position value
+        // This provides a small speedup at the cost of an instruction or two instead of
+        // having to use unsafe.
+        let pos = pos & LZ_DICT_SIZE_MASK;
         let bytes: [u8; 8] = self.b.dict[pos..pos + 8].try_into().unwrap();
         u64::from_le_bytes(bytes)
     }