From ffc0299cfd9bf32e02b61d15e52904d7ce268565 Mon Sep 17 00:00:00 2001
From: Daniel Schadt <kingdread@gmx.de>
Date: Fri, 13 Jun 2025 17:35:46 +0200
Subject: use SIMD for left-shift

This saves the integer conversion. Here, for the 1024 byte benchmark, I
get an improvement of ~4% in runtime
---
 src/block.rs | 8 ++++++++
 1 file changed, 8 insertions(+)

(limited to 'src/block.rs')

diff --git a/src/block.rs b/src/block.rs
index bde60fc..0e64bbe 100644
--- a/src/block.rs
+++ b/src/block.rs
@@ -148,6 +148,14 @@ impl BitXor<Block> for Block {
 impl Shl<u32> for Block {
     type Output = Block;
     fn shl(self, rhs: u32) -> Block {
+        // We often use a shift by one, for example in the multiplication. We therefore optimize
+        // for this special case.
+        #[cfg(feature = "simd")]
+        {
+            if rhs == 1 {
+                return Block((self.0 << 1) | (self.0.shift_elements_left::<1>(0) >> 7));
+            }
+        }
         Block::from(self.to_int() << rhs)
     }
 }
-- 
cgit v1.2.3