reduce the CHUNK_LEN from 2048 bytes to 1024 bytes

Smaller chunk sizes are a big benefit for parallelism at shorter input
lengths, and recent benchmarks show that this reduction has a relative
small cost in terms of peak throughput. It's also a nice round number.

1 files changed, 1 insertions, 1 deletions

diff --git a/reference_impl/reference_impl.rs b/reference_impl/reference_impl.rs
index f90b07d..07cbd80 100644
--- a/reference_impl/reference_impl.rs
+++ b/reference_impl/reference_impl.rs
@@ -4,7 +4,7 @@ use core::convert::TryInto;
 const OUT_LEN: usize = 32;
 const KEY_LEN: usize = 32;
 const BLOCK_LEN: usize = 64;
-const CHUNK_LEN: usize = 2048;
+const CHUNK_LEN: usize = 1024;
 const ROUNDS: usize = 7;
 
 const CHUNK_START: u32 = 1 << 0;