diff --git a/src/aead/poly1305/arm_neon.rs b/src/aead/poly1305/arm_neon.rs
index e9d500018f..af22c37dbc 100644
--- a/src/aead/poly1305/arm_neon.rs
+++ b/src/aead/poly1305/arm_neon.rs
@@ -188,19 +188,12 @@ impl fe1305x2 {
     }
 }
 
-// TODO: Does this really need to be `#[repr(C)]` or `#[repr(align(16))`]? We
-// need to read the assembly code and see what assumptions it makes on the
-// layout of its inputs. We've already made `fe1305x2` 16-byte aligned and that
-// might be all we need.
-//
 // XXX/TODO(MSRV): change to `pub(super)`.
-#[repr(C, align(16))]
 pub(in super::super) struct State {
     r: fe1305x2,
     h: fe1305x2,
     c: fe1305x2,
     precomp: [fe1305x2; 2],
-    data: [u8; data_len()], // TODO: Does the assembly code use this?
 
     buf: [u8; 2 * BLOCK_LEN],
     buf_used: c::size_t,
@@ -231,7 +224,6 @@ impl State {
             h: fe1305x2 { v: [ZERO; 12] },
             c: fe1305x2 { v: [ZERO; 12] },
             precomp: [fe1305x2 { v: [ZERO; 12] }; 2],
-            data: [0u8; data_len()],
 
             buf: Default::default(),
             buf_used: 0,
@@ -336,10 +328,6 @@ impl State {
     }
 }
 
-const fn data_len() -> usize {
-    128
-}
-
 fn store32<'o>(output: &'o mut [u8; BLOCK_LEN], i: usize, value: W32) {
     let output: &'o mut [u8; 4] = (&mut output[i..][..4]).try_into().unwrap();
     *output = u32::to_le_bytes(value.0);