Clarify admitted SIMD cleanup labels

Author: eldryoth

CHANGELOG.md

@@ -59,13 +59,13 @@
   standard checks so the future encode-dispatch admission contract remains
   packaged and machine-checked.
 - Wiped the small unpadded in-place decode tail buffer before return, expanded
-  test-only AArch64 NEON prototype cleanup to all vector registers, and
+  AArch64 NEON encode cleanup to all vector registers, and
   documented the wrapped slice encoder's temporary in-buffer staging behavior.
 - Added `base64-ng-subtle` as an optional companion crate for projects that
   already admit `subtle` and want reviewed `ConstantTimeEq` comparisons for
   `base64-ng` buffers.
-- Inlined the test-only AArch64 NEON register cleanup macro into the prototype
-  path so callee-saved `v8..v15` are not restored by a separate helper frame.
+- Inlined the AArch64 NEON register cleanup macro into the encode block path so
+  callee-saved `v8..v15` are not restored by a separate helper frame.
 - Added a real non-dispatchable AVX2 fixed-block encode prototype for Standard
   and URL-safe alphabets. The prototype remains test-only and is not reachable
   from runtime backend selection.

docs/UNSAFE.md

@@ -749,7 +749,7 @@ Safety argument:
   NEON is part of the target contract. Direct tests use the same availability
   precondition.
 - Register-retention note: the AArch64 vector path loads caller bytes into NEON
-  state and expands `clear_neon_registers_for_test_prototype!` directly inside
+  state and expands `clear_neon_registers_after_encode_block!` directly inside
   the block function before return. This is retention reduction for the
   admitted encode block, not a formal microarchitectural side-channel proof.
 
@@ -782,7 +782,7 @@ Unsafe operation:
 - `encode_standard_family_indices_neon` maps those indices to Standard or
   URL-safe alphabet bytes with NEON comparisons and bit selects.
 - `vst1q_u8` stores the 16 encoded bytes into the output buffer.
-- `clear_neon_registers_for_test_prototype!` clears `v0` through `v31` inside
+- `clear_neon_registers_after_encode_block!` clears `v0` through `v31` inside
   the block function before return.
 - The local staging array is wiped with the crate cleanup primitive before the
   function returns.
@@ -830,7 +830,7 @@ Safety argument:
   six-bit Base64 value.
 - The helper is private to the Standard-family NEON encode path.
 
-### `clear_neon_registers_for_test_prototype!`
+### `clear_neon_registers_after_encode_block!`
 
 Location: `src/simd/mod.rs`
 
@@ -843,9 +843,9 @@ Purpose:
 
 Preconditions:
 
-- Called only after the prototype has stored its output and no later NEON value
+- Called only after the encode block has stored its output and no later NEON value
   is needed by the function.
-- Expanded directly inside the prototype function. It must not be moved to a
+- Expanded directly inside the encode block function. It must not be moved to a
   separate function because an AArch64 helper can save and restore callee-saved
   `v8` through `v15`, undoing register clearing in the helper frame.
 

src/simd/neon.rs

@@ -191,7 +191,7 @@ where
     target_arch = "aarch64",
     any(test, all(feature = "std", feature = "simd"))
 ))]
-macro_rules! clear_neon_registers_for_test_prototype {
+macro_rules! clear_neon_registers_after_encode_block {
     () => {{
         // SAFETY: This cleanup is expanded directly inside the block encoder
         // after it stores its output. There is no separate helper frame whose
@@ -304,7 +304,7 @@ where
 
         let encoded = encode_standard_family_indices_neon::<A>(indices);
         vst1q_u8(output.as_mut_ptr(), encoded);
-        clear_neon_registers_for_test_prototype!();
+        clear_neon_registers_after_encode_block!();
     }
     crate::wipe_bytes(&mut staged);
 }

src/simd/x86/mod.rs

@@ -237,7 +237,6 @@ where
     crate::wipe_bytes(&mut staged);
 }
 
-#[allow(dead_code, reason = "inactive prototype is not dispatchable yet")]
 #[expect(
     clippy::cast_ptr_alignment,
     reason = "_mm256_storeu_si256 accepts unaligned pointers"