Add SIMD to add_shift_none

ext/crates/fp/src/limb.rs

@@ -1,7 +1,7 @@
 use std::ops::Range;
 
 pub(crate) use crate::constants::Limb;
-use crate::{constants::BITS_PER_LIMB, prime::ValidPrime};
+use crate::{constants::BITS_PER_LIMB, prime::ValidPrime, simd};
 
 /// A struct containing the information required to access a specific entry in an array of `Limb`s.
 #[derive(Copy, Clone)]
@@ -231,6 +231,19 @@ pub(crate) const fn add<const P: u32>(limb_a: Limb, limb_b: Limb, coeff: u32) ->
     }
 }
 
+/// Add (`c` times) all of the limbs in `rhs` to the limbs in `lhs`. This is optimized to use SIMD
+/// when `P == 2`.
+pub(crate) fn add_all<const P: u32>(lhs: &mut [Limb], rhs: &[Limb], c: u32) {
+    if P == 2 {
+        simd::add_simd(lhs, rhs, 0);
+    } else {
+        for (left, right) in lhs.iter_mut().zip(rhs) {
+            *left = add::<P>(*left, *right, c);
+            *left = reduce::<P>(*left);
+        }
+    }
+}
+
 /// Return the `Limb` whose entries are the entries of `limb` reduced modulo `P`.
 ///
 /// Contributed by Robert Burklund.

ext/crates/fp/src/vector/internal/mod.rs

@@ -205,13 +205,11 @@ pub trait InternalBaseVectorMutP<const P: u32>: InternalBaseVectorP<P> {
         let target_inner_range = self._len().limb_range_inner();
         let source_inner_range = other._len().limb_range_inner();
         if !source_inner_range.is_empty() {
-            for (left, right) in self._limbs_mut()[target_inner_range]
-                .iter_mut()
-                .zip(&other_limbs[source_inner_range])
-            {
-                *left = limb::add::<P>(*left, *right, c);
-                *left = limb::reduce::<P>(*left);
-            }
+            limb::add_all::<P>(
+                &mut self._limbs_mut()[target_inner_range],
+                &other_limbs[source_inner_range],
+                c,
+            );
         }
         if source_range.len() > 1 {
             // The first and last limbs are distinct, so we process the last.