Optimize loading of batch_bool from memory on arm

Use the same approach as #1172

Author: serge-sans-paille

include/xsimd/arch/xsimd_neon.hpp

@@ -573,6 +573,52 @@ namespace xsimd
             return vld1q_f32(src);
         }
 
+        /* batch bool version */
+        template <class T, class A, detail::enable_sized_t<T, 1> = 0>
+        XSIMD_INLINE batch_bool<T, A> load_unaligned(bool const* mem, batch_bool<T, A>, requires_arch<neon>) noexcept
+        {
+            auto vmem = load_unaligned<A>((unsigned char const*)mem, convert<unsigned char> {}, A {});
+            return batch_bool<T, A>(bitwise_cast<typename std::make_unsigned<T>::type>(0 - vmem).data);
+        }
+        template <class T, class A, detail::enable_sized_t<T, 1> = 0>
+        XSIMD_INLINE batch_bool<T, A> load_aligned(bool const* mem, batch_bool<T, A> t, requires_arch<neon> r) noexcept
+        {
+            return load_unaligned(mem, t, r);
+        }
+
+        template <class T, class A, detail::enable_sized_t<T, 2> = 0>
+        XSIMD_INLINE batch_bool<T, A> load_unaligned(bool const* mem, batch_bool<T, A>, requires_arch<neon>) noexcept
+        {
+            batch<unsigned short, neon> vmem = vmovl_u8(vld1_u8((unsigned char const*)mem));
+            return batch_bool<T, A>(bitwise_cast<typename std::make_unsigned<T>::type>(0 - vmem).data);
+        }
+
+        template <class T, class A, detail::enable_sized_t<T, 2> = 0>
+        XSIMD_INLINE batch_bool<T, A> load_aligned(bool const* mem, batch_bool<T, A> t, requires_arch<neon> r) noexcept
+        {
+            return load_unaligned(mem, t, r);
+        }
+
+        template <class T, class A, detail::enable_sized_t<T, 4> = 0>
+        XSIMD_INLINE batch_bool<T, A> load_unaligned(bool const* mem, batch_bool<T, A>, requires_arch<neon>) noexcept
+        {
+            uint32x2_t tmp = vset_lane_u32(*(unsigned int*)mem, vdup_n_u32(0), 0);
+            return batch_bool<T, A>(bitwise_cast<typename std::make_unsigned<T>::type>(0 - vmovl_u16(vget_low_u16(vmovl_u8(tmp)))));
+        }
+
+        template <class T, class A, detail::enable_sized_t<T, 4> = 0>
+        XSIMD_INLINE batch_bool<T, A> load_aligned(bool const* mem, batch_bool<T, A> t, requires_arch<neon> r) noexcept
+        {
+            return load_unaligned(mem, t, r);
+        }
+
+        template <class A> = 0>
+        XSIMD_INLINE batch_bool<float, A> load_aligned(bool const* mem, batch_bool<float, A> t, requires_arch<neon> r) noexcept
+        {
+            uint32x2_t tmp = vset_lane_u32(*(unsigned int*)mem, vdup_n_u32(0), 0);
+            return batch_bool<float, A>(0 - vmovl_u16(vget_low_u16(vmovl_u8(tmp))));
+        }
+
         /*********
          * store *
          *********/