sequenzo 0.1.17__cp311-cp311-win_amd64.whl → 0.1.19__cp311-cp311-win_amd64.whl

sequenzo/dissimilarity_measures/src/xsimd/include/xsimd/arch/xsimd_neon.hpp

@@ -13,6 +13,7 @@
 #define XSIMD_NEON_HPP
 
 #include <algorithm>
+#include <array>
 #include <complex>
 #include <tuple>
 #include <type_traits>
@@ -717,16 +718,10 @@ namespace xsimd
             return vnegq_s32(rhs);
         }
 
-        template <class A, class T, detail::enable_sized_unsigned_t<T, 8> = 0>
-        XSIMD_INLINE batch<T, A> neg(batch<T, A> const& rhs, requires_arch<neon>) noexcept
-        {
-            return batch<T, A> { -rhs.get(0), -rhs.get(1) };
-        }
-
-        template <class A, class T, detail::enable_sized_signed_t<T, 8> = 0>
+        template <class A, class T, detail::enable_sized_integral_t<T, 8> = 0>
         XSIMD_INLINE batch<T, A> neg(batch<T, A> const& rhs, requires_arch<neon>) noexcept
         {
-            return batch<T, A> { -rhs.get(0), -rhs.get(1) };
+            return 0 - rhs;
         }
 
         template <class A>
@@ -923,16 +918,28 @@ namespace xsimd
             return dispatcher.apply(register_type(lhs), register_type(rhs));
         }
 
-        template <class A, class T, detail::enable_sized_integral_t<T, 8> = 0>
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 8> = 0>
         XSIMD_INLINE batch_bool<T, A> eq(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
         {
-            return batch_bool<T, A>({ lhs.get(0) == rhs.get(0), lhs.get(1) == rhs.get(1) });
+            auto eq32 = vceqq_u32(vreinterpretq_u32_u64(lhs.data), vreinterpretq_u32_u64(rhs.data));
+            auto rev32 = vrev64q_u32(eq32);
+            auto eq64 = vandq_u32(eq32, rev32);
+            return batch_bool<T, A>(vreinterpretq_u64_u32(eq64));
+        }
+
+        template <class A, class T, detail::enable_sized_signed_t<T, 8> = 0>
+        XSIMD_INLINE batch_bool<T, A> eq(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
+        {
+            auto eq32 = vceqq_u32(vreinterpretq_u32_s64(lhs.data), vreinterpretq_u32_s64(rhs.data));
+            auto rev32 = vrev64q_u32(eq32);
+            auto eq64 = vandq_u32(eq32, rev32);
+            return batch_bool<T, A>(vreinterpretq_u64_u32(eq64));
         }
 
         template <class A, class T, detail::enable_sized_integral_t<T, 8> = 0>
         XSIMD_INLINE batch_bool<T, A> eq(batch_bool<T, A> const& lhs, batch_bool<T, A> const& rhs, requires_arch<neon>) noexcept
         {
-            return batch_bool<T, A>({ lhs.get(0) == rhs.get(0), lhs.get(1) == rhs.get(1) });
+            return eq(batch<T, A> { lhs.data }, batch<T, A> { rhs.data }, A {});
         }
 
         /*************
@@ -985,10 +992,19 @@ namespace xsimd
             return dispatcher.apply(register_type(lhs), register_type(rhs));
         }
 
-        template <class A, class T, detail::enable_sized_integral_t<T, 8> = 0>
+        template <class A, class T, detail::enable_sized_signed_t<T, 8> = 0>
+        XSIMD_INLINE batch_bool<T, A> lt(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
+        {
+            using register_type = typename batch<T, A>::register_type;
+            return batch_bool<T, A>(vreinterpretq_u64_s64(vshrq_n_s64(vqsubq_s64(register_type(lhs), register_type(rhs)), 63)));
+        }
+
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 8> = 0>
         XSIMD_INLINE batch_bool<T, A> lt(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
         {
-            return batch_bool<T, A>({ lhs.get(0) < rhs.get(0), lhs.get(1) < rhs.get(1) });
+            using register_type = typename batch<T, A>::register_type;
+            register_type acc = { 0x7FFFFFFFFFFFFFFFull, 0x7FFFFFFFFFFFFFFFull };
+            return batch_bool<T, A>(vreinterpretq_u64_s64(vshrq_n_s64(vreinterpretq_s64_u64(vqaddq_u64(vqsubq_u64(register_type(rhs), register_type(lhs)), acc)), 63)));
         }
 
         /******
@@ -1012,12 +1028,24 @@ namespace xsimd
         template <class A, class T, detail::enable_sized_integral_t<T, 8> = 0>
         XSIMD_INLINE batch_bool<T, A> le(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
         {
-            return batch_bool<T, A>({ lhs.get(0) <= rhs.get(0), lhs.get(1) <= rhs.get(1) });
+            return !(lhs > rhs);
         }
 
         /******
          * gt *
          ******/
+        namespace detail
+        {
+            XSIMD_INLINE int64x2_t bitwise_not_s64(int64x2_t arg) noexcept
+            {
+                return vreinterpretq_s64_s32(vmvnq_s32(vreinterpretq_s32_s64(arg)));
+            }
+
+            XSIMD_INLINE uint64x2_t bitwise_not_u64(uint64x2_t arg) noexcept
+            {
+                return vreinterpretq_u64_u32(vmvnq_u32(vreinterpretq_u32_u64(arg)));
+            }
+        }
 
         WRAP_BINARY_INT_EXCLUDING_64(vcgtq, detail::comp_return_type)
         WRAP_BINARY_FLOAT(vcgtq, detail::comp_return_type)
@@ -1033,10 +1061,19 @@ namespace xsimd
             return dispatcher.apply(register_type(lhs), register_type(rhs));
         }
 
-        template <class A, class T, detail::enable_sized_integral_t<T, 8> = 0>
+        template <class A, class T, detail::enable_sized_signed_t<T, 8> = 0>
         XSIMD_INLINE batch_bool<T, A> gt(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
         {
-            return batch_bool<T, A>({ lhs.get(0) > rhs.get(0), lhs.get(1) > rhs.get(1) });
+            using register_type = typename batch<T, A>::register_type;
+            return batch_bool<T, A>(vreinterpretq_u64_s64(vshrq_n_s64(vqsubq_s64(register_type(rhs), register_type(lhs)), 63)));
+        }
+
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 8> = 0>
+        XSIMD_INLINE batch_bool<T, A> gt(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
+        {
+            using register_type = typename batch<T, A>::register_type;
+            register_type acc = { 0x7FFFFFFFFFFFFFFFull, 0x7FFFFFFFFFFFFFFFull };
+            return batch_bool<T, A>(vreinterpretq_u64_s64(vshrq_n_s64(vreinterpretq_s64_u64(vqaddq_u64(vqsubq_u64(register_type(lhs), register_type(rhs)), acc)), 63)));
         }
 
         /******
@@ -1060,7 +1097,7 @@ namespace xsimd
         template <class A, class T, detail::enable_sized_integral_t<T, 8> = 0>
         XSIMD_INLINE batch_bool<T, A> ge(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
         {
-            return batch_bool<T, A>({ lhs.get(0) >= rhs.get(0), lhs.get(1) >= rhs.get(1) });
+            return !(lhs < rhs);
         }
 
         /*******************
@@ -1212,16 +1249,6 @@ namespace xsimd
 
         namespace detail
         {
-            XSIMD_INLINE int64x2_t bitwise_not_s64(int64x2_t arg) noexcept
-            {
-                return vreinterpretq_s64_s32(vmvnq_s32(vreinterpretq_s32_s64(arg)));
-            }
-
-            XSIMD_INLINE uint64x2_t bitwise_not_u64(uint64x2_t arg) noexcept
-            {
-                return vreinterpretq_u64_u32(vmvnq_u32(vreinterpretq_u32_u64(arg)));
-            }
-
             XSIMD_INLINE float32x4_t bitwise_not_f32(float32x4_t arg) noexcept
             {
                 return vreinterpretq_f32_u32(vmvnq_u32(vreinterpretq_u32_f32(arg)));
@@ -1314,7 +1341,7 @@ namespace xsimd
         template <class A, class T, detail::enable_sized_integral_t<T, 8> = 0>
         XSIMD_INLINE batch<T, A> min(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
         {
-            return { std::min(lhs.get(0), rhs.get(0)), std::min(lhs.get(1), rhs.get(1)) };
+            return select(lhs > rhs, rhs, lhs);
         }
 
         /*******
@@ -1338,7 +1365,7 @@ namespace xsimd
         template <class A, class T, detail::enable_sized_integral_t<T, 8> = 0>
         XSIMD_INLINE batch<T, A> max(batch<T, A> const& lhs, batch<T, A> const& rhs, requires_arch<neon>) noexcept
         {
-            return { std::max(lhs.get(0), rhs.get(0)), std::max(lhs.get(1), rhs.get(1)) };
+            return select(lhs > rhs, lhs, rhs);
         }
 
         /*******
@@ -1678,14 +1705,21 @@ namespace xsimd
          * reduce_max *
          **************/
 
-        // Using common implementation because ARM doe snot provide intrinsics
+        // Using common implementation because ARM does not provide intrinsics
         // for this operation
 
         /**************
          * reduce_min *
          **************/
 
-        // Using common implementation because ARM doe snot provide intrinsics
+        // Using common implementation because ARM does not provide intrinsics
+        // for this operation
+
+        /**************
+         * reduce_mul *
+         **************/
+
+        // Using common implementation because ARM does not provide intrinsics
         // for this operation
 
         /**********
@@ -2280,6 +2314,55 @@ namespace xsimd
             return vshlq_s64(lhs, rhs);
         }
 
+        // immediate variant
+        template <size_t shift, class A, class T, detail::enable_sized_unsigned_t<T, 1> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_lshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshlq_n_u8(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_signed_t<T, 1> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_lshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshlq_n_s8(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_unsigned_t<T, 2> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_lshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshlq_n_u16(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_signed_t<T, 2> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_lshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshlq_n_s16(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_unsigned_t<T, 4> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_lshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshlq_n_u32(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_signed_t<T, 4> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_lshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshlq_n_s32(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_unsigned_t<T, 8> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_lshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshlq_n_u64(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_signed_t<T, 8> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_lshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshlq_n_s64(x, shift);
+        }
+
         /******************
          * bitwise_rshift *
          ******************/
@@ -2455,6 +2538,110 @@ namespace xsimd
             return vshlq_s32(lhs, vnegq_s32(rhs));
         }
 
+        // immediate variant
+        template <size_t shift, class A, class T, detail::enable_sized_unsigned_t<T, 1> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshrq_n_u8(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_signed_t<T, 1> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshrq_n_s8(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_unsigned_t<T, 2> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshrq_n_u16(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_signed_t<T, 2> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshrq_n_s16(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_unsigned_t<T, 4> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshrq_n_u32(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_signed_t<T, 4> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshrq_n_s32(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_unsigned_t<T, 8> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshrq_n_u64(x, shift);
+        }
+
+        template <size_t shift, class A, class T, detail::enable_sized_signed_t<T, 8> = 0>
+        XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& x, requires_arch<neon>) noexcept
+        {
+            return vshrq_n_s64(x, shift);
+        }
+
+        // first
+        template <class A>
+        XSIMD_INLINE float first(batch<float, A> const& self, requires_arch<neon>) noexcept
+        {
+            return vgetq_lane_f32(self, 0);
+        }
+
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 1> = 0>
+        XSIMD_INLINE T first(batch<T, A> val, requires_arch<neon>) noexcept
+        {
+            return vgetq_lane_u8(val, 0);
+        }
+
+        template <class A, class T, detail::enable_sized_signed_t<T, 1> = 0>
+        XSIMD_INLINE T first(batch<T, A> val, requires_arch<neon>) noexcept
+        {
+            return vgetq_lane_s8(val, 0);
+        }
+
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 2> = 0>
+        XSIMD_INLINE T first(batch<T, A> val, requires_arch<neon>) noexcept
+        {
+            return vgetq_lane_u16(val, 0);
+        }
+
+        template <class A, class T, detail::enable_sized_signed_t<T, 2> = 0>
+        XSIMD_INLINE T first(batch<T, A> val, requires_arch<neon>) noexcept
+        {
+            return vgetq_lane_s16(val, 0);
+        }
+
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 4> = 0>
+        XSIMD_INLINE T first(batch<T, A> val, requires_arch<neon>) noexcept
+        {
+            return vgetq_lane_u32(val, 0);
+        }
+
+        template <class A, class T, detail::enable_sized_signed_t<T, 4> = 0>
+        XSIMD_INLINE T first(batch<T, A> val, requires_arch<neon>) noexcept
+        {
+            return vgetq_lane_s32(val, 0);
+        }
+
+        template <class A, class T, detail::enable_sized_unsigned_t<T, 8> = 0>
+        XSIMD_INLINE T first(batch<T, A> val, requires_arch<neon>) noexcept
+        {
+            return vgetq_lane_u64(val, 0);
+        }
+
+        template <class A, class T, detail::enable_sized_signed_t<T, 8> = 0>
+        XSIMD_INLINE T first(batch<T, A> val, requires_arch<neon>) noexcept
+        {
+            return vgetq_lane_s64(val, 0);
+        }
+
         // Overloads of bitwise shifts accepting two batches of uint64/int64 are not available with ARMv7
 
         /*******
@@ -2771,10 +2958,11 @@ namespace xsimd
         XSIMD_INLINE batch<T, A> rotate_left(batch<T, A> const& a, requires_arch<neon>) noexcept
         {
             using register_type = typename batch<T, A>::register_type;
+            // Adding modulo to avoid warning.
             const detail::neon_dispatcher::binary dispatcher = {
-                std::make_tuple(wrap::rotate_left_u8<N>, wrap::rotate_left_s8<N>, wrap::rotate_left_u16<N>, wrap::rotate_left_s16<N>,
-                                wrap::rotate_left_u32<N>, wrap::rotate_left_s32<N>, wrap::rotate_left_u64<N>, wrap::rotate_left_s64<N>,
-                                wrap::rotate_left_f32<N>)
+                std::make_tuple(wrap::rotate_left_u8<N>, wrap::rotate_left_s8<N>, wrap::rotate_left_u16<N % 8>, wrap::rotate_left_s16<N % 8>,
+                                wrap::rotate_left_u32<N % 4>, wrap::rotate_left_s32<N % 4>, wrap::rotate_left_u64<N % 2>, wrap::rotate_left_s64<N % 2>,
+                                wrap::rotate_left_f32<N % 4>)
             };
             return dispatcher.apply(register_type(a), register_type(a));
         }
@@ -2799,6 +2987,147 @@ namespace xsimd
             self.store_aligned(data.data());
             return set(batch<T, A>(), A(), data[idx]...);
         }
+
+        template <class A, uint64_t V0, uint64_t V1>
+        XSIMD_INLINE batch<uint64_t, A> swizzle(batch<uint64_t, A> const& self,
+                                                batch_constant<uint64_t, A, V0, V1>,
+                                                requires_arch<neon>) noexcept
+        {
+            XSIMD_IF_CONSTEXPR(V0 == 0 && V1 == 0)
+            {
+                auto lo = vget_low_u64(self);
+                return vcombine_u64(lo, lo);
+            }
+            XSIMD_IF_CONSTEXPR(V0 == 1 && V1 == 1)
+            {
+                auto hi = vget_high_u64(self);
+                return vcombine_u64(hi, hi);
+            }
+            XSIMD_IF_CONSTEXPR(V0 == 0 && V1 == 1)
+            {
+                return self;
+            }
+            else
+            {
+                return vextq_u64(self, self, 1);
+            }
+        }
+
+        template <class A, uint64_t V0, uint64_t V1>
+        XSIMD_INLINE batch<int64_t, A> swizzle(batch<int64_t, A> const& self,
+                                               batch_constant<int64_t, A, V0, V1> mask,
+                                               requires_arch<neon>) noexcept
+        {
+            return vreinterpretq_s64_u64(swizzle(vreinterpretq_u64_s64(self), mask, A {}));
+        }
+
+        namespace detail
+        {
+            template <uint32_t Va, uint32_t Vb>
+            XSIMD_INLINE uint8x8_t make_mask()
+            {
+                uint8x8_t res = {
+                    static_cast<uint8_t>((Va % 2) * 4 + 0),
+                    static_cast<uint8_t>((Va % 2) * 4 + 1),
+                    static_cast<uint8_t>((Va % 2) * 4 + 2),
+                    static_cast<uint8_t>((Va % 2) * 4 + 3),
+                    static_cast<uint8_t>((Vb % 2) * 4 + 0),
+                    static_cast<uint8_t>((Vb % 2) * 4 + 1),
+                    static_cast<uint8_t>((Vb % 2) * 4 + 2),
+                    static_cast<uint8_t>((Vb % 2) * 4 + 3),
+                };
+                return res;
+            }
+        }
+
+        template <class A, uint32_t V0, uint32_t V1, uint32_t V2, uint32_t V3>
+        XSIMD_INLINE batch<uint32_t, A> swizzle(batch<uint32_t, A> const& self,
+                                                batch_constant<uint32_t, A, V0, V1, V2, V3> mask,
+                                                requires_arch<neon>) noexcept
+        {
+            constexpr bool is_identity = detail::is_identity(mask);
+            constexpr bool is_dup_lo = detail::is_dup_lo(mask);
+            constexpr bool is_dup_hi = detail::is_dup_hi(mask);
+
+            XSIMD_IF_CONSTEXPR(is_identity)
+            {
+                return self;
+            }
+            XSIMD_IF_CONSTEXPR(is_dup_lo)
+            {
+                XSIMD_IF_CONSTEXPR(V0 == 0 && V1 == 1)
+                {
+                    return vreinterpretq_u32_u64(vdupq_lane_u64(vget_low_u64(vreinterpretq_u64_u32(self)), 0));
+                }
+                XSIMD_IF_CONSTEXPR(V0 == 1 && V1 == 0)
+                {
+                    return vreinterpretq_u32_u64(vdupq_lane_u64(vreinterpret_u64_u32(vrev64_u32(vget_low_u32(self))), 0));
+                }
+                return vdupq_n_u32(vgetq_lane_u32(self, V0));
+            }
+            XSIMD_IF_CONSTEXPR(is_dup_hi)
+            {
+                XSIMD_IF_CONSTEXPR(V0 == 2 && V1 == 3)
+                {
+                    return vreinterpretq_u32_u64(vdupq_lane_u64(vget_high_u64(vreinterpretq_u64_u32(self)), 0));
+                }
+                XSIMD_IF_CONSTEXPR(V0 == 3 && V1 == 2)
+                {
+                    return vreinterpretq_u32_u64(vdupq_lane_u64(vreinterpret_u64_u32(vrev64_u32(vget_high_u32(self))), 0));
+                }
+                return vdupq_n_u32(vgetq_lane_u32(self, V0));
+            }
+            XSIMD_IF_CONSTEXPR(V0 < 2 && V1 < 2 && V2 < 2 && V3 < 2)
+            {
+                uint8x8_t low = vreinterpret_u8_u64(vget_low_u64(vreinterpretq_u64_u32(self)));
+                uint8x8_t mask_lo = detail::make_mask<V0, V1>();
+                uint8x8_t mask_hi = detail::make_mask<V2, V3>();
+                uint8x8_t lo = vtbl1_u8(low, mask_lo);
+                uint8x8_t hi = vtbl1_u8(low, mask_hi);
+                return vreinterpretq_u32_u8(vcombine_u8(lo, hi));
+            }
+            XSIMD_IF_CONSTEXPR(V0 >= 2 && V1 >= 2 && V2 >= 2 && V3 >= 2)
+            {
+                uint8x8_t high = vreinterpret_u8_u64(vget_high_u64(vreinterpretq_u64_u32(self)));
+                uint8x8_t mask_lo = detail::make_mask<V0, V1>();
+                uint8x8_t mask_hi = detail::make_mask<V2, V3>();
+                uint8x8_t lo = vtbl1_u8(high, mask_lo);
+                uint8x8_t hi = vtbl1_u8(high, mask_hi);
+                return vreinterpretq_u32_u8(vcombine_u8(lo, hi));
+            }
+
+            uint8x8_t mask_lo = detail::make_mask<V0, V1>();
+            uint8x8_t mask_hi = detail::make_mask<V2, V3>();
+
+            uint8x8_t low = vreinterpret_u8_u64(vget_low_u64(vreinterpretq_u64_u32(self)));
+            uint8x8_t lol = vtbl1_u8(low, mask_lo);
+            uint8x8_t loh = vtbl1_u8(low, mask_hi);
+            uint32x4_t true_br = vreinterpretq_u32_u8(vcombine_u8(lol, loh));
+
+            uint8x8_t high = vreinterpret_u8_u64(vget_high_u64(vreinterpretq_u64_u32(self)));
+            uint8x8_t hil = vtbl1_u8(high, mask_lo);
+            uint8x8_t hih = vtbl1_u8(high, mask_hi);
+            uint32x4_t false_br = vreinterpretq_u32_u8(vcombine_u8(hil, hih));
+
+            batch_bool_constant<uint32_t, A, (V0 < 2), (V1 < 2), (V2 < 2), (V3 < 2)> blend_mask;
+            return select(blend_mask, batch<uint32_t, A>(true_br), batch<uint32_t, A>(false_br), A {});
+        }
+
+        template <class A, uint32_t V0, uint32_t V1, uint32_t V2, uint32_t V3>
+        XSIMD_INLINE batch<int32_t, A> swizzle(batch<int32_t, A> const& self,
+                                               batch_constant<int32_t, A, V0, V1, V2, V3> mask,
+                                               requires_arch<neon>) noexcept
+        {
+            return vreinterpretq_s32_u32(swizzle(vreinterpretq_u32_s32(self), mask, A {}));
+        }
+
+        template <class A, uint32_t V0, uint32_t V1, uint32_t V2, uint32_t V3>
+        XSIMD_INLINE batch<float, A> swizzle(batch<float, A> const& self,
+                                             batch_constant<uint32_t, A, V0, V1, V2, V3> mask,
+                                             requires_arch<neon>) noexcept
+        {
+            return vreinterpretq_f32_u32(swizzle(batch<uint32_t, A>(vreinterpretq_u32_f32(self)), mask, A {}));
+        }
     }
 
 }

sequenzo/dissimilarity_measures/src/xsimd/include/xsimd/arch/xsimd_neon64.hpp

@@ -28,6 +28,13 @@ namespace xsimd
     {
         using namespace types;
 
+        // first
+        template <class A>
+        XSIMD_INLINE double first(batch<double, A> const& self, requires_arch<neon64>) noexcept
+        {
+            return vgetq_lane_f64(self, 0);
+        }
+
         /*******
          * all *
          *******/

sequenzo/dissimilarity_measures/src/xsimd/include/xsimd/arch/xsimd_rvv.hpp

@@ -1335,6 +1335,19 @@ namespace xsimd
             return result;
         }
 
+        // first
+        template <class A, class T, detail::rvv_enable_all_t<T> = 0>
+        XSIMD_INLINE T first(batch<T, A> const& arg, requires_arch<rvv>) noexcept
+        {
+            return detail::rvvmv_lane0(arg);
+        }
+
+        template <class A, class T, detail::rvv_enable_all_t<T> = 0>
+        XSIMD_INLINE std::complex<T> first(batch<std::complex<T>, A> const& arg, requires_arch<rvv>) noexcept
+        {
+            return std::complex<T> { detail::rvvmv_lane0(arg.real()), detail::rvvmv_lane0(arg.imag()) };
+        }
+
         // insert
         template <class A, class T, size_t I, detail::rvv_enable_all_t<T> = 0>
         XSIMD_INLINE batch<T, A> insert(batch<T, A> const& arg, T val, index<I>, requires_arch<rvv>) noexcept

sequenzo/dissimilarity_measures/src/xsimd/include/xsimd/arch/xsimd_scalar.hpp

@@ -300,12 +300,29 @@ namespace xsimd
         return x << shift;
     }
 
+    template <size_t shift, class T>
+    XSIMD_INLINE typename std::enable_if<std::is_integral<T>::value, T>::type
+    bitwise_lshift(T x) noexcept
+    {
+        constexpr auto bits = std::numeric_limits<T>::digits + std::numeric_limits<T>::is_signed;
+        static_assert(shift < bits, "Count must be less than the number of bits in T");
+        return x << shift;
+    }
+
     template <class T0, class T1>
     XSIMD_INLINE typename std::enable_if<std::is_integral<T0>::value && std::is_integral<T1>::value, T0>::type
     bitwise_rshift(T0 x, T1 shift) noexcept
     {
         return x >> shift;
     }
+    template <size_t shift, class T>
+    XSIMD_INLINE typename std::enable_if<std::is_integral<T>::value, T>::type
+    bitwise_rshift(T x) noexcept
+    {
+        constexpr auto bits = std::numeric_limits<T>::digits + std::numeric_limits<T>::is_signed;
+        static_assert(shift < bits, "Count must be less than the number of bits in T");
+        return x >> shift;
+    }
 
     template <class T>
     XSIMD_INLINE typename std::enable_if<std::is_integral<T>::value, T>::type
@@ -447,16 +464,32 @@ namespace xsimd
     XSIMD_INLINE typename std::enable_if<std::is_integral<T0>::value && std::is_integral<T1>::value, T0>::type
     rotl(T0 x, T1 shift) noexcept
     {
-        constexpr auto N = std::numeric_limits<T0>::digits;
-        return (x << shift) | (x >> (N - shift));
+        constexpr auto bits = std::numeric_limits<T0>::digits + std::numeric_limits<T0>::is_signed;
+        return (x << shift) | (x >> (bits - shift));
+    }
+    template <size_t count, class T>
+    XSIMD_INLINE typename std::enable_if<std::is_integral<T>::value, T>::type
+    rotl(T x) noexcept
+    {
+        constexpr auto bits = std::numeric_limits<T>::digits + std::numeric_limits<T>::is_signed;
+        static_assert(count < bits, "Count must be less than the number of bits in T");
+        return (x << count) | (x >> (bits - count));
     }
 
     template <class T0, class T1>
     XSIMD_INLINE typename std::enable_if<std::is_integral<T0>::value && std::is_integral<T1>::value, T0>::type
     rotr(T0 x, T1 shift) noexcept
     {
-        constexpr auto N = std::numeric_limits<T0>::digits;
-        return (x >> shift) | (x << (N - shift));
+        constexpr auto bits = std::numeric_limits<T0>::digits + std::numeric_limits<T0>::is_signed;
+        return (x >> shift) | (x << (bits - shift));
+    }
+    template <size_t count, class T>
+    XSIMD_INLINE typename std::enable_if<std::is_integral<T>::value, T>::type
+    rotr(T x) noexcept
+    {
+        constexpr auto bits = std::numeric_limits<T>::digits + std::numeric_limits<T>::is_signed;
+        static_assert(count < bits, "Count must be less than the number of bits in T");
+        return (x >> count) | (x << (bits - count));
     }
 
     template <class T>
@@ -510,7 +543,11 @@ namespace xsimd
     template <class T, class = typename std::enable_if<std::is_scalar<T>::value>::type>
     XSIMD_INLINE bool is_flint(const T& x) noexcept
     {
+#ifdef __FAST_MATH__
+        return (x - std::trunc(x)) == T(0);
+#else
         return std::isnan(x - x) ? false : (x - std::trunc(x)) == T(0);
+#endif
     }
 
     template <class T, class = typename std::enable_if<std::is_scalar<T>::value>::type>