sequenzo 0.1.17__cp39-cp39-macosx_10_9_universal2.whl → 0.1.18__cp39-cp39-macosx_10_9_universal2.whl

sequenzo/dissimilarity_measures/src/xsimd/include/xsimd/arch/xsimd_sse2.hpp

@@ -292,6 +292,36 @@ namespace xsimd
                 return {};
             }
         }
+        template <size_t shift, class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
+        XSIMD_INLINE batch<T, A> bitwise_lshift(batch<T, A> const& self, requires_arch<sse2>) 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");
+            XSIMD_IF_CONSTEXPR(shift == 0)
+            {
+                return self;
+            }
+            else XSIMD_IF_CONSTEXPR(sizeof(T) == 1)
+            {
+                // 8-bit left shift via 16-bit shift + mask
+                __m128i shifted = _mm_slli_epi16(self, static_cast<int>(shift));
+                __m128i mask = _mm_set1_epi8(static_cast<char>(0xFF << shift));
+                return _mm_and_si128(shifted, mask);
+            }
+            else XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
+            {
+                return _mm_slli_epi16(self, static_cast<int>(shift));
+            }
+            else XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
+            {
+                return _mm_slli_epi32(self, static_cast<int>(shift));
+            }
+            else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)
+            {
+                return _mm_slli_epi64(self, static_cast<int>(shift));
+            }
+            return bitwise_lshift<shift>(self, common {});
+        }
 
         // bitwise_not
         template <class A>
@@ -420,6 +450,63 @@ namespace xsimd
                 }
             }
         }
+        template <size_t shift, class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
+        XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& self, requires_arch<sse2>) noexcept
+        {
+            constexpr auto bits = std::numeric_limits<T>::digits + std::numeric_limits<T>::is_signed;
+            static_assert(shift < bits,
+                          "Shift must be less than the number of value bits in the type");
+
+            XSIMD_IF_CONSTEXPR(shift == 0)
+            {
+                return self;
+            }
+
+            XSIMD_IF_CONSTEXPR(std::is_signed<T>::value)
+            {
+                XSIMD_IF_CONSTEXPR(sizeof(T) == 1)
+                {
+                    // 8-bit arithmetic right shift via 16-bit shift + sign-extension handling.
+                    __m128i shifted = _mm_srai_epi16(self, static_cast<int>(shift));
+                    __m128i sign_mask = _mm_set1_epi16(static_cast<short>(0xFF00 >> shift));
+                    __m128i cmp_negative = _mm_cmpgt_epi8(_mm_setzero_si128(), self);
+                    return _mm_or_si128(_mm_and_si128(sign_mask, cmp_negative),
+                                        _mm_andnot_si128(sign_mask, shifted));
+                }
+                else XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
+                {
+                    return _mm_srai_epi16(self, static_cast<int>(shift));
+                }
+                else XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
+                {
+                    return _mm_srai_epi32(self, static_cast<int>(shift));
+                }
+                // No 64-bit arithmetic right shift in SSE2; fall back
+                return bitwise_rshift<shift>(self, common {});
+            }
+            else // unsigned / logical right shift
+            {
+                XSIMD_IF_CONSTEXPR(sizeof(T) == 1)
+                {
+                    // Emulate byte-wise logical right shift using 16-bit shifts + per-byte mask.
+                    __m128i s16 = _mm_srli_epi16(self, static_cast<int>(shift));
+                    __m128i mask = _mm_set1_epi8(static_cast<char>(0xFFu >> shift));
+                    return _mm_and_si128(s16, mask);
+                }
+                else XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
+                {
+                    return _mm_srli_epi16(self, static_cast<int>(shift));
+                }
+                else XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
+                {
+                    return _mm_srli_epi32(self, static_cast<int>(shift));
+                }
+                else // sizeof(T) == 8
+                {
+                    return _mm_srli_epi64(self, static_cast<int>(shift));
+                }
+            }
+        }
 
         // bitwise_xor
         template <class A>
@@ -673,6 +760,53 @@ namespace xsimd
             return _mm_castsi128_pd(_mm_cmpeq_epi32(_mm_castpd_si128(self), _mm_castpd_si128(other)));
         }
 
+        // first
+        template <class A>
+        XSIMD_INLINE float first(batch<float, A> const& self, requires_arch<sse2>) noexcept
+        {
+            return _mm_cvtss_f32(self);
+        }
+
+        template <class A>
+        XSIMD_INLINE double first(batch<double, A> const& self, requires_arch<sse2>) noexcept
+        {
+            return _mm_cvtsd_f64(self);
+        }
+
+        template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
+        XSIMD_INLINE T first(batch<T, A> const& self, requires_arch<sse2>) noexcept
+        {
+            XSIMD_IF_CONSTEXPR(sizeof(T) == 1)
+            {
+                return static_cast<T>(_mm_cvtsi128_si32(self) & 0xFF);
+            }
+            else XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
+            {
+                return static_cast<T>(_mm_cvtsi128_si32(self) & 0xFFFF);
+            }
+            else XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
+            {
+                return static_cast<T>(_mm_cvtsi128_si32(self));
+            }
+            else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)
+            {
+#if defined(__x86_64__)
+                return static_cast<T>(_mm_cvtsi128_si64(self));
+#else
+                __m128i m;
+                _mm_storel_epi64(&m, self);
+                int64_t i;
+                std::memcpy(&i, &m, sizeof(i));
+                return i;
+#endif
+            }
+            else
+            {
+                assert(false && "unsupported arch/op combination");
+                return {};
+            }
+        }
+
         // from_mask
         template <class A>
         XSIMD_INLINE batch_bool<float, A> from_mask(batch_bool<float, A> const&, uint64_t mask, requires_arch<sse2>) noexcept
@@ -1090,7 +1224,7 @@ namespace xsimd
         template <class A>
         XSIMD_INLINE batch<float, A> max(batch<float, A> const& self, batch<float, A> const& other, requires_arch<sse2>) noexcept
         {
-            return _mm_max_ps(self, other);
+            return _mm_max_ps(other, self);
         }
         template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
         XSIMD_INLINE batch<T, A> max(batch<T, A> const& self, batch<T, A> const& other, requires_arch<sse2>) noexcept
@@ -1100,14 +1234,14 @@ namespace xsimd
         template <class A>
         XSIMD_INLINE batch<double, A> max(batch<double, A> const& self, batch<double, A> const& other, requires_arch<sse2>) noexcept
         {
-            return _mm_max_pd(self, other);
+            return _mm_max_pd(other, self);
         }
 
         // min
         template <class A>
         XSIMD_INLINE batch<float, A> min(batch<float, A> const& self, batch<float, A> const& other, requires_arch<sse2>) noexcept
         {
-            return _mm_min_ps(self, other);
+            return _mm_min_ps(other, self);
         }
         template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
         XSIMD_INLINE batch<T, A> min(batch<T, A> const& self, batch<T, A> const& other, requires_arch<sse2>) noexcept
@@ -1117,7 +1251,7 @@ namespace xsimd
         template <class A>
         XSIMD_INLINE batch<double, A> min(batch<double, A> const& self, batch<double, A> const& other, requires_arch<sse2>) noexcept
         {
-            return _mm_min_pd(self, other);
+            return _mm_min_pd(other, self);
         }
 
         // mul
@@ -1243,7 +1377,7 @@ namespace xsimd
             }
             else
             {
-                return hadd(self, common {});
+                return reduce_add(self, common {});
             }
         }
 
@@ -1269,10 +1403,10 @@ namespace xsimd
             batch<T, A> step2 = _mm_shufflelo_epi16(acc1, mask2);
             batch<T, A> acc2 = max(acc1, step2);
             if (sizeof(T) == 2)
-                return acc2.get(0);
+                return first(acc2, A {});
             batch<T, A> step3 = bitwise_cast<T>(bitwise_cast<uint16_t>(acc2) >> 8);
             batch<T, A> acc3 = max(acc2, step3);
-            return acc3.get(0);
+            return first(acc3, A {});
         }
 
         // reduce_min
@@ -1291,10 +1425,56 @@ namespace xsimd
             batch<T, A> step2 = _mm_shufflelo_epi16(acc1, mask2);
             batch<T, A> acc2 = min(acc1, step2);
             if (sizeof(T) == 2)
-                return acc2.get(0);
+                return first(acc2, A {});
             batch<T, A> step3 = bitwise_cast<T>(bitwise_cast<uint16_t>(acc2) >> 8);
             batch<T, A> acc3 = min(acc2, step3);
-            return acc3.get(0);
+            return first(acc3, A {});
+        }
+
+        // reduce_mul
+        template <class A>
+        XSIMD_INLINE float reduce_mul(batch<float, A> const& self, requires_arch<sse2>) noexcept
+        {
+            __m128 tmp0 = _mm_mul_ps(self, _mm_movehl_ps(self, self));
+            __m128 tmp1 = _mm_mul_ss(tmp0, _mm_shuffle_ps(tmp0, tmp0, 1));
+            return _mm_cvtss_f32(tmp1);
+        }
+
+        template <class A>
+        XSIMD_INLINE double reduce_mul(batch<double, A> const& self, requires_arch<sse2>) noexcept
+        {
+            return _mm_cvtsd_f64(_mm_mul_sd(self, _mm_unpackhi_pd(self, self)));
+        }
+
+        template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
+        XSIMD_INLINE T reduce_mul(batch<T, A> const& self, requires_arch<sse2>) noexcept
+        {
+            XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
+            {
+                batch<T, A> tmp1 = _mm_shuffle_epi32(self, _MM_SHUFFLE(0, 1, 2, 3));
+                tmp1 = tmp1 * self;
+                batch<T, A> tmp2 = _mm_unpackhi_epi32(tmp1, tmp1);
+                tmp2 = tmp2 * tmp1;
+                return _mm_cvtsi128_si32(tmp2);
+            }
+            else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)
+            {
+                batch<T, A> tmp1 = _mm_unpackhi_epi64(self, self);
+                auto tmp2 = tmp1 * self;
+#if defined(__x86_64__)
+                return _mm_cvtsi128_si64(tmp2);
+#else
+                __m128i m;
+                _mm_storel_epi64(&m, tmp2);
+                int64_t i;
+                std::memcpy(&i, &m, sizeof(i));
+                return i;
+#endif
+            }
+            else
+            {
+                return reduce_mul(self, common {});
+            }
         }
 
         // rsqrt
@@ -1641,22 +1821,78 @@ namespace xsimd
         }
 
         template <class A, uint16_t V0, uint16_t V1, uint16_t V2, uint16_t V3, uint16_t V4, uint16_t V5, uint16_t V6, uint16_t V7>
-        XSIMD_INLINE batch<uint16_t, A> swizzle(batch<uint16_t, A> const& self, batch_constant<uint16_t, A, V0, V1, V2, V3, V4, V5, V6, V7>, requires_arch<sse2>) noexcept
+        XSIMD_INLINE batch<uint16_t, A> swizzle(batch<uint16_t, A> const& self, batch_constant<uint16_t, A, V0, V1, V2, V3, V4, V5, V6, V7> mask, requires_arch<sse2>) noexcept
         {
-            // permute within each lane
+            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)
+            {
+                // permute the low half
+                constexpr int imm = detail::mod_shuffle(V0, V1, V2, V3);
+                const auto lo = _mm_shufflelo_epi16(self, imm);
+                // broadcast that 64-bit low half into both halves
+                const auto lo_all = _mm_unpacklo_epi64(lo, lo);
+                return lo_all;
+            }
+            XSIMD_IF_CONSTEXPR(is_dup_hi)
+            {
+                // permute the high half
+                constexpr int imm = detail::mod_shuffle(V4, V5, V6, V7);
+                const auto hi = _mm_shufflehi_epi16(self, imm);
+                // broadcast that 64-bit high half into both halves
+                const auto hi_all = _mm_unpackhi_epi64(hi, hi);
+                return hi_all;
+            }
+            // Only pick elements from the low lane
+            XSIMD_IF_CONSTEXPR((V0 < 4) && (V1 < 4) && (V2 < 4) && (V3 < 4) && (V4 < 4) && (V5 < 4) && (V6 < 4) && (V7 < 4))
+            {
+                // permute within each sub lane
+                constexpr auto mask_lo = detail::mod_shuffle(V0, V1, V2, V3);
+                constexpr auto mask_hi = detail::mod_shuffle(V4, V5, V6, V7);
+                __m128i lol = _mm_shufflelo_epi16(self, mask_lo);
+                __m128i loh = _mm_shufflelo_epi16(self, mask_hi);
+
+                // generate temporary lanes
+                return _mm_unpacklo_epi64(lol, loh);
+            }
+            // Only pick elements from the high lane
+            XSIMD_IF_CONSTEXPR((V0 >= 4) && (V1 >= 4) && (V2 >= 4) && (V3 >= 4) && (V4 >= 4) && (V5 >= 4) && (V6 >= 4) && (V7 >= 4))
+            {
+                // permute within each sub lane
+                constexpr auto mask_lo = detail::mod_shuffle(V0, V1, V2, V3);
+                constexpr auto mask_hi = detail::mod_shuffle(V4, V5, V6, V7);
+                __m128i hil = _mm_shufflehi_epi16(self, mask_lo);
+                __m128i hih = _mm_shufflehi_epi16(self, mask_hi);
+
+                // generate temporary lanes
+                return _mm_unpackhi_epi64(hil, hih);
+            }
+
+            // Generic case
+
+            // permute within each sub lane
             constexpr auto mask_lo = detail::mod_shuffle(V0, V1, V2, V3);
             constexpr auto mask_hi = detail::mod_shuffle(V4, V5, V6, V7);
-            __m128i lo = _mm_shufflelo_epi16(self, mask_lo);
-            __m128i hi = _mm_shufflehi_epi16(self, mask_hi);
+            __m128i lol = _mm_shufflelo_epi16(self, mask_lo);
+            __m128i loh = _mm_shufflelo_epi16(self, mask_hi);
+            __m128i hil = _mm_shufflehi_epi16(self, mask_lo);
+            __m128i hih = _mm_shufflehi_epi16(self, mask_hi);
 
-            __m128i lo_lo = _mm_castpd_si128(_mm_shuffle_pd(_mm_castsi128_pd(lo), _mm_castsi128_pd(lo), _MM_SHUFFLE2(0, 0)));
-            __m128i hi_hi = _mm_castpd_si128(_mm_shuffle_pd(_mm_castsi128_pd(hi), _mm_castsi128_pd(hi), _MM_SHUFFLE2(1, 1)));
+            // generate temporary lanes
+            __m128i lo = _mm_unpacklo_epi64(lol, loh);
+            __m128i hi = _mm_unpackhi_epi64(hil, hih);
 
             // mask to choose the right lane
             batch_bool_constant<uint16_t, A, (V0 < 4), (V1 < 4), (V2 < 4), (V3 < 4), (V4 < 4), (V5 < 4), (V6 < 4), (V7 < 4)> blend_mask;
 
             // blend the two permutes
-            return select(blend_mask, batch<uint16_t, A>(lo_lo), batch<uint16_t, A>(hi_hi));
+            return select(blend_mask, batch<uint16_t, A>(lo), batch<uint16_t, A>(hi));
         }
 
         template <class A, uint16_t V0, uint16_t V1, uint16_t V2, uint16_t V3, uint16_t V4, uint16_t V5, uint16_t V6, uint16_t V7>

sequenzo/dissimilarity_measures/src/xsimd/include/xsimd/arch/xsimd_sse3.hpp

@@ -51,6 +51,15 @@ namespace xsimd
             return _mm_cvtss_f32(tmp1);
         }
 
+        // reduce_mul
+        template <class A>
+        XSIMD_INLINE float reduce_mul(batch<float, A> const& self, requires_arch<sse3>) noexcept
+        {
+            __m128 tmp1 = _mm_mul_ps(self, _mm_movehl_ps(self, self));
+            __m128 tmp2 = _mm_mul_ps(tmp1, _mm_movehdup_ps(tmp1));
+            return _mm_cvtss_f32(tmp2);
+        }
+
     }
 
 }

sequenzo/dissimilarity_measures/src/xsimd/include/xsimd/arch/xsimd_ssse3.hpp

@@ -107,11 +107,22 @@ namespace xsimd
 
         // rotate_left
         template <size_t N, class A>
-        XSIMD_INLINE batch<uint16_t, A> rotate_left(batch<uint16_t, A> const& self, requires_arch<ssse3>) noexcept
+        XSIMD_INLINE batch<uint8_t, A> rotate_left(batch<uint8_t, A> const& self, requires_arch<ssse3>) noexcept
         {
             return _mm_alignr_epi8(self, self, N);
         }
         template <size_t N, class A>
+        XSIMD_INLINE batch<int8_t, A> rotate_left(batch<int8_t, A> const& self, requires_arch<ssse3>) noexcept
+        {
+            return bitwise_cast<int8_t>(rotate_left<N, A>(bitwise_cast<uint8_t>(self), ssse3 {}));
+        }
+
+        template <size_t N, class A>
+        XSIMD_INLINE batch<uint16_t, A> rotate_left(batch<uint16_t, A> const& self, requires_arch<ssse3>) noexcept
+        {
+            return _mm_alignr_epi8(self, self, 2 * N);
+        }
+        template <size_t N, class A>
         XSIMD_INLINE batch<int16_t, A> rotate_left(batch<int16_t, A> const& self, requires_arch<ssse3>) noexcept
         {
             return bitwise_cast<int16_t>(rotate_left<N, A>(bitwise_cast<uint16_t>(self), ssse3 {}));

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

@@ -949,6 +949,13 @@ namespace xsimd
             return svsel(index_predicate, broadcast<A, T>(val, sve {}), arg);
         }
 
+        // first
+        template <class A, class T, detail::sve_enable_all_t<T> = 0>
+        XSIMD_INLINE T first(batch<T, A> const& self, requires_arch<sve>) noexcept
+        {
+            return self.data[0];
+        }
+
         // all
         template <class A, class T, detail::sve_enable_all_t<T> = 0>
         XSIMD_INLINE bool all(batch_bool<T, A> const& arg, requires_arch<sve>) noexcept