react-native-executorch 0.10.5 → 0.10.7

package/third-party/common/phonemis/src/third-party/xsimd/arch/xsimd_avx512ifma.hpp

@@ -0,0 +1,17 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_AVX512IFMA_HPP
+#define XSIMD_AVX512IFMA_HPP
+
+#include "../types/xsimd_avx512ifma_register.hpp"
+
+#endif

package/third-party/common/phonemis/src/third-party/xsimd/arch/xsimd_avx512pf.hpp

@@ -0,0 +1,17 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_AVX512PF_HPP
+#define XSIMD_AVX512PF_HPP
+
+#include "../types/xsimd_avx512pf_register.hpp"
+
+#endif

package/third-party/common/phonemis/src/third-party/xsimd/arch/xsimd_avx512vbmi.hpp

@@ -0,0 +1,76 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_AVX512VBMI_HPP
+#define XSIMD_AVX512VBMI_HPP
+
+#include <type_traits>
+
+#include "../types/xsimd_avx512vbmi_register.hpp"
+
+namespace xsimd
+{
+
+    namespace kernel
+    {
+        using namespace types;
+
+        // slide_left
+        template <size_t N, class A, class T, class = std::enable_if_t<(N & 3) != 0 && (N < 64)>>
+        XSIMD_INLINE batch<T, A> slide_left(batch<T, A> const& x, requires_arch<avx512vbmi>) noexcept
+        {
+            static_assert((N & 3) != 0 && N < 64, "The AVX512F implementation may have a lower latency.");
+
+            __mmask64 mask = 0xFFFFFFFFFFFFFFFFull << (N & 63);
+            auto slide_pattern = make_batch_constant<uint8_t, detail::make_slide_left_pattern<N>, A>();
+            return _mm512_maskz_permutexvar_epi8(mask, slide_pattern.as_batch(), x);
+        }
+
+        // slide_right
+        template <size_t N, class A, class T, class = std::enable_if_t<(N & 3) != 0 && (N < 64)>>
+        XSIMD_INLINE batch<T, A> slide_right(batch<T, A> const& x, requires_arch<avx512vbmi>) noexcept
+        {
+            static_assert((N & 3) != 0 && N < 64, "The AVX512F implementation may have a lower latency.");
+
+            __mmask64 mask = 0xFFFFFFFFFFFFFFFFull >> (N & 63);
+            auto slide_pattern = make_batch_constant<uint8_t, detail::make_slide_right_pattern<N>, A>();
+            return _mm512_maskz_permutexvar_epi8(mask, slide_pattern.as_batch(), x);
+        }
+
+        // swizzle (dynamic version)
+        template <class A>
+        XSIMD_INLINE batch<uint8_t, A> swizzle(batch<uint8_t, A> const& self, batch<uint8_t, A> mask, requires_arch<avx512vbmi>) noexcept
+        {
+            return _mm512_permutexvar_epi8(mask, self);
+        }
+
+        template <class A>
+        XSIMD_INLINE batch<int8_t, A> swizzle(batch<int8_t, A> const& self, batch<uint8_t, A> mask, requires_arch<avx512vbmi>) noexcept
+        {
+            return bitwise_cast<int8_t>(swizzle(bitwise_cast<uint8_t>(self), mask, avx512vbmi {}));
+        }
+
+        // swizzle (static version)
+        template <class A, uint8_t... Vs>
+        XSIMD_INLINE batch<uint8_t, A> swizzle(batch<uint8_t, A> const& self, batch_constant<uint8_t, A, Vs...> mask, requires_arch<avx512vbmi>) noexcept
+        {
+            return swizzle(self, mask.as_batch(), avx512vbmi {});
+        }
+
+        template <class A, uint8_t... Vs>
+        XSIMD_INLINE batch<int8_t, A> swizzle(batch<int8_t, A> const& self, batch_constant<uint8_t, A, Vs...> mask, requires_arch<avx512vbmi>) noexcept
+        {
+            return swizzle(self, mask.as_batch(), avx512vbmi {});
+        }
+    }
+}
+
+#endif

package/third-party/common/phonemis/src/third-party/xsimd/arch/xsimd_avx512vbmi2.hpp

@@ -0,0 +1,130 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_AVX512VBMI2_HPP
+#define XSIMD_AVX512VBMI2_HPP
+
+#include <type_traits>
+
+#include "../types/xsimd_avx512vbmi2_register.hpp"
+
+namespace xsimd
+{
+
+    namespace kernel
+    {
+        using namespace types;
+
+        // compress
+        template <class A>
+        XSIMD_INLINE batch<int16_t, A> compress(batch<int16_t, A> const& self, batch_bool<int16_t, A> const& mask, requires_arch<avx512vbmi2>) noexcept
+        {
+            return _mm512_maskz_compress_epi16((__mmask32)mask.mask(), self);
+        }
+        template <class A>
+        XSIMD_INLINE batch<uint16_t, A> compress(batch<uint16_t, A> const& self, batch_bool<uint16_t, A> const& mask, requires_arch<avx512vbmi2>) noexcept
+        {
+            return _mm512_maskz_compress_epi16((__mmask32)mask.mask(), self);
+        }
+        template <class A>
+        XSIMD_INLINE batch<int8_t, A> compress(batch<int8_t, A> const& self, batch_bool<int8_t, A> const& mask, requires_arch<avx512vbmi2>) noexcept
+        {
+            return _mm512_maskz_compress_epi8(mask.mask(), self);
+        }
+        template <class A>
+        XSIMD_INLINE batch<uint8_t, A> compress(batch<uint8_t, A> const& self, batch_bool<uint8_t, A> const& mask, requires_arch<avx512vbmi2>) noexcept
+        {
+            return _mm512_maskz_compress_epi8(mask.mask(), self);
+        }
+
+        // expand
+        template <class A>
+        XSIMD_INLINE batch<int16_t, A> expand(batch<int16_t, A> const& self, batch_bool<int16_t, A> const& mask, requires_arch<avx512vbmi2>) noexcept
+        {
+            return _mm512_maskz_expand_epi16((__mmask32)mask.mask(), self);
+        }
+        template <class A>
+        XSIMD_INLINE batch<uint16_t, A> expand(batch<uint16_t, A> const& self, batch_bool<uint16_t, A> const& mask, requires_arch<avx512vbmi2>) noexcept
+        {
+            return _mm512_maskz_expand_epi16((__mmask32)mask.mask(), self);
+        }
+        template <class A>
+        XSIMD_INLINE batch<int8_t, A> expand(batch<int8_t, A> const& self, batch_bool<int8_t, A> const& mask, requires_arch<avx512vbmi2>) noexcept
+        {
+            return _mm512_maskz_expand_epi8(mask.mask(), self);
+        }
+        template <class A>
+        XSIMD_INLINE batch<uint8_t, A> expand(batch<uint8_t, A> const& self, batch_bool<uint8_t, A> const& mask, requires_arch<avx512vbmi2>) noexcept
+        {
+            return _mm512_maskz_expand_epi8(mask.mask(), self);
+        }
+
+        // rotl
+        template <class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> rotl(batch<T, A> const& self, int32_t other, requires_arch<avx512vbmi2>) noexcept
+        {
+            XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
+            {
+                return _mm512_shldv_epi16(self, self, _mm512_set1_epi16(static_cast<uint16_t>(other)));
+            }
+            else
+            {
+                return rotl(self, other, avx512bw {});
+            }
+        }
+
+        template <size_t count, class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> rotl(batch<T, A> const& self, requires_arch<avx512vbmi2>) 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");
+            XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
+            {
+                return _mm512_shldi_epi16(self, self, count);
+            }
+            else
+            {
+                return rotl<count>(self, avx512bw {});
+            }
+        }
+
+        // rotr
+        template <class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> rotr(batch<T, A> const& self, int32_t other, requires_arch<avx512vbmi2>) noexcept
+        {
+            XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
+            {
+                return _mm512_shrdv_epi16(self, self, _mm512_set1_epi16(static_cast<uint16_t>(other)));
+            }
+            else
+            {
+                return rotr(self, other, avx512bw {});
+            }
+        }
+
+        template <size_t count, class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> rotr(batch<T, A> const& self, requires_arch<avx512vbmi2>) 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");
+            XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
+            {
+                return _mm512_shrdi_epi16(self, self, count);
+            }
+            else
+            {
+                return rotr<count>(self, avx512bw {});
+            }
+        }
+    }
+}
+
+#endif

package/third-party/common/phonemis/src/third-party/xsimd/arch/xsimd_avx512vnni_avx512bw.hpp

@@ -0,0 +1,17 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_AVX512VNNI_AVX512_BW_HPP
+#define XSIMD_AVX512VNNI_AVX512_BW_HPP
+
+#include "../types/xsimd_avx512vnni_avx512bw_register.hpp"
+
+#endif

package/third-party/common/phonemis/src/third-party/xsimd/arch/xsimd_avx512vnni_avx512vbmi2.hpp

@@ -0,0 +1,17 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_AVX512VNNI_AVX512VBMI2_HPP
+#define XSIMD_AVX512VNNI_AVX512VBMI2_HPP
+
+#include "../types/xsimd_avx512vnni_avx512vbmi2_register.hpp"
+
+#endif

package/third-party/common/phonemis/src/third-party/xsimd/arch/xsimd_avx_128.hpp

@@ -0,0 +1,164 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ * Copyright (c) Marco Barbone                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_AVX_128_HPP
+#define XSIMD_AVX_128_HPP
+
+#include <type_traits>
+
+#include "../types/xsimd_avx_register.hpp"
+#include "../types/xsimd_batch_constant.hpp"
+
+namespace xsimd
+{
+    namespace kernel
+    {
+        using namespace types;
+
+        // broadcast
+        template <class A, class T, class = std::enable_if_t<std::is_same<T, float>::value>>
+        XSIMD_INLINE batch<T, A> broadcast(T val, requires_arch<avx_128>) noexcept
+        {
+            return _mm_broadcast_ss(&val);
+        }
+
+        // eq
+        template <class A>
+        XSIMD_INLINE batch_bool<float, A> eq(batch<float, A> const& self, batch<float, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_ps(self, other, _CMP_EQ_OQ);
+        }
+        template <class A>
+        XSIMD_INLINE batch_bool<double, A> eq(batch<double, A> const& self, batch<double, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_pd(self, other, _CMP_EQ_OQ);
+        }
+
+        // gt
+        template <class A>
+        XSIMD_INLINE batch_bool<float, A> gt(batch<float, A> const& self, batch<float, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_ps(self, other, _CMP_GT_OQ);
+        }
+        template <class A>
+        XSIMD_INLINE batch_bool<double, A> gt(batch<double, A> const& self, batch<double, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_pd(self, other, _CMP_GT_OQ);
+        }
+
+        // ge
+        template <class A>
+        XSIMD_INLINE batch_bool<float, A> ge(batch<float, A> const& self, batch<float, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_ps(self, other, _CMP_GE_OQ);
+        }
+        template <class A>
+        XSIMD_INLINE batch_bool<double, A> ge(batch<double, A> const& self, batch<double, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_pd(self, other, _CMP_GE_OQ);
+        }
+
+        // lt
+        template <class A>
+        XSIMD_INLINE batch_bool<float, A> lt(batch<float, A> const& self, batch<float, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_ps(self, other, _CMP_LT_OQ);
+        }
+        template <class A>
+        XSIMD_INLINE batch_bool<double, A> lt(batch<double, A> const& self, batch<double, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_pd(self, other, _CMP_LT_OQ);
+        }
+
+        // le
+        template <class A>
+        XSIMD_INLINE batch_bool<float, A> le(batch<float, A> const& self, batch<float, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_ps(self, other, _CMP_LE_OQ);
+        }
+        template <class A>
+        XSIMD_INLINE batch_bool<double, A> le(batch<double, A> const& self, batch<double, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_pd(self, other, _CMP_LE_OQ);
+        }
+
+        // neq
+        template <class A>
+        XSIMD_INLINE batch_bool<float, A> neq(batch<float, A> const& self, batch<float, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_ps(self, other, _CMP_NEQ_UQ);
+        }
+        template <class A>
+        XSIMD_INLINE batch_bool<double, A> neq(batch<double, A> const& self, batch<double, A> const& other, requires_arch<avx_128>) noexcept
+        {
+            return _mm_cmp_pd(self, other, _CMP_NEQ_UQ);
+        }
+
+        // load_masked
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE batch<float, A> load_masked(float const* mem, batch_bool_constant<float, A, Values...> mask, convert<float>, Mode, requires_arch<avx_128>) noexcept
+        {
+            return _mm_maskload_ps(mem, mask.as_batch());
+        }
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE batch<double, A> load_masked(double const* mem, batch_bool_constant<double, A, Values...> mask, convert<double>, Mode, requires_arch<avx_128>) noexcept
+        {
+            return _mm_maskload_pd(mem, mask.as_batch());
+        }
+
+        // store_masked
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE void store_masked(float* mem, batch<float, A> const& src, batch_bool_constant<float, A, Values...> mask, Mode, requires_arch<avx_128>) noexcept
+        {
+            return _mm_maskstore_ps(mem, mask.as_batch(), src);
+        }
+
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE void store_masked(double* mem, batch<double, A> const& src, batch_bool_constant<double, A, Values...> mask, Mode, requires_arch<avx_128>) noexcept
+        {
+            return _mm_maskstore_pd(mem, mask.as_batch(), src);
+        }
+
+        // swizzle (dynamic mask)
+        template <class A, class T, class ITy, class = std::enable_if_t<std::is_floating_point<T>::value && sizeof(T) == sizeof(ITy)>>
+        XSIMD_INLINE batch<T, A> swizzle(batch<T, A> const& self, batch<ITy, A> mask, requires_arch<avx_128>) noexcept
+        {
+            XSIMD_IF_CONSTEXPR(std::is_same<T, float>::value)
+            {
+                return _mm_permutevar_ps(self, mask);
+            }
+            else
+            {
+                // VPERMILPD's variable control reads bit 1 of each 64-bit selector
+                // (bit 0 is ignored), so a {0,1} index needs to become {0,2}.
+                // Negation is a cheap alternative to a left shift by 1.
+                return _mm_permutevar_pd(self, -mask);
+            }
+        }
+
+        // swizzle (constant mask)
+        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>, requires_arch<avx_128>) noexcept
+        {
+            return _mm_permute_ps(self, detail::mod_shuffle(V0, V1, V2, V3));
+        }
+
+        template <class A, uint32_t V0, uint32_t V1>
+        XSIMD_INLINE batch<double, A> swizzle(batch<double, A> const& self, batch_constant<uint64_t, A, V0, V1>, requires_arch<avx_128>) noexcept
+        {
+            return _mm_permute_pd(self, detail::mod_shuffle(V0, V1));
+        }
+
+    }
+}
+
+#endif

package/third-party/common/phonemis/src/third-party/xsimd/arch/xsimd_avxvnni.hpp

@@ -0,0 +1,17 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_AVXVNNI_HPP
+#define XSIMD_AVXVNNI_HPP
+
+#include "../types/xsimd_avxvnni_register.hpp"
+
+#endif

package/third-party/common/phonemis/src/third-party/xsimd/arch/xsimd_common.hpp

@@ -0,0 +1,26 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_COMMON_HPP
+#define XSIMD_COMMON_HPP
+
+#include "./common/xsimd_common_arithmetic.hpp"
+#include "./common/xsimd_common_bit.hpp"
+#include "./common/xsimd_common_cast.hpp"
+#include "./common/xsimd_common_complex.hpp"
+#include "./common/xsimd_common_logical.hpp"
+#include "./common/xsimd_common_math.hpp"
+#include "./common/xsimd_common_memory.hpp"
+#include "./common/xsimd_common_rounding.hpp"
+#include "./common/xsimd_common_swizzle.hpp"
+#include "./common/xsimd_common_trigo.hpp"
+
+#endif

package/third-party/common/phonemis/src/third-party/xsimd/arch/xsimd_common_fwd.hpp

@@ -0,0 +1,127 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ * Copyright (c) Marco Barbone                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_COMMON_FWD_HPP
+#define XSIMD_COMMON_FWD_HPP
+
+#include <cstdint>
+#include <type_traits>
+
+namespace xsimd
+{
+    // Minimal forward declarations used in this header
+    template <class T, class A>
+    class batch;
+    template <class T, class A>
+    class batch_bool;
+    template <class T, class A, T... Vs>
+    struct batch_constant;
+    template <class T, class A, bool... Vs>
+    struct batch_bool_constant;
+    template <class T>
+    struct convert;
+    template <class A>
+    struct requires_arch;
+    struct aligned_mode;
+    struct unaligned_mode;
+
+    namespace types
+    {
+        template <typename T, class A>
+        struct has_simd_register;
+    }
+
+    namespace kernel
+    {
+        // forward declaration
+        template <class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> abs(batch<T, A> const& self, requires_arch<common>) noexcept;
+        template <class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> bitwise_lshift(batch<T, A> const& self, batch<T, A> const& other, requires_arch<common>) noexcept;
+        template <size_t shift, class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> bitwise_lshift(batch<T, A> const& self, requires_arch<common>) noexcept;
+        template <class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& self, batch<T, A> const& other, requires_arch<common>) noexcept;
+        template <size_t shift, class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> bitwise_rshift(batch<T, A> const& self, requires_arch<common>) noexcept;
+        template <class A, class T>
+        XSIMD_INLINE batch_bool<T, A> gt(batch<T, A> const& self, batch<T, A> const& other, requires_arch<common>) noexcept;
+        template <class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> mul(batch<T, A> const& self, batch<T, A> const& other, requires_arch<common>) noexcept;
+        template <class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> sadd(batch<T, A> const& self, batch<T, A> const& other, requires_arch<common>) noexcept;
+        template <class A, class T, class = std::enable_if_t<std::is_integral<T>::value>>
+        XSIMD_INLINE batch<T, A> ssub(batch<T, A> const& self, batch<T, A> const& other, requires_arch<common>) noexcept;
+        template <class A, class T, class = std::enable_if_t<std::is_scalar<T>::value>>
+        XSIMD_INLINE T reduce_add(batch<T, A> const& self, requires_arch<common>) noexcept;
+        template <class A, class T, class = std::enable_if_t<std::is_scalar<T>::value>>
+        XSIMD_INLINE T reduce_mul(batch<T, A> const& self, requires_arch<common>) noexcept;
+        template <class A, class T, class STy>
+        XSIMD_INLINE batch<T, A> rotl(batch<T, A> const& self, STy other, requires_arch<common>) noexcept;
+        template <size_t count, class A, class T>
+        XSIMD_INLINE batch<T, A> rotl(batch<T, A> const& self, requires_arch<common>) noexcept;
+        template <class A, class T, class STy>
+        XSIMD_INLINE batch<T, A> rotr(batch<T, A> const& self, STy other, requires_arch<common>) noexcept;
+        template <size_t count, class A, class T>
+        XSIMD_INLINE batch<T, A> rotr(batch<T, A> const& self, requires_arch<common>) noexcept;
+        template <class A, class T>
+        XSIMD_INLINE batch<T, A> load(T const* mem, aligned_mode, requires_arch<A>) noexcept;
+        template <class A, class T>
+        XSIMD_INLINE batch<T, A> load(T const* mem, unaligned_mode, requires_arch<A>) noexcept;
+        template <class A, class T_in, class T_out, bool... Values, class alignment>
+        XSIMD_INLINE batch<T_out, A> load_masked(T_in const* mem, batch_bool_constant<T_out, A, Values...> mask, convert<T_out>, alignment, requires_arch<common>) noexcept;
+        template <class A, class T_in, class T_out, bool... Values, class alignment>
+        XSIMD_INLINE void store_masked(T_out* mem, batch<T_in, A> const& src, batch_bool_constant<T_in, A, Values...> mask, alignment, requires_arch<common>) noexcept;
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE batch<int32_t, A> load_masked(int32_t const* mem, batch_bool_constant<int32_t, A, Values...> mask, convert<int32_t>, Mode, requires_arch<A>) noexcept;
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE batch<uint32_t, A> load_masked(uint32_t const* mem, batch_bool_constant<uint32_t, A, Values...> mask, convert<uint32_t>, Mode, requires_arch<A>) noexcept;
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE std::enable_if_t<types::has_simd_register<double, A>::value, batch<int64_t, A>> load_masked(int64_t const*, batch_bool_constant<int64_t, A, Values...>, convert<int64_t>, Mode, requires_arch<A>) noexcept;
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE std::enable_if_t<types::has_simd_register<double, A>::value, batch<uint64_t, A>> load_masked(uint64_t const*, batch_bool_constant<uint64_t, A, Values...>, convert<uint64_t>, Mode, requires_arch<A>) noexcept;
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE void store_masked(int32_t* mem, batch<int32_t, A> const& src, batch_bool_constant<int32_t, A, Values...> mask, Mode, requires_arch<A>) noexcept;
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE void store_masked(uint32_t* mem, batch<uint32_t, A> const& src, batch_bool_constant<uint32_t, A, Values...> mask, Mode, requires_arch<A>) noexcept;
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE std::enable_if_t<types::has_simd_register<double, A>::value> store_masked(int64_t*, batch<int64_t, A> const&, batch_bool_constant<int64_t, A, Values...>, Mode, requires_arch<A>) noexcept;
+        template <class A, bool... Values, class Mode>
+        XSIMD_INLINE std::enable_if_t<types::has_simd_register<double, A>::value> store_masked(uint64_t*, batch<uint64_t, A> const&, batch_bool_constant<uint64_t, A, Values...>, Mode, requires_arch<A>) noexcept;
+
+        // Forward declarations for pack-level helpers
+        namespace detail
+        {
+            template <class T>
+            XSIMD_INLINE void reassociation_barrier(T& x, const char*) noexcept;
+            template <class T, class A>
+            XSIMD_INLINE void reassociation_barrier(batch<T, A>& b, const char* reason) noexcept;
+
+            template <typename T, T... Vs>
+            XSIMD_INLINE constexpr bool is_identity() noexcept;
+            template <typename T, class A, T... Vs>
+            XSIMD_INLINE constexpr bool is_identity(batch_constant<T, A, Vs...>) noexcept;
+            template <typename T, class A, T... Vs>
+            XSIMD_INLINE constexpr bool is_dup_lo(batch_constant<T, A, Vs...>) noexcept;
+            template <typename T, class A, T... Vs>
+            XSIMD_INLINE constexpr bool is_dup_hi(batch_constant<T, A, Vs...>) noexcept;
+            template <typename T, class A, T... Vs>
+            XSIMD_INLINE constexpr bool is_cross_lane(batch_constant<T, A, Vs...>) noexcept;
+            template <typename T, class A, T... Vs>
+            XSIMD_INLINE constexpr bool is_only_from_lo(batch_constant<T, A, Vs...>) noexcept;
+            template <typename T, class A, T... Vs>
+            XSIMD_INLINE constexpr bool is_only_from_hi(batch_constant<T, A, Vs...>) noexcept;
+        }
+    }
+}
+
+#endif