diff --git a/third_party/xsimd/include/xsimd/arch/xsimd_sse2.hpp b/third_party/xsimd/include/xsimd/arch/xsimd_sse2.hpp
--- a/third_party/xsimd/include/xsimd/arch/xsimd_sse2.hpp
+++ b/third_party/xsimd/include/xsimd/arch/xsimd_sse2.hpp
@@ -1638,10 +1638,35 @@
         XSIMD_INLINE batch<int32_t, A> swizzle(batch<int32_t, A> const& self, batch_constant<uint32_t, A, V0, V1, V2, V3> mask, requires_arch<sse2>) noexcept
         {
             return bitwise_cast<int32_t>(swizzle(bitwise_cast<uint32_t>(self), mask, sse2 {}));
         }
 
+        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
+        {
+            // permute within each 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 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)));
+
+            // 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));
+        }
+
+        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<int16_t, A> swizzle(batch<int16_t, A> const& self, batch_constant<uint16_t, A, V0, V1, V2, V3, V4, V5, V6, V7> mask, requires_arch<sse2>) noexcept
+        {
+            return bitwise_cast<int16_t>(swizzle(bitwise_cast<uint16_t>(self), mask, sse2 {}));
+        }
+
         // transpose
         template <class A>
         XSIMD_INLINE void transpose(batch<float, A>* matrix_begin, batch<float, A>* matrix_end, requires_arch<sse2>) noexcept
         {
             assert((matrix_end - matrix_begin == batch<float, A>::size) && "correctly sized matrix");
diff --git a/third_party/xsimd/include/xsimd/memory/xsimd_aligned_allocator.hpp b/third_party/xsimd/include/xsimd/memory/xsimd_aligned_allocator.hpp
--- a/third_party/xsimd/include/xsimd/memory/xsimd_aligned_allocator.hpp
+++ b/third_party/xsimd/include/xsimd/memory/xsimd_aligned_allocator.hpp
@@ -37,11 +37,11 @@
      * performs memory allocation aligned by the specified value.
      *
      * @tparam T type of objects to allocate.
      * @tparam Align alignment in bytes.
      */
-    template <class T, size_t Align>
+    template <class T, size_t Align = std::conditional<std::is_same<unsupported, default_arch>::value, generic, default_arch>::type::alignment()>
     class aligned_allocator
     {
     public:
         using value_type = T;
         using pointer = T*;
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx2_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx2_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx2_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx2_register.hpp
@@ -27,10 +27,15 @@
         static constexpr bool available() noexcept { return true; }
         static constexpr char const* name() noexcept { return "avx2"; }
     };
 
 #if XSIMD_WITH_AVX2
+
+#if !XSIMD_WITH_AVX
+#error "architecture inconsistency: avx2 requires avx"
+#endif
+
     namespace types
     {
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(avx2, avx);
     }
 #endif
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx512bw_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx512bw_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx512bw_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx512bw_register.hpp
@@ -29,10 +29,14 @@
         static constexpr char const* name() noexcept { return "avx512bw"; }
     };
 
 #if XSIMD_WITH_AVX512BW
 
+#if !XSIMD_WITH_AVX512DQ
+#error "architecture inconsistency: avx512bw requires avx512dq"
+#endif
+
     namespace types
     {
         template <class T>
         struct get_bool_simd_register<T, avx512bw>
         {
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx512cd_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx512cd_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx512cd_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx512cd_register.hpp
@@ -29,10 +29,14 @@
         static constexpr char const* name() noexcept { return "avx512cd"; }
     };
 
 #if XSIMD_WITH_AVX512CD
 
+#if !XSIMD_WITH_AVX512F
+#error "architecture inconsistency: avx512bw requires avx512f"
+#endif
+
     namespace types
     {
         template <class T>
         struct get_bool_simd_register<T, avx512cd>
         {
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx512dq_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx512dq_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx512dq_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx512dq_register.hpp
@@ -29,10 +29,14 @@
         static constexpr char const* name() noexcept { return "avx512dq"; }
     };
 
 #if XSIMD_WITH_AVX512DQ
 
+#if !XSIMD_WITH_AVX512CD
+#error "architecture inconsistency: avx512dq requires avx512cd"
+#endif
+
     namespace types
     {
         template <class T>
         struct get_bool_simd_register<T, avx512dq>
         {
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx512er_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx512er_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx512er_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx512er_register.hpp
@@ -29,10 +29,14 @@
         static constexpr char const* name() noexcept { return "avx512er"; }
     };
 
 #if XSIMD_WITH_AVX512ER
 
+#if !XSIMD_WITH_AVX512CD
+#error "architecture inconsistency: avx512er requires avx512cd"
+#endif
+
     namespace types
     {
         template <class T>
         struct get_bool_simd_register<T, avx512er>
         {
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx512f_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx512f_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx512f_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx512f_register.hpp
@@ -31,10 +31,14 @@
         static constexpr char const* name() noexcept { return "avx512f"; }
     };
 
 #if XSIMD_WITH_AVX512F
 
+#if !XSIMD_WITH_AVX2
+#error "architecture inconsistency: avx512f requires avx2"
+#endif
+
     namespace types
     {
         template <class T>
         struct simd_avx512_bool_register
         {
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx512ifma_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx512ifma_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx512ifma_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx512ifma_register.hpp
@@ -29,10 +29,14 @@
         static constexpr char const* name() noexcept { return "avx512ifma"; }
     };
 
 #if XSIMD_WITH_AVX512IFMA
 
+#if !XSIMD_WITH_AVX512BW
+#error "architecture inconsistency: avx512ifma requires avx512bw"
+#endif
+
     namespace types
     {
         template <class T>
         struct get_bool_simd_register<T, avx512ifma>
         {
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx512pf_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx512pf_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx512pf_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx512pf_register.hpp
@@ -29,10 +29,14 @@
         static constexpr char const* name() noexcept { return "avx512pf"; }
     };
 
 #if XSIMD_WITH_AVX512PF
 
+#if !XSIMD_WITH_AVX512ER
+#error "architecture inconsistency: avx512pf requires avx512er"
+#endif
+
     namespace types
     {
         template <class T>
         struct get_bool_simd_register<T, avx512pf>
         {
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx512vbmi_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx512vbmi_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx512vbmi_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx512vbmi_register.hpp
@@ -29,10 +29,14 @@
         static constexpr char const* name() noexcept { return "avx512vbmi"; }
     };
 
 #if XSIMD_WITH_AVX512VBMI
 
+#if !XSIMD_WITH_AVX512IFMA
+#error "architecture inconsistency: avx512vbmi requires avx512ifma"
+#endif
+
     namespace types
     {
         template <class T>
         struct get_bool_simd_register<T, avx512vbmi>
         {
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx512vnni_avx512bw_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx512vnni_avx512bw_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx512vnni_avx512bw_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx512vnni_avx512bw_register.hpp
@@ -32,10 +32,14 @@
         static constexpr char const* name() noexcept { return "avx512vnni+avx512bw"; }
     };
 
 #if XSIMD_WITH_AVX512VNNI_AVX512BW
 
+#if !XSIMD_WITH_AVX512BW
+#error "architecture inconsistency: avx512vnni+avx512bw requires avx512bw"
+#endif
+
     namespace types
     {
         template <class T>
         struct get_bool_simd_register<T, avx512vnni<avx512bw>>
         {
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx512vnni_avx512vbmi_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx512vnni_avx512vbmi_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx512vnni_avx512vbmi_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx512vnni_avx512vbmi_register.hpp
@@ -32,10 +32,14 @@
         static constexpr char const* name() noexcept { return "avx512vnni+avx512vbmi"; }
     };
 
 #if XSIMD_WITH_AVX512VNNI_AVX512VBMI
 
+#if !XSIMD_WITH_AVX512VBMI
+#error "architecture inconsistency: avx512vnni+avx512vbmi requires avx512vbmi"
+#endif
+
     namespace types
     {
         template <class T>
         struct get_bool_simd_register<T, avx512vnni<avx512vbmi>>
         {
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avx_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avx_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avx_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avx_register.hpp
@@ -32,10 +32,14 @@
     };
 }
 
 #if XSIMD_WITH_AVX
 
+#if !XSIMD_WITH_SSE4_2
+#error "architecture inconsistency: avx requires sse4.2"
+#endif
+
 #include <immintrin.h>
 
 namespace xsimd
 {
     namespace types
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_avxvnni_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_avxvnni_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_avxvnni_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_avxvnni_register.hpp
@@ -27,10 +27,15 @@
         static constexpr bool available() noexcept { return true; }
         static constexpr char const* name() noexcept { return "avxvnni"; }
     };
 
 #if XSIMD_WITH_AVXVNNI
+
+#if !XSIMD_WITH_AVX2
+#error "architecture inconsistency: avxvnni requires avx2"
+#endif
+
     namespace types
     {
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(avxvnni, avx2);
     }
 #endif
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_fma3_avx2_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_fma3_avx2_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_fma3_avx2_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_fma3_avx2_register.hpp
@@ -31,10 +31,15 @@
         static constexpr bool available() noexcept { return true; }
         static constexpr char const* name() noexcept { return "fma3+avx2"; }
     };
 
 #if XSIMD_WITH_FMA3_AVX2
+
+#if !XSIMD_WITH_AVX2
+#error "architecture inconsistency: fma3+avx2 requires avx2"
+#endif
+
     namespace types
     {
 
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(fma3<avx2>, avx2);
 
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_fma3_avx_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_fma3_avx_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_fma3_avx_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_fma3_avx_register.hpp
@@ -31,10 +31,15 @@
         static constexpr bool available() noexcept { return true; }
         static constexpr char const* name() noexcept { return "fma3+avx"; }
     };
 
 #if XSIMD_WITH_FMA3_AVX
+
+#if !XSIMD_WITH_AVX
+#error "architecture inconsistency: fma3+avx requires avx"
+#endif
+
     namespace types
     {
 
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(fma3<avx>, avx);
 
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_fma3_sse_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_fma3_sse_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_fma3_sse_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_fma3_sse_register.hpp
@@ -31,10 +31,15 @@
         static constexpr bool available() noexcept { return true; }
         static constexpr char const* name() noexcept { return "fma3+sse4.2"; }
     };
 
 #if XSIMD_WITH_FMA3_SSE
+
+#if !XSIMD_WITH_SSE4_2
+#error "architecture inconsistency: fma3+sse4.2 requires sse4.2"
+#endif
+
     namespace types
     {
 
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(fma3<sse4_2>, sse4_2);
 
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_fma4_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_fma4_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_fma4_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_fma4_register.hpp
@@ -31,10 +31,15 @@
         static constexpr bool available() noexcept { return true; }
         static constexpr char const* name() noexcept { return "fma4"; }
     };
 
 #if XSIMD_WITH_FMA4
+
+#if !XSIMD_WITH_SSE4_2
+#error "architecture inconsistency: fma4 requires sse4.2"
+#endif
+
     namespace types
     {
 
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(fma4, sse4_2);
 
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_i8mm_neon64_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_i8mm_neon64_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_i8mm_neon64_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_i8mm_neon64_register.hpp
@@ -31,10 +31,15 @@
         static constexpr bool available() noexcept { return true; }
         static constexpr char const* name() noexcept { return "i8mm+neon64"; }
     };
 
 #if XSIMD_WITH_I8MM_NEON64
+
+#if !XSIMD_WITH_NEON64
+#error "architecture inconsistency: i8mm+neon64 requires neon64"
+#endif
+
     namespace types
     {
 
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(i8mm<neon64>, neon64);
 
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_neon64_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_neon64_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_neon64_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_neon64_register.hpp
@@ -30,10 +30,14 @@
         static constexpr char const* name() noexcept { return "arm64+neon"; }
     };
 
 #if XSIMD_WITH_NEON64
 
+#if !XSIMD_WITH_NEON
+#error "architecture inconsistency: neon64 requires neon"
+#endif
+
     namespace types
     {
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(neon64, neon);
         XSIMD_DECLARE_SIMD_REGISTER(double, neon64, float64x2_t);
 
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_sse3_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_sse3_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_sse3_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_sse3_register.hpp
@@ -31,10 +31,15 @@
         static constexpr bool available() noexcept { return true; }
         static constexpr char const* name() noexcept { return "sse3"; }
     };
 
 #if XSIMD_WITH_SSE3
+
+#if !XSIMD_WITH_SSE2
+#error "architecture inconsistency: sse3 requires sse2"
+#endif
+
     namespace types
     {
 
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(sse3, sse2);
     }
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_sse4_1_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_sse4_1_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_sse4_1_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_sse4_1_register.hpp
@@ -31,10 +31,15 @@
         static constexpr bool available() noexcept { return true; }
         static constexpr char const* name() noexcept { return "sse4.1"; }
     };
 
 #if XSIMD_WITH_SSE4_1
+
+#if !XSIMD_WITH_SSSE3
+#error "architecture inconsistency: sse4.1 requires ssse3"
+#endif
+
     namespace types
     {
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(sse4_1, ssse3);
     }
 #endif
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_sse4_2_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_sse4_2_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_sse4_2_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_sse4_2_register.hpp
@@ -31,10 +31,15 @@
         static constexpr bool available() noexcept { return true; }
         static constexpr char const* name() noexcept { return "sse4.2"; }
     };
 
 #if XSIMD_WITH_SSE4_2
+
+#if !XSIMD_WITH_SSE4_1
+#error "architecture inconsistency: sse4.2 requires sse4.1"
+#endif
+
     namespace types
     {
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(sse4_2, sse4_1);
     }
 #endif
diff --git a/third_party/xsimd/include/xsimd/types/xsimd_ssse3_register.hpp b/third_party/xsimd/include/xsimd/types/xsimd_ssse3_register.hpp
--- a/third_party/xsimd/include/xsimd/types/xsimd_ssse3_register.hpp
+++ b/third_party/xsimd/include/xsimd/types/xsimd_ssse3_register.hpp
@@ -31,10 +31,15 @@
         static constexpr bool available() noexcept { return true; }
         static constexpr char const* name() noexcept { return "ssse3"; }
     };
 
 #if XSIMD_WITH_SSSE3
+
+#if !XSIMD_WITH_SSE3
+#error "architecture inconsistency: ssse3 requires sse3"
+#endif
+
     namespace types
     {
         XSIMD_DECLARE_SIMD_REGISTER_ALIAS(ssse3, sse3);
     }
 #endif
diff --git a/third_party/xsimd/moz.yaml b/third_party/xsimd/moz.yaml
--- a/third_party/xsimd/moz.yaml
+++ b/third_party/xsimd/moz.yaml
@@ -8,12 +8,12 @@
   name: xsimd
   description: C++ wrappers for SIMD intrinsics
 
   url: https://github.com/QuantStack/xsimd
 
-  release: 297dd76efd0978b3152e5bbd7df9d52680905c4b (2025-01-02T09:12:22Z).
-  revision: 297dd76efd0978b3152e5bbd7df9d52680905c4b
+  release: 6d6067a049987803ae07dad62f66f603773324b9 (2025-01-26T21:05:43Z).
+  revision: 6d6067a049987803ae07dad62f66f603773324b9
 
   license: BSD-3-Clause
 
 vendoring:
   url: https://github.com/QuantStack/xsimd