fix style issues and fix the implementation of first() for 32-bit architectures in windows

Author: emrys53

include/xsimd/arch/xsimd_avx.hpp

@@ -1891,16 +1891,15 @@ namespace xsimd
             }
             else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)
             {
-                __m128i low = _mm256_castsi256_si128(self);
-                return static_cast<T>(_mm_cvtsi128_si64(low));
+                batch<T, sse4_2> low = _mm256_castsi256_si128(self);
+                return first(low, sse4_2 {});
             }
             else
             {
                 assert(false && "unsupported arch/op combination");
                 return {};
             }
         }
-
     }
 }
 

include/xsimd/arch/xsimd_avx512f.hpp

@@ -2369,7 +2369,8 @@ namespace xsimd
             }
             else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)
             {
-                return static_cast<T>(_mm_cvtsi128_si64(_mm512_castsi512_si128(self)));
+                batch<T, sse4_2> low = _mm512_castsi256_si128(self);
+                return first(low, sse4_2 {});
             }
             else
             {

include/xsimd/arch/xsimd_sse2.hpp

@@ -1813,7 +1813,15 @@ namespace xsimd
             }
             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
             {

include/xsimd/types/xsimd_batch.hpp

@@ -1100,7 +1100,6 @@ namespace xsimd
     {
     }
 
-
     template <class T, class A>
     template <class U, class... V, size_t I, size_t... Is>
     XSIMD_INLINE auto batch_bool<T, A>::make_register(detail::index_sequence<I, Is...>, U u, V... v) noexcept -> register_type