Improve generic version of complex masked load

Use the usual kernel mechanism which allows for specialization.
Implement specialization for avx.

Follow-up to #1391

Author: serge-sans-paille

include/xsimd/arch/common/xsimd_common_memory.hpp

@@ -450,6 +450,29 @@ namespace xsimd
             return batch<T, A>::load_aligned(buffer.data());
         }
 
+        template <class A, class T, class Mode>
+        XSIMD_INLINE batch<std::complex<T>, A>
+        load_complex_masked(std::complex<T> const* mem, batch_bool<T, A> mask, Mode, requires_arch<common>) noexcept
+        {
+            // Scalar fallback: only active lanes are touched. Arches with
+            // hardware predicated loads should override this.
+            constexpr std::size_t size = batch<T, A>::size;
+            alignas(A::alignment()) std::array<T, size> buffer_real;
+            alignas(A::alignment()) std::array<T, size> buffer_imag;
+            for (std::size_t i = 0; i < size; ++i)
+                if (mask.get(i))
+                {
+                    buffer_real[i] = mem[i].real();
+                    buffer_imag[i] = mem[i].imag();
+                }
+                else
+                {
+                    buffer_real[i] = T(0);
+                    buffer_imag[i] = T(0);
+                }
+            return batch<std::complex<T>, A>::load_aligned(buffer_real.data(), buffer_imag.data());
+        }
+
         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 mode, requires_arch<common>) noexcept

include/xsimd/arch/xsimd_avx.hpp

@@ -1025,6 +1025,19 @@ namespace xsimd
             return _mm256_maskload_pd(mem, _mm256_castpd_si256(mask));
         }
 
+        template <class A, class T, class Mode>
+        XSIMD_INLINE batch<std::complex<T>, A>
+        load_complex_masked(std::complex<T> const* mem, batch_bool<T, A> mask, Mode mode, requires_arch<avx>) noexcept
+        {
+            using mask_register_type = typename batch_bool<T, A>::register_type;
+            mask_register_type nmask = mask.to_native();
+            batch_bool<T, A> lo_mask = zip_lo(batch<T, A>(nmask), batch<T, A>(nmask)).to_native();
+            batch_bool<T, A> hi_mask = zip_hi(batch<T, A>(nmask), batch<T, A>(nmask)).to_native();
+            batch<T, A> res_lo = batch<T, A>::load(reinterpret_cast<T const*>(mem), lo_mask, mode);
+            batch<T, A> res_hi = batch<T, A>::load(reinterpret_cast<T const*>(mem) + mask.size, hi_mask, mode);
+            return detail::load_complex(res_lo, res_hi, A{});
+        }
+
         // 4/8-byte ints: bitcast to same-width float, reuse the vmaskmov path.
         template <class A, class T, class Mode>
         XSIMD_INLINE std::enable_if_t<std::is_integral_v<T> && (sizeof(T) == 4 || sizeof(T) == 8), batch<T, A>>

include/xsimd/arch/xsimd_avx512f.hpp

@@ -1655,6 +1655,27 @@ namespace xsimd
             }
         }
 
+        template <class A, class T, class Mode>
+        XSIMD_INLINE batch<std::complex<T>, A>
+        load_complex_masked(std::complex<T> const* mem, batch_bool<T, A> mask, Mode mode, requires_arch<avx512f>) noexcept
+        {
+            using mask_register_type = typename batch_bool<T, A>::register_type;
+            mask_register_type nmask = mask.to_native();
+
+            // manually zip mask
+            constexpr mask_register_type lo_bitmask = xsimd::utils::make_low_mask<mask_register_type>(mask.size / 2);
+            mask_register_type lo_mask = nmask & lo_bitmask;
+            lo_mask |= lo_mask << (mask.size / 2);
+
+            constexpr mask_register_type hi_bitmask = lo_bitmask << (mask.size / 2);
+            mask_register_type hi_mask = nmask & hi_bitmask;
+            hi_mask |= hi_mask >> (mask.size / 2);
+
+            batch<T, A> res_lo = batch<T, A>::load(reinterpret_cast<T const*>(mem), lo_mask, mode);
+            batch<T, A> res_hi = batch<T, A>::load(reinterpret_cast<T const*>(mem) + mask.size, hi_mask, mode);
+            return detail::load_complex(res_lo, res_hi, A{});
+        }
+
         // load_unaligned
         template <class A, class T, class = std::enable_if_t<std::is_integral_v<T>>>
         XSIMD_INLINE batch<T, A> load_unaligned(T const* mem, convert<T>, requires_arch<avx512f>) noexcept

include/xsimd/types/xsimd_batch.hpp

@@ -1557,12 +1557,9 @@ namespace xsimd
 
     template <class T, class A>
     template <class Mode>
-    XSIMD_INLINE batch<std::complex<T>, A> batch<std::complex<T>, A>::load(value_type const* mem, batch_bool<T, A> mask, Mode) noexcept
+    XSIMD_INLINE batch<std::complex<T>, A> batch<std::complex<T>, A>::load(value_type const* mem, batch_bool<T, A> mask, Mode mode) noexcept
     {
-        alignas(A::alignment()) std::array<value_type, size> buffer {};
-        for (std::size_t i = 0; i < size; ++i)
-            buffer[i] = mask.get(i) ? mem[i] : value_type(0);
-        return load_aligned(buffer.data());
+        return kernel::load_complex_masked<A>(mem, mask, mode, A {});
     }
 
     template <class T, class A>