WIP

Author: serge-sans-paille

include/xsimd/arch/xsimd_neon.hpp

@@ -2953,6 +2953,33 @@ namespace xsimd
                 }
                 return vdupq_n_u32(vgetq_lane_u32(self, V0));
             }
+            XSIMD_IF_CONSTEXPR(V0 < 2 && V1 < 2 && V2 < 2 && V3 < 2)
+            {
+                uint8x8_t low = vreinterpret_u8_u64(vget_low_u64(vreinterpretq_u64_u32(self)));
+                uint8x8_t mask_lo = {
+                    V0 * 4 + 0,
+                    V0 * 4 + 1,
+                    V0 * 4 + 2,
+                    V0 * 4 + 3,
+                    V1 * 4 + 0,
+                    V1 * 4 + 1,
+                    V1 * 4 + 2,
+                    V1 * 4 + 3,
+                };
+                uint8x8_t mask_hi = {
+                    V2 * 4 + 0,
+                    V2 * 4 + 1,
+                    V2 * 4 + 2,
+                    V2 * 4 + 3,
+                    V3 * 4 + 0,
+                    V3 * 4 + 1,
+                    V3 * 4 + 2,
+                    V3 * 4 + 3,
+                };
+                uint8x8_t lo = vtbl1_u8(low, mask_lo);
+                uint8x0_t hi = vtbl1_u8(low, mask_hi);
+                return vreinterpretq_u32_u8(vcombine_u8(lo, hi));
+            }
             std::array<uint32_t, 4> data;
             self.store_aligned(data.data());
             return set(batch<uint32_t, A>(), A(), data[V0], data[V1], data[V2], data[V3]);