1273   const uint16x8_t L0 = vmovl_u8(vld1_u8(dst + 0 * BPS - 1));  in DC4()
1274   const uint16x8_t L1 = vmovl_u8(vld1_u8(dst + 1 * BPS - 1));  in DC4()
1275   const uint16x8_t L2 = vmovl_u8(vld1_u8(dst + 2 * BPS - 1));  in DC4()
1276   const uint16x8_t L3 = vmovl_u8(vld1_u8(dst + 3 * BPS - 1));  in DC4()
1277   const uint16x8_t s0 = vaddq_u16(L0, L1);  in DC4()
1278   const uint16x8_t s1 = vaddq_u16(L2, L3);  in DC4()
1279   const uint16x8_t s01 = vaddq_u16(s0, s1);  in DC4()
1280   const uint16x8_t sum = vaddq_u16(s01, vcombine_u16(p1, p1));  in DC4()
1411   uint16x8_t sum_top;  in DC8()
1412   uint16x8_t sum_left;  in DC8()
1424     const uint16x8_t L0 = vmovl_u8(vld1_u8(dst + 0 * BPS - 1));  in DC8()
1425     const uint16x8_t L1 = vmovl_u8(vld1_u8(dst + 1 * BPS - 1));  in DC8()
1426     const uint16x8_t L2 = vmovl_u8(vld1_u8(dst + 2 * BPS - 1));  in DC8()
1427     const uint16x8_t L3 = vmovl_u8(vld1_u8(dst + 3 * BPS - 1));  in DC8()
1428     const uint16x8_t L4 = vmovl_u8(vld1_u8(dst + 4 * BPS - 1));  in DC8()
1429     const uint16x8_t L5 = vmovl_u8(vld1_u8(dst + 5 * BPS - 1));  in DC8()
1430     const uint16x8_t L6 = vmovl_u8(vld1_u8(dst + 6 * BPS - 1));  in DC8()
1431     const uint16x8_t L7 = vmovl_u8(vld1_u8(dst + 7 * BPS - 1));  in DC8()
1432     const uint16x8_t s0 = vaddq_u16(L0, L1);  in DC8()
1433     const uint16x8_t s1 = vaddq_u16(L2, L3);  in DC8()
1434     const uint16x8_t s2 = vaddq_u16(L4, L5);  in DC8()
1435     const uint16x8_t s3 = vaddq_u16(L6, L7);  in DC8()
1436     const uint16x8_t s01 = vaddq_u16(s0, s1);  in DC8()
1437     const uint16x8_t s23 = vaddq_u16(s2, s3);  in DC8()
1442     const uint16x8_t sum = vaddq_u16(sum_left, sum_top);  in DC8()
1489   uint16x8_t sum_top;  in DC16()
1490   uint16x8_t sum_left;  in DC16()
1495     const uint16x8_t p0 = vpaddlq_u8(A);  // cascading summation of the top  in DC16()
1506       const uint16x8_t L0 = vmovl_u8(vld1_u8(dst + (i + 0) * BPS - 1));  in DC16()
1507       const uint16x8_t L1 = vmovl_u8(vld1_u8(dst + (i + 1) * BPS - 1));  in DC16()
1508       const uint16x8_t L2 = vmovl_u8(vld1_u8(dst + (i + 2) * BPS - 1));  in DC16()
1509       const uint16x8_t L3 = vmovl_u8(vld1_u8(dst + (i + 3) * BPS - 1));  in DC16()
1510       const uint16x8_t L4 = vmovl_u8(vld1_u8(dst + (i + 4) * BPS - 1));  in DC16()
1511       const uint16x8_t L5 = vmovl_u8(vld1_u8(dst + (i + 5) * BPS - 1));  in DC16()
1512       const uint16x8_t L6 = vmovl_u8(vld1_u8(dst + (i + 6) * BPS - 1));  in DC16()
1513       const uint16x8_t L7 = vmovl_u8(vld1_u8(dst + (i + 7) * BPS - 1));  in DC16()
1514       const uint16x8_t s0 = vaddq_u16(L0, L1);  in DC16()
1515       const uint16x8_t s1 = vaddq_u16(L2, L3);  in DC16()
1516       const uint16x8_t s2 = vaddq_u16(L4, L5);  in DC16()
1517       const uint16x8_t s3 = vaddq_u16(L6, L7);  in DC16()
1518       const uint16x8_t s01 = vaddq_u16(s0, s1);  in DC16()
1519       const uint16x8_t s23 = vaddq_u16(s2, s3);  in DC16()
1520       const uint16x8_t sum = vaddq_u16(s01, s23);  in DC16()
1526     const uint16x8_t sum = vaddq_u16(sum_left, sum_top);  in DC16()