39     const int16x8_t vxa0 = vreinterpretq_s16_u16(vsubl_u8(vget_low_u8(va01), va_zero_point));  in xnn_q8_vadd_ukernel__neon()
40     const int16x8_t vxb0 = vreinterpretq_s16_u16(vsubl_u8(vget_low_u8(vb01), vb_zero_point));  in xnn_q8_vadd_ukernel__neon()
41     const int16x8_t vxa1 = vreinterpretq_s16_u16(vsubl_u8(vget_high_u8(va01), va_zero_point));  in xnn_q8_vadd_ukernel__neon()
42     const int16x8_t vxb1 = vreinterpretq_s16_u16(vsubl_u8(vget_high_u8(vb01), vb_zero_point));  in xnn_q8_vadd_ukernel__neon()
43     const int16x8_t vxa2 = vreinterpretq_s16_u16(vsubl_u8(vget_low_u8(va23), va_zero_point));  in xnn_q8_vadd_ukernel__neon()
44     const int16x8_t vxb2 = vreinterpretq_s16_u16(vsubl_u8(vget_low_u8(vb23), vb_zero_point));  in xnn_q8_vadd_ukernel__neon()
45     const int16x8_t vxa3 = vreinterpretq_s16_u16(vsubl_u8(vget_high_u8(va23), va_zero_point));  in xnn_q8_vadd_ukernel__neon()
46     const int16x8_t vxb3 = vreinterpretq_s16_u16(vsubl_u8(vget_high_u8(vb23), vb_zero_point));  in xnn_q8_vadd_ukernel__neon()
109     const int16x8_t vxa0 = vreinterpretq_s16_u16(vsubl_u8(vget_low_u8(va01), va_zero_point));  in xnn_q8_vadd_ukernel__neon()
110     const int16x8_t vxb0 = vreinterpretq_s16_u16(vsubl_u8(vget_low_u8(vb01), vb_zero_point));  in xnn_q8_vadd_ukernel__neon()
111     const int16x8_t vxa1 = vreinterpretq_s16_u16(vsubl_u8(vget_high_u8(va01), va_zero_point));  in xnn_q8_vadd_ukernel__neon()
112     const int16x8_t vxb1 = vreinterpretq_s16_u16(vsubl_u8(vget_high_u8(vb01), vb_zero_point));  in xnn_q8_vadd_ukernel__neon()
155     const int16x8_t vxa = vreinterpretq_s16_u16(vsubl_u8(va, va_zero_point));  in xnn_q8_vadd_ukernel__neon()
156     const int16x8_t vxb = vreinterpretq_s16_u16(vsubl_u8(vb, vb_zero_point));  in xnn_q8_vadd_ukernel__neon()
198     const int16x8_t vxa = vreinterpretq_s16_u16(vsubl_u8(va, va_zero_point));  in xnn_q8_vadd_ukernel__neon()
199     const int16x8_t vxb = vreinterpretq_s16_u16(vsubl_u8(vb, vb_zero_point));  in xnn_q8_vadd_ukernel__neon()