89       int32x4_t vacc0123 = vld1q_s32(w); w = (const void*) ((uintptr_t) w + 4 * sizeof(int32_t));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()  local
96       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi0x01234567), vget_low_s16(vk0x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
102       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi1x01234567), vget_low_s16(vk1x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
108       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi2x01234567), vget_low_s16(vk2x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
114       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi3x01234567), vget_low_s16(vk3x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
120       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi4x01234567), vget_low_s16(vk4x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
126       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
132       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi6x01234567), vget_low_s16(vk6x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
138       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi7x01234567), vget_low_s16(vk7x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
144       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi8x01234567), vget_low_s16(vk8x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
147       vacc0123 = vqrdmulhq_s32(vacc0123, vmultiplier);  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
150       vacc0123 = vsraq_n_s32(vacc0123, vbicq_s32(vacc0123, vzero_shift_mask), 31);  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
153       vacc0123 = vrshlq_s32(vacc0123, vright_shift);  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
157 …const int16x8_t vacc01234567 = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(vacc0123), vacc4567), voutput…  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
161 …const int16x8_t vacc01234567 = vqaddq_s16(vcombine_s16(vqmovn_s32(vacc0123), vqmovn_s32(vacc4567))…  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
174         int32x4_t vacc0123 = vld1q_s32(w); w = (const void*) ((uintptr_t) w + 4 * sizeof(int32_t));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()  local
180         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi0x01234567), vget_low_s16(vk0x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
185         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi1x01234567), vget_low_s16(vk1x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
190         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi2x01234567), vget_low_s16(vk2x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
195         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi3x01234567), vget_low_s16(vk3x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
200         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi4x01234567), vget_low_s16(vk4x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
205         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
210         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi6x01234567), vget_low_s16(vk6x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
215         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi7x01234567), vget_low_s16(vk7x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
220         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi8x01234567), vget_low_s16(vk8x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
223         vacc0123 = vqrdmulhq_s32(vacc0123, vmultiplier);  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
226         vacc0123 = vsraq_n_s32(vacc0123, vbicq_s32(vacc0123, vzero_shift_mask), 31);  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
229         vacc0123 = vrshlq_s32(vacc0123, vright_shift);  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
233 …const int16x8_t vacc01234567 = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(vacc0123), vacc4567), voutput…  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()
237 …const int16x8_t vacc01234567 = vqaddq_s16(vcombine_s16(vqmovn_s32(vacc0123), vqmovn_s32(vacc4567))…  in xnn_qs8_dwconv_minmax_ukernel_up8x9__neon_mul16()