89       int32x4_t vacc0123 = vld1q_s32(w); w = (const void*) ((uintptr_t) w + 4 * sizeof(int32_t));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()  local
100       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi0x01234567), vget_low_s16(vk0x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
110       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi1x01234567), vget_low_s16(vk1x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
120       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi2x01234567), vget_low_s16(vk2x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
130       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi3x01234567), vget_low_s16(vk3x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
140       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi4x01234567), vget_low_s16(vk4x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
150       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
160       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi6x01234567), vget_low_s16(vk6x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
170       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi7x01234567), vget_low_s16(vk7x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
180       vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi8x01234567), vget_low_s16(vk8x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
185       vacc0123 = vqrdmulhq_s32(vacc0123, vmultiplier);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
190       vacc0123 = vsraq_n_s32(vacc0123, vbicq_s32(vacc0123, vzero_shift_mask), 31);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
195       vacc0123 = vrshlq_s32(vacc0123, vright_shift);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
201 …const int16x8_t vacc01234567 = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(vacc0123), vacc4567), voutput…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
206 …const int16x8_t vacc01234567 = vqaddq_s16(vcombine_s16(vqmovn_s32(vacc0123), vqmovn_s32(vacc4567))…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
221         int32x4_t vacc0123 = vld1q_s32(w); w = (const void*) ((uintptr_t) w + 4 * sizeof(int32_t));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()  local
227         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi0x01234567), vget_low_s16(vk0x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
232         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi1x01234567), vget_low_s16(vk1x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
237         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi2x01234567), vget_low_s16(vk2x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
242         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi3x01234567), vget_low_s16(vk3x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
247         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi4x01234567), vget_low_s16(vk4x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
252         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi5x01234567), vget_low_s16(vk5x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
257         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi6x01234567), vget_low_s16(vk6x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
262         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi7x01234567), vget_low_s16(vk7x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
267         vacc0123 = vmlal_s16(vacc0123, vget_low_s16(vi8x01234567), vget_low_s16(vk8x01234567));  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
270         vacc0123 = vqrdmulhq_s32(vacc0123, vmultiplier);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
273         vacc0123 = vsraq_n_s32(vacc0123, vbicq_s32(vacc0123, vzero_shift_mask), 31);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
276         vacc0123 = vrshlq_s32(vacc0123, vright_shift);  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
280 …const int16x8_t vacc01234567 = vqaddq_s16(vqmovn_high_s32(vqmovn_s32(vacc0123), vacc4567), voutput…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()
284 …const int16x8_t vacc01234567 = vqaddq_s16(vcombine_s16(vqmovn_s32(vacc0123), vqmovn_s32(vacc4567))…  in xnn_qs8_dwconv_minmax_ukernel_up16x9__neon_mul16()