103         int8x8_t va2x1 = vld1_s8(a2); a2 += 8;  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()  local
128         vprod2x01c0 = vmlal_s8(vprod2x01c0, vb01c0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
138         vprod2x23c0 = vmlal_s8(vprod2x23c0, vb23c0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
148         vprod2x45c0 = vmlal_s8(vprod2x45c0, vb45c0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
158         vprod2x67c0 = vmlal_s8(vprod2x67c0, vb67c0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
168         vprod2x89c0 = vmlal_s8(vprod2x89c0, vb89c0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
178         vprod2xABc0 = vmlal_s8(vprod2xABc0, vbABc0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
188         vprod2xCDc0 = vmlal_s8(vprod2xCDc0, vbCDc0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
198         vprod2xEFc0 = vmlal_s8(vprod2xEFc0, vbEFc0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
207         va2x1 = vext_s8(va2x1, va2x1, 4);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
214         vprod2x01c1 = vmlal_s8(vprod2x01c1, vb01c1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
224         vprod2x23c1 = vmlal_s8(vprod2x23c1, vb23c1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
234         vprod2x45c1 = vmlal_s8(vprod2x45c1, vb45c1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
244         vprod2x67c1 = vmlal_s8(vprod2x67c1, vb67c1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
254         vprod2x89c1 = vmlal_s8(vprod2x89c1, vb89c1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
264         vprod2xABc1 = vmlal_s8(vprod2xABc1, vbABc1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
274         vprod2xCDc1 = vmlal_s8(vprod2xCDc1, vbCDc1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()
284         vprod2xEFc1 = vmlal_s8(vprod2xEFc1, vbEFc1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c4s2__neon_mlal()