91         int8x8_t va2x1 = vld1_s8(a2); a2 += 8;  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()  local
116         vprod2x0123c0 = vmlal_s8(vprod2x0123c0, vb0123c0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
126         vprod2x4567c0 = vmlal_s8(vprod2x4567c0, vb4567c0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
136         vprod2x89ABc0 = vmlal_s8(vprod2x89ABc0, vb89ABc0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
146         vprod2xCDEFc0 = vmlal_s8(vprod2xCDEFc0, vbCDEFc0x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
155         va2x1 = vext_s8(va2x1, va2x1, 2);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
162         vprod2x0123c1 = vmlal_s8(vprod2x0123c1, vb0123c1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
172         vprod2x4567c1 = vmlal_s8(vprod2x4567c1, vb4567c1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
182         vprod2x89ABc1 = vmlal_s8(vprod2x89ABc1, vb89ABc1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
192         vprod2xCDEFc1 = vmlal_s8(vprod2xCDEFc1, vbCDEFc1x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
201         va2x1 = vext_s8(va2x1, va2x1, 2);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
208         vprod2x0123c2 = vmlal_s8(vprod2x0123c2, vb0123c2x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
218         vprod2x4567c2 = vmlal_s8(vprod2x4567c2, vb4567c2x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
228         vprod2x89ABc2 = vmlal_s8(vprod2x89ABc2, vb89ABc2x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
238         vprod2xCDEFc2 = vmlal_s8(vprod2xCDEFc2, vbCDEFc2x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
247         va2x1 = vext_s8(va2x1, va2x1, 2);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
254         vprod2x0123c3 = vmlal_s8(vprod2x0123c3, vb0123c3x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
264         vprod2x4567c3 = vmlal_s8(vprod2x4567c3, vb4567c3x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
274         vprod2x89ABc3 = vmlal_s8(vprod2x89ABc3, vb89ABc3x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()
284         vprod2xCDEFc3 = vmlal_s8(vprod2xCDEFc3, vbCDEFc3x1, va2x1);  in xnn_qs8_igemm_minmax_rndnu_ukernel_3x16c2s4__neon_mlal()