117         int8x8_t va3x0 = vld1_s8(a3); a3 += 8;  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()  local
139         int16x8_t vprod3x01c0 = vmull_s8(vb01c0x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
147         int16x8_t vprod3x23c0 = vmull_s8(vb23c0x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
155         int16x8_t vprod3x45c0 = vmull_s8(vb45c0x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
163         int16x8_t vprod3x67c0 = vmull_s8(vb67c0x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
171         int16x8_t vprod3x89c0 = vmull_s8(vb89c0x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
179         int16x8_t vprod3xABc0 = vmull_s8(vbABc0x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
187         int16x8_t vprod3xCDc0 = vmull_s8(vbCDc0x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
195         int16x8_t vprod3xEFc0 = vmull_s8(vbEFc0x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
203         va3x0 = vext_s8(va3x0, va3x0, 4);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
207         int16x8_t vprod3x01c1 = vmull_s8(vb01c1x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
215         int16x8_t vprod3x23c1 = vmull_s8(vb23c1x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
223         int16x8_t vprod3x45c1 = vmull_s8(vb45c1x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
231         int16x8_t vprod3x67c1 = vmull_s8(vb67c1x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
239         int16x8_t vprod3x89c1 = vmull_s8(vb89c1x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
247         int16x8_t vprod3xABc1 = vmull_s8(vbABc1x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
255         int16x8_t vprod3xCDc1 = vmull_s8(vbCDc1x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()
263         int16x8_t vprod3xEFc1 = vmull_s8(vbEFc1x0, va3x0);  in xnn_qs8_igemm_minmax_rndnu_ukernel_4x16c4s2__neon_mull()