175     const int32x4_t vsum0x01 = vpaddq_s32(vacc0x0, vacc0x1);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
176     const int32x4_t vsum0x23 = vpaddq_s32(vacc0x2, vacc0x3);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
177     const int32x4_t vsum0x45 = vpaddq_s32(vacc0x4, vacc0x5);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
178     const int32x4_t vsum0x67 = vpaddq_s32(vacc0x6, vacc0x7);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
179     const int32x4_t vsum0x89 = vpaddq_s32(vacc0x8, vacc0x9);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
180     const int32x4_t vsum0xAB = vpaddq_s32(vacc0x10, vacc0x11);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
181     const int32x4_t vsum0xCD = vpaddq_s32(vacc0x12, vacc0x13);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
182     const int32x4_t vsum0xEF = vpaddq_s32(vacc0x14, vacc0x15);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
183     const int32x4_t vsum1x01 = vpaddq_s32(vacc1x0, vacc1x1);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
184     const int32x4_t vsum1x23 = vpaddq_s32(vacc1x2, vacc1x3);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
185     const int32x4_t vsum1x45 = vpaddq_s32(vacc1x4, vacc1x5);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
186     const int32x4_t vsum1x67 = vpaddq_s32(vacc1x6, vacc1x7);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
187     const int32x4_t vsum1x89 = vpaddq_s32(vacc1x8, vacc1x9);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
188     const int32x4_t vsum1xAB = vpaddq_s32(vacc1x10, vacc1x11);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
189     const int32x4_t vsum1xCD = vpaddq_s32(vacc1x12, vacc1x13);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
190     const int32x4_t vsum1xEF = vpaddq_s32(vacc1x14, vacc1x15);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
191     int32x4_t vacc0x0123 = vpaddq_s32(vsum0x01, vsum0x23);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
192     int32x4_t vacc0x4567 = vpaddq_s32(vsum0x45, vsum0x67);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
193     int32x4_t vacc0x89AB = vpaddq_s32(vsum0x89, vsum0xAB);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
194     int32x4_t vacc0xCDEF = vpaddq_s32(vsum0xCD, vsum0xEF);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
195     int32x4_t vacc1x0123 = vpaddq_s32(vsum1x01, vsum1x23);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
196     int32x4_t vacc1x4567 = vpaddq_s32(vsum1x45, vsum1x67);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
197     int32x4_t vacc1x89AB = vpaddq_s32(vsum1x89, vsum1xAB);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()
198     int32x4_t vacc1xCDEF = vpaddq_s32(vsum1xCD, vsum1xEF);  in xnn_qs8_gemm_minmax_ukernel_2x16c8__neon_mull_padal()