174     const int32x4_t vsum0x01 = vpaddq_s32(vacc0x0, vacc0x1);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
175     const int32x4_t vsum0x23 = vpaddq_s32(vacc0x2, vacc0x3);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
176     const int32x4_t vsum0x45 = vpaddq_s32(vacc0x4, vacc0x5);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
177     const int32x4_t vsum0x67 = vpaddq_s32(vacc0x6, vacc0x7);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
178     const int32x4_t vsum1x01 = vpaddq_s32(vacc1x0, vacc1x1);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
179     const int32x4_t vsum1x23 = vpaddq_s32(vacc1x2, vacc1x3);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
180     const int32x4_t vsum1x45 = vpaddq_s32(vacc1x4, vacc1x5);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
181     const int32x4_t vsum1x67 = vpaddq_s32(vacc1x6, vacc1x7);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
182     const int32x4_t vsum2x01 = vpaddq_s32(vacc2x0, vacc2x1);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
183     const int32x4_t vsum2x23 = vpaddq_s32(vacc2x2, vacc2x3);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
184     const int32x4_t vsum2x45 = vpaddq_s32(vacc2x4, vacc2x5);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
185     const int32x4_t vsum2x67 = vpaddq_s32(vacc2x6, vacc2x7);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
186     int32x4_t vacc0x0123 = vpaddq_s32(vsum0x01, vsum0x23);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
187     int32x4_t vacc0x4567 = vpaddq_s32(vsum0x45, vsum0x67);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
188     int32x4_t vacc1x0123 = vpaddq_s32(vsum1x01, vsum1x23);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
189     int32x4_t vacc1x4567 = vpaddq_s32(vsum1x45, vsum1x67);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
190     int32x4_t vacc2x0123 = vpaddq_s32(vsum2x01, vsum2x23);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()
191     int32x4_t vacc2x4567 = vpaddq_s32(vsum2x45, vsum2x67);  in xnn_qs8_gemm_minmax_ukernel_3x8c16__neon_mlal_padal()