137         const float16x4_t va7 = vld1_f16(a7); a7 += 4;  in xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64()  local
149           vacc7x01234567 = vfmaq_lane_f16(vacc7x01234567, vb01234567c0, va7, 0);  in xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64()
158           const float16x8_t va7c0 = vdupq_lane_f16(va7, 0);  in xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64()
179           vacc7x01234567 = vfmaq_lane_f16(vacc7x01234567, vb01234567c1, va7, 1);  in xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64()
188           const float16x8_t va7c1 = vdupq_lane_f16(va7, 1);  in xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64()
209           vacc7x01234567 = vfmaq_lane_f16(vacc7x01234567, vb01234567c2, va7, 2);  in xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64()
218           const float16x8_t va7c2 = vdupq_lane_f16(va7, 2);  in xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64()
239           vacc7x01234567 = vfmaq_lane_f16(vacc7x01234567, vb01234567c3, va7, 3);  in xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64()
248           const float16x8_t va7c3 = vdupq_lane_f16(va7, 3);  in xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64()
269           const float16x8_t va7 = vld1q_dup_f16(a7); a7 += 1;  in xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64()  local
280           vacc7x01234567 = vfmaq_f16(vacc7x01234567, va7, vb01234567);  in xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64()