64       const float32x4_t vi1x0123 = vld1q_f32(i1); i1 += 4;  in xnn_f32_prelu_ukernel__neon_4x8()  local
75       float32x4_t vacc1x0123 = vmulq_f32(vi1x0123, vw0123);  in xnn_f32_prelu_ukernel__neon_4x8()
76       const uint32x4_t vm1x0123 = vcltq_s32(vreinterpretq_s32_f32(vi1x0123), vmovq_n_s32(0));  in xnn_f32_prelu_ukernel__neon_4x8()
90       vacc1x0123 = vbslq_f32(vm1x0123, vacc1x0123, vi1x0123);  in xnn_f32_prelu_ukernel__neon_4x8()
111       const float32x4_t vi1x0123 = vld1q_f32(i1);  in xnn_f32_prelu_ukernel__neon_4x8()  local
120       float32x4_t vacc1x0123 = vmulq_f32(vi1x0123, vw0123);  in xnn_f32_prelu_ukernel__neon_4x8()
121       const uint32x4_t vm1x0123 = vcltq_s32(vreinterpretq_s32_f32(vi1x0123), vmovq_n_s32(0));  in xnn_f32_prelu_ukernel__neon_4x8()
128       vacc1x0123 = vbslq_f32(vm1x0123, vacc1x0123, vi1x0123);  in xnn_f32_prelu_ukernel__neon_4x8()
142       const float32x4_t vi1x0123 = vld1q_f32(i1);  in xnn_f32_prelu_ukernel__neon_4x8()  local
151       float32x4_t vacc1x0123 = vmulq_f32(vi1x0123, vw0123);  in xnn_f32_prelu_ukernel__neon_4x8()
152       const uint32x4_t vm1x0123 = vcltq_s32(vreinterpretq_s32_f32(vi1x0123), vmovq_n_s32(0));  in xnn_f32_prelu_ukernel__neon_4x8()
159       vacc1x0123 = vbslq_f32(vm1x0123, vacc1x0123, vi1x0123);  in xnn_f32_prelu_ukernel__neon_4x8()