86       float32x4_t vacc0123p0 = vld1q_f32(w); w += 4;  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()  local
91       vacc0123p0 = vmlaq_f32(vacc0123p0, vi0x0123, vk0x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
95       vacc0123p0 = vmlaq_f32(vacc0123p0, vi1x0123, vk1x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
99       vacc0123p0 = vmlaq_f32(vacc0123p0, vi2x0123, vk2x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
103       vacc0123p0 = vmlaq_f32(vacc0123p0, vi3x0123, vk3x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
107       vacc0123p0 = vmlaq_f32(vacc0123p0, vi4x0123, vk4x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
111       vacc0123p0 = vmlaq_f32(vacc0123p0, vi5x0123, vk5x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
115       vacc0123p0 = vmlaq_f32(vacc0123p0, vi6x0123, vk6x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
119       vacc0123p0 = vmlaq_f32(vacc0123p0, vi7x0123, vk7x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
123       vacc0123p0 = vmlaq_f32(vacc0123p0, vi8x0123, vk8x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
126       float32x4_t vacc0123 = vmaxq_f32(vacc0123p0, vmin);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
132       float32x4_t vacc0123p0 = vld1q_f32(w); w += 4;  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()  local
137       vacc0123p0 = vmlaq_f32(vacc0123p0, vi0x0123, vk0x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
141       vacc0123p0 = vmlaq_f32(vacc0123p0, vi1x0123, vk1x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
145       vacc0123p0 = vmlaq_f32(vacc0123p0, vi2x0123, vk2x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
149       vacc0123p0 = vmlaq_f32(vacc0123p0, vi3x0123, vk3x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
153       vacc0123p0 = vmlaq_f32(vacc0123p0, vi4x0123, vk4x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
157       vacc0123p0 = vmlaq_f32(vacc0123p0, vi5x0123, vk5x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
161       vacc0123p0 = vmlaq_f32(vacc0123p0, vi6x0123, vk6x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
165       vacc0123p0 = vmlaq_f32(vacc0123p0, vi7x0123, vk7x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
169       vacc0123p0 = vmlaq_f32(vacc0123p0, vi8x0123, vk8x0123);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()
172       float32x4_t vacc0123 = vmaxq_f32(vacc0123p0, vmin);  in xnn_f32_dwconv_minmax_ukernel_up4x9__neon()