134         const int16x8_t vxa0 = vreinterpretq_s16_u16(vmovl_u8(va0));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
135         const int16x8_t vxa1 = vreinterpretq_s16_u16(vmovl_u8(va1));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
136         const int16x8_t vxa2 = vreinterpretq_s16_u16(vmovl_u8(va2));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
137         const int16x8_t vxa3 = vreinterpretq_s16_u16(vmovl_u8(va3));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
138         const int16x8_t vxa4 = vreinterpretq_s16_u16(vmovl_u8(va4));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
139         const int16x8_t vxa5 = vreinterpretq_s16_u16(vmovl_u8(va5));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
140         const int16x8_t vxa6 = vreinterpretq_s16_u16(vmovl_u8(va6));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
141         const int16x8_t vxa7 = vreinterpretq_s16_u16(vmovl_u8(va7));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
330         const int16x8_t vxa0 = vreinterpretq_s16_u16(vmovl_u8(va0));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
331         const int16x8_t vxa1 = vreinterpretq_s16_u16(vmovl_u8(va1));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
332         const int16x8_t vxa2 = vreinterpretq_s16_u16(vmovl_u8(va2));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
333         const int16x8_t vxa3 = vreinterpretq_s16_u16(vmovl_u8(va3));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
334         const int16x8_t vxa4 = vreinterpretq_s16_u16(vmovl_u8(va4));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
335         const int16x8_t vxa5 = vreinterpretq_s16_u16(vmovl_u8(va5));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
336         const int16x8_t vxa6 = vreinterpretq_s16_u16(vmovl_u8(va6));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()
337         const int16x8_t vxa7 = vreinterpretq_s16_u16(vmovl_u8(va7));  in xnn_qu8_igemm_minmax_ukernel_8x8__neon()