102       const int16x8_t vxa0 = vreinterpretq_s16_u16(vmovl_u8(va0)); a0 += 8;  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
104       const int16x8_t vxa1 = vreinterpretq_s16_u16(vmovl_u8(va1)); a1 += 8;  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
106       const int16x8_t vxa2 = vreinterpretq_s16_u16(vmovl_u8(va2)); a2 += 8;  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
108       const int16x8_t vxa3 = vreinterpretq_s16_u16(vmovl_u8(va3)); a3 += 8;  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
110       const int16x8_t vxa4 = vreinterpretq_s16_u16(vmovl_u8(va4)); a4 += 8;  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
112       const int16x8_t vxa5 = vreinterpretq_s16_u16(vmovl_u8(va5)); a5 += 8;  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
114       const int16x8_t vxa6 = vreinterpretq_s16_u16(vmovl_u8(va6)); a6 += 8;  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
116       const int16x8_t vxa7 = vreinterpretq_s16_u16(vmovl_u8(va7)); a7 += 8;  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
282       const int16x8_t vxa0 = vreinterpretq_s16_u16(vmovl_u8(va0));  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
284       const int16x8_t vxa1 = vreinterpretq_s16_u16(vmovl_u8(va1));  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
286       const int16x8_t vxa2 = vreinterpretq_s16_u16(vmovl_u8(va2));  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
288       const int16x8_t vxa3 = vreinterpretq_s16_u16(vmovl_u8(va3));  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
290       const int16x8_t vxa4 = vreinterpretq_s16_u16(vmovl_u8(va4));  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
292       const int16x8_t vxa5 = vreinterpretq_s16_u16(vmovl_u8(va5));  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
294       const int16x8_t vxa6 = vreinterpretq_s16_u16(vmovl_u8(va6));  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()
296       const int16x8_t vxa7 = vreinterpretq_s16_u16(vmovl_u8(va7));  in xnn_qu8_gemm_minmax_ukernel_8x8__neon()