1  /* { dg-do assemble { target { aarch64*-*-* } } } */
       2  /* { dg-require-effective-target arm_v8_2a_bf16_neon_ok } */
       3  /* { dg-add-options arm_v8_2a_bf16_neon }  */
       4  /* { dg-additional-options "-O2  --save-temps" } */
       5  
       6  #include <arm_neon.h>
       7  
       8  void
       9  test_vst2_lane_bf16 (bfloat16_t *ptr, bfloat16x4x2_t b)
      10  {
      11    vst2_lane_bf16 (ptr, b, 2);
      12  }
      13  
      14  void
      15  test_vst2q_lane_bf16 (bfloat16_t *ptr, bfloat16x8x2_t b)
      16  {
      17    vst2q_lane_bf16 (ptr, b, 2);
      18  }
      19  
      20  /* { dg-final { scan-assembler-times "st2\\t{v\[0-9\]+.h - v\[0-9\]+.h}\\\[2\\\], \\\[x\[0-9\]+\\\]" 2 } } */
      21  
      22  void
      23  test_vst3_lane_bf16 (bfloat16_t *ptr, bfloat16x4x3_t b)
      24  {
      25    vst3_lane_bf16 (ptr, b, 2);
      26  }
      27  
      28  void
      29  test_vst3q_lane_bf16 (bfloat16_t *ptr, bfloat16x8x3_t b)
      30  {
      31    vst3q_lane_bf16 (ptr, b, 2);
      32  }
      33  
      34  /* { dg-final { scan-assembler-times "st3\\t{v\[0-9\]+.h - v\[0-9\]+.h}\\\[2\\\], \\\[x\[0-9\]+\\\]" 2 } } */
      35  
      36  void
      37  test_vst4_lane_bf16 (bfloat16_t *ptr, bfloat16x4x4_t b)
      38  {
      39    vst4_lane_bf16 (ptr, b, 2);
      40  }
      41  
      42  void
      43  test_vst4q_lane_bf16 (bfloat16_t *ptr, bfloat16x8x4_t b)
      44  {
      45    vst4q_lane_bf16 (ptr, b, 2);
      46  }
      47  
      48  /* { dg-final { scan-assembler-times "st4\\t{v\[0-9\]+.h - v\[0-9\]+.h}\\\[2\\\], \\\[x\[0-9\]+\\\]" 2 } } */