1  /* { dg-do assemble { target { arm*-*-* } } } */
       2  /* { dg-require-effective-target arm_v8_2a_bf16_neon_ok } */
       3  /* { dg-add-options arm_v8_2a_bf16_neon }  */
       4  /* { dg-additional-options "-save-temps -march=armv8.2-a+bf16+fp16" } */
       5  
       6  #include "arm_neon.h"
       7  
       8  float32x2_t
       9  test_vbfdot_vcreate (float32x2_t r, uint64_t a, uint64_t b)
      10  {
      11    bfloat16x4_t _a = vcreate_bf16(a);
      12    bfloat16x4_t _b = vcreate_bf16(b);
      13  
      14    return vbfdot_f32 (r, _a, _b);
      15  }
      16  /* { dg-final { scan-assembler {vdot.bf16\td[0-9]+, d[0-9]+, d[0-9]+} } } */
      17  
      18  bfloat16x8_t test_vcombine_bf16 (bfloat16x4_t a, bfloat16x4_t b)
      19  {
      20    return vcombine_bf16 (a, b);
      21  }
      22  
      23  bfloat16x4_t test_vget_high_bf16 (bfloat16x8_t a)
      24  {
      25    return vget_high_bf16 (a);
      26  }
      27  
      28  bfloat16x4_t test_vget_low_bf16 (bfloat16x8_t a)
      29  {
      30    return vget_low_bf16 (a);
      31  }
      32  
      33  bfloat16_t test_vget_lane_bf16 (bfloat16x4_t a)
      34  {
      35    return vget_lane_bf16 (a, 1);
      36  }
      37  
      38  bfloat16_t test_vgetq_lane_bf16 (bfloat16x8_t a)
      39  {
      40    return vgetq_lane_bf16 (a, 7);
      41  }
      42  
      43  bfloat16x4_t test_vset_lane_bf16 (bfloat16_t a, bfloat16x4_t b)
      44  {
      45    return vset_lane_bf16 (a, b, 1);
      46  }
      47  
      48  bfloat16x8_t test_vsetq_lane_bf16 (bfloat16_t a, bfloat16x8_t b)
      49  {
      50    return vsetq_lane_bf16 (a, b, 7);
      51  }
      52  
      53  bfloat16x4_t vdup_test (bfloat16_t a)
      54  {
      55    return vdup_n_bf16 (a);
      56  }
      57  /* { dg-final { scan-assembler {vdup\.16\td[0-9]+, r[0-9]+} } }  */
      58  
      59  bfloat16x8_t vdupq_test (bfloat16_t a)
      60  {
      61    return vdupq_n_bf16 (a);
      62  }
      63  /* { dg-final { scan-assembler {vdup\.16\tq[0-9]+, r[0-9]+} } }  */
      64  
      65  
      66  bfloat16x4_t test_vdup_lane_bf16 (bfloat16x4_t a)
      67  {
      68    return vdup_lane_bf16 (a, 1);
      69  }
      70  /* { dg-final { scan-assembler-times {vdup\.16\td[0-9]+, d[0-9]+\[1\]} 1 } }  */
      71  
      72  bfloat16x8_t test_vdupq_lane_bf16 (bfloat16x4_t a)
      73  {
      74    return vdupq_lane_bf16 (a, 1);
      75  }
      76  /* { dg-final { scan-assembler-times {vdup\.16\tq[0-9]+, d[0-9]+\[1\]} 1 } }  */
      77  
      78  bfloat16x4_t test_vdup_laneq_bf16 (bfloat16x8_t a)
      79  {
      80    return vdup_laneq_bf16 (a, 3);
      81  }
      82  
      83  bfloat16x8_t test_vdupq_laneq_bf16 (bfloat16x8_t a)
      84  {
      85    return vdupq_laneq_bf16 (a, 3);
      86  }
      87  
      88  bfloat16_t test_vduph_lane_bf16 (bfloat16x4_t a)
      89  {
      90    return vduph_lane_bf16 (a, 1);
      91  }
      92  
      93  bfloat16_t test_vduph_laneq_bf16 (bfloat16x8_t a)
      94  {
      95    return vduph_laneq_bf16 (a, 7);
      96  }