1  #include <arm_neon.h>
       2  #include "arm-neon-ref.h"
       3  #include "compute-ref-data.h"
       4  
       5  #if defined(__aarch64__) && defined(__ARM_FEATURE_FMA)
       6  /* Expected results.  */
       7  VECT_VAR_DECL(expected,hfloat,32,2) [] = { 0x4438ca3d, 0x44390a3d };
       8  VECT_VAR_DECL(expected,hfloat,32,4) [] = { 0x44869eb8, 0x4486beb8, 0x4486deb8, 0x4486feb8 };
       9  VECT_VAR_DECL(expected,hfloat,64,2) [] = { 0x408906e1532b8520, 0x40890ee1532b8520 };
      10  
      11  #define VECT_VAR_ASSIGN(S,Q,T1,W) S##Q##_##T1##W
      12  #define ASSIGN(S, Q, T, W, V) T##W##_t S##Q##_##T##W = V
      13  #define TEST_MSG "VFMA_N/VFMAQ_N"
      14  
      15  void exec_vfma_n (void)
      16  {
      17    /* Basic test: v4=vfma_n(v1,v2), then store the result.  */
      18  #define TEST_VFMA_N(Q, T1, T2, W, N)					\
      19    VECT_VAR(vector_res, T1, W, N) =					\
      20      vfma##Q##_n_##T2##W(VECT_VAR(vector1, T1, W, N),			\
      21  			VECT_VAR(vector2, T1, W, N),			\
      22  			VECT_VAR_ASSIGN(scalar, Q, T1, W));		\
      23    vst1##Q##_##T2##W(VECT_VAR(result, T1, W, N), VECT_VAR(vector_res, T1, W, N))
      24  
      25  #define CHECK_VFMA_N_RESULTS(test_name,comment)				\
      26    {									\
      27      CHECK_FP(test_name, float, 32, 2, PRIx32, expected, comment);	\
      28      CHECK_FP(test_name, float, 32, 4, PRIx32, expected, comment);	\
      29      CHECK_FP(test_name, float, 64, 2, PRIx64, expected, comment);	\
      30    }
      31  
      32  #define DECL_VFMA_N_VAR(VAR)			\
      33    DECL_VARIABLE(VAR, float, 32, 2);		\
      34    DECL_VARIABLE(VAR, float, 32, 4);		\
      35    DECL_VARIABLE(VAR, float, 64, 2);
      36  
      37    DECL_VFMA_N_VAR(vector1);
      38    DECL_VFMA_N_VAR(vector2);
      39    DECL_VFMA_N_VAR(vector3);
      40    DECL_VFMA_N_VAR(vector_res);
      41  
      42    clean_results ();
      43  
      44    /* Initialize input "vector1" from "buffer".  */
      45    VLOAD(vector1, buffer, , float, f, 32, 2);
      46    VLOAD(vector1, buffer, q, float, f, 32, 4);
      47    VLOAD(vector1, buffer, q, float, f, 64, 2);
      48  
      49    /* Choose init value arbitrarily.  */
      50    VDUP(vector2, , float, f, 32, 2, 9.3f);
      51    VDUP(vector2, q, float, f, 32, 4, 29.7f);
      52    VDUP(vector2, q, float, f, 64, 2, 15.8f);
      53    
      54    /* Choose init value arbitrarily.  */
      55    ASSIGN(scalar, , float, 32, 81.2f);
      56    ASSIGN(scalar, q, float, 32, 36.8f);
      57    ASSIGN(scalar, q, float, 64, 51.7f);
      58  
      59    /* Execute the tests.  */
      60    TEST_VFMA_N(, float, f, 32, 2);
      61    TEST_VFMA_N(q, float, f, 32, 4);
      62    TEST_VFMA_N(q, float, f, 64, 2);
      63  
      64    CHECK_VFMA_N_RESULTS (TEST_MSG, "");
      65  }
      66  #endif
      67  
      68  int main (void)
      69  {
      70  #if defined(__aarch64__) && defined(__ARM_FEATURE_FMA)
      71    exec_vfma_n ();
      72  #endif
      73    return 0;
      74  }