1  /* { dg-do run } */
       2  /* { dg-require-effective-target p9vector_hw } */
       3  /* { dg-options "-O2 -mdejagnu-cpu=power9" } */
       4  
       5  #include <altivec.h>
       6  #include <stdio.h>
       7  
       8  void abort (void);
       9  
      10  int main() {
      11    int i;
      12    vector float vfa, vfb;
      13    vector unsigned short vresult, vexpected;
      14  
      15    vfa = (vector float){0.4, 1.6, 20.0, 99.9 };
      16    vfb = (vector float){10.0, -2.0, 70.0, 999.0 };
      17  
      18    /* Expected results.  */
      19  #ifdef __BIG_ENDIAN__
      20    vexpected = (vector unsigned short) { 0x4900, 0xc000, 0x5460, 0x63ce,
      21  					0x3666, 0x3e66, 0x4d00, 0x563e };
      22  #else
      23    vexpected = (vector unsigned short) { 0x3666, 0x3e66, 0x4d00, 0x563e,
      24  					0x4900, 0xc000, 0x5460, 0x63ce };
      25  #endif
      26  
      27  /*
      28       vresult = vec_pack_to_short_fp32 (vfa, vfb);
      29    This built-in converts a pair of vector floats into a single vector of
      30    packed half-precision (F16) values.  The result type is a vector of
      31    signed shorts.
      32    The expected codegen for this builtin is
      33      xvcvsphp t, vfa
      34      xvcvsphp u, vfb
      35      if (little endian)
      36        vpkuwum vresult, t, u
      37      else
      38        vpkuwum vresult, u, t
      39  */
      40  
      41    vresult = vec_pack_to_short_fp32 (vfa, vfb);
      42  
      43  #ifdef DEBUG
      44    for(i = 0; i< 4; i++) { printf("i=[%d] %f  \n",i,vfa[i]); }
      45    for(i = 0; i< 4; i++) { printf("i=[%d] %f  \n",i+4,vfb[i]); }
      46    for(i = 0; i< 8; i++) { printf("i=[%d] %d  \n",i,vresult[i]); }
      47  #endif
      48  
      49    for(i = 0; i< 8; i++) {
      50      if (vresult[i] != vexpected[i]) {
      51  	printf("i=[%d] 0x%x != 0x%x \n",i,vresult[i],vexpected[i]);
      52  	abort();
      53      }
      54    }
      55  }