1  /* { dg-do compile } */
       2  /* { dg-skip-if "" { powerpc*-*-darwin* } } */
       3  /* { dg-options "-O3 -mdejagnu-cpu=power7" } */
       4  /* { dg-require-effective-target lp64 } */
       5  /* { dg-require-effective-target powerpc_vsx_ok } */
       6  
       7  #define NO_WARN_X86_INTRINSICS 1
       8  #include <x86intrin.h>
       9  
      10  unsigned long long
      11  test__pexp_cmask_u64 (unsigned long long a[4])
      12  {
      13    /* The _pext implmentation is nominally a popcount of the mask,
      14       followed by a loop using count leading zeros to find the
      15       next bit to process.
      16       If the mask is a const, the popcount should be folded and
      17       the constant propagation should eliminate the mask
      18       generation loop and produce a single constant bpermd permute
      19       control word.
      20       This test verifies that the compiler is replacing the mask
      21       popcount and loop with a const bperm control and generating
      22       the bpermd for this case.  */
      23    const unsigned long mask = 0x00000000100000a4UL;
      24    unsigned long res;
      25    res = _pext_u64 (a[0], mask);
      26    res = (res << 8) | _pext_u64 (a[1], mask);
      27    res = (res << 8) | _pext_u64 (a[2], mask);
      28    res = (res << 8) | _pext_u64 (a[3], mask);
      29    return (res);
      30  }
      31  /* the resulting assembler should have 4 X bpermd and no popcntd or
      32     cntlzd instructions.  */
      33  
      34  /* { dg-final { scan-assembler-times "bpermd" 4 } } */
      35  /* { dg-final { scan-assembler-not "popcntd" } } */
      36  /* { dg-final { scan-assembler-not "cntlzd" } } */