gcc/testsuite/gcc.target/powerpc/bmi2-pext64-1a.c

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