1  /* { dg-skip-if "" { *-*-* } { "-O0" } { "" } } */
       2  /* { dg-skip-if "" { *-*-* } { "*" } { "-DACC_MEM_SHARED=0" } } */
       3  /* { dg-additional-options "-std=c99 -ftree-slp-vectorize -foffload=-fdump-tree-slp1" } */
       4  
       5  #include <stdio.h>
       6  #include <sys/time.h>
       7  
       8  long long int p[32 *1000] __attribute__((aligned(16)));
       9  long long int p2[32 *1000] __attribute__((aligned(16)));
      10  
      11  int
      12  main (void)
      13  {
      14  #pragma acc parallel num_gangs(1) num_workers(1) vector_length(32)
      15    {
      16      if (((unsigned long int)p & (0xfULL)) != 0)
      17        __builtin_abort ();
      18      if (((unsigned long int)p2 & (0xfULL)) != 0)
      19        __builtin_abort ();
      20  
      21      for (unsigned int k = 0; k < 10000; k += 1)
      22        {
      23  #pragma acc loop vector
      24  	for (unsigned long long int j = 0; j < 32; j += 1)
      25  	  {
      26  	    unsigned long long a, b;
      27  	    unsigned long long *p3, *p4;
      28  	    p3 = (unsigned long long *)((unsigned long long int)p & (~0xfULL));
      29  	    p4 = (unsigned long long *)((unsigned long long int)p2 & (~0xfULL));
      30  
      31  	    for (unsigned int i = 0; i < 1000; i += 2)
      32  	      {
      33  		a = p3[j * 1000 + i];
      34  		b = p3[j * 1000 + i + 1];
      35  		
      36  		p4[j * 1000 + i] = a;
      37  		p4[j * 1000 + i + 1] = b;
      38  	      }
      39  	  }
      40        }
      41    }
      42  
      43    return 0;
      44  }
      45  
      46  /* { dg-final { scan-offload-tree-dump "vector\\(2\\) long long unsigned int" "slp1" } } */