1  /* { dg-do compile } */
       2  /* { dg-require-effective-target powerpc_p8vector_ok } */
       3  /* { dg-options "-mdejagnu-cpu=power8 -O3 " } */
       4  
       5  /* Previous versions of this test required that the assembler does not
       6     contain xxpermdi or xxswapd.  However, with the more sophisticated
       7     code generation used today, it is now possible that xxpermdi (aka
       8     xxswapd) show up without being part of a lxvd2x or stxvd2x
       9     sequence.  */
      10  
      11  #include <altivec.h>
      12  
      13  extern void abort (void);
      14  
      15  vector short x;
      16  const vector short y = { 0, 1, 2, 3, 4, 5, 6, 7 };
      17  vector short z;
      18  
      19  vector short
      20  foo (void)
      21  {
      22    return y;			/* Remove 1 swap and use lvx.  */
      23  }
      24  
      25  vector short
      26  foo1 (void)
      27  {
      28    x = y;			/* Remove 2 redundant swaps here.  */
      29    return x;			/* Remove 1 swap and use lvx.  */
      30  }
      31  
      32  void __attribute__ ((noinline))
      33  fill_local (vector short *vp)
      34  {
      35    *vp = x;			/* Remove 2 redundant swaps here.  */
      36  }
      37  
      38  /* Test aligned load from local.  */
      39  vector short
      40  foo2 (void)
      41  {
      42    vector short v;
      43  
      44    /* Need to be clever here because v will normally reside in a
      45       register rather than memory.  */
      46    fill_local (&v);
      47    return v;			/* Remove 1 swap and use lvx.  */
      48  }
      49  
      50  
      51  /* Test aligned load from pointer.  */
      52  vector short
      53  foo3 (vector short *arg)
      54  {
      55    return *arg;			/* Remove 1 swap and use lvx.  */
      56  }
      57  
      58  /* In this structure, the compiler should insert padding to assure
      59     that a_vector is properly aligned.  */
      60  struct bar {
      61    short a_field;
      62    vector short a_vector;
      63  };
      64  
      65  vector short
      66  foo4 (struct bar *bp)
      67  {
      68    return bp->a_vector;		/* Remove 1 swap and use lvx.  */
      69  }
      70  
      71  /* Test aligned store to global.  */
      72  void
      73  baz (vector short arg)
      74  {
      75    x = arg;			/* Remove 1 swap and use stvx.  */
      76  }
      77  
      78  void __attribute__ ((noinline))
      79  copy_local (vector short *arg)
      80  {
      81    x = *arg;			/* Remove 2 redundant swaps.  */
      82  }
      83  
      84  
      85  /* Test aligned store to local.  */
      86  void
      87  baz1 (vector short arg)
      88  {
      89    vector short v;
      90  
      91    /* Need cleverness, because v will normally reside in a register
      92       rather than memory.  */
      93    v = arg;			/* Aligned store to local: remove 1
      94  				   swap and use stvx.  */
      95    copy_local (&v);
      96  }
      97  
      98  /* Test aligned store to pointer.  */
      99  void
     100  baz2 (vector short *arg1, vector short arg2)
     101  {
     102    /* Assume arg2 resides in register.  */
     103    *arg1 = arg2;			/* Remove 1 swap and use stvx.  */
     104  }
     105  
     106  void
     107  baz3 (struct bar *bp, vector short v)
     108  {
     109    /* Assume v resides in register.  */
     110    bp->a_vector = v;		/* Remove 1 swap and use stvx.  */
     111  }
     112  
     113  int
     114  main (int argc, short *argv[])
     115  {
     116    vector short fetched_value = foo ();
     117    if (fetched_value[0] != 0 || fetched_value[7] != 7)
     118      abort ();
     119  
     120    fetched_value = foo1 ();
     121    if (fetched_value[1] != 1 || fetched_value[6] != 6)
     122      abort ();
     123  
     124    fetched_value = foo2 ();
     125    if (fetched_value[2] != 2 || fetched_value[5] != 5)
     126      abort ();
     127  
     128    fetched_value = foo3 (&x);
     129    if (fetched_value[3] != 3 || fetched_value[4] != 4)
     130      abort ();
     131  
     132    struct bar a_struct;
     133    a_struct.a_vector = x;	/* Remove 2 redundant swaps.  */
     134    fetched_value = foo4 (&a_struct);
     135    if (fetched_value[4] != 4 || fetched_value[3] != 3)
     136      abort ();
     137  
     138    for (int i = 0; i < 8; i++)
     139      z[i] = 7 - i;
     140  
     141    baz (z);
     142    if (x[0] != 7 || x[7] != 0)
     143      abort ();
     144  
     145    vector short source = { 8, 7, 6, 5, 4, 3, 2, 1 };
     146  
     147    baz1 (source);
     148    if (x[3] != 5 || x[7] != 1)
     149      abort ();
     150  
     151    vector short dest;
     152    baz2 (&dest, source);
     153    if (dest[4] != 4 || dest[1] != 7)
     154      abort ();
     155  
     156    baz3 (&a_struct, source);
     157    if (a_struct.a_vector[7] != 1 || a_struct.a_vector[5] != 3)
     158      abort ();
     159  
     160    return 0;
     161  }