gcc/testsuite/gcc.target/powerpc/swaps-p8-37.c

   1 /* { dg-do run { target { powerpc*-*-* } } } */
   2 /* { dg-require-effective-target p8vector_hw } */
   3 /* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power8" } } */
   4 /* { dg-options "-mcpu=power8 -O3 " } */
   5
   6 #include <altivec.h>
   7
   8 extern void abort (void);
   9
  10 vector float x;
  11 vector float y = { 0.0F, 0.1F, 0.2F, 0.3F };
  12 vector float z;
  13
  14 vector float
  15 foo (void)
  16 {
  17   return y;                     /* Remove 1 swap and use lvx.  */
  18 }
  19
  20 vector float
  21 foo1 (void)
  22 {
  23   x = y;                        /* Remove 2 redundant swaps here.  */
  24   return x;                     /* Remove 1 swap and use lvx.  */
  25 }
  26
  27 void __attribute__ ((noinline))
  28 fill_local (vector float *vp)
  29 {
  30   *vp = x;                      /* Remove 2 redundant swaps here.  */
  31 }
  32
  33 /* Test aligned load from local.  */
  34 vector float
  35 foo2 (void)
  36 {
  37   vector float v;
  38
  39   /* Need to be clever here because v will normally reside in a
  40      register rather than memory.  */
  41   fill_local (&v);
  42   return v;                     /* Remove 1 swap and use lvx.  */
  43 }
  44
  45
  46 /* Test aligned load from pointer.  */
  47 vector float
  48 foo3 (vector float *arg)
  49 {
  50   return *arg;                  /* Remove 1 swap and use lvx.  */
  51 }
  52
  53 /* In this structure, the compiler should insert padding to assure
  54    that a_vector is properly aligned.  */
  55 struct bar {
  56   short a_field;
  57   vector float a_vector;
  58 };
  59
  60 vector float
  61 foo4 (struct bar *bp)
  62 {
  63   return bp->a_vector;          /* Remove 1 swap and use lvx.  */
  64 }
  65
  66 /* Test aligned store to global.  */
  67 void
  68 baz (vector float arg)
  69 {
  70   x = arg;                      /* Remove 1 swap and use stvx.  */
  71 }
  72
  73 void __attribute__ ((noinline))
  74 copy_local (vector float *arg)
  75 {
  76   x = *arg;                     /* Remove 2 redundant swaps.  */
  77 }
  78
  79
  80 /* Test aligned store to local.  */
  81 void
  82 baz1 (vector float arg)
  83 {
  84   vector float v;
  85
  86   /* Need cleverness, because v will normally reside in a register
  87      rather than memory.  */
  88   v = arg;                      /* Aligned store to local: remove 1
  89                                    swap and use stvx.  */
  90   copy_local (&v);
  91 }
  92
  93 /* Test aligned store to pointer.  */
  94 void
  95 baz2 (vector float *arg1, vector float arg2)
  96 {
  97   /* Assume arg2 resides in register.  */
  98   *arg1 = arg2;                 /* Remove 1 swap and use stvx.  */
  99 }
 100
 101 void
 102 baz3 (struct bar *bp, vector float v)
 103 {
 104   /* Assume v resides in register.  */
 105   bp->a_vector = v;             /* Remove 1 swap and use stvx.  */
 106 }
 107
 108 int
 109 main (float argc, float *argv[])
 110 {
 111   vector float fetched_value = foo ();
 112   if (fetched_value[0] != 0.0F || fetched_value[3] != 0.3F)
 113     abort ();
 114
 115   fetched_value = foo1 ();
 116   if (fetched_value[1] != 0.1F || fetched_value[2] != 0.2F)
 117     abort ();
 118
 119   fetched_value = foo2 ();
 120   if (fetched_value[2] != 0.2F || fetched_value[1] != 0.1F)
 121     abort ();
 122
 123   fetched_value = foo3 (&x);
 124   if (fetched_value[3] != 0.3F || fetched_value[0] != 0.0F)
 125     abort ();
 126
 127   struct bar a_struct;
 128   a_struct.a_vector = x;        /* Remove 2 redundant swaps.  */
 129   fetched_value = foo4 (&a_struct);
 130   if (fetched_value[2] != 0.2F || fetched_value[3] != 0.3F)
 131     abort ();
 132
 133   z[0] = 0.7F;
 134   z[1] = 0.6F;
 135   z[2] = 0.5F;
 136   z[3] = 0.4F;
 137
 138   baz (z);
 139   if (x[0] != 0.7F || x[3] != 0.4F)
 140     abort ();
 141
 142   vector float source = { 0.8F, 0.7F, 0.6F, 0.5F };
 143
 144   baz1 (source);
 145   if (x[2] != 0.6F || x[1] != 0.7F)
 146     abort ();
 147
 148   vector float dest;
 149   baz2 (&dest, source);
 150   if (dest[0] != 0.8F || dest[1] != 0.7F)
 151     abort ();
 152
 153   baz3 (&a_struct, source);
 154   if (a_struct.a_vector[3] != 0.5F || a_struct.a_vector[0] != 0.8F)
 155     abort ();
 156
 157   return 0;
 158 }